BOOK OF ABSTRACTS

TIES 2009 – the 20th Annual Conference of The International Environmetrics Society, a Section of

the ISI and

GRASPA Conference 2009

HANDLING COMPLEXITY AND UNCERTAINTY IN ENVIRONMENTAL STUDIES

July 5-9, 2009 University of Bologna, Italy

The University of Bologna hosts two joint conferences under the common heading of “Handling complexity and uncertainty in environmental studies”: TIES 2009 conference, the annual Meeting of The International Environmetrics Society GRASPA 2009, the final meeting of two Italian Projects (PRIN-MIUR-2006) on environmental statistics: "Statistical analysis and modelling of impact and risk for environmental phenomena in space and time" (project n. 2006131039) and "Methods for collecting and analyzing environmental data" (project n. 2006139812).

Daniela Cocchi Chair of the Scientific Committee

Conference Scientific Committee

Daniela Cocchi (Chair) Italy Lucio Barabesi Italy David Brillinger USA Abdel El-Shaarawi Canada Alessandro Fassò Italy Carlo Gaetan Italy Bronwyn Harch Australia Ulla Holst Sweden Giovanna Jona Lasinio Italy Gianfranco Lovison Italy Alexandra Schmidt Brazil Rongvald Smith UK Don Stevens USA

Local Organizing Committee Michele Scagliarini (Co-chair) Italy Carlo Trivisano (Co-chair) Italy Francesca Bruno Italy Michela Cameletti Italy Fedele Greco Italy Rossella Miglio Italy Orietta Nicolis Italy Marilena Pillati Italy Meri Raggi Italy

Acknowledgments The organizers thank for the support: the University of Bologna, the Faculty of Statistical Sciences of the University of Bologna, the Department of Statistics “Paolo Fortunati” of the University of Bologna, Agenzia Regionale Prevenzione Ambiente dell´Emilia-Romagna (ARPA), Margherita srl and the Italian Statistical Society (SIS).

i

TIES 2009 – GRASPA 2009 MEETING List of Sessions

PLENARY SESSIONS TIES J. Stuart Hunter Lecture Chair: Don Stevens Douglas W. Nychka, Director of the Institute for Mathematics Applied to Geosciences National, Center for Atmospheric Research “Spatial statistics, computer models and regional climate change”. TIES President's Invited Lecture Chair: Daniela Cocchi Walter Radermacher, Chief Statistician of the European Union and Director-General of Eurostat: “Sustainable Economics: the contribution of official statistics” GRASPA Invited Lecture Project “Statistical analysis and modelling of impact and risk for environmental phenomena in space and time” Chair : Gianfranco Lovison Giovanna Jona Lasinio, University of Roma "La Sapienza" , “Spatio temporal data modeling in environmental sciences a review.” GRASPA Invited Lecture. Project “Methods for collecting and analyzing environmental data” Chair : Alessandro Fassò Lorenzo Fattorini, University of Siena, “Multi-phase sampling strategies for large-scale environmental surveys”

TIES SESSIONS

T1: Statistical issues in assessing forest sustainability (TIES/ISARA) Organizer: Ronald Mc Roberts - Steen Magnussen, Ronald Mc Roberts, Canadian Forest Service, British Columbia, Canada: ”A

bootstrap variance estimator for the observed species richness in quadrat sampling from finite populations”

- Erkki Tomppo, Finnish Forest Research Institute (Metla), Finland: ”Predicting tree level variables using airborne LiDAR data and field observations”

- Ronald Mc Roberts, U.S. Forest Service, Minnesota USA: “Satellite Image-based Maps: Scientific Inference or Just Pretty Pictures?”

ii

T2: Change-Point Methods for Environmental Data Analysis Organizer: Venkata K. Jandhyala - Ian B. MacNeill, Venkata K. Jandhyala, Elena Naumova, University of Western Ontario,

Canada: ”Monitoring using Changepoints” - Anders Grimvall, Sackmone Sirisack, Linköping Universitet, Sweden: “Integrating smoothing

and regression trees for change-point detection in environmental data” - Venkata K. Jandhyala, Stergios Fotopoulos, Elena Khapalova, Washington State University,

USA: “Earthquakes from the Indonesian region: an application of exact computable expressions for the asymptotic distribution of change-point mle in the exponential case”

- Hyune-Ju Kim, Jun Luo, Michael Barrett, Eric Feuer, Syracuse University, USA: ”Comparing piecewise linear trends”

T3: Palaeoclimate reconstruction: statistical challenges Organizer: John Haslett - John Haslett, Trinity College Dublin, Ireland: “Statistical Methods in the Reconstruction of

Paleoclimate” - Andrew Parnell, John Haslett, Michael Salter-Townshend, University College Dublin, Ireland:

“Bayesian methods for reconstructing past climate histories” - Vincent Garreta, Joël Guiot, CEREGE in Aix-en-Provence, France, “Climate reconstructed

from pollen data using a dynamic vegetation model” - Michael Crucifix, Rouger Jonathn, Université catholique de Louvain, Belgie: ”Long-range

climate reconstructions with dynamical systems”

T4: Uncertainty in Hydrological forecasting Organizer: Daniela Cocchi-Ezio Todini - Ezio Todini, University of Bologna, Italy: “Predictive Uncertainty in Hydrological Forecasting” - Alexandra Schmidt, Romy Ravines, Helio Migon, Federal University of Rio de Janeiro, Brazil:

“Modelling multiple series of runoff: the case of Rio Grande Basin” - Pierre Aillot, Craig Thompson, Peter Thomson, University of Brest, France: “Space time

modeling of precipitation using hidden Markov models” T5: Ocean Climatologies Organizer: Alexandra Schmidt - Bruno Sanso, Ricardo Lemos, University of California, USA: ”Spatio-temporal models for

oceanic variables” - Lelys Guenni, Gabriel Huerta, Bruno Sansò, Universidad Simón Bolívar, Venezuela: “Detection

of oceanic influence on the precipitation of the central Venezuelan coast using time-varying models”

- Nadia Pinardi, Srdjan Dobiricic, Ralph Milliff, University of Bologna, Italy: “Operational Oceanography: the science based approach to marine management problems”

T6: Fast computation for spatial data Organizer: Ulla Holst - Nicolas Verzelen, LM-Orsay, INRIA Saclay-Ile de France: “Data-driven neighborhood

selection of a Gaussian field” - Finn Lindgren, Håvard Rue, Johan Lindström, David Bolin, Lund University, Sweden:

“Eliminating the practical boundary between Markov and other Gaussian random fields” - Johan Lindström, Finn Lindgren, Peter Jonsson, David Bolin, Håvard Rue, Lund University,

Sweden: “Fast estimation of non-stationary Gaussian Markov Random Fields” - David Bolin, Finn Lindgren, Lund University, Sweden: ”Non-stationary spatial ARMA models

applied to global ozone data”

iii

T7: Risk and extremes in climate research Organizer: Peter Guttorp - Ola Haug, Norwegian Computing Center, Norway: ”Projections of future insurance losses from

climate model data” - George Lindgren, Lund University, Sweden: “Possible climate change effects on marine safety” - Peter Guttorp, University of Washington, USA: ”Looking for climate change signals in extreme

temperatures” T8: Random effects modeling of environmental data Organizer: Renjun Ma - Francesca Dominici, Roger Peng, Michelle Bell, Johns Hopkins Bloomberg School of Public

Health, USA: “A Bayesian hierarchical model for estimating health effects of chemical constituents of particulate matter”

- Ronghui Xu, Michael Donohue, Florin Vaida, Rosanna Haut, University of California, USA: “Mixed-effects model selection”

- Guohua Yan, Renjun Ma, University of New Brunswick, Canada: “Analysis of clustered environmental multinomial data with random cluster sizes”

T9: Computational Methods for Large Spatial Data Organizer: Hao Zhang - Abdel El-Shaarawi, National Water Research Institute and McMaster University, Canada:

”Matrix inversion and statistical data analysis” - Pascal Monestiez, David Nerini, INRA, France: ”Functional kriging of ocean profile data” - Hao Zhang, Purdue University, USA: ”Dealing with large covariance matrices for spatial data” T10: Forest Fire and Weather Organizer: W. John Braun - Valentin Rousson, Juhyun Park, Theo Gasser, University of Lausanne - Switzerland, “On the

concept of structural components with an application to weather functional data” - Alisha Albert-Green, W. John Braun, David L. Martell, Douglas G. Woolford, University of

Western Ontario, Canada: “Modelling the Ontario Fire Weather Index” - Sylvia Esterby, Zuzana Hrdlickova, Steve Taylor, University of British Columbia-Okanagan,

Canada, “Characterizing spatial patterns of fire weather using historical data” - Jonathan Lee, W. John Braun, Bruce Jones, Doug Woolford, Mike Wotton, University of

Western Ontario, Canada: “Fire risk assessment in Muskoka, Ontario” T11: Modeling count environmental data Organizer: Abdel El-Shaarawi - Renjun Ma, University of New Brunswick Fredericton, Canada: “Poisson nonlinear mixed

models for environmental data” - Rhong Zhu, Abdel El-Shaarawi, Harry Joe, McMaster University, Canada: “Modelling bacterial

density count data with various overdispersion and tail heaviness” - Dianliang Deng, University of Regina, Canada: “The testing of Zero-inflation and over-

dispersion for the environmental count data”

iv

T12: Chemiometrics Organizer: Abdel El-Shaarawi - Clifford Spiegelman, Abdel El-Shaarawi, Texas A&M University Center for Statistical

Bioinformatics, USA: “Chemometrics and Environmetrics: Tall Shoulders, Accomplishments, and Future Directions”.

- Maria Chiara Pietrogrande, Mattia Mercuriali, Nicola Marchetti, Luisa Pasti, Dimitri Bacco, Gaetano Zanghirati, Francesco Dondi, University of Ferrara, Italy: “A chemometric approach based on the autocovariance function for handling complex signals from environmental monitoring”.

- Ester Papa, Paola Gramatica, University of Insubria, Italy: “QSAR modelling and multivariate analysis of the environmental behaviour of organic pollutants”

T13: Estimating personal exposures to air pollution Organizer: Gavin Shaddik - Marta Blangiardo, Sylvia Richardson, Imperial College, UK: “A Bayesian model of time

activity data for ecological studies with implications to the bias of disease risks” - Duncan Lee, Gavin Shaddik, Ruth Salway, Jim Zidek, University of Glasgow, UK: “Using

estimated personal exposures in studies of the effects of air pollution on health” - Gulliver John, Blangiardo Marta, Briggs David, Hansell Anna, University of the West of

Scotland, UK: “Simultaneous modelling of spatial and temporal variations in air pollution exposures for health risk assessment and epidemiological analysis”

T14: Assessment of ecosystem status Organizer: Ron Smith - Anne E Magurran, Stephen R Baillie, Steve T Buckland, Jan McP Dick, David A Elston, E

Marian Scott, Ron Smith, Paul J. Somerfield, Allan Watt, University of St Andrews, :” Measuring Biodiversity in the context of Ecosystem Services”;

- Paul Warren, University of Albury-Wodonga, Australia: “A Causal Modelling Approach to Spatial and Temporal Confounding in Environmental Impact Studies”

- Ron Smith, Jan Dick, Centre for Ecology and Hydrology, Bush Estate: “Linking Statistics to Ecosystem Services Frameworks”

GRASPA/TIES SESSIONS

GT1: Bayesian multivariate spatial models Organizer: Francesca Bruno - Fedele Greco, Carlo Trivisano, Daniela Cocchi, University of Bologna, Italy: “A multivariate

CAR model and its applications” - Sudipto Banerjee, Yufen Zhang, James Hodges, University of Minnesota, USA: ” Smoothed

ANOVA with spatial effects as a competitor to MCAR in multivariate spatial smoothing” - Steve Sain, Geophysical Statistics Project–NCAR: ”Spatial Analysis of Regional Climate

Model Ensembles”

v

GT2: Environmental Sampling Applications Organizer: Lucio Barabesi, Don Stevens - Dick Brus, Jaap de Gruijter, Wageningen University and Research Centre, The Netherlands: “A

mixed, design-based model-based sampling approach for estimating global quantities in space-time”

- Don Stevens, Oregon State University, USA, “Using imputation to estimate trend and abundance in Coho Salmon numbers using a multi-period rotating panel sampling design”

- Hailemariam Temesgen, Vicente Monleon, Aaron Weiskittel, Duncan Wilson, Oregon State University, USA, “A tail of two phases: design and estimation of three foliage biomass”

- Sara Franceschi, University of Siena, Italy: “Sampling properties of spatial total estimators under tessellation stratified designs”

GT3: Statistical Modeling for Water Resources Organizer: Sylvia Esterby - Vito Muggeo, Gianfranco Lovison, University of Palermo, Italy: “Score and quasi-score

inference for change-points, with applications in marine ecology and groundwater monitoring” - Gonçalves A. Manuela, Marco Costa, Universidade do Minho, Portugal: "Prediction of Water

Quality Variables using State-Space and Linear Models for River Network Data " - Guillaume Evin, Anne-Catherine Favre, University of Québec, Canada: “A non-stationary

Neyman-Scott model for rainfall”

GRASPA SESSIONS

G1: Spatiotemporal models Organizer: Rosaria Ignaccolo - Yongku Kim, Institute for Mathematics Applied to Geosciences, NCAR, USA: “Change of

Spatiotemporal Scale in Dynamic Models” - Veronica Berrocal, Alan Gelfand, Duke University, USA: ”A multivariate spatio-temporal

downscaler for output from numerical models” - Michela Cameletti, Rosaria Ignaccolo, Stefano Bande, University of Bergamo, Italy:

“Comparing air quality statistical models” - Pancrazio Bertaccini, Vanja Dukic, Rosaria Ignaccolo, University of Turin, Italy: “Air

pollution: meteorology or traffic, what does really matter?” G2: Environmental indices Organizer: Alessandro Fassò - Antonello Maruotti, Francesco Lagona, University of Rome, Italy: ”A Hurdle Markov model for

pollutants concentrations” - Antonella Plaia, Mariantonietta Ruggieri, Anna Lisa Bondì, University of Palermo, Italy: “An

aggregate air quality index considering interactions among pollutants” - Marian Scott, Duncan Lee, Claire Ferguson, Ron Smith, University of Glasgow, UK: ”Simple

metrics, complex environmental systems”

vi

G3: Air quality modeling Organizer: Alessandro Fassò - Paul D.Sampson, Adam Szpiro, Lianne Sheppard, Johan Lindstrom, University of Washington,

USA: “Spatial regression modeling with nonstationary spatial covariance structure for air quality exposure from complex spatio-temporal monitoring and GIS-based covariates”

- Francesco Finazzi, Cinzia D’Ariano, Alessandro Fassò, Gianandrea Mannarini, Orietta Nicolis, University of Bergamo, Italy: “Integrating satellite and ground level data for air quality monitoring and dynamical mapping”

- Wolfgang Schmid, Olha Bodnar, Universität Viadrina Frankfurt an der Oder, Germany: “Local Approaches for interpolating Air Pollution Processes”

G4: Source apportionment with multivariate receptor models Organizer: Alessio Pollice - Alessio Pollice, University of Bari, Italy: ”A critical review of some statistical issues implied by

the use of multivariate receptor models” - Eleonora Andriani, Maurizio Caselli, Gianluigi De Gennaro, University of Bari, Italy:

“Synergistic use of several receptor models (CMB, APCS and PMF) to interpret air quality data”

- Bo Larsen, IES, EU Joint Research Center, Ispra: ”A critical look at the fulfillment of basic assumptions for the application of the two common receptor models CMB and PMF for source apportionment of PM10”

- William F. Christensen, Basil Williams, Shane Reese, Brigham Young University, USA: “Identifying pollution source directions for pollution source apportionment”

G5: Environmental point process Organizer: Marcello Chiodi - Rick Schoenberg, University of California, USA: “Separable conditional intensity estimates for

space-time point processes with application to Los Angeles County wildfires” - Crescenza Calculli, Alessio Pollice, University of Bari, Italy: “Joint incidence of various

diseases in the presence of risk source” - Renata Rotondi, Elisa Varini, CNR-IMATI, Italy: “Bayesian estimation of the conditional

intensity function in self-correcting point processes applied to the seismic activity of Italian tectonic regions”

- Giada Adelfio, Marcello Chiodi, University of Palermo, Italy: “Semi-parametric estimation of the intensity function in space-time point processes”

G6: Extreme values for spatio-temporal data Organizer: Jean-Noel Bacro, Liliane Bel - Simone Padoan, Anthony Davison, Mathieu Ribatet, Ecole Polytechnique Federale de

Lausanne, Switzerland: “Modelling of spatial extremes: a review” - Gwladys Toulemonde, Armelle Guillou, Philippe Naveau, Mathieu Vrac, Frédéric Chevallier,

Université Montpellier II, France: “State-space models in extreme value theory” - Mathieu Ribatet, Simone A. Padoan, Scott A. Sisson, EPFL, Lausanne,

Switzerland:”Likelihood-based inference for max-stable processes” - Pierre Ribereau, Armelle Guillou, Philippe Naveau, Université de Montpellier 2, France:

“Generalized Probability Weighted Moments Methods in Extreme Value Theory”

vii

G7: Nonparametric statistics and functional analysis in environmental problems Organizer: Carlo Gaetan - Stefano Tonellato, Stefano Ciavatta, Andrea Pastore, Roberto Pastres, University of Venice,

Italy: ”Clustering of monitoring stations in Venice Lagoon” - Sonia Petrone, Michele Guindani, Alan Gelfand, Bocconi University, Italy: ”Bayesian

nonparametric mixtures for local clustering of functional data” - Jan-Michel Poggi, Francois-Xavier Jollois, Bruno Portier, Université Paris-Sud, France: “Three

non-linear statistical methods to analyze PM10 pollution in Rouen area” - David Dunson, Duke University, USA: "Bayesian density regression and mixtures with

environmental applications" G8: Wavelet-based analysis of environmental data Organizer: Orietta Nicolis - Jorge Mateu, Orietta Nicolis, Universitat Jaume I, Spain: ”Wavelet-based analysis of the spatial

structure of point patterns” - Josè Miguel Angulo, Ana E. Madrid, Universidad de Granada Facultad de Ciencias, Spain:

”Wavelet-based multiscale intermittency analysis in environmental applications” - Orietta Nicolis, George Christakos, University of Bergamo, Italy: “A unified framework of

fractal and wavelet random fields” - Emilio Porcu, Maria Dolores Ruiz Medina, Rosaura Ferndandez Pascual, Universitat Göttingen,

Germany: "Functional estimation of Gaussian Dagum random fields and related models" G9: Epidemiology Organizer: Emanuela Dreassi - Leonhard Held, Birgit Schrödle, University of Zurich, Switzerland: “Spatio-temporal disease

mapping using INLA” - Rossella Miglio, Francesca Bruno, University of Bologna, Italy, “Flexible statistical models in

the study of vulnerability to environmental exposure” - Dolores Catelan, Annibale Biggeri, Corrado Lagazio, University of Florence, Italy:

“Epidemiologic surveillance and impact evaluation: the false discovery rate” - Adam A. Szpiro, Lianne Sheppard, Thomas Lumley, University of Washington, USA:

”Accounting for Exposure Measurement Error in Environmental Epidemiology”

CONTRIBUTED SESSIONS

C1: Time series models for environmental applications − Marcus Herbst, Markus C. Casper, Jens Grundmann, Oliver Buchholz; Economic and Social

Statistic Dept. and Centre for Regional and Environmental Statistics, University of Trier: 'Comparative analysis of model behaviour for flood prediction purposes using Self-Organizing Maps'

− Rey Decastro, Timothy Buckley, Lu Wang, Jana Mihalic, Patrick Breysse, Alison Geyh; WESTAT: 'The Longitudinal Dependence of Black Carbon Concentration on Traffic Volume in an Urban Environment'

− Helena Mouriño, Sofia Palma, Maria Teresa Moita, Maria Isabel Barão; Universidade de Lisboa, Portugal: 'Zero-Inflated Generalised Poisson Regression Model to Describe Pseudo-nitzschia Concentration in Lisbon Bay'

viii

C2: Models for marine applications − Clarissa Ferrari, Alan E. Gelfand, Giovanna Jona Lasinio, Daniela Cocchi; University of

Bologna: 'Approximated wrapped distributions for modeling circular data' − Valérie Monbet, Pierre Ailliot, Anne Cuzol, Nicolas Raillard; Université de Bretagne Sud:

'Space-time models for moving fields. Application to significant wave height.' − Francesco Lagona, Marco Picone; DIPES - University of Roma Tre: 'Multiple imputation of

incomplete oceanographic linear-circular data using multivariate mixture models'

C3: Fuzzy statistics and its extension − Ip Wai Cheung, Baoqing Hu, Heung Wong, Jun Xia; The Hong Kong Polytechnic University:

'Applications of Grey Relational Method to River Environment Quality Evaluation in China' − Reinhard Viertl; TU Wien: 'Analysis of fuzzy environmental data' − Christophe Faust, Peter Gemmar, Oliver Gronz, Markus Casper; University of Applied Sciences

(FH), Trier Germany: 'Generating Fuzzy Logic-Based Rainfall-Runoff Models Using Self-Organizing Maps'

C4: Environmental monitoring − Michele Scagliarini, Daniela Cocchi; University of Bologna: 'Gauge imprecision effect on

environmental monitoring algorithms' − Massimiliano Lega, Luca d'Antonio, Rodolfo Napoli; DiSAm-Dipartimento di Scienze per

l'Ambiente-University of Naples Parthenope, Italy : 'Solid Waste Landfills Monitoring by Aerial Infrared Thermography'

− Giuseppe Persechino, Pasquale Schiano, Massimiliano Lega, Rodolfo Napoli; CIRA-Italian Aerospace Research Centre: 'Environment Monitoring performed by Advanced Hybrid Airship at low altitude'

C5: Air quality monitoring and assessment − Tarana A. Solaiman; Civil and Environmental Engineering, The University of Western Ontario,

Ontario, Canada: 'Partial Least Square (PLS) Model in Air Quality Modelling' − Francesca Di Salvo, Gianna Agro', Mariantonietta Ruggieri, Antonella Plaia; DSSM, University

of Palermo: 'Air quality assessment via Functional Principal Component Analysis' − Walter Di Nicolantonio, Alessandra Cacciari, Gabriele Curci, Paolo Stocchi, Ezio Bolzacchini,

Luca Ferrero, Claudio Tomasi; Carlo Gavazzi Space at ISAC-CNR, Bologna, Italy: 'Air Quality monitoring fusing satellite remote sensing, ground-based measurements and meteorological modeling in Northern Italy'

− Claire Faucheux, Chantal de Fouquet, Giovanni Cárdenas, Laure Malherbe; Mines ParisTech-Centre de Géosciences-Géostatistique: 'How to build an initial sampling scheme: recommendations for measurement surveys of air quality'

ix

C6: Spatio-temporal models − Juan Antonio Luque-Espinar, Mario Chica-Olmo, Eulogio Pardo-Igúzquiza, María José García-

Soldado, Juan Grima-Olmedo; Geological Survey of Spain (IGME): 'Analysis of the piezometric spatial distribution based on the estimation of piezometric differences'

− Maura Mezzetti, Samantha Leorato; Facoltà Economia, Università "Tor Vergata", Roma: 'Bayesian Spatial Panel Data'

− Alexandre Rodrigues, Peter Diggle; Lancaster University: 'A Spatio-Temporal Model for Point-Source Modelling in Criminology: A CCTV Application'

− Cinzia D'Ariano, Alessandro Fassò, Francesco Finazzi; Dipartimento di Ingegneria dell'informazione e metodi matematici- Università degli studi di Bergamo: 'Estimation of hierarchical spatio-temporal coregionalization models with the EM algorithm'

C7: Modelling and monitoring soil data − Jacqueline Potts; Biomathematics and Statistics Scotland: 'Options for the Design of a Soil

Monitoring Scheme' − Petra Kuhnert, Anne Kinsey-Henderson, Rebecca Bartley, Alexander Herr; CSIRO

Mathematical and Information Sciences: 'Incorporating uncertainty into gully erosion calculations: a random forest approach'

− Jiri Komprda, Klara Kubosova, Milan Sanka, Ondrej Hajek, Ivan Holoubek; Research Centre for Environmental Chemistry and Toxicology, Masaryk University: 'Spatial model of persistent organic pollutants volatilization from soil: connection of deterministic and stochastic approach'

C8: Statistics for planning and decisions − Eva Horvathova; Charles University in Prague: 'Does environmental performance affect

financial performance? A meta-analysis' − Petr Pecha, Radek Hofman, Emily Pechova; Institute of Information Theory and Automation:

'HARP-A Software Tool for Decision Support during Nuclear Emergencies ' − Maria Franco Villoria, Marian Scott, Trevor Hoey, Denis Smith, Alistair Cargill; University of

Glasgow: 'Extreme Events and business continuity planning - exploration in flood risk strategies within Scotland'

C9: Forest and Fire modelling − Andrew Finley, Sudipto Banerjee; Department of Forestry, Michigan State University: 'A

hierarchical mixture model for estimating zero inflated continuous forest variables' − Piermaria Corona, Mariagrazia Agrimi, Federica Baffetta, Anna Barbati, Lorenzo Fattorini,

Enrico Pompei, Walter Mattioli; University of Tuscia - DISAFRI: 'Urban Forest Assessment in Italy under National Forest Inventory framework'

− Frederic Schoenberg; UCLA Statistics: 'Separable conditional intensity estimates for space-time point processes with application to Los Angeles County wildfires'

x

C10: Hydrological applications − Maged Daoud, Saleh Magram; Civil Engineering Dept., King Abdulaziz University: 'Impact of

residential underground water storage tanks on drinking Water quality of Jeddah Supply System, Saudi Arabia'

− Oliver Gronz, Peter Gemmar, Markus Casper; University of Applied Sciences (FH), Trier, Germany: 'Restricting Parameter Space in Mesoscale Water Balance Models using Hydrological Soil Maps'

− Enrico Guastaldi, Andrea Carloni, Vincenzo Ferrara, Claudio Gallo; Centro di GeoTecnologie, Università degli Studi di Siena: 'Stochastic groundwater modelling for a fracturated aquifer in Augusta area (Syracuse, Italy)'

C11: Environmental regulation and sustainability − Suresh Deman; Centre for Economics & Finance: 'Environmental Compliance, Social

Regulation: a Game Theoretic Model' − Kristina Voigt, Hagen Scherb; Helmholtz Zentrum Muenchen: 'The Interplay of Environmetrics

with other Environmentally-Oriented Societies' − Pancrazio Bertaccini, Marco Bagliani; IRES Piemonte: 'An estimate of the industrial

metabolism of the Piedmont region (Italy) using the environmental input-output analysis and the ecological footprint'

C12: Modeling health effects − Duncan Lee, Gavin Shaddick; University of Glasgow: 'Spatial modelling and ecological bias

within air pollution and health studies' − Anne Knol, Arthur Petersen, Jeroen van der Sluijs; RIVM: 'Dealing with uncertainties in

environmental burden of disease assessment' − Erik-A. Sauleau, Silvia Columbu, Monica Musio; Faculte de Medecine, Universite de

Strasbourg, France: 'Risk standardisation and Poisson regression in environmental cancer incidence studies'

− Michael Salter-Townshend, John Haslett; University College Dublin: 'The INLA Method for Multivariate Counts Data'

C13: Ozone modelling − José Pires, F.G. Martins, M.C.M. Alvim-Ferraz, M.C. Pereira; LEPAE, Faculdade de

Engenharia da Universidade do Porto: 'Darwinian Theory based technique to predict air pollutant concentrations'

− Dariusz Kayzer, Klaudia Borowiak, Anna Budka, Janina Zbierska; Poznan University of Life Sciences, Department of Mathematical and Statistical Methods: 'Effect of tropospheric ozone on tobacco plants in various exposure series during growing seasons '

− Armando Pelliccioni, Stefano Lucidi, Vittorio La Torre, Fabrizio Pungì; Ispesl: 'Forecast of Ozone pollutant up to 5 days in advance in Rome Urban Area by means Neural Net'

− Rossana Cotroneo, Silvia Bartoletti, Armando Pelliccioni; Ispesl: 'Connection between meteorological and Ozone scenarios in different urban areas'

xi

C14: Modeling Extremes − Zaher Khraibani, Hussein M. Badran, Hussein Khraibani; Université Paris 10: 'Records method

for the natural disasters: Application to the Storm events' − Lee Fawcett, David Walshaw; Newcastle University: 'Bayesian inference for clustered extremes' − Olivia Grigg, Jonathan Tawn; Lancaster University: 'Modelling joint extremes: application to

UK river flow data'

C15: Statistical methods for disease mapping − Massimo Ventrucci; Dipartimento di Scienze Statistiche "Paolo Fortunati": 'Multiple testing on

Standardized Mortality Ratios: a Bayesian Hierarchical model for False Discovery Rate estimation'

− Tomás Goicoa, M.D. Ugarte, Jaione Etxeberria, F. Militino Ana; Universidad Pública de Navarra: 'Detecting high-risk regions in disease mapping using spatial P-spline models'

− Monica Musio, Erik Sauleau, Nicole Augustin; University of Cagliari: 'Modelling Space-time variation of cancer incidence data: a case study'

C16: Climate modelling − Juan Antonio Luque-Espinar, Eulogio Pardo-Igúzquiza, Mario Chica-Olmo, María José García-

Soldado; Geological Survey of Spain (IGME): 'Temporal and spatial analysis of climatic cycles in a detritic aquifer: behaviour of recharge '

− Lasse Holmström, Leena Pasanen; University of Oulu: 'Bayesian scale space analysis with application to remote sensing and climate modeling'

− Hans Visser; Netherlands Environmental Assessment Agency: 'Dendroclimatological reconstructions of past climates: is the present the key to the past?'

− Milena Kovárová; University of South Bohemia,Institute of Physical Biology: 'The causes of climate change'

C17: Time series models for environmental applications II − Esmail Amiri; Department of Statistics IKIU : 'Bayesian modelling volatility of growth rate in

atmospheric carbon dioxide concentrations' − Grace Chiu, Shahedul Ahsan Khan, Joel Dubin; University of Waterloo: 'Monitoring

Atmospheric Chlorofluorocarbons by the Longitudinal Bent-Cable Model' − Daniele Imparato, Mauro Gasparini; Department of Mathematics - Politecnico di Torino:

'Reconstructing hourly PM10 gravimetric measurements through transfer function models'

C18: Multivariate analysis for water quality − Huseyin Ankara, Suheyla Yerel; Eskisehir Osmangazi University: 'Assessment of Surface

Water Quality by Multivariate Statistical Analysis Techniques' − Ruth Haggarty, Marian Scott, Claire Ferguson; University of Glasgow: 'Extreme Value Theory

Applied to the Defnition of Bathing Water Discounting Limits' − Hans Visser; Netherlands Environmental Assessment Agency: 'Steering factors and the

ecological quality of regional surface waters: A successful application of regression-tree analysis'

xii

C19: Air pollution and health − Steven Roberts, Michael Martin; Australian National University: 'Does ignoring model selection

effects in assessing the effect of PM on mortality make us too vigilant?' − Borek Puza, Terence O'Neill; Australian National University: 'Optimal constrained confidence

estimation via tail functions with applications to environmental data' − Erik-A. Sauleau, Agnes Fromont, Laurence Clerc, Audrey Bellisario, Claire Bonithon-Kopp,

Thibault Moreau, Christine Binquet; Faculty of medicine, University of Strasbourg, Strasbourg, France: 'Correction of edge effect in spatial generalized additive mixed models'

− Gavin Shaddick, Duncan Lee, Ruth Salway, Stephen Walker; University of Bath: 'Bayesian latent variable modelling in studies of air pollution and health. '

C20: Models for precipitation data − Antonella Bodini, Bruno Betrò, Antonio Cossu; Institute of Applied Mathematics and

Information Technology: 'Analysis of extreme events in Sardinia (Italy) via a hidden Markov model'

− Erdem Albek; Anadolu University: 'Correlation between Precipitation at two Stations in Turkey and NAO/MO indices '

− Claudie Beaulieu, Taha B.M.J. Ouarda, Ousmane Seidou; Princeton University: 'Bayesian detection of inhomogeneities in precipitation series'

C21: Change point detection in metereology − Daniela Jaruskova; Czech Technical University, Prague: 'A test for change in mean of random

vectors with application to temperature series' − Patricia Menéndez, Sucharita Ghosh, Hans Rudolf Künsch, Willy Tinner; Biometris,

Wageningen University & Swiss Federal Research Institute WSL, Switzerland: 'Smoothing and change point estimation for irregularly spaced palaeo proxy time series'

− Hagen Scherb, Kristina Voigt; Institute of Biomathematics and Biometry Helmholtz Zentrum München-German Research Center for Environmental Health: 'Radiation-induced genetic effects and ecological dose-response analyses'

− Laimonis Kavalieris; University of Otago: 'Changepoint detection in autocorrelated time series'

C22: Multivariate techniques − Mokhtar Elatrash, Aly Okasha, Hesham Ibrahim; Al-Mergheb University: 'Copper Removal

From Aqueous Solutions By Using Portland Cement Kiln Dust' − Annalina Sarra, Gianfranco Iurisci; Dipartimento di Metodi Quantitativi e Teoria Economica

Università "G.d'Annunzio" di Pescara: 'Solar photovoltaic in Italy: a statistical spatial analysis for accurate energy planning'

− Alessandro Zini; Università degli studi di Milano-Bicocca: 'Testing Isotropy of Space Covariance Functions'

xiii

C23: Marine environment − Juan Grima, Bruno Ballesteros, José Antonio Domínguez, Juan Antonio Luque; Geological

Survey of Spain (IGME): 'Data analysis of seawater intrusion in the Pego-Oliva marshland' − Scott D. Foster, Mark V. Bravington; CSIRO Mathematical and Information Sciences: 'A

Markov Model Approach to the Analysis of Video Transect Data from the Marine Environment' − Natalie Kelly, David Peel, Mark Bravington, Sharon Hedley; CSIRO: 'Design of Antarctic

circumpolar whale research cruises: a model-based approach'

POSTER SESSIONS POSTER SESSION 1 1) Suheyla Yerel, Huseyin Ankara; Bilecik University: 'Surface Water Quality Assessment using

Statistical Techniques' 2) Ivan Klozyatnyk, Natalia Klymenko; Institute of Colloid Chemistry and Chemistry of Water,

National Academy of Sciences of Ukraine: 'Impact of seasonal changes of the river water on drinking water quality'

3) Olena Samsoni-Todorova, Natalia Klymenko, Ivan Kozyatnyk; National Technical University of Ukraine Kiev Polytechnic Institute: 'Changing of biodegradable organic carbon in some processes of drinking water preparation'

4) Duncan Lee, Tereza Neocleous; University of Glasgow: 'Using quantile regression in environmental epidemiology'

5) Ali Gargoum; UAE University Professor: 'Bayesian forecasting algorithm for accommodating non-Gaussian air concentration observations '

6) Ramón Giraldo, Pedro Delicado, Jorge Mateu; Universitat Politècnica de Catalunya: 'Hierarchical clustering of spatially correlated functional data'

7) Eva Mª Ramos-Ábalos, Ramón Gutiérrez-Jaimez, Ramón Gutiérrez-Sánchez, Ahmed Nafidi; Universidad de Granada: 'Tri-Parameter Lognormal Diffusion Process with Exogenous Factors: Inference Based in Simulated Data'

8) H.I. Calvete, J.A. Carrión, C. Galé, E. García, R. Núñez-Lagos, C. Pérez, J. Puimedón, S. Rodríguez, M.L. Sarsa, J.A. Villar; Dpto. de Métodos Estadísticos. Universidad de Zaragoza: 'Environmental radioactivity monitoring in Aragón (Spain)'

9) Ronald E. McRoberts, Susan M. Stein, Lisa G. Mahal; U.S. Forest Service: 'Forests on the Edge' 10) Elizabeth Mannshardt-Shamseldin, Eric Gilleland, Harold Brooks; Duke University and the

Statistical and Applied Mathematical Sciences Institute: 'Severe Weather under a Changing Climate: Large Scale Indicators of Extreme Events'

11) Baba Thiam, Sophie Dabo-Niang; University Lille 3: 'Robust quantile regression estimation for spatial processes'

12) Serdar Göncü, Erdem Ahmet Albek, Ömer Güngör; Anadolu University: 'Spatial Variation of Factors Obtained from Two Monitoring Stations Along a Stream by Factor Analysis'

13) Nuria Rico, Desiree Romero, Francisco Torres, Patricia Román; Universidad de Granada, Spain: 'Gompertz-lognormal diffusion process for modelling the accumulated nutrients dissolved in a cultivation of Capiscum Annuum.'

14) Zbigniew Ziembik, Agnieszka Dolhaczuk-Srodka, Maria Waclawek; Opole University: 'Application of robust Principal Component Analysis for assessment of Cs-137 transport in forest soil'

15) Sophie Dabo-Niang, Anne-Françoise Yao, Mustapha Rachdi; University Lille 3: 'Space regression estimation for functional data'

xiv

16) Mine Albek, Erdem Albek; Anadolu University: 'Investigation of the interactions among water quality variables in a stream with Partial Correlations'

17) Ottorino-Luca Pantani, Irene Lozzi, Luca Calamai, Marinella Bosetto, Ettore Capri; Dip di Scienza del Suolo e Nutrizione della Pianta, Univ. di Firenze: 'Modelling sorption of tricyclazole on rice paddy sediments by statistical analysis of lab-sorption data'

18) Rossella Onorati, Paul D. Sampson, Peter Guttorp; University of Washington: 'A Spatio-Temporal Model based on the SVD to analyze large Spatio-Temporal datasets'

19) Lee Fawcett, David Walshaw; Newcastle University: 'A Hierarchical Model for Extreme Wind Speeds'

20) Alessandro Vagheggini, Francesca Bruno; Dipartimento di Scienze Statistiche, Università di Bologna: 'A Geographically Weighted Regression-based Generalized Regression estimator for spatial data'

21) Massimiliano Lega, Claudia Ferrara, Patrizia Manganiello, Vincenzo Severino; DiSAm-Dipartimento di Scienze per l'Ambiente - University of Naples Parthenope (Italy) : 'Soil Monitoring by aerial infrared thermography'

22) Senatro Di Leo, Carmelina Cosmi, Maria Macchiato, Maria Ragosta; Dipartimento di Ingegneria e Fisica dell'Ambiente, Università della Basilicata, Potenza Italy: 'Sustainability indicators from multidimensional analysis of partial equilibrium model data'

23) Francesca Bruno, Daniela Cocchi; Department of Statistics "P.Fortunati", University of Bologna: 'Air pollution indices: comparisons of their uncertainty'

24) Scott D. Foster, Piers K. Dunstan; CSIRO Mathematical and Information Sciences: 'Biodiversity Analysis Using Rank Abundance Distributions'

25) Claire Ferguson, Marian Scott, Laurence Carvalho, Geoffrey A. Codd, Andrew Tyler; Department of Statistics, University of Glasgow, Glasgow, UK: 'Monitoring and Managing Cyanobacterial Risks in Freshwater Lakes'

26) Monica Musio, Nicole Augustin, Klaus H. von Wilpert, Edgar Kublin, Simon Wood, Martin Schumacher; University of Cagliari: 'Modelling Spatio-temporal forest health data'

POSTER SESSION 2 27) Livia Trizio, Maurizio Caselli, Gianluigi de Gennaro, Pierina Ielpo; University of Bari: 'A

Simple Feedforward Neural Network for the PM10 Forecasting: Comparison with a Radial Basis Function Network and a Multivariate Linear Regression Model'

28) Maria Kozlowska, Agnieszka Lacka, Roman Krawczyk, Radoslaw J. Kozlowski; Department of Mathematical and Statistical Methods, Poznan University of Life Sciences: 'Some block designs with nested rows and columns for research on pesticide dose limitation'

29) Jean Sanderson, Michel Crucifix, Piotr Fryzlewicz, Jonathan Rougier; University of Bristol: 'Spectral analysis of irregularly spaced paleoclimatic time series using empirical mode decomposition and wavelet lifting.'

30) Helen Powell, Duncan Lee, Adrian Bowman; University of Glasgow: 'Do ozone concentrations affect respiratory health in Scotland?'

31) Petra Kuhnert, Shane Griffiths, William Venables, Stephen Blaber; CSIRO Mathematical and Information Sciences: 'Estimating abundance of pelagic fishes using gillnet catch data in data-limited fisheries: a Bayesian approach'

32) Barbara Cafarelli, Annamaria Castrignano', Antonio Troccoli, Salvatore Colecchia; Università degli Studi di Foggia: 'The use of geoadditive models to estimate the spatial distribution of grain weight in an agronomic field: a comparison with kriging with external drift'

33) Mattia Mercuriali, Maria Grazia Perrone, Luca Ferrero, Ezio Bolzacchini, Maria Chiara Pietrogrande; Department of Chemistry, University of Ferrara: 'Data handling based on AutoCoVariance Function for decoding complex signals from environmental monitoring: identification of organic tracers in atmospheric aerosol'

xv

34) Zdenek Pospisil, Jitka Kuhnova; Masaryk University, Brno, Czech Republic: 'Bio-diversity in vineyards with conventional, biological and integrated treatment'

35) Stan Yip; National Centre for Atmospheric Science and University of Exeter: 'On the partitioning uncertainty in global climate predictions'

36) James Sweeney, John Haslett; Trinity College Dublin: 'Multivariate Non-Parametric Regression via Gaussian Markov Random Fields'

37) Sara Focaccia; DICMA: 'The problem of the complex hydrocarbon pollution in stockage sites: the importance of geostatistics'

38) Claire Faucheux, Yves Benoit, Claire Carpentier, Chantal de Fouquet, Bruno Fricaudet, Jean-Christophe Gourry, Edwige Polus-Lefebvre; Mines ParisTech-Centre de Géosciences-Géostatistique: 'Spatial variability of hydrocarbon polluted soils: main contributions of the LOQUAS project'

39) Gavin Shaddick, Marta Blangiardo, Ruth Salway, Alex Zenie, Bruce Denby, Edzer Pebesma; University of Bath: 'Uncertainty analysis within the HEIMTSA (Health and Environment Integrated Methodology and Toolbox for Scenario Assessment) project. '

40) Klara Kubosova, Jiri Jarkovsky, Karel Brabec, Svetlana Zahradkova, Jindriska Bojkova, Pavel Bartusek; RECETOX (Research Centre for Environmental Chemistry and Toxicology), Masaryk University: 'Analysis of stream macroinvertebrates response to environmental conditions: research support of the Water Framework Directive implementation in the Czech Republic'

41) Marco A. Rodríguez, Clarice G.B. Demétrio, Silvio S. Zocchi, Roseli A. Leandro, Julie Deschênes; Université du Québec à Trois-Rivières, Trois-Rivières, Canada: 'Multilevel zero-inflated regression for modelling species abundance in relation to habitat: a Bayesian approach'

42) Raquel Menezes, Ana Cristina Fernandes; Minho University, Portugal: 'Data analysis on environmental monitoring networks'

43) Katarzyna Sobiech-Matura, Maria A. Olech, Jerzy W. Mietelski, Anna Masniak; Institute of Botany, Department of Biology and Earth Science, Jagiellonian University, Poland: '137CS, 40K, 238PU, 239+240PU and 90SR in environmental material originatig from king george island (South Shetlands, Antarctica)'

44) Ramos M. do Rosário; Universidade Aberta(PT) and CMAF, University of Lisbon: 'Trend testing: searching for a better estimation of the variance of the test statistic.'

45) María C. Bueso, José M. Angulo, María D. Ruiz-Medina; Department of Applied Mathematics and Statistics, Technical University of Cartagena, Spain: 'Multi-scale clustering of spatio-temporal observations'

46) Beverly McNaughton; Science and Technology Branch, Department of the Environment: 'National Freshwater Quality Indicator-A Canadian Environmental Sustainability Indicator'

47) Lee YoungSaeng, Yoon SangHoo, Park Jeong Soo; Chonnam National University: 'Non-stationary Frequency Analysis of Extreme Precipitation in South Korea'

48) Lee Jung Hee, Jeong Bo-Yoon, Park Jeong Soo; Chonnam national university: ' Fisher information matrix and Method of L-moments on the Generalized Gumbel Distribution'

49) Francesca Bruno, Rodolfo Rosa, Luca Talenti; Department of Statistics "P.Fortunati", University of Bologna, Italy: 'Parametric spatial bootstrap and moving block bootstrap: a comparative study'

50) Mario Lloyd Virgilio Martina, Sara Pignone, Ezio Todini; University of Bologna: 'A Bayesian approach to determine the rainfall thresholds for landslides triggering '

Abstracts of Papers and Posters

Listed in alphabetic order (by presenting author)

Contents

SEMI-PARAMETRIC ESTIMATION OF THE INTENSITY FUNCTION IN SPACE-TIME POINT PROCESSESGiada Adel�o, Marcello Chiodi 16

SPACE TIME MODELING OF PRECIPITATION USING HIDDEN MARKOV MODELSPierre Ailliot, Craig Thompson, Peter Thomson 16

CORRELATION BETWEEN PRECIPITATION AT TWO STATIONS IN TURKEYAND NAO/MO INDICESErdem Albek 17

INVESTIGATION OF THE INTERACTIONS AMONGWATER QUALITY VARIABLESIN A STREAM WITH PARTIAL CORRELATIONSMine Albek, Erdem Albek 17

MODELLING THE ONTARIO FIRE WEATHER INDEXAlisha Albert-Green, W. John Braun, David L. Martell, Douglas G. Woolford 18

BAYESIAN MODELLING VOLATILITY OF GROWTH RATE IN ATMOSPHERICCARBON DIOXIDE CONCENTRATIONSEsmail Amiri 18

SYNERGISTIC USE OF SEVERAL RECEPTOR MODELS (CMB, APCS AND PMF)TO INTERPRET AIR QUALITY DATAEleonora Andriani, Maurizio Caselli, Gianluigi de Gennaro 18

WAVELET-BASEDMULTISCALE INTERMITTENCYANALYSIS IN ENVIRONMENTALAPPLICATIONSJosé M. Angulo, Ana E. Madrid 19

ASSESSMENT OF SURFACE WATER QUALITY BY MULTIVARIATE STATISTICALANALYSIS TECHNIQUESHuseyin Ankara, Suheyla Yerel 19

SMOOTHED ANOVA WITH SPATIAL EFFECTS AS A COMPETITOR TO MCAR INMULTIVARIATE SPATIAL SMOOTHINGSudipto Banerjee, Yufen Zhang, James Hodges 20

BAYESIAN DETECTION OF INHOMOGENEITIES IN PRECIPITATION SERIESClaudie Beaulieu, Taha B.M.J. Ouarda, Ousmane Seidou 20

A MULTIVARIATE SPATIO-TEMPORAL DOWNSCALER FOR OUTPUT FROMNUMERICAL MODELSVeronica Berrocal, Alan Gelfand 21

AN ESTIMATE OF THE INDUSTRIAL METABOLISM OF THE PIEDMONT REGION(ITALY) USING THE ENVIRONMENTAL INPUT-OUTPUT ANALYSIS AND THEECOLOGICAL FOOTPRINTPancrazio Bertaccini, Marco Bagliani 21

AIR POLLUTION: METEOROLOGY OR TRAFFIC, WHAT DOES REALLY MATTER?Pancrazio Bertaccini, Vanja Dukic, Rosaria Ignaccolo 22

A BAYESIAN MODEL OF TIME ACTIVITY DATA FOR ECOLOGICAL STUDIESWITH IMPLICATIONS ON THE BIAS OF DISEASE RISKSMarta Blangiardo, Sylvia Richardson 22

1

2

ANALYSIS OF EXTREME EVENTS IN SARDINIA (ITALY) VIA A HIDDENMARKOVMODELAntonella Bodini, Bruno Betrò , Q. Antonio Cossu 23

NON-STATIONARY SPATIAL ARMA MODELS APPLIED TO GLOBAL OZONEDATADavid Bolin, Finn Lindgren 23

AIR POLLUTION INDICES: COMPARISONS OF THEIR UNCERTAINTYFrancesca Bruno, Daniela Cocchi 23

PARAMETRIC SPATIAL BOOTSTRAP AND MOVING BLOCK BOOTSTRAP: ACOMPARATIVE STUDYFrancesca Bruno, Rodolfo Rosa, Luca Talenti 24

A MIXED, DESIGN-BASED MODEL-BASED SAMPLING APPROACH FORESTIMATING GLOBAL QUANTITIES IN SPACE-TIMEDick Brus, Jaap de Gruijter 24

MULTI-SCALE CLUSTERING OF SPATIO-TEMPORAL OBSERVATIONSMaría C. Bueso, José M. Angulo, María D. Ruiz-Medina 25

THE USE OF GEOADDITIVEMODELS TO ESTIMATE THE SPATIAL DISTRIBUTIONOF GRAIN WEIGHT IN AN AGRONOMIC FIELD: A COMPARISONWITH KRIGINGWITH EXTERNAL DRIFTBarbara Cafarelli, Annamaria Castrignano', Antonio Troccoli, Salvatore Colecchia 25

JOINT INCIDENCE OF VARIOUS DISEASES IN THE PRESENCE OF A RISKSOURCECrescenza Calculli, Alessio Pollice 26

ENVIRONMENTAL RADIOACTIVITY MONITORING IN ARAGóN (SPAIN)H.I. Calvete, J.A. Carrión, C. Galé, E. García, R. Núñez-Lagos, C. Pérez, J. Puimedón, S.Rodríguez, M.L. Sarsa, J.A. Villar 26

COMPARING AIR QUALITY STATISTICAL MODELSMichela Cameletti, Rosaria Ignaccolo, Stefano Bande 27

EPIDEMIOLOGIC SURVEILLANCE AND IMPACT EVALUATION: THE FALSEDISCOVERY RATEDolores Catelan, Annibale Biggeri, Corrado Lagazio 27

MONITORING ATMOSPHERIC CHLOROFLUOROCARBONS BY THE LONGITUDINALBENT-CABLE MODELGrace Chiu, Shahedul Ahsan Khan, Joel Dubin 28

IDENTIFYING POLLUTION SOURCE DIRECTIONS FOR POLLUTION SOURCEAPPORTIONMENTWilliam Christensen, Basil Williams, Shane Reese 28

URBAN FOREST ASSESSMENT IN ITALY UNDER NATIONAL FORESTINVENTORY FRAMEWORKPiermaria Corona, Mariagrazia Agrimi, Federica Ba�etta, Anna Barbati, Lorenzo Fattorini,Enrico Pompei, Walter Mattioli 28

CONNECTION BETWEEN METEOROLOGICAL AND OZONE SCENARIOS INDIFFERENT URBAN AREASRossana Cotroneo, Silvia Bartoletti, Armando Pelliccioni 29

3

LONG-RANGE CLIMATE RECONSTRUCTIONS WITH DYNAMICAL SYSTEMSMichel Cruci�x, Rougier Jonathan 29

SPACE REGRESSION ESTIMATION FOR FUNCTIONAL DATASophie Dabo-Niang, Anne-Françoise Yao, Mustapha Rachdi 30

IMPACT OF RESIDENTIAL UNDERGROUND WATER STORAGE TANKS ONDRINKING WATER QUALITY OF JEDDAH SUPPLY SYSTEM, SAUDI ARABIAMaged Daoud, Saleh Magram 30

THE LONGITUDINAL DEPENDENCE OF BLACK CARBON CONCENTRATION ONTRAFFIC VOLUME IN AN URBAN ENVIRONMENTRey Decastro, Timothy Buckley, Lu Wang, Jana Mihalic, Patrick Breysse, Alison Geyh 31

ENVIRONMENTAL COMPLIANCE, SOCIAL REGULATION A GAME THEORETICMODELDr Suresh Deman 31

THE TESTING OF ZERO-INFLATION AND OVER-DISPERSION FOR THEENVIRONMENTAL COUNT DATADianliang Deng 32

SUSTAINABILITY INDICATORS FROM MULTIDIMENSIONAL ANALYSIS OFPARTIAL EQUILIBRIUM MODEL DATASenatro Di Leo, Carmelina Cosmi, Maria Macchiato, Maria Ragosta 32

AIR QUALITY MONITORING FUSING SATELLITE REMOTE SENSING, GROUND-BASED MEASUREMENTS AND METEOROLOGICAL MODELING IN NORTHERNITALYWalter Di Nicolantonio, Alessandra Cacciari, Gabriele Curci, Paolo Stocchi, Ezio Bolzacchini,Luca Ferrero, Claudio Tomasi 33

AIR QUALITY ASSESSMENT VIA FUNCTIONAL PRINCIPAL COMPONENTANALYSISFrancesca Di Salvo, Gianna Agro', Mariantonietta Ruggieri, Antonella Plaia 33

A BAYESIAN HIERARCHICAL MODEL FOR ESTIMATING THE HEALTH EFFECTSOF CHEMICAL CONSTITUENTS OF PARTICULATE MATTERFrancesca Dominici, Roger Peng, Michelle Bell 34

BAYESIAN DENSITY REGRESSION AND MIXTURES WITH ENVIRONMENTALAPPLICATIONSDavid Dunson 34

ESTIMATION OF HIERARCHICAL SPATIO-TEMPORAL COREGIONALIZATIONMODELS WITH THE EM ALGORITHMCinzia D'Ariano, Alessandro Fassò, Francesco Finazzi 34

MATRIX INVERSION AND STATISTICAL DATA ANALYSISAbdel El-Shaarawi 35

COPPER REMOVAL FROM AQUEOUS SOLUTIONS BY USING PORTLANDCEMENT KILN DUSTMokhtar Elatrash, Aly Okasha, Hesham Ibrahim 35

CHARACTERIZING SPATIAL PATTERNS OF FIRE WEATHER USING HISTORICALDATASylvia Esterby, Zuzana Hrdlickova, Steve Taylor 36

4

A NON-STATIONARY NEYMAN-SCOTT MODEL FOR RAINFALLGuillaume Évin, Anne-Catherine Favre 36

SAMPLING STRATEGIES FOR THE ASSESSMENT OF ECOLOGICAL DIVERSITYLorenzo Fattorini 37

SPATIAL VARIABILITY OF HYDROCARBON POLLUTED SOILS: MAINCONTRIBUTIONS OF THE LOQUAS PROJECTClaire Faucheux, Yves Benoit, Claire Carpentier, Chantal de Fouquet, Bruno Fricaudet,Jean-Christophe Gourry, Edwige Polus-Lefebvre 37

HOW TO BUILD AN INITIAL SAMPLING SCHEME: RECOMMENDATIONS FORMEASUREMENT SURVEYS OF AIR QUALITYClaire Faucheux, Chantal de Fouquet, Giovanni Cárdenas, Laure Malherbe 38

GENERATING FUZZY LOGIC-BASED RAINFALL-RUNOFF MODELS USING SELFORGANIZING MAPSChristophe Faust, Peter Gemmar, Oliver Gronz, Markus Casper 38

BAYESIAN INFERENCE FOR CLUSTERED EXTREMESLee Fawcett, David Walshaw 39

A HIERARCHICAL MODEL FOR EXTREME WIND SPEEDSLee Fawcett, David Walshaw 39

MONITORING AND MANAGING CYANOBACTERIAL RISKS IN FRESHWATERLAKESClaire Ferguson, E. Marian Scott, Laurence Carvalho, Geo�rey A. Codd, Andrew Tyler 39

APPROXIMATED WRAPPED DISTRIBUTIONS FOR MODELING CIRCULAR DATAClarissa Ferrari, Alan E. Gelfand, Giovanna Jona Lasinio, Daniela Cocchi 40

INTEGRATING SATELLITE AND GROUND LEVEL DATA FOR AIR QUALITYMONITORING AND DYNAMICAL MAPPINGFrancesco Finazzi, Cinzia D'Ariano, Alessandro Fassò, Gianandrea Mannarini, Orietta Nicolis 40

A HIERARCHICAL MIXTURE MODEL FOR ESTIMATING ZERO INFLATEDCONTINUOUS FOREST VARIABLESAndrew Finley, Sudipto Banerjee 41

THE PROBLEM OF THE COMPLEX HYDROCARBON POLLUTION IN STOCKAGESITES: THE IMPORTANCE OF GEOSTATISTICSSara Focaccia 41

A MARKOV MODEL APPROACH TO THE ANALYSIS OF VIDEO TRANSECT DATAFROM THE MARINE ENVIRONMENTScott D. Foster, Mark V. Bravington 42

BIODIVERSITY ANALYSIS USING RANK ABUNDANCE DISTRIBUTIONSScott D. Foster, Piers K. Dunstan 42

SAMPLING PROPERTIES OF SPATIAL TOTAL ESTIMATORS UNDER TESSELLATIONSTRATIFIED DESIGNSSara Franceschi 43

EXTREME EVENTS AND BUSINESS CONTINUITY PLANNING - EXPLORATIONIN FLOOD RISK STRATEGIES WITHIN SCOTLANDMaria Franco Villoria, Marian Scott, Trevor Hoey, Denis Smith, Alistair Cargill 43

5

BAYESIAN FORECASTING ALGORITHM FOR ACCOMMODATING NON-GAUSSIAN AIR CONCENTRATION OBSERVATIONSAli Gargoum 43

CLIMATE RECONSTRUCTED FROM POLLEN DATA USING A DYNAMICVEGETATION MODELVincent Garreta, Joël Guiot 44

HIERARCHICAL CLUSTERING OF SPATIALLY CORRELATED FUNCTIONALDATARamón Giraldo, Pedro Delicado, Jorge Mateu 44

DETECTING HIGH-RISK REGIONS IN DISEASE MAPPING USING SPATIALP-SPLINE MODELSTomás Goicoa, M.D. Ugarte, Jaione Etxeberria, Ana F. Militino 45

PREDICTION OF WATER QUALITY VARIABLES USING STATE-SPACE ANDLINEAR MODELS FOR RIVER NETWORK DATAA. Manuela Gonçalves, Marco Costa 45

A MULTIVARIATE CAR MODEL AND ITS APPLICATIONSFedele Greco, Daniela Cocchi, Carlo Trivisano 46

MODELLING JOINT EXTREMES: APPLICATION TO UK RIVER FLOW DATAOlivia Grigg, Jonathan Tawn 46

DATA ANALYSIS OF SEAWATER INTRUSION IN THE PEGO-OLIVA MARSHLANDJuan Grima, Bruno Ballesteros, José Antonio Domínguez, Juan Antonio Luque 46

INTEGRATING SMOOTHING AND REGRESSION TREES FOR CHANGE-POINTDETECTION IN ENVIRONMENTAL DATAAnders Grimvall, Sackmone Sirisack 47

RESTRICTING PARAMETER SPACE IN MESOSCALE WATER BALANCE MODELSUSING HYDROLOGICAL SOIL MAPSOliver Gronz, Peter Gemmar, Markus Casper 47

STOCHASTIC GROUNDWATER MODELLING FOR A FRACTURATED AQUIFERIN AUGUSTA AREA (SYRACUSE, ITALY)Enrico Guastaldi, Andrea Carloni, Vincenzo Ferrara, Claudio Gallo 48

DETECTION OF OCEANIC INFLUENCE ON THE PRECIPITATION OF THECENTRAL VENEZUELAN COAST USING TIME-VARYING MODELSLelys Guenni, Gabriel Huerta, Bruno Sansó 48

SIMULTANEOUS MODELLING OF SPATIAL AND TEMPORAL VARIATIONSIN AIR POLLUTION EXPOSURES FOR HEALTH RISK ASSESSMENT ANDEPIDEMIOLOGICAL ANALYSISJohn Gulliver, Marta Blangiardo, David Briggs, Anna Hansell 49

LOOKING FOR CLIMATE CHANGE SIGNALS IN EXTREME TEMPERATURESPeter Guttorp 49

SPATIAL VARIATION OF FACTORS OBTAINED FROM TWO MONITORINGSTATIONS ALONG A STREAM BY FACTOR ANALYSISSerdar Göncü, Erdem Ahmet Albek, Ömer Güngör 50

6

EXTREME VALUE THEORY APPLIED TO THE DEFNITION OF BATHINGWATERDISCOUNTING LIMITSRuth Haggarty, Marian Scott, Claire Ferguson 50

A TALE OF TWO PHASES: DESIGN AND ESTIMATION OF TREE FOLIAGEBIOMASSTemesgen Hailemariam, Vicente Monleon, Aaron Weiskittel, Duncan Wilson 51

STATISTICAL METHODS IN THE RECONSTRUCTION OF PALAEOCLIMATEJohn Haslett 51

PROJECTIONS OF FUTURE INSURANCE LOSSES FROM CLIMATE MODEL DATAOla Haug 52

SPATIO-TEMPORAL DISEASE MAPPING USING INLALeonhard Held, Birgit Schrödle 52

COMPARATIVE ANALYSIS OF MODEL BEHAVIOUR FOR FLOOD PREDICTIONPURPOSES USING SELF-ORGANIZING MAPSMarcus Herbst, Markus C. Casper, Jens Grundmann, Oliver Buchholz 53

BAYESIAN SCALE SPACE ANALYSIS WITH APPLICATION TO REMOTE SENSINGAND CLIMATE MODELINGLasse Holmström, Leena Pasanen 53

DOES ENVIRONMENTAL PERFORMANCE AFFECT FINANCIAL PERFORMANCE?A META-ANALYSISEva Horvathova 54

RECONSTRUCTING HOURLY PM10 GRAVIMETRIC MEASUREMENTS THROUGHTRANSFER FUNCTION MODELSDaniele Imparato, Mauro Gasparini 54

APPLICATIONS OF GREY RELATIONAL METHOD TO RIVER ENVIRONMENTQUALITY EVALUATION IN CHINAWai Cheung Ip, Baoqing Hu, Heung Wong, Jun Xia 55

EARTHQUAKES FROM THE INDONESIAN REGION: AN APPLICATION OFEXACT COMPUTABLE EXPRESSIONS FOR THE ASYMPTOTIC DISTRIBUTIONOF CHANGE-POINT MLE IN THE EXPONENTIAL CASEVenkata K. Jandhyala, Stergios B. Fotopoulos, Elena Khapalova 55

A TEST FOR CHANGE IN MEAN OF RANDOM VECTORS WITH APPLICATIONTO TEMPERATURE SERIESDaniela Jaruskova 55

SPATIO TEMPORAL DATA MODELING IN ENVIRONMENTAL SCIENCES AREVIEW.Giovanna Jona Lasinio 56

CHANGEPOINT DETECTION IN AUTOCORRELATED TIME SERIESLaimonis Kavalieris 56

EFFECT OF TROPOSPHERIC OZONE ON TOBACCO PLANTS IN VARIOUSEXPOSURE SERIES DURING GROWING SEASONSDariusz Kayzer, Klaudia Borowiak, Anna Budka, Janina Zbierska 57

7

DESIGN OF ANTARCTIC CIRCUMPOLAR WHALE RESEARCH CRUISES: AMODEL-BASED APPROACHNatalie Kelly, David Peel, Mark Bravington, Sharon Hedley 57

RECORDS METHOD FOR THE NATURAL DISASTERS APPLICATION TO THESTORM EVENTSZaher Khraibani, Hussein M. Badran, Hussein Khraibani 58

COMPARING PIECEWISE LINEAR TRENDSHyune-Ju Kim, Jun Luo, Michael Barrett, Eric Feuer 58

IMPACT OF SEASONAL CHANGES OF THE RIVERWATER ON DRINKINGWATERQUALITYIvan Klozyatnyk, Natalia Klymenko 58

DEALING WITH UNCERTAINTIES IN ENVIRONMENTAL BURDEN OF DISEASEASSESSMENTAnne Knol, Arthur Petersen, Jeroen van der Sluijs 59

SPATIAL MODEL OF PERSISTENT ORGANIC POLLUTANTS VOLATILIZATIONFROM SOIL: CONNECTION OF DETERMINISTIC AND STOCHASTIC APPROACHJiri Komprda, Klara Kubosova, Milan Sanka, Ondrej Hajek, Ivan Holoubek 59

THE CAUSES OF CLIMATE CHANGEMilena Kovárová 60

SOME BLOCK DESIGNS WITH NESTED ROWS AND COLUMNS FOR RESEARCHON PESTICIDE DOSE LIMITATIONMaria Kozlowska, Agnieszka Lacka, Roman Krawczyk, Radoslaw J. Kozlowski 60

ANALYSIS OF STREAMMACROINVERTEBRATES RESPONSE TO ENVIRONMENTALCONDITIONS: RESEARCH SUPPORT OF THE WATER FRAMEWORK DIRECTIVEIMPLEMENTATION IN THE CZECH REPUBLICKlara Kubosova, Jiri Jarkovsky, Karel Brabec, Svetlana Zahradkova, Jindriska Bojkova, PavelBartusek 60

INCORPORATING UNCERTAINTY INTO GULLY EROSION CALCULATIONS: ARANDOM FOREST APPROACHPetra Kuhnert, Anne Kinsey-Henderson, Rebecca Bartley, Alexander Herr 61

ESTIMATING ABUNDANCE OF PELAGIC FISHES USING GILLNET CATCH DATAIN DATA-LIMITED FISHERIES: A BAYESIAN APPROACHPetra Kuhnert, Shane Gri�ths, William Venables, Stephen Blaber 61

MULTIPLE IMPUTATION OF INCOMPLETE OCEANOGRAPHIC LINEAR-CIRCULAR DATA USING MULTIVARIATE MIXTURE MODELSFrancesco Lagona, Marco Picone 62

A CRITICAL LOOK AT THE FULFILLMENT OF BASIC ASSUMPTIONS FOR THEAPPLICATION OF TWO COMMON RECEPTOR MODELS CMB AND PMF FORSOURCE APPORTIONMENT OF PM10Bo R. Larsen 62

NON-STATIONARY FREQUENCY ANALYSIS OF EXTREME PRECIPITATION INSOUTH KOREAYoungSaeng Lee, SangHoo Yoon, Jeong Soo Park 63

8

FIRE RISK ASSESSMENT IN MUSKOKA, ONTARIOJonathan Lee, W. John Braun, Bruce Jones, Doug Woolford, Mike Wotton 63

SPATIAL MODELLING AND ECOLOGICAL BIAS WITHIN AIR POLLUTION ANDHEALTH STUDIESDuncan Lee, Gavin Shaddick 63

USING QUANTILE REGRESSION IN ENVIRONMENTAL EPIDEMIOLOGYDuncan Lee, Tereza Neocleous 64

USING ESTIMATED PERSONAL EXPOSURES IN STUDIES OF THE EFFECTS OFAIR POLLUTION ON HEALTHDuncan Lee, Gavin Shaddick, Ruth Salway, Jim Zidek 64

SOLIDWASTE LANDFILLSMONITORING BYAERIAL INFRARED THERMOGRAPHYMassimiliano Lega, Luca d'Antonio, Rodolfo M.A. Napoli 65

SOIL MONITORING BY AERIAL INFRARED THERMOGRAPHYMassimiliano Lega, Claudia Ferrara, Patrizia Manganiello, Vincenzo Severino 65

ELIMINATING THE PRACTICAL BOUNDARY BETWEEN MARKOV AND OTHERGAUSSIAN RANDOM FIELDSFinn Lindgren, Håvard Rue, Johan Lindström, David Bolin 66

POSSIBLE CLIMATE CHANGE EFFECTS ON MARINE SAFETYGeorg Lindgren 66

FAST ESTIMATION OF NON-STATIONARY GAUSSIAN MARKOV RANDOMFIELDSJohan Lindström, Finn Lindgren, Peter Jonsson, David Bolin, Håvard Rue 66

TEMPORAL AND SPATIAL ANALYSIS OF CLIMATIC CYCLES IN A DETRITICAQUIFER: BEHAVIOUR OF RECHARGEJuan Antonio Luque-Espinar, Eulogio Pardo-Igúzquiza, Mario Chica-Olmo, María JoséGarcía-Soldado 67

ANALYSIS OF THE PIEZOMETRIC SPATIAL DISTRIBUTION BASED ON THEESTIMATION OF PIEZOMETRIC DIFFERENCESJuan Antonio Luque-Espinar, Mario Chica-Olmo, Eulogio Pardo-Igúzquiza, María JoséGarcía-Soldado, Juan Grima-Olmedo 67

POISSON NONLINEAR MIXED MODELS FOR ENVIRONMENTAL DATARenjun Ma 68

MONITORING USING CHANGEPOINTSIan MacNeill, Krishna Jandhyala, Elena Naumova 68

A BOOTSTRAP VARIANCE ESTIMATOR FOR THE OBSERVED SPECIES RICHNESSIN QUADRAT SAMPLING FROM FINITE POPULATIONSSteen Magnussen, Ron McRoberts 68

SEVEREWEATHER UNDER A CHANGING CLIMATE: LARGE SCALE INDICATORSOF EXTREME EVENTSElizabeth Mannshardt-Shamseldin, Eric Gilleland, Harold Brooks 69

A BAYESIAN APPROACH TO DETERMINE THE RAINFALL THRESHOLDS FORLANDSLIDES TRIGGERINGMario Lloyd Virgilio Martina, Sara Pignone, Ezio Todini 69

9

A HURDLE MARKOV MODEL FOR POLLUTANTS CONCENTRATIONSAntonello Maruotti, Francesco Lagona 70

WAVELET-BASED ANALYSIS OF THE SPATIAL STRUCTURE OF POINTPATTERNSJorge Mateu, Orietta Nicolis 70

NATIONAL FRESHWATERQUALITY INDICATOR-A CANADIAN ENVIRONMENTALSUSTAINABILITY INDICATORBeverly McNaughton 71

SATELLITE IMAGE-BASED MAPS: SCIENTIFIC INFERENCE OR JUST PRETTYPICTURES?Ronald E. McRoberts 71

FORESTS ON THE EDGERonald E. McRoberts, Susan M. Stein, Lisa G. Mahal 71

SMOOTHING AND CHANGE POINT ESTIMATION FOR IRREGULARLY SPACEDPALAEO PROXY TIME SERIESPatricia Menéndez, Sucharita Ghosh, Hans Rudolf Künsch, Willy Tinner 72

DATA ANALYSIS ON ENVIRONMENTAL MONITORING NETWORKSRaquel Menezes, Ana Cristina Fernandes 72

DATA HANDLING BASED ON AUTOCOVARIANCE FUNCTION FOR DECODINGCOMPLEX SIGNALS FROM ENVIRONMENTAL MONITORING: IDENTIFICATIONOF ORGANIC TRACERS IN ATMOSPHERIC AEROSOLMattia Mercuriali, Maria Grazia Perrone, Luca Ferrero, Ezio Bolzacchini, Maria ChiaraPietrogrande 73

BAYESIAN SPATIAL PANEL DATAMaura Mezzetti, Samantha Leorato 73

FLEXIBLE STATISTICAL MODELS IN THE STUDY OF VULNERABILITY TOENVIRONMENTAL EXPOSURERossella Miglio, Francesca Bruno 74

SPACE-TIME MODELS FOR MOVING FIELDS. APPLICATION TO SIGNIFICANTWAVE HEIGHTValérie Monbet, Pierre Ailliot, Anne Cuzol, Nicolas Raillard 74

FUNCTIONAL KRIGING OF OCEAN PROFILE DATAPascal Monestiez, David Nerini 75

ZERO-INFLATED GENERALISED POISSON REGRESSION MODEL TO DESCRIBEPSEUDO-NITZSCHIA CONCENTRATION IN LISBON BAYHelena Mouriño, So�a Palma, Maria Teresa Moita, Maria Isabel Barão 75

SCORE AND QUASI-SCORE INFERENCE FOR CHANGE-POINTS, WITHAPPLICATIONS IN MARINE ECOLOGY AND GROUNDWATER MONITORINGVito Muggeo, Gianfranco Lovison 76

MODELLING SPATIO-TEMPORAL FOREST HEALTH DATAMonica Musio, Nicole, H. Augustin, Klaus von Wilpert, Edgar Kublin, Simon Wood, MartinSchumacher 76

10

MODELLING SPACE-TIME VARIATION OF CANCER INCIDENCE DATA: A CASESTUDYMonica Musio, Erik Sauleau, Nicole, H. Augustin 77

A UNIFIED FRAMEWORK OF FRACTAL AND WAVELET RANDOM FIELDSOrietta Nicolis, George Christakos 77

SPATIAL STATISTICS, COMPUTER MODELS AND REGIONAL CLIMATE CHANGEDouglas Nychka, Stephen Sain, Linda Mearns, Cari Kaufman 77

A SPATIO-TEMPORAL MODEL BASED ON THE SVD TO ANALYZELARGESPATIO-TEMPORAL DATASETSRossella Onorati, Paul D. Sampson, Peter Guttorp 78

MODELLING OF SPATIAL EXTREMES: A REVIEWSimone Padoan, Anthony Davison, Mathieu Ribatet 78

MODELLING SORPTION OF TRICYCLAZOLE ON RICE PADDY SEDIMENTS BYSTATISTICAL ANALYSIS OF LAB-SORPTION DATAOttorino-Luca Pantani, Irene Lozzi, Luca Calamai, Marinella Bosetto, Ettore Capri 79

QSAR MODELLING AND MULTIVARIATE ANALYSIS OF THE ENVIRONMENTALBEHAVIOUR OF ORGANIC POLLUTANTSEster Papa, Paola Gramatica 79

BAYESIAN METHODS FOR RECONSTRUCTING PAST CLIMATE HISTORIESAndrew Parnell, John Haslett, Michael Salter-Townshend 80

A CAUSALMODELLING APPROACH TO SPATIAL AND TEMPORAL CONFOUNDINGIN ENVIRONMENTAL IMPACT STUDIESWarren Paul 80

HARP-A SOFTWARE TOOL FOR DECISION SUPPORT DURING NUCLEAREMERGENCIESPetr Pecha, Radek Hofman, Emily Pechova 81

FORECAST OF OZONE POLLUTANT UP TO 5 DAYS IN ADVANCE IN ROMEURBAN AREA BY MEANS NEURAL NETArmando Pelliccioni, Stefano Lucidi, Vittorio La Torre, Fabrizio Pungì 81

ENVIRONMENT MONITORING PERFORMED BY ADVANCED HYBRID AIRSHIPAT LOW ALTITUDEGiuseppe Persechino, Pasquale Schiano, Massimiliano Lega, Rodolfo Napoli 82

BAYESIAN NONPARAMETRIC MIXTURES FOR LOCAL CLUSTERING OFFUNCTIONAL DATASonia Petrone, Michele Guindani, Alan E. Gelfand 82

A CHEMOMETRIC APPROACH BASED ON THE AUTOCOVARIANCE FUNCTIONFOR HANDLING COMPLEX SIGNALS FROM ENVIRONMENTAL MONITORINGMaria Chiara Pietrogrande, Mattia Mercuriali, Nicola Marchetti, Luisa Pasti, Dimitri Bacco,Gaetano Zanghirati, Francesco Dondi 83

OPERATIONAL OCEANOGRAPHY: THE SCIENCE BASED APPROACH TOMARINE MANAGEMENT PROBLEMSNadia Pinardi, Srdjan Dobricic, Ralph Milli� 83

11

DARWINIAN THEORY BASED TECHNIQUE TO PREDICT AIR POLLUTANTCONCENTRATIONSJosé Pires, F.G. Martins, M.C.M. Alvim-Ferraz, M.C. Pereira 84

AN AGGREGATE AIR QUALITY INDEX CONSIDERING INTERACTIONS AMONGPOLLUTANTSAntonella Plaia, Mariantonietta Ruggieri, Anna Lisa Bondì 84

THREE NONLINEAR STATISTICAL METHODS TO ANALYZE PM10 POLLUTIONIN ROUEN AREAJean-Michel Poggi , Francois-Xavier Jollois, Bruno Portier 85

A CRITICAL REVIEW OF SOME STATISTICAL ISSUES IMPLIED BY THE USE OFMULTIVARIATE RECEPTOR MODELSAlessio Pollice 85

FUNCTIONAL ESTIMATION OF GAUSSIAN DAGUM RANDOM FILEDS ANDRELATED MODELSEmilio Porcu, Maria Dolores Ruiz Medina, Rosaura Fernàndez Pascual 86

BIO-DIVERSITY IN VINEYARDS WITH CONVENTIONAL, BIOLOGICAL ANDINTEGRATED TREATMENTZdenek Pospisil, Jitka Kuhnova 86

OPTIONS FOR THE DESIGN OF A SOIL MONITORING SCHEMEJacqueline Potts 86

DO OZONE CONCENTRATIONS AFFECT RESPIRATORY HEALTH IN SCOTLAND?Helen Powell, Duncan Lee, Adrian Bowman 87

OPTIMAL CONSTRAINED CONFIDENCE ESTIMATION VIA TAIL FUNCTIONSWITH APPLICATIONS TO ENVIRONMENTAL DATABorek Puza, Terence O'Neill 87

SUSTAINABLE ECONOMICS: THE CONTRIBUTION OF OFFICIAL STATISTICSWalter J. Radermacher 88

TREND TESTING: SEARCHING FOR A BETTER ESTIMATION OF THE VARIANCEOF THE TEST STATISTICM. do Rosário Ramos 88

TRI-PARAMETER LOGNORMAL DIFFUSION PROCESS WITH EXOGENOUSFACTORS: INFERENCE BASED IN SIMULATED DATAEva Mª Ramos-Ábalos, Ramón Gutiérrez-Jaimez, Ramón Gutiérrez-Sánchez, Ahmed Na�di 89

LIKELIHOOD-BASED INFERENCE FOR MAX-STABLE PROCESSESMathieu Ribatet, Simone, A. Padoan, Scott, A. Sisson 89

GENERALIZED PROBABILITY WEIGHTED MOMENTS METHODS IN EXTREMEVALUE THEORYPierre Ribereau, Armelle Guillou, Philippe Naveau 89

GOMPERTZ-LOGNORMAL DIFFUSION PROCESS FOR MODELLING THEACCUMULATED NUTRIENTS DISSOLVED IN A CULTIVATION OF CAPISCUMANNUUMNuria Rico, Desiree Romero, Francisco Torres, Patricia Román 90

12

DOES IGNORING MODEL SELECTION EFFECTS IN ASSESSING THE EFFECT OFPM ON MORTALITY MAKE US TOO VIGILANT?Steven Roberts, Michael Martin 90

A SPATIO-TEMPORALMODEL FOR POINT-SOURCEMODELLING IN CRIMINOLOGY:A CCTV APPLICATIONAlexandre Rodrigues, Peter Diggle 91

MULTILEVEL ZERO-INFLATED REGRESSION FOR MODELLING SPECIESABUNDANCE IN RELATION TO HABITAT: A BAYESIAN APPROACHMarco A. Rodríguez, Clarice G.B. Demétrio, Silvio S. Zocchi, Roseli A. Leandro, Julie Deschênes 91

BAYESIAN ESTIMATION OF THE CONDITIONAL INTENSITY FUNCTION INSELF-CORRECTING POINT PROCESSES APPLIED TO THE SEISMIC ACTIVITYOF ITALIAN TECTONIC REGIONSRenata Rotondi, Elisa Varini 91

ON THE CONCEPT OF STRUCTURAL COMPONENTS WITH AN APPLICATIONTO WEATHER FUNCTIONAL DATAValentin Rousson, Juhyun Park, Theo Gasser 92

SPATIAL ANALYSIS OF REGIONAL CLIMATE MODEL ENSEMBLESSteve Sain 92

THE INLA METHOD FOR MULTIVARIATE COUNTS DATAMichael Salter-Townshend, John Haslett 93

SPATIAL REGRESSION MODELING WITH NONSTATIONARY SPATIALCOVARIANCE STRUCTURE FOR AIR QUALITY EXPOSURE FROM COMPLEXSPATIO-TEMPORAL MONITORING AND GIS-BASED COVARIATESPaul D Sampson, Adam Szpiro, Lianne Sheppard, Johan Lindstrom 93

CHANGING OF BIODEGRADABLE ORGANIC CARBON IN SOME PROCESSES OFDRINKING WATER PREPARATIONOlena Samsoni-Todorova, Natalia Klymenko, Ivan Kozyatnyk 94

SPECTRAL ANALYSIS OF IRREGULARLY SPACED PALEOCLIMATIC TIMESERIES USING EMPIRICAL MODE DECOMPOSITION AND WAVELET LIFTINGJean Sanderson, Michel Cruci�x, Piotr Fryzlewicz, Jonathan Rougier 94

SPATIO-TEMPORAL MODELS FOR OCEANIC VARIABLESBruno Sanso, Ricardo Lemos 95

SOLAR PHOTOVOLTAIC IN ITALY: A STATISTICAL SPATIAL ANALYSIS FORACCURATE ENERGY PLANNINGAnnalina Sarra, Gianfranco Iurisci 95

RISK STANDARDISATION AND POISSON REGRESSION IN ENVIRONMENTALCANCER INCIDENCE STUDIESErik-A. Sauleau, Silvia Columbu, Monica Musio 96

CORRECTION OF EDGE EFFECT IN SPATIAL GENERALIZED ADDITIVE MIXEDMODELSErik-A. Sauleau, Agnes Fromont, Laurence Clerc, Audrey Bellisario, Claire Bonithon-Kopp ,Thibault Moreau, Christine Binquet 96

GAUGE IMPRECISION EFFECT ON ENVIRONMENTALMONITORING ALGORITHMSMichele Scagliarini, Daniela Cocchi 97

13

RADIATION-INDUCED GENETIC EFFECTS AND ECOLOGICAL DOSE-RESPONSEANALYSESHagen Scherb, Kristina Voigt 97

LOCAL APPROACHES FOR INTERPOLATING AIR POLLUTION PROCESSESWolfgang Schmid, Olha Bodnar 98

MODELLING MULTIPLE SERIES OF RUNOFF: THE CASE OF RIO GRANDE BASINAlexandra Schmidt, Romy Ravines, Helio Migon 98

SEPARABLE CONDITIONAL INTENSITY ESTIMATES FOR SPACE-TIME POINTPROCESSES WITH APPLICATION TO LOS ANGELES COUNTY WILDFIRESFrederic Schoenberg 98

SIMPLE METRICS, COMPLEX ENVIRONMENTAL SYSTEMSMarian Scott, Duncan Lee, Claire Ferguson, Ron Smith 99

BAYESIAN LATENT VARIABLE MODELLING IN STUDIES OF AIR POLLUTIONAND HEALTHGavin Shaddick, Duncan Lee, Ruth Salway, Stephen Walker 99

UNCERTAINTY ANALYSISWITHIN THE HEIMTSA (HEALTHAND ENVIRONMENTINTEGRATED METHODOLOGY AND TOOLBOX FOR SCENARIO ASSESSMENT)PROJECTGavin Shaddick, Marta Blangiardo, Ruth Salway, Alex Zenie, Bruce Denby, Edzer Pebesma 100

LINKING STATISTICS TO ECOSYSTEM SERVICES FRAMEWORKSRon Smith, Jan Dick 100

137CS, 40K, 238PU, 239+240PU AND 90SR IN ENVIRONMENTAL MATERIALORIGINATIG FROMKING GEORGE ISLAND (SOUTH SHETLANDS, ANTARCTICA)Katarzyna Sobiech-Matura, Maria A. Olech, Jerzy W. Mietelski, Anna Masniak 101

PARTIAL LEAST SQUARE (PLS) MODEL IN AIR QUALITY MODELLINGTarana A. Solaiman 101

CHEMOMETRICS AND ENVIRONMETRICS: TALL SHOULDERS, ACCOMPLISHMENTS,AND FUTURE DIRECTIONSCli�ord H. Spiegelman, Abdel El-Shaarawi 102

USING IMPUTATION TO ESTIMATE TREND AND ABUNDANCE IN COHOSALMON NUMBERS USING A MULTI-PERIOD ROTATING PANEL SAMPLINGDESIGNDon Stevens 102

MULTIVARIATE NON-PARAMETRIC REGRESSION VIA GAUSSIAN MARKOVRANDOM FIELDSJames Sweeney, John Haslett 102

ACCOUNTING FOR EXPOSURE MEASUREMENT ERROR IN ENVIRONMENTALEPIDEMIOLOGYAdam Szpiro, Lianne Sheppard, Thomas Lumley 103

ROBUST QUANTILE REGRESSION ESTIMATION FOR SPATIAL PROCESSESBaba Thiam, Dabo-Niang Sophie 103

PREDICTIVE UNCERTAINTY IN HYDROLOGICAL FORECASTINGEzio Todini 104

14

PREDICTING TREE LEVEL VARIABLES USING AIR-BORNE LIDAR DATA ANDFIELD OBSERVATIONSErkki Tomppo, Mari Myllymäki, Antti Penttinen 104

CLUSTERING OF MONITORING STATIONS IN VENICE LAGOONStefano Tonellato, Stefano Ciavatta , Andrea Pastore, Roberto Pastres 105

STATE-SPACE MODELS IN EXTREME VALUE THEORYGwladys Toulemonde, Armelle Guillou, Philippe Naveau, Mathieu Vrac, Frédéric Chevallier 105

A SIMPLE FEEDFORWARD NEURAL NETWORK FOR THE PM10FORECASTING:COMPARISON WITH A RADIAL BASIS FUNCTIONNETWORK AND AMULTIVARIATE LINEAR REGRESSION MODELLivia Trizio, Maurizio Caselli, Gianluigi de Gennaro, Pierina Ielpo 106

A GEOGRAPHICALLY WEIGHTED REGRESSION-BASED GENERALIZEDREGRESSION ESTIMATOR FOR SPATIAL DATAAlessandro Vagheggini, Francesca Bruno 106

MULTIPLE TESTING ON STANDARDIZED MORTALITY RATIOS: A BAYESIANHIERARCHICAL MODEL FOR FALSE DISCOVERY RATE ESTIMATIONMassimo Ventrucci 107

DATA-DRIVEN NEIGHBORHOOD SELECTION OF A GAUSSIAN FIELDNicolas Verzelen 107

ANALYSIS OF FUZZY ENVIRONMENTAL DATAReinhard VIERTL 107

DENDROCLIMATOLOGICAL RECONSTRUCTIONS OF PAST CLIMATES: IS THEPRESENT THE KEY TO THE PAST?Hans Visser 108

STEERING FACTORS AND THE ECOLOGICAL QUALITY OF REGIONAL SURFACEWATERS:A SUCCESSFUL APPLICATION OF REGRESSION-TREE ANALYSISHans Visser 108

THE INTERPLAY OF ENVIRONMETRICS WITH OTHER ENVIRONMENTALLY-ORIENTED SOCIETIESKristina Voigt, Scherb Hagen 109

MIXED-EFFECTS MODEL SELECTIONRonghui Xu, Michael Donohue, Florin Vaida, Rosanna Haut 109

ANALYSIS OF CLUSTERED ENVIRONMENTAL MULTINOMIAL DATA WITHRANDOM CLUSTER SIZESGuohua Yan, Renjun Ma 110

SURFACE WATER QUALITY ASSESSMENT USING STATISTICAL TECHNIQUESSuheyla Yerel, Huseyin Ankara 110

ON THE PARTITIONING UNCERTAINTY IN GLOBAL CLIMATE PREDICTIONSStan Yip 110

DEALING WITH LARGE COVARIANCE MATRICES FOR SPATIAL DATAHao Zhang 111

15

MODELLING BACTERIAL DENSITY COUNTDATAWITHVARIOUS OVERDISPERSIONAND TAIL HEAVINESSRong Zhu, Abdel H. El-Shaarawi, Harry Joe 111

APPLICATION OF ROBUST PRINCIPAL COMPONENT ANALYSIS FORASSESSMENT OF CS-137 TRANSPORT IN FOREST SOILZbigniew Ziembik, Agnieszka Dolhaczuk-Srodka, Maria Waclawek 111

TESTING ISOTROPY OF SPACE COVARIANCE FUNCTIONSAlessandro Zini 112

Bologna, Italy TIES 2009

Oral Presentation

SEMI-PARAMETRIC ESTIMATION OF THE INTENSITY FUNCTION IN SPACE-TIMEPOINT PROCESSES.

Giada Adel�o1, Marcello Chiodi1

1University of Palermo

Point processes are well studied objects in probability theory and a powerful tool in statistics for modelingand analyzing the distribution of real phenomena. Point processes can be speci�ed mathematically in severalways, e.g. by considering the joint distributions of the counts of points in arbitrary sets or de�ning acomplete intensity function, that is a function of the points history that generalizes the rate function of aPoisson process. Here some techniques for estimating the intensity function of space-time point processesare developed, by following semi-parametric approaches.Parametric estimation has some drawbacks, mostlyrelated to the de�nition of a reliable mathematical model from the speci�c-�eld theory.Disadvantages ofparametric modeling can be avoided by making use of a �exible procedure, i.e. nonparametric, based onkernel density methods.Kernel estimators are here used to describe space-time variations of the seismicactivity of an observed area. For this issue, the choice of the windows width is often crucial; therefore we usean estimation procedure choosing the bandwidth in order to have good predictive properties of the estimatedintensity. Since a direct ML approach is not feasible in this context, an estimation procedure based on thesubsequent increments of likelihood is introduced. Similarly to CV, we consider additive contributions givenby the log-likelihood of the (m+1)th observed point given the nonparametric estimation based on the previousobservations, providing a measure of the ability of the previous m observations to give information to future.keywords: Intensity function, Forward estimate, Kernel estimatorsGiada Adel�o, Dipartimento Scienze Statistiche e Matematiche "S. Vianelli", viale delle Scienze, ed. 13. 90128PalermoE-mail address:adel�o@dssm.unipa.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

SPACE TIME MODELING OF PRECIPITATION USING HIDDEN MARKOV MODELS.

Pierre Ailliot1, Craig Thompson2, Peter Thomson3

1Université de Brest, France, 2National Institute of Water and Atmospheric Research, New Zealand,3Statistics Research Associates Ltd, New Zealand

In this talk, I will introduce a space-time model for daily precipitation over mesoscale spatial areas. Suchmodels have important applications. For example, they can be used as stochastic rainfall generators toprovide realistic inputs to �ooding, runo� and crop growth models, and also as components within generalcirculation models. A variety of stochastic models have been proposed in the literature, and among them theweather type models play an important role. The basic idea of these models consists of introducing an extravariable to describe the meteorological regime (weather type), and assume that this variable explains mostof the space-time structure of the data.In the model proposed in this talk, the weather type is introducedas a hidden Markov chain, and precipitation within weather types are described using censored power-transformed Gaussian distributions. The latter provide �exible and interpretable multivariate models for themixed discrete-continuous variables that describe both precipitation, when it occurs, and no precipitation.Finally, the proposed model is a hidden Markov model, in which the hidden process has a discrete component(the weather type) and a continuous component (due to the censoring).The parameters will be estimatedusing a Monte-Carlo EM algorithm and the �tted model will then be validated on rainfall data from a smallnetwork of stations in New Zealand encompassing a diverse range of orographic e�ects. We will show that itprovides a better description of the spatial structure of precipitation than a more conventional HMM.keywords: Space-time model, Precipitation, Hidden Markov model, Censored Gaussian distribution, Monte CarloEM algorithm.Pierre Ailliot, 6, Avenue Victor Le Gorgeu B.P. 809 29285 BREST Cedex, FranceE-mail address:pierre.ailliot@univ-brest.fr

16

Bologna, Italy TIES 2009

Oral Presentation

CORRELATION BETWEEN PRECIPITATION AT TWO STATIONS IN TURKEY ANDNAO/MO INDICES.

Erdem Albek1

1Anadolu University

This study investigates the relationship between precipitation recorded at two neighbour meteorologicalstations in northwestern Inner Anatolia, Turkey and the NAO (North Atlantic Oscillation) and MO (Mediter-ranean Oscillation) indices. The time series regarding precipitation and air temperature are on a monthlybasis and span a period of twenty-�ve years beginning in 1975. The NAO and MO indices reduce the quitecomplex spatial and temporal variability of climate over the Atlantic Ocean and Mediterranean Sea whichgoverns the large scale precipitation regimes over Europe into simpler measures. The series are computedfrom normalized sea level pressure anomalies between pressure centers. Precipitation and temperature at thestations are found to be negatively correlated with both indices while the association with NAO is strongerthan with MO. Semi-partial correlations computed while keeping the e�ects of temperature on precipitationconstant are found to be higher than the zero order correlations. Seasonal groupings based on precipitationvariations, and partitioning of data with respect to positive and negative indices and breakpoints in the tem-perature series are also investigated. Results have shown that precipitation at these two stations is a�ectedto di�erent degrees by the indices which points towards di�erent mechanisms of precipitation generation. Ithas been found that the relationships have a seasonal dependence. Di�erencing between consequent recordshas shown larger correlations than absolute records.keywords: Precipitation, Partial Correlation, Climate Change, NAO Index, MO IndexErdem Albek, Anadolu University Environmental Engineering Department Iki Eylul Campus 26555 Eskisehir, TurkeyE-mail address:ealbek@anadolu.edu.tr

Supporting grant: Anadolu University, Turkey

Poster Presentation

INVESTIGATION OF THE INTERACTIONS AMONG WATER QUALITY VARIABLESIN A STREAM WITH PARTIAL CORRELATIONS.

Mine Albek1, Erdem Albek1

1Anadolu University

This study investigates the correlations between water quality variables and stream�ow recorded at suc-cessive monitoring stations on a stream and meteorological conditions prevailing on the streams watershed.The stream in question is the Porsuk Stream located in northwestern Inner Anatolia, Turkey and is heavilypolluted by both point and di�use sources. The water quality variables are water temperature, dissolvedoxygen concentration and Biochemical Oxygen Demand (BOD). The meteorological variables are precipita-tion and air temperature. Partial correlations are investigated besides zero-order correlations to adjust forthe e�ects of exogenous variables. With this procedure, it is aimed to determine the e�ects of water temper-ature, air temperature, stream�ow and temperature on the dissolved oxygen concentration and extract thedependence of the biochemical oxygen demand on this water quality variable. Moreover, it is also aimed toobserve the pro�le of this correlation and dependence along the stream whose natural �ow is interrupted by amajor reservoir and a regulator. The equalization e�ect of the reservoir on the respective concentrations andthe in�uences of the point and di�use sources are investigated. Results have shown signi�cant correlationsamong the majority of the variables which explain a large portion of variability which the dissolved oxygenconcentration exhibits. When the e�ects of the intervening variables like temperature and �ow are removed,the variation explained by the biochemical oxygen demand is reduced. The �ndings can be utilized in thepreparation and implementation of control measures to increase the quality of the stream.keywords: Water Quality, Partial Correlation, Stream pollutionMine Albek, Anadolu University Environmental Engineering Department Iki Eylul Campus 26555 Eskisehir, TurkeyE-mail address:mlule@anadolu.edu.tr

Supporting grant: Anadolu University, Turkey

17

Bologna, Italy TIES 2009

Oral Presentation

MODELLING THE ONTARIO FIRE WEATHER INDEX.

Alisha Albert-Green1, W. John Braun1, David L. Martell2, Douglas G. Woolford3

1University of Western Ontario, 2University of Toronto, 3Simon Fraser University

In Canada, the Fire Weather Index (FWI) represents the potential frontal intensity of a forest �re basedon relative risk of spread and vegetation available for combustion. This paper describes and assesses severalmodels for day to day changes in FWI using 42 years of observations from a sample of �re-weather stationsin Ontario, Canada. Our models are motivated by the simple Markov chain models proposed and �t byMartell (International Journal of Wildland Fire, 1999), which we describe and evaluate. We discuss non-negative time series models with atoms at zero, present possible methods for modelling transitions betweendocumented mesoscale weather patterns, and present the following three models for phase-dependent non-nilFWI behaviour: exponentially distributed increments, an autoregressive moving average (ARMA) process,and an exponential bilinear process. We evaluate these candidates, conclude that the ARMA process maybe most appropriate at this stage, and test for intra and inter annual trends using parametric bootstrapmethods. No seasonality was observed for any of the ARMA parameters for modelling non-nil FWI, someseasonality for switching between nil and non-nil states was noted among stations in the northwest, and themajority of the stations do not appear to have any signi�cant annual trend.keywords: non-negative time series, parametric bootstrap hypothesis testingAlisha Albert-Green, 93 Castle�eld Ave.Toronto, OntarioM4R 1G5E-mail address:aalbertg@uwo.ca

Supporting grant: Natural Sciences and Engineering Research Council

Oral Presentation

BAYESIAN MODELLING VOLATILITY OF GROWTH RATE IN ATMOSPHERIC CAR-BON DIOXIDE CONCENTRATIONS.

Esmail Amiri1

1Department of Statistics IKIU

Atmospheric gases, such as carbon dioxide, ozone, methane, nitrous oxide,and etc., create a natural green-house e�ect and cause climate change. Therefore,modelling behavior of these gases could help policy makersto control greenhousee�ects.In a Bayesian frame work, we analyse and model volatility of growth rate inat-mospheric carbon dioxide concentrations using monthly data from January 1965 toDecember 2002. The dataare a subset of the well known Mauna Loa atmosphere carbondioxide record obtained through the CarbonDioxide Information Analysis Center. Theconditional variance of ACDC monthly growth rate is modelledusing theautoregressive conditional heteroscedasticity (ARCH), generalized ARCH model andtwo variants ofstochastic volatility(sv) models. The latter models are shown tobe able to capture the dynamics in the con-ditional variance in ACDC growth rate andto improve the out-of-sample forecast accuracy of ACDC growthrate.keywords: Stochastic volatility, Smooth transition autoregressive, Markov chain Monte Carlo methods, Bayesian,GARCHEsmail Amiri, Department of Statistics, Imam Khomeini International University, Ghazvin, IranE-mail address:e_amiri@yahoo.com

Oral Presentation

SYNERGISTIC USE OF SEVERAL RECEPTOR MODELS (CMB, APCS AND PMF) TOINTERPRET AIR QUALITY DATA.

Eleonora Andriani1, Maurizio Caselli1, Gianluigi de Gennaro1

1Dipartimento di Chimica, Università degli Studi di Bari

Receptor models are useful to provide a correct daily source apportionment for developing e�ective controlstrategies. These models can be categorized into di�erent types based on whether PM chemical characteristicsfrom emission sources are required to be known before the source apportionment. On one hand, ChemicalMass Balance (CMB) requires 'a priori' knowledge of major sources and their emission pro�les in the areaunder investigation; on the other hand, model as Absolute Principal Component Scores (APCS) and PositiveMatrix Factorization (PMF) require ambient measurement data only to perform source apportionment. CMB

18

Bologna, Italy TIES 2009

consists of a least squares solution to a set of linear equations that express each chemical concentration asa linear sum of products of source pro�les and source contributions. The model needs as input data speciesconcentrations at receptor site and source pro�les for sources taken into consideration; statistical parameterscontrol the performance of CMB run.APCS model decomposes data matrix into the product of two matricescontaining source's patterns and contributions; these latent matrices are then rescaled using linear regressionto produce real contributions and patterns having physical meaning. PMF, leaving from random positivevalues for contributes matrix (F) and pro�les one (A), minimizes the di�erence between data matrix and theproduct of FxA by an iterative process. The goodness of the apportion procedures is both for APCS andPMF the best reconstruction of the data matrix.The aim of this work is to apply models above mentionedto PM samples and to compare obtained resultskeywords: PM, Receptor model, Source apportionmentEleonora Andriani, Università degli Studi di Bari, Dipartimento di Chimica. Laboratorio 20 Via Orabona, 470126BariE-mail address:eleonora.andriani@chimica.uniba.it

Oral Presentation

WAVELET-BASED MULTISCALE INTERMITTENCY ANALYSIS IN ENVIRONMENTALAPPLICATIONS.

José M. Angulo1, Ana E. Madrid1

1Department of Statistics and O.R., University of Granada, Spain

Intermittency, generally understood as pseudo-periodic occurrence of high level or variation episodes withina certain regular behaviour, is considered a phenomenon of interest in very diverse �elds of applications(e.g. seismology, turbulence, hydrology, astronomy, �nance, insurance, epidemiology, etc.), related to whicha variety of speci�c formalizations and measures have been derived. Structural characteristics associatedwith such e�ect, in relation to the underlying generating process, often constitute a primary objective inenvironmental studies, as they provide relevant information for detection and prediction of critical events andfor risk assessment. Essential aspects intrinsic to the concept of intermittency such as 'scales of variation'and 'localization', and related interactive dynamics, have led to the use of wavelets as a suitable functionaltool for technical analysis. In this paper, di�erent wavelet-based approaches proposed in this context arereviewed. The study is then focused on the analysis of signi�cant aspects related to random �eld deformations,meaningful both from physical considerations and for methodological purposes, and structural implicationsof such transformations on intermittency, as well as concerning wavelet-related methods.keywords: deformation, intermittency, multiscale analysis, risk assessment, waveletsJosé M. Angulo, Universidad de Granada, Departamento de Estadística e Investigación Operativa, Facultad de Cien-cias, Campus Fuente Nueva s/n, E-18071 Granada, SpainE-mail address:jmangulo@ugr.es

Supporting grant: Work partially supported by project P08-FQM-03834 of the Andalusian CICE, and project MTM2008-03903of the DGPTC, Ministerio de Ciencia e Innovación, Spain.

Oral Presentation

ASSESSMENT OF SURFACEWATER QUALITY BYMULTIVARIATE STATISTICAL ANAL-YSIS TECHNIQUES.

Huseyin Ankara1, Suheyla Yerel2

1Eskisehir Osmangazi University, 2Bilecik University

In this study, the surface water quality of Porsuk River in Turkey is evaluated by using the multivariatestatistical methods including factor analysis, cluster analysis and discriminant analysis. When factor analysesas applied to the surface water quality data obtain from the nine di�erent observation stations and threefactors were determined of the surface water quality. Cluster analysis and discriminant analysis grouped nineobservation stations into two grouped based on the similarity of surface water quality characteristics. Basedon the locations of the stations consist by each group and variable concentrations at these stations, it wasconcluded that urban, industrial and agricultural discharge strongly a�ected east part of the region. Thus,

19

Bologna, Italy TIES 2009

this study show that the multivariate statistical methods are useful for interpreting complex datasets in theanalysis of spatial variations in surface water quality and the optimization of river water quality monitoringnetwork.keywords: Factor analysis, cluster analysis, discriminant analysis, water qualityHuseyin Ankara, Eskisehir Osmangazi University, Engineering and Architecture Faculty, Eskisehir, Turkey.E-mail address:hankara@ogu.edu.tr

Oral Presentation

SMOOTHED ANOVA WITH SPATIAL EFFECTS AS A COMPETITOR TO MCAR INMULTIVARIATE SPATIAL SMOOTHING.

Sudipto Banerjee1, Yufen Zhang2, James Hodges1

1University of Minnesota, 2Novartis, NJ.

Rapid developments in geographical information systems (GIS) and advanced spatial statistics continue togenerate interest in analyzingcomplex spatial datasets. One area of activity is in creating smoothed diseasemaps to describe the geographic variation of disease and generate hypotheses for apparent di�erences inrisk. With multiple diseases, a multivariate conditionally autoregressive (MCAR) model is often used tosmooth across space while accounting for associations between the diseases. The MCAR, however, imposescomplex covariance structures that are di�cult to interpret and estimate. This article develops a muchsimpler alternative approach building upon the techniques of smoothed ANOVA (SANOVA). Instead ofsimply shrinkinge�ects without any structure, here we use SANOVA to smooth spatial random e�ects bytaking advantage of the spatial structure. This paperextends SANOVA to cases in which one factor is a spatiallattice, which is smoothed using a CAR model, and a second factor is, for example,type of cancer. Datasetsroutinely lack enough information to identify the additional structure of MCAR. SANOVA o�ers a simplerand more intelligible structure than the MCAR while performing as well. We demonstrate our approachwith simulation studies designed to compareSANOVA with di�erent design matrices versus MCAR withdi�erent priors. Subsequently a cancer-surveillance dataset, describing incidence of 3cancers in Minnesota's87 counties, is analyzed using both approaches, showing the competitiveness of the SANOVA approach.keywords: Analysis of variance, Bayesian inference, conditionally autoregressive model, hierarchical model, smooth-ingSudipto Banerjee, 420 Delaware Street SE.A460 Mayo Bldg. MMC303. Division of Biostatistics, Minneapolis, MN55455. USA.E-mail address:sudiptob@biostat.umn.edu

Supporting grant: NIH grant 1�R01�CA95995

Oral Presentation

BAYESIAN DETECTION OF INHOMOGENEITIES IN PRECIPITATION SERIES.

Claudie Beaulieu1, Taha B.M.J. Ouarda2, Ousmane Seidou3

1Princeton University, 2INRS-ETE, University of Quebec, 3University of Ottawa

Hydroclimatic data records often undergo arti�cial disturbances that do not re�ect the real climate vari-ations. These changes can be due to station relocation, instrument replacement, change of observer ormodi�cation in the immediate environment of the site. Such arti�cial changes are called inhomogeneities.If these inhomogeneities are not retrieved from the data, there is a risk to interpret them as regional cli-mate change. A Bayesian approach for the detection of inhomogeneities in precipitation series is presented.The technique allows the integration of di�erent sources of information (metadata, expert belief, regionalinformation) in the analysis. Furthermore, this approach allows detecting changes in precipitation series atdi�erent time scales. Its ability to discriminate homogeneous and inhomogeneous series was evaluated in asimulation study on synthetic series with similar statistical properties as observed total annual precipitationin the southern and central parts of the province of Quebec, Canada. Di�erent priors were used to investigatethe sensitivity of the test to the choice of priors. It was found that a high prior probability of no-change yieldslow false detection rates on the homogeneous series and that the test has a very high power of detection onseries with a single or multiple shifts. The test was also applied to analyze the homogeneity of precipitation

20

Bologna, Italy TIES 2009

totals in a real case study in the province of Quebec, Canada, at di�erent time scales. It was shown that thetest is able to detect inhomogeneities in the annual, seasonal and monthly series.keywords: Bayesian technique, changepoint detection, homogenization, precipitation seriesClaudie Beaulieu, Program in Atmospheric and Oceanic Sciences, Princeton University,404B Sayre Hall,300 ForrestalRoad, Princeton, NJ, 08540 USAE-mail address:beaulieu@princeton.edu

Oral Presentation

AMULTIVARIATE SPATIO-TEMPORAL DOWNSCALER FOROUTPUT FROMNUMER-ICAL MODELS.

Veronica Berrocal1, Alan Gelfand1

1Duke University

Numerical models are mathematical deterministic models used to predict spatio-temporal processes. Formodel evaluation and calibration, it is necessary to solve the spatial misalignment between the predictionsyielded by numerical models, given in terms of averages over grid cells, with observations collected at sites.Here, we propose a multivariate spatio-temporal model that allows to downscale to point level multipleoutputs from a numerical model, and addresses not only the spatial misalignment between the numericalmodel output and the observational data, but also the spatial and temporal misalignment between thedi�erent observational data.The model regresses the observational data on the numerical model outputsusing correlated Gaussian processes as coe�cients in the linear regression. We explore di�erent ways tocharacterize the correlation among the Gaussian processes. As an example, we apply our model to ozone andPM 2.5 concentration for the Eastern US.keywords: spatio-temporal misalignment, spatially varying coe�cient model, calibrationVeronica Berrocal, Department of Statistical Science Box 90251223A Old Chemistry Building Duke University Durham,NC 27708USE-mail address: vjb2@stat.duke.edu

Oral Presentation

AN ESTIMATE OF THE INDUSTRIAL METABOLISM OF THE PIEDMONT REGION(ITALY) USING THE ENVIRONMENTAL INPUT-OUTPUT ANALYSIS AND THE ECO-LOGICAL FOOTPRINT.

Pancrazio Bertaccini1, Marco Bagliani1

1IRES Piemonte

In this paper we propose the combined use of two di�erent methodologies: the Leontiev's input-outputformalism and the ecological footprint metric, in order to jointly analyse the metabolism of a territoryand to describe both the economic and the environmental aspects, with high coherence and accuracy. Theenvironmental extension of the input-output analysis is a well established formalism that links the �naldemand for goods and services in an economy to the related environmental impact, or, in other words, to the�ow of necessary natural resources. To measure the natural resources required by the economic metabolismof a region we use the ecological footprint. This accounting system quanti�es the use of renewable resourcesby estimating the amount of terrestrial and marine ecosystems that should be available to produce the relatedamount of natural goods. The joint use of input-output and ecological footprint allows the evaluation of thetotal requirement of ecological resources and the estimate of the environmental burden of the �nal demand ofgoods per industrial sector. We apply the combined analysis to the Piedmont region, starting from the italiansupply and use tables (SUT), and focus on the year 2001. We also propose a novel algorithm speci�callydesigned for and implemented in the R statistical software; the algorithm easily manages the large matricessuch as the supply and use tables, and provides results for di�erent types and numbers of impacts or categoriesof the �nal demand.keywords: input-output analysis, ecological footprint, Piedmont region, industrial metabolism, R statistical softwarePancrazio Bertaccini, c/o IRES Piemonte, Via Nizza, 18, 10125 Torino TO (Italy)E-mail address:bertaccini@ires.piemonte.it

Supporting grant: The work was partly funded by the Regione Piemonte CIPE project 2004

21

Bologna, Italy TIES 2009

Oral Presentation

AIR POLLUTION: METEOROLOGY OR TRAFFIC, WHAT DOES REALLY MATTER?.

Pancrazio Bertaccini1, Vanja Dukic2, Rosaria Ignaccolo1

1Dep. of Statistics and Applied Mathematics - D. De Castro (Università degli studi di Torino), 2Dep. ofHealth Studies University of Chicago

The variation of the air pollution concentration in urban area is arguably correlated with the meteorologicalconditions and the vehicular tra�c volumes. We model the behavior of some important pollutants, such asnitrogen oxides (NO, NO2 and NOx), carbon oxide (CO) and particulate matter (PM), on the Turin urbanarea considering the spatial variability of the pollutant concentration and the related covariates, such asvehicular tra�c and meteorology. This work is aimed to evaluate the role of the tra�c and the meteorologicalvariables highlighting how it changes with the di�erent seasons. To take into account the di�erences amongthe monitoring sites we set our model in the framework of the mixed-e�ect models. Hence we considera generalized additive mixed model (GAMM), that allows to analyze the relationships between pollutant,tra�c and meteorological covariates, focusing on tra�c along with the meteorological predictors during theperiod from January 2004 to April 2005. The tra�c volumes are measured by a loop induction sensornetwork, whereas the meteorological data are provided by a computer model chain and the pollution data areobserved hourly or daily. Applying sensitivity analysis techniques we can investigate the relative importanceof tra�c and meteorology, and then give support to policy makers in understanding when a tra�c regulationis necessary.keywords: generalized additive mixed models (GAMM), atmospheric pollution, vehicular tra�c, Turin urban area,meteorologyPancrazio Bertaccini, Corso Unione Sovietica 218/bis - 10134 TorinoE-mail address:bertaccini@econ.unito.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time", by Regione Piemonte CIPE project 2004 and ICER

Oral Presentation

A BAYESIAN MODEL OF TIME ACTIVITY DATA FOR ECOLOGICAL STUDIES WITHIMPLICATIONS ON THE BIAS OF DISEASE RISKS.

Marta Blangiardo1, Sylvia Richardson1

1Department of Epidemiology and Public Health, Imperial College

There now remains little doubt about the negative e�ect of short term air pollution on the health ofindividuals; typically an ecological study considers measures of ambient concentration (either point or areasources) as an approximation of the personal exposure. This approach relies on the unrealistic assumption thatpeople spend all their time outside, leading to a bias in the exposure. Recently the use of time activity data hasbeen introduced as a way to estimate the exposure based on the time people spend in di�erent environments,characterised by a di�erent degree of concentration for pollutants.All the approaches proposed so far usesimulators to estimate the personal exposure for a particular individual, through the information obtainedfrom diaries of people's activities. In this work we propose an alternative to the simulative approach, with ahierarchical Bayesian model that uses time activity as a source of information to estimate the distribution ofthe exposure in an ecological perspective. We investigated the bias occurring when the ambient concentrationwas used instead of the personal exposure in a simulation study and applied the model to a case study usingthe US Consolidated Human Activities database (CHAD) [Stallings et al.(2000), EPA National ExposureResearch Laboratory] and air pollution and mortality data from the US National Morbidity, Mortality, andAir Pollution Study (NMMAPS)[Samet et al., Res Rep Health E� Inst (2000): 5-70].keywords: Personal exposure, Ecological model, Bayesian model, Ambient concentrationMarta Blangiardo, EPH, Imperial College, St Mary's hospital, Norfolk Place, Paddington, London W2 1PGE-mail address:m.blangiardo@imperial.ac.uk

22

Bologna, Italy TIES 2009

Oral Presentation

ANALYSIS OF EXTREME EVENTS IN SARDINIA (ITALY) VIA A HIDDEN MARKOVMODEL.

Antonella Bodini1, Bruno Betrò 1, Q. Antonio Cossu2

1Institute of Applied Mathematics and Information Technology, 2Sardinian Environmental ProtectionAgency

In Betrò et al. [Environmetrics (2008): 702-713)], the capability of a hidden Markov model in identifyingpossible recurrent patterns in the occurrence of extreme events has been established by analyzing data from4 stations in a small area of Central-East Sardinia (Italy). In particular, mixture of Weibull distributionshave been used to model (positive) rainfall amounts: these distributions proved to provide an adequaterepresentation of extreme events. In this work, estimation techniques have been improved, and the analysishas been extend to a set of stations located in all the island and characterized by frequently occurring extremeevents.keywords: Extreme events, hidden Markov model, Weibull distributionsAntonella Bodini, via E. Bassini 15, 20133 Milano (Italy)E-mail address:antonella.bodini@mi.imati.cnr.it

Oral Presentation

NON-STATIONARY SPATIAL ARMA MODELS APPLIED TO GLOBAL OZONE DATA.

David Bolin1, Finn Lindgren1

1Lund University

Building spatial models for environmental data is a challenging problem. The data sets are often large,which makes standard covariance models computationally infeasible, and non-stationary covariance structuresare often needed to capture the global behavior of the data. It has previously been shown that computationallye�cient Markov models with approximate Matérn covariance functions can be obtained by solving a particularAR-like stochastic di�erential equation using a �nite element method. In this work, we extend that idea byconsidering a class of spatial models obtained as solutions to more general ARMA-like stochastic di�erentialequations. These models will not have Markov structures, but will have the same computational bene�tsand contain a large �exible class of covariance functions that can easily be made non-stationary. Resultsare illustrated with a large data set of spatially irregular global Total Column Ozone (TCO) data, and themodels are compared to some other non-stationary covariance models previously suggested for TCO data.David Bolin, Mathematical StatisticsCentre for Mathematical Sciences Box 118, SE-221 00 Lund, SwedenE-mail address:bolin@maths.lth.se

Poster Presentation

AIR POLLUTION INDICES: COMPARISONS OF THEIR UNCERTAINTY.

Francesca Bruno1, Daniela Cocchi1

1Department of Statistics "P.Fortunati", University of Bologna

We propose air quality indices uncertainty as a tool for comparing air quality situations. Usually, syn-thetic environmental indices do not deal uncertainty explicitly, but they ought to be considered as randomquantities since they are functions of components that should be seen as random. Moreover, their complexstructure is better represented by means of a probabilistic model rather than by a descriptive function. Thisinformation can be recovered from synthetic indices by studying their distributions. Focusing on indices thatcapture the less favorable pollution situations, the Generalized Extreme Value (GEV) distribution is a suit-able probabilistic model. In this work, we explore the construction of con�dence bands of GEV cumulativedensity functions (CDFs) and its contribution for comparing di�erent pollution situations. When we comparecouples of di�erent pollution situations, three di�erent cases are possible: the �rst shows the situation whereCDFs can be de�nitely considered as di�erent (one of the CDFs can always be considered lower than theother); the second case concerns percentiles that are di�erent until a benchmark value, while above it CDFsare not signi�cantly di�erent. In this case we are able to identify the benchmark value below which a location

23

Bologna, Italy TIES 2009

shows a maximum pollution situation that is less serious than its competitor; above this benchmark valuethe pollution situations in the two locations are always separate. The last case describes situations thatnever can be distinguished; this clear �nding is hard to achieve by means of con�dence intervals on the mleparameter estimates.keywords: Air quality, GEV distribution, con�dence bands, uncertaintyFrancesca Bruno, via Belle Arti 41, 40126 BolognaE-mail address:francesca.bruno@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Poster Presentation

PARAMETRIC SPATIAL BOOTSTRAP AND MOVING BLOCK BOOTSTRAP: A COM-PARATIVE STUDY.

Francesca Bruno1, Rodolfo Rosa1, Luca Talenti1

1Department of Statistics "P.Fortunati", University of Bologna, Italy

The classic bootstrap uses valid resamples whenever the observation are independent and identically dis-tributed. In a spatial simulation study if the iid bootstrap is applied many information on spatial correlationcan be lost. In this work, we compare three di�erent spatial simulating procedures. The �rst is the para-metric spatial bootstrap (Solow [Journal of the International Association for Mathematical Geology (1985),769-775]) based on resamples by considering a Cholesky decomposition for isolate the spatial correlation ofthe process and then using bootstrap on decorrelated data. The second approach (Tang et al. [TechincalReport 337 (2006), 1-23]) combines spatial modeling and the parametric bootstrap in order to obtain validresamples of spatially correlated normal data. The moving block (adjusted for spatially correlated data) wasthe last method considered. The comparison between the three methods described is based on coverage ofcon�dence interval.Di�erences in grid size and variogram models are then compared.keywords: moving block bootstrap, spatial correlation, Parametric spatial bootstrapFrancesca Bruno, via Belle Arti 41, 40126 Bologna, ItalyE-mail address:francesca.bruno@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

A MIXED, DESIGN-BASED MODEL-BASED SAMPLING APPROACH FOR ESTIMAT-ING GLOBAL QUANTITIES IN SPACE-TIME.

Dick Brus1, Jaap de Gruijter2

1Wageningen University and research Centre, 2Retired

A major decision in designing sampling schemes for monitoring is the choice between a design-based and amodel-based sampling strategy (Brus and de Gruijter, Environmetrics 4 (1993): 123-152). In space-time thereare four combinations of the design-based and model-based approaches. In the fully design-based approachboth sampling locations and sampling times are selected by probability sampling, and the inference is basedon the spatial and temporal sampling designs. In this approach no model of variation is used, enhancingthe validity of the result, see Brus and Knotters [Water Resources Research 44 (2008)] for an example. TerBraak et al. (Journ Agric. Bolog. Env. Stat, 13 (2008): 159-176) used a fully model-based approach tocompare pattern types of observations in space-time for predicting the spatial mean temporal trend. In thispresentation we will launch a new mixed, design-based model-based sampling approach, in which samplinglocations are selected by probability sampling but sampling times are not. Contrary to the fully model-basedapproach, the stochastic space-time process is only partly described by a model of the temporal variation ofthe spatial means. In quantifying the uncertainty, the random selection of the sampling locations and the

24

Bologna, Italy TIES 2009

stochastic space-time process is accounted for. We will illustrate the new approach with optimization of thepattern type of observations in space-time and of the sample size for estimating the temporal tend of thespatial mean.keywords: monitoring, temporal trend, static pattern, serially alternating pattern, space-time modelDick Brus, P.O. Box 47, 6700 AA Wageningen, the NetherlandsE-mail address:dick.brus@wur.nl

Poster Presentation

MULTI-SCALE CLUSTERING OF SPATIO-TEMPORAL OBSERVATIONS.

María C. Bueso1, José M. Angulo2, María D. Ruiz-Medina 2

1Department of Applied Mathematics and Statistics, Technical University of Cartagena, Spain, 2Departmentof Statistics and O.R., University of Granada, Spain

Environmental applications often involve the collection of large amounts of spatio-temporal data sets. Highdimensionality and associated lack of e�ciency in the exploitation of sample information due to redundancycan be overcome by applying procedures for data selection which take into account the underlying dependencestructure, allowing for a minimum cost in terms of information loss. Depending on the case, this canbe achieved by prior or adaptive design of sampling strategies or by re-processing the available sampleinformation. When data are collected at speci�c sites through time, identi�cation of global similaritiesin relation to temporal evolution for each speci�c site provides useful information for data reduction andsubsequent inference.In this study we propose a clustering procedure to classify monitoring sites based onwavelet representations of the temporal observations obtained for each sampling location. Di�erences in thegrouping structures derived for each resolution level, as well as in terms of inter- and intra-cluster informationindicators, are analyzed in relation to multi-scale characteristics of the underlying process. Non-parametricestimation based on the empirical wavelet spectrum is performed to infer the dependence structure of thegenerating process under each scenario, particularly in relation to selection of resolution levels and intra-cluster simpli�cations.keywords: cluster analysis, dimension reduction, non-parametric estimation, spatio-temporal observation, wavelettransformMaría C. Bueso, Universidad Politécnica de Cartagena, Departamento de Matemática Aplicada y Estadística,CampusMuralla del Mar, Doctor Fleming s/n, 30202 Cartagena, Murcia,SpainE-mail address:mcarmen.bueso@upct.es

Supporting grant: Work partially supported by project P08-FQM-03834 of the Andalusian CICE, and project MTM2008-03903of the DGPTC, Ministerio de Ciencia e Innovación, Spain.

Poster Presentation

THE USE OF GEOADDITIVE MODELS TO ESTIMATE THE SPATIAL DISTRIBUTIONOF GRAIN WEIGHT IN AN AGRONOMIC FIELD: A COMPARISON WITH KRIGINGWITH EXTERNAL DRIFT.

Barbara Cafarelli1, Annamaria Castrignano'2, Antonio Troccoli3, Salvatore Colecchia3

1Università degli Studi di Foggia, 2CRA-SCA, 3CRA-CER

Precision Agriculture is an ecological management strategy based on the use of several sources of infor-mation to support decisions concerning the agricultural practice with the aim of optimizing the use of soiland water resources and chemical inputs on a site speci�c basis. In this way the adoption of soil manage-ment practices and natural resources conservation policies can take advantage of relevant spatial statisticalmethods, which can be helpful in a di�erential farm management calibrating di�erent actions according toagricultural practices and soil conditions. In geostatistics the use of kriging with external drift to buildmaps of grain weight in presence of other covariates is common. Nevertheless such a predictor requires thecovariates to have a linear e�ect and this assumption is seldom veri�ed and often violated. In this paper ageoadditive model is used to analyse the spatial distribution of a grain weight and the nonlinear relationswith some explanatory covariates. Such a model is the result of the sum of two components: an additivemodel expressing the nonlinear relation between response and covariates and a lme model which takes accountof the spatial correlation. Both components are given a low-rank formulation, so that the whole model isestimated by REML. The estimated model is used to obtain predictions for the plot under investigation andthese results are compared with those suggested by kriging with external drift. The accuracy and precision of

25

Bologna, Italy TIES 2009

grain feature estimates obtained from each of the estimation methodologies is evaluated by cross-validationtechniques.keywords: geoadditive models, kriging with externale drift, cross validation, precision agricoltureBarbara Cafarelli, Dipartimento di scienze economiche, matematiche e statistiche, Università di Foggia, LARGOPAPA GIOVANNI PAOLO II, 1, 71100 FOGGIA, ITALIAE-mail address:b.cafarelli@unifg.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

JOINT INCIDENCE OF VARIOUS DISEASES IN THE PRESENCE OF A RISK SOURCE.

Crescenza Calculli1, Alessio Pollice1

1Dipartimento di Scienze Statistiche ed Economiche "Carlo Cecchi", Università degli Studi di Bari

In Epidemiology the analysis of the spatial variation of the incidence of a disease in relation with a speci-�ed environmental risk source is a well known issue and methods based on spatial point patterns have longbeen used to deal with single pathologies. To outrun this limitation we propose a statistical model for jointlyanalysing spatial distributions of multiple pathologies at the same time. Our approach is based on a logitmodel considering an extension to more than two diseases where the Odds are decomposed additively into atrend and an error process. The e�ects of relevant covariates and of the point source on the trend is modelledby parametric and semi-parametric terms in the context of Generalized Additive Models (GAM). The overallerror process can be partitioned into random e�ects that capture several sources of variation including thevariability due to the point source e�ect (by an autoregressive point source model), the spatial variability (bya conditional autoregressive model) and a baseline residual component representing the null e�ect of sourceon the system. Inference on the previous model proceeds according to the Bayesian hierarchical approachimplemented via MCMC. An example with data from a spatial case-control study on the Brindisi environ-mental risk area is considered as an application of the proposed methodology. Points (cases and controls)corresponding to di�erent neoplasm typologies are geocoded and located within the area characterized bythe presence of a petrochemical plant.keywords: environmental epidemiology, point pattern analysis, bayesian hierarchical modelsCrescenza Calculli, Università degli Studi di Bari,via C. Rosalba 53, 70124 BariE-mail address:e.calculli@dss.uniba.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Poster Presentation

ENVIRONMENTAL RADIOACTIVITY MONITORING IN ARAGóN (SPAIN).

H.I. Calvete1, J.A. Carrión2, C. Galé1, E. García3, R. Núñez-Lagos2, C. Pérez2, J. Puimedón3, S.Rodríguez2, M.L. Sarsa3, J.A. Villar3

1Dpto. de Métodos Estadísticos. Universidad de Zaragoza, 2Lab. de Bajas Actividades. Universidad deZaragoza, 3Lab. de Física Nuclear y Astropartículas. Universidad de Zaragoza

Natural radioactivity is an inherent phenomenon to environment. There are no places without radioactivityand all surrounded materials are, to a greater o lesser extent, radioactive. The radioactivity level varies overtime and from a place to another. Land composition, height above sea level, meteorological condition, sunactivity, etc. are factors which a�ects radioactivity variability. In this respect, in order to detect abnormalradioactivity levels in a certain place, it is necessary to have a previous knowing of standard values. Thisongoing research aims to provide information on environmental radioactivity in Aragon, as the �rst stepin the process of establishing a monitoring system of it. Measurements will be carried out using TDL

26

Bologna, Italy TIES 2009

(termoluminiscent) dosimeters and obtaining complementary samples of land in selected points. Severaldosimeters will be placed in each point. The methodology used and preliminary results obtained will bepresented.keywords: Environmental radioactivityH.I. Calvete, Dpto. de Metodos Estadisticos.Facultad de Ciencias. Universidad de Zaragoza. Pedro Cerbuna 12.50009Zaragoza, Spain.E-mail address:herminia@unizar.es

Oral Presentation

COMPARING AIR QUALITY STATISTICAL MODELS.

Michela Cameletti1, Rosaria Ignaccolo2, Stefano Bande3

1University of Bergamo, 2University of Torino, 3University of Torino & Area Previsione e MonitoraggioAmbientale, ARPA Piemonte

Particulate matter is one of the more important air pollutants in Europe and despite the improvementsdue to the European Union legislation it still has a heavy toll on human life and health. From the statis-tical perspective, in the last years many works have been proposed for modelling PM concentrations andunderstanding the complex underlying spatio-temporal dynamics of the phenomenon. In particular, spatialprediction techniques are used in order to obtain concentration maps that are useful for evaluating the healthrisk and assessing the compliance with European and national directives also where no measurement sta-tions are displaced.In this work we consider some statistical models for spatial mapping of daily ParticulateMatter concentrations in the Piemonte region (Italy). The considered models belong to the hierarchicalspatio-temporal model class and combine observed and simulated data obtained by a deterministic atmo-spheric dispersion model. The aim of this work is to compare the considered models on the basis of theircomplexity and spatial prediction capability evaluated using some prediction error based indexes.keywords: hierarchical spatio-temporal models, spatio-temporal covariance function, Markov chain Monte CarloMethodsMichela Cameletti, Via dei Caniana, 2 25127 Bergamo (BG) ItalyE-mail address:michela.cameletti@unibg.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

EPIDEMIOLOGIC SURVEILLANCE AND IMPACT EVALUATION: THE FALSE DISCOV-ERY RATE.

Dolores Catelan1, Annibale Biggeri1, Corrado Lagazio2

1Univeristy of Florence, 2Univeristy of Udine

Population Health pro�ling is an important phase in Environmental Epidemiology investigations. It con-sists in identifying altered rate of a disease among many diseases or, for a given disease, of an area withina region. In the context of scienti�c reporting and Policy making control of False Discovery Rate (FDR) isappropriate. Two Bayesian approaches will be discussed. We developed a Bayesian approach to model under-lying risk pattern under the null and we used cross-validatory predictive distributions to generate model-basedp-values (Ohlssen JRSSA 2007). Alternatively we speci�ed a three level hierarchical Bayesian model and usethe posterior classi�cation probabilities as local FDR (Efron JASA 2001).We used a real examples and asimulation study. Our approach di�ers from thresholding q-values because it takes into account the wholeset of data. We compare the results with other approaches in the literature (posterior probabilities underdi�erent models, simple FDR procedures for Poisson data).keywords: Environmental Epidemiology, False Discovery Rate, Hierarchical Bayesian ModelsDolores Catelan, Dep. of Statistics "G. Parenti", viale Morgagni, 59 50134 Florence, ItalyE-mail address:catelan@ds.uni�.it

Supporting grant: PRIN 20072S2HT8

27

Bologna, Italy TIES 2009

Oral Presentation

MONITORING ATMOSPHERIC CHLOROFLUOROCARBONS BY THE LONGITUDINALBENT-CABLE MODEL.

Grace Chiu1, Shahedul Ahsan Khan1, Joel Dubin1

1University of Waterloo

The recent steady decline in atmospheric chloro�uorocarbon (CFC) concentrations could be a direct resultof the Montréal Protocol on Substances That Deplete the Ozone Layer, in e�ect since 1989. To study theextent of the decline, we apply the longitudinal bent-cable model to describe CFC concentrations observedover a global detection network. The bent cable is a parametric regression model to study data that exhibita trend change. It comprises two linear segments to describe the incoming and outgoing phases, joined bya quadratic bend to model the transition period. For longitudinal data, measurements taken over time arenested within observational units drawn from some population of interest. Here, it is useful to develop amixed-e�ects model extension of existing (frequentist) bent-cable methodology for a single time series. Wedo so in a hierarchical Bayesian framework, where each observational unit is associated with a random bentcable and within-unit serial correlation. We also discuss extensions of the model to account for changepointdata that may be characterized by a special case of the bent cable.keywords: Bayesian hierarchical model, Changepoint model, Longitudinal data, Mixed e�ects, Segmented regressionGrace Chiu, Dept. of Stat. & Act. Sci., U of Waterloo, 200 University Ave. W., Waterloo, Ontario, N2L 3G1, CanadaE-mail address:gchiu@uwaterloo.ca

Supporting grant: Canadian sources: NSERC Discovery Grants (to G. Chiu; J. Dubin) and OGS (to S. Khan)

Oral Presentation

IDENTIFYING POLLUTION SOURCE DIRECTIONS FOR POLLUTION SOURCE AP-PORTIONMENT.

William Christensen1, Basil Williams1, Shane Reese1

1Brigham Young University

The identi�cation of pollution source directions is an important part of the source apportionment problem.Estimated source directions are used both as inputs to a Bayesian source apportionment analysis, and as partof a post-analysis check to associate identi�ed pollution factors with potential pollution sources. We considertwo approaches for source location identi�cation which can be used in di�erent settings. The �rst requireswind direction data measured at the air quality receptor and makes use of statistical and/or deterministic(AERMOD) models for chemical transport of particulate matter from source to receptor. The second makesuse of HYSPLIT back-trajectory estimates and a kriging estimator which �lters heterogeneous measurementerrors.keywords: bayesian hierarchical model, receptor model, dispersion model, krigingWilliam Christensen, Brigham Young University Department of Statistics 230 TMCB Provo, UT 84602E-mail address:william@stat.byu.edu

Supporting grant:U.S. EPA Star Grant #RD-83216001-0

Oral Presentation

URBAN FOREST ASSESSMENT IN ITALY UNDER NATIONAL FOREST INVENTORYFRAMEWORK.

Piermaria Corona1, Mariagrazia Agrimi2, Federica Ba�etta3, Anna Barbati2, Lorenzo Fattorini3, EnricoPompei4, Walter Mattioli2

1University of Tuscia - DISAFRI, 2University of Tuscia, 3University of Siena, 4Corpo Forestale dello Stato

In post-industrial societies, urban areas are continuously expanding, thus extending their in�uence on anincreasingly large proportion of woods and trees located in or nearby urban and urbanizing areas. Theseforests under urban land use, known as urban forests, have the potential for signi�cantly improving the qualitythe urban environment. Data to quantify the extent and characteristics of urban forests are still lacking orfragmentary. This calls for an expansion of the domain of forest inventories, towards non traditional objects,like urban forests. It would be convenient to exploit the same sampling scheme used for large-scale forestinventories . Most forest surveys performed over large areas involve several sampling phases. The �rst phase

28

Bologna, Italy TIES 2009

is usually performed by partitioning the study region into quadrats or other regular shape units of the samesize; then sampling points are randomly or systematically assigned to each unit. In most cases, the spatiallayout of �rst-phase sampling points is carried out on orthophotos or very high resolution satellite imagesavailable for the whole study area. Whatever the design of the subsequent phases is, the �rst phase samplingpoints can be e�ectively used to assess the basic features of urban forests. This paper proposes approximatelyunbiased estimators of abundance and coverage of urban forests, together with conservative or approximatelyconservative estimator of the corresponding variances, which can be applied in the �rst-phase systematicsampling design adopted by most large-scale forest inventories.keywords: urban forestry, national forest inventory, probabilistic estimationPiermaria Corona, University of Tuscia - DISAFRI - Via San Camillo de Lellis, 01100 Viterbo (Italy)E-mail address:piermaria.corona@unitus.it

Oral Presentation

CONNECTION BETWEEN METEOROLOGICAL AND OZONE SCENARIOS IN DIFFER-ENT URBAN AREAS.

Rossana Cotroneo1, Silvia Bartoletti2, Armando Pelliccioni1

1Ispesl, 2Ispra

In areas with a high population density, the air quality is important to evaluate the e�ects of pollutanton human health. Ozone is one of the most critical pollutants and causes health problems and damageto ecosystems. It becomes meaningful in regions during summer under stable turbulence conditions. Weconsider three urban areas (Roma, Milano and Palermo) and 8760 hourly patterns during all 2007. The maintarget of our work concerns the use of data mining to determine di�erent scenarios linked to ozone levels. Wehighlight our attention on time series and cluster analysis. First, we use time series to analyse the in�uence ofmeteorology on ozone. Than we apply cluster analysis to examine meteorological conditions linked to ozone,so to incorporate the results into scenario analysis. Our results individuated twelve clusters planning scenariosbetween ozone and di�erent meteorological conditions for urban areas under examination. We found seasonalscenarios and nocturnal/diurnal scenarios. Inside the clusters, we observe similar meteorological scenarioslinked to di�erent levels of ozone. During summer season and diurnal day, Palermo presents maximum levelof ozone (95µg/m3), Roma 75µg/m3 and Milano 98µg/m3. An interesting result concerns winter scenario: interm of ozone, we have about 84µg/m3 for Palermo and 41µg/m3 for Roma and 79µg/m3 for Milano. At last,our study demonstrates levels of ozone are strictly connected to geographic and topographic characteristicsand local air quality of each city.keywords: Environmental statistics, Data mining, Cluster analysis, Time seriesRossana Cotroneo, Ispesl-Dipia, Via Fontana Candida 1, 00040 Monteporzio Catone, RomaE-mail address:rossana.cotroneo@ispesl.it

Oral Presentation

LONG-RANGE CLIMATE RECONSTRUCTIONS WITH DYNAMICAL SYSTEMS.

Michel Cruci�x1, Rougier Jonathan2

1Université de Louvain, 2University of Bristol

Climate exhibits a vast range of modes of variability. Some have characteristic times of a few days; othersevolve on thousands of years. All these structures are interdependent; in other words, they communicate.It is often considered that the only way to cope with climate complexity is to integrate the equations ofatmospheric and oceanic motion with the �ner possible mesh. Is this the sole strategy? Aren't we missinganother characteristic of the climate system: its ability to destroy and generate information at the macroscopicscale? Paleoclimatologists consider that much of this information is present in palaeoclimate archives. It istherefore natural to build climate models such as to get the most of these archives.The strategy proposedhere is based on Bayesian statistics and low-order non-linear dynamical systems, in a modelling approachthat explicitly includes the e�ects of uncertainties.Its practical interest is illustrated through the problem ofthe timing of the next great glaciation. Is glacial inception overdue, or do we need to wait for another 50,000years before ice caps grow again? Our (provisional) results indicate a glaciation inception in 50,000 years.keywords: Bayesian methods, Palaeoclimates, Non-linear dynamicsMichel Cruci�x, Institut d'Astronomie et de Géophysique G. Lemaître, 2, chemin du Cyclotron, BE-1348 Louvain-la-Neuve, BelgiumE-mail address:michel.Cruci�x@uclouvain.be

29

Bologna, Italy TIES 2009

Poster Presentation

SPACE REGRESSION ESTIMATION FOR FUNCTIONAL DATA.

Sophie Dabo-Niang1, Anne-Françoise Yao2, Mustapha Rachdi3

1University Lille 3, 2University Aix-Marseille 2, 3University Grenoble 2

Spatial regression estimation as well as prediction is an interesting and crucial problem in statistical in-ference for a number of applications, where the in�uence of a vector ofcovariates on some response variableis to be studied in a context of spatial dependence. Spatial data are modeled as �niterealizations of random�elds and are collected from di�erent spatial location on the earth, as in a variety of �elds, including soil sci-ence, geology, oceanography, econometrics, epidemiology,environmental science, forestry and many others.Inour knowledge, although potential applications of spatial nonparametric regression estimation to functionaldata are without number, no theoretical works have been devoted so farto nonparametric space regressionestimation and prediction for functional data.We extend here some of the existing results on functional re-gression estimations for non-spatialdata to the spatial case.We give some asymptotic properties (with rates)of the regression estimate.Then, we compare through some simulations the results obtained by taking or not-into account the spatial aspect of the random �eld.Special attention is paid to apply theseresults to predictun-sampled locations of a continuously indexedrandom �eld.keywords: regression, non-parametric, functional data, random �eldsSophie Dabo-Niang, university Lille 3, laboratory EQUIPPE, domaine du pont de bois, BP 60149, 59653 Villeneuved'ascq cedex FranceE-mail address:sophie.dabo@univ-lille3.fr

Oral Presentation

IMPACT OF RESIDENTIAL UNDERGROUND WATER STORAGE TANKS ON DRINK-ING WATER QUALITY OF JEDDAH SUPPLY SYSTEM, SAUDI ARABIA.

Maged Daoud1, Saleh Magram1

1 Civil Engineering Dept., King Abdulaziz University

The lack of sewerage systems in large areas of Jeddah City in Saudi Arabia has led to the emergenceof the problem of the high water level and groundwater contamination as a result of leakage from thesewage groves and the houses septic tanks. The Red Sea water intrusion increases the water table levels ofcontaminated groundwater. A rapid erosion of di�erent underground tanks walls can be occurred. Due to theirregular pumping of water to the water distribution networks in Jeddah City, the people installed residentialunderground water storage tanks to meet their water requirements during the o� periods of pumping water.The in�ltration of contaminated groundwater to water supply pipes during the o� periods can be also a sourceof pollution of water supply system and residential facilities. One hundred and twenty water samples werecollected from the residential underground water storage tanks from three regions of Jeddah City. Sampleswere analyzed for physical, chemical, and biological characteristics in order to determine the quality andvalidity of the water for human use and to compare the results with the standards of drinking water in SaudiArabia, and to analyze the results and know the causes of pollution and to make suggestions and appropriatesolutions for this problem.keywords: Water supply system, Underground storage tank, Assessment, Water qualityMaged Daoud, Civil Engineering Dept., Faculty of Engineering, King Abdulaziz University P.O. Box 80204 Jeddah21589 Saudi ArabiaE-mail address:sawmag@hotmail.com

30

Bologna, Italy TIES 2009

Oral Presentation

THE LONGITUDINAL DEPENDENCE OF BLACK CARBON CONCENTRATION ONTRAFFIC VOLUME IN AN URBAN ENVIRONMENT.

Rey Decastro1, Timothy Buckley2, Lu Wang3, Jana Mihalic4, Patrick Breysse4, Alison Geyh4

1WESTAT, 2Ohio State University School of Public Health, Division of Environmental Health Sciences,3Johns Hopkins Bloomberg School of Public Health, Department of Epidemiology, 4Johns HopkinsBloomberg School of Public Health, Department of Environmental Health Sciences

Purpose: Model the dependence of ambient black carbon concentration on tra�c volume in an inner cityneighborhood while accounting for weather and time. Methodology: Continuous monitoring at 5 minuteintervals was conducted for 12 months at the Baltimore Tra�c Study site surrounded by major streets thattogether carry over 150,000 vehicles per day. Missing data were imputed and all data were normalized toa 5 minute observational interval (n = 105,120). Time series modeling accounted for autoregressively (AR)correlated errors. Results: Outdoor black carbon is positively correlated at a statistically signi�cant level withneighborhood level vehicle counts, which contributed at a rate of 66 [±SE: 10] ng/m3 per 100 vehicles every�ve minutes. Winds from the SW-S-SE quarter are associated with the greatest increases in black carbon(376-612 ng/m3), with a background exposure of 905 ng/m3. These winds would have entrained black carbonfrom Baltimore's densely tra�cked downtown and a nearby interstate highway. Dew point, mixing height,wind speed, season, and workday are also statistically signi�cant predictors. The optimal representation ofblack carbon's autocorrelation is AR([1:6])*(288)*(2016), where the �rst factor demonstrates black carbon'sautocorrelation up to 30 minutes, while the second and third factors indicate diurnal and weekly cycles,respectively. Conclusions: Local exposure to black carbon from mobile sources is substantially modi�ed bymeteorologic and temporal conditions, including atmospheric transport processes. Black carbon concentrationalso demonstrates statistically signi�cant autocorrelation at several timescales.keywords: Air quality monitoring, Environmental monitoring, Autocorrelation, Imputation, Time series modellingRey Decastro, 1650 RESEARCH BOULEVARD WB 246 ROCKVILLE,MD 20850E-mail address:reyDecastro@westat.com

Oral Presentation

ENVIRONMENTAL COMPLIANCE, SOCIAL REGULATION AGAME THEORETICMODEL.

Suresh Deman1

1Centre for Economics & Finance

There can be little question that environmental regulation has gained increasing prominence in recentyears. Environmental regulation can be particularly di�cult to deal with and its impact particularly di�cultto account for at the enterprise level. A number of factors contribute to di�culties that the regulationsthemselves and the vigour with which they are enforced tend to vary with political trends. The costsassociated with compliance of environmental regulation are often di�cult to quantify and the bene�ts ofcompliance are almost always speculative in the short run as they are di�cult to quantify with any degreeof certainty in a foreseeable time horizon, but can be very meaningful in the longer run.The principal-agentframework is a hierarchy consisting of three players namely, Regulator (government), Regulated (agents, e.g.Board of Directors), and Managers at Plants (with third grade education). The Board of Directors sets thecontract for the managers, and the Regulator determines rewards for compliance and penalties for violatingenvironmental standards, as well as government policy. A government policy of incentive mechanism is basedon two things: (a) standards of compliance, and (b) investigative rules, which are enforced by inspectors. Thepaper will generate a new awareness on the part of top managers (and managers in general) to "go beyondthe numbers" in formulating environmental policy must be fostered. Education as to the bene�ts, potentialliabilities, and new SEC reporting requirements in the area will be an important �rst step in that direction.keywords: Environment, Game, SEC, Regulations, ComplianceDr Suresh Deman, Director & Visiting Professor Centre for Economics & Finance PO Box 17517, London SE9 2ZPE-mail address:s_deman2000@yahoo.co.uk

31

Bologna, Italy TIES 2009

Oral Presentation

THE TESTING OF ZERO-INFLATION AND OVER-DISPERSION FOR THE ENVIRON-MENTAL COUNT DATA.

Dianliang Deng1

1University of Regina

Discrete data in the form of counts sometimes exhibit more zeros than what can be predicted by a simplemodel. Also, these data show more zeros as well as extra variation. Therefore a discrete Poisson model mayfail to �t a set of discrete data either because of zero-in�ation or because of over-dispersion or because thereare both zero-in�ation and over-dispersion in the data. In this talk, we will deal with the class of zero-in�atedover-dispersed generalized linear models and propose procedures based on score tests for selecting a modelthat �ts such data. Empirical levels and power properties of the tests are examined by a limited simulationstudy. Two illustrative examples are given.Dianliang Deng, Department of Mathematics and Statistics, University of Regina, Saskatchewan, S4S 0A2 CanadaE-mail address:deng@uregina.ca

Poster Presentation

SUSTAINABILITY INDICATORS FROMMULTIDIMENSIONAL ANALYSIS OF PARTIALEQUILIBRIUM MODEL DATA.

Senatro Di Leo1, Carmelina Cosmi2, Maria Macchiato3, Maria Ragosta1

1Dipartimento di Ingegneria e Fisica dell'Ambiente, Università della Basilicata, Potenza Italy, 2Istituto diMetodologie di Analisi Ambientale IMAA-CNR, 3CNISM, UdR-NA, Napoli, Italy

The objective of this study is to set up a procedure for characterizing data correlation structure andidentifying sustainability indicators to improve the interpretation of the results of partial equilibrium energymodels, making easier both data calibration and policy strategies identi�cation. The energy sector is of out-standing importance for de�ning coherent planning strategies because of its close relationships with economyand environment. In this framework, technical economic models (based on the MARKAL-TIMES generatorsdeveloped by ETSAP-IEA) allow a comprehensive representation of energy systems and the individuation ofoptimal energy-technology con�gurations in compliance with di�erent exogenous constraints, at the minimumfeasible cost. The results provide a detailed breakdown of energy consumptions, technologies, emissions, andresources equilibrium prices. However, the large amount of data should be aggregated into homogenous setsto identify the key variables for de�ning operative priorities in the implementation of energy-environmentalstrategies. A preliminary application of multivariate statistical techniques to models outputs has provenuseful for improving the post optimal analysis, allowing to identify the statistically signi�cant variables andto �nd out the parameters that plays a key role in contrasting development scenarios. Starting from resultsof such analysis, it is possible to identify operational indicators to handle multidimensional and multiscaledata and to de�ne sustainability indexes that ease the planning strategies implementation. Here we presentthe analysis of the NEEDS TIMES country models' results for the electric power generation sector in 29European countries, for three reference time periods and three policy scenarios.keywords: Aggregate indeces, Energy-environmental strategies, Multivariate stistical analysisSenatro Di Leo, Dipartimento di Ingegneria e Fisica dell'Ambiente, Università della Basilicata, V.le dell'AteneoLucano, 85100, Potenza ItalyE-mail address:dileo@imaa.cnr.it

32

Bologna, Italy TIES 2009

Oral Presentation

AIR QUALITYMONITORING FUSING SATELLITE REMOTE SENSING, GROUND-BASEDMEASUREMENTS AND METEOROLOGICAL MODELING IN NORTHERN ITALY.

Walter Di Nicolantonio1, Alessandra Cacciari1, Gabriele Curci2, Paolo Stocchi2, Ezio Bolzacchini3, LucaFerrero3, Claudio Tomasi4

1Carlo Gavazzi Space at ISAC-CNR, Bologna, Italy, 2CETEMPS, Department of Physics, University ofL'Aquila, Italy, 3Department of Environmental Science, University of Milano Bicocca, Italy, 4Institute ofAtmospheric Science and Climate, ISAC-CNR,Bologna, Italy

Satellite remote sensing of Aerosol Optical Properties (AOP) can be pro�tably exploited in Air Qualitymonitoring in terms of estimates of particulate matter (PM) concentration at surface level. Actually, thesynoptic viewing of satellite sensor enables monitoring of PM concentration spatial distribution fostering thecompliance with the new EC Directive on ambient air quality and cleaner air for Europe (2008/50/EC).In particular, Aerosol Optical Depth at 550 nm wavelength, providing a measure of the columnar aerosolextinction in the visible spectral range, can be related to the ground-level concentration of PM havingdiameter ranging between 0.1 and 2.5 µm. In this study, the actual potential role of satellite observations ishighlighted combined with regional meteorological modelling and ground-based measurements in the contextof Air Quality monitoring. The capability of MODIS sensors (Terra and Aqua/NASA platforms) to retrieveAOP has been used in a semi-empirical approach to estimate PM2.5 content at the ground over a domaincontaining whole Northern Italy. Daily PM2.5 concentration samplings collected in several sites are employedin order to infer AOP to PM conversion parameters on a monthly basis. Furthermore, meteorological featuresof the geographical domain are provided by MM5 simulated �elds. Thus, daily maps of satellite-based PM2.5concentrations over Northern Italy are derived. Monthly averaged values have been compared to in-situPM2.5 sampling providing a good agreement.keywords: Aerosol Remote Sensing, Earth Observation satellite, Particulate MatterWalter Di Nicolantonio, Carlo Gavazzi Space S.p.A.at Institute of Atmospheric Sciences and Climate, CNR-ISAC,viaP. Gobetti, 101, 40129 Bologna BO-ItalyE-mail address:wdinicolantonio@cgspace.it

Oral Presentation

AIR QUALITY ASSESSMENT VIA FUNCTIONAL PRINCIPAL COMPONENT ANALY-SIS.

Francesca Di Salvo1, Gianna Agro'1, Mariantonietta Ruggieri1, Antonella Plaia1

1DSSM - University of Palermo

The knowledge of the global urban air quality situation represents the �rst step to face air pollution issues.For the last decades many urban areas can rely on a monitoring network, recording hourly data for the mainpollutants. Such data need to be aggregated according to di�erent dimensions, such as time, space and typeof pollutant, in order to provide a synthetic air quality index which takes into account interactions amongpollutants and correlation among monitoring sites.This paper focuses on Functional Principal Componenttechniques for the statistical analysis of a set of environmental data x(spt), where s stands for the monitoringsite, p for the pollutant and t for time, usually days (after the aggregation according to national agencyguidelines). This approach could highlight some relevant statistical features of time series from an explorativepoint of view, and, consequently, new opportunities to obtain a synthetic AQI. The analysis will be illustratedby considering the data concerning the daily values of the 5 main pollutants collected in Palermo during 2006.keywords: Air quality, Functional Principal Component AnalysisFrancesca Di Salvo, Dipartimento di Scienze Statistiche e Matematiche Viale delle Scienze ed. 1390128 Palermo -ItalyE-mail address:disalvo@dssm.unipa.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

33

Bologna, Italy TIES 2009

Oral Presentation

A BAYESIAN HIERARCHICAL MODEL FOR ESTIMATING THE HEALTH EFFECTS OFCHEMICAL CONSTITUENTS OF PARTICULATE MATTER.

Francesca Dominici1, Roger Peng1, Michelle Bell2

1Johns Hopkins, 2Yale

Population-based studies have estimated health risks of short-term exposure to �ne particles using mass ofPM2.5 (particulate matter < 2.5 micrometers in aerodynamic diameter) as the indicator. Evidence regardingthe toxicity of the chemical components of the PM2.5 mixture is limited. We used a national databasecomprising daily data for 2000-2006 on hospital admissions for cardiovascular and respiratory outcomes,ambient levels of major PM2.5 chemical components (sulfate, nitrate, silicon, elemental carbon, organiccarbon matter, sodium and ammonium ions), and weather.We develop Bayesian Hierarchical Models forspatio-temporal data to estimate the association between hospital admission for cardiovascular diseases andthe chemical components of �ne particles in the United States. Statistical challenges include: 1) exposuremeasurement error that vary across constituents; 2) large correlation among the predictors.Francesca Dominici, 615 N Wolfe Street21205 Baltimore MD, USADepartment of BiostatisticsBloomberg School ofPublic HealthJohns Hopkins UniversityE-mail address:fdominic@jhsph.edu

Supporting grant: EPA and NIH

Oral Presentation

BAYESIAN DENSITY REGRESSION AND MIXTURES WITH ENVIRONMENTAL AP-PLICATIONS.

David Dunson1

1Duke University

In many environmental applications, the shape of the response density can vary as environmental condi-tions, spatial location and time change. It is particularly important to accurately characterize such changeswhen there is interest in studying the tails of the response distribution, which may correspond to adversehealth conditions or unusual levels of a pollutant. In this talk, I describe recently developed methods forBayesian density regression, which allow a response density to change nonparametrically with multiple pre-dictors, including time and space. The proposed methods rely on �exible semiparametric Bayes mixturemodels, which allow the mixture weights to be predictor-dependent. A particularly appealing formulationcomputationally is the probit stick-breaking process (PSBP), which facilitates automated variable selectionand hypothesis testing through a conjugate structure after data augmentation. MCMC and sequential MonteCarlo algorithms are described and are applied to an epidemiologic study of diabetes and a hurricane activitydata set.keywords: Bayesian nonparametrics, Mixture model , Variable selection , Spatio-temporal , EpidemiologyDavid Dunson, Department of Statistical Science, Box 90251, Duke University, Durham, NC 27708-0251E-mail address:dunson@stat.duke.edu

Oral Presentation

ESTIMATION OF HIERARCHICAL SPATIO-TEMPORAL COREGIONALIZATION MOD-ELS WITH THE EM ALGORITHM.

Cinzia D'Ariano1, Alessandro Fassò1, Francesco Finazzi1

1Dipartimento di Ingegneria dell'informazione e metodi matematici- Università degli studi di Bergamo

Statistical models of spatio-temporal data are useful tools for environmental science (Fassò & Cameletti.,2009a). Linear models of coregionalization have been used to deal with multiple spatial variables in order tomodel permissible correlation and cross-correlation structures (Zhang, 2007).In this work we discuss spatio-temporal coregionalization model able to manage several sampling variables collected over a time period. Themodel, for its �exible formulation, can be applied either in the isotopic, partially heterotopic or completelyheterotopic case. In order to improve calibration capability, the model can handle several �elds.Maximum like-lihood estimation of model parameters are provided within the generalized EM framework (Fassò et al., 2007,2009b) mostly in closed form.In order to discuss the properties of the introduced model, some realistic dataconcerning airbone particulate matters are simulated.Fassò, A., & Cameletti, M., (2009a) A uni�ed statistical

34

Bologna, Italy TIES 2009

approach for simulation, modeling, analysis and mapping of environmental data, Simulation: Transactionsof the Society for Modeling and Simulation International, accepted.ISSN: 0037-5497.Fassò, A., & Cameletti,M., (2009b) The EM algorithm in a distributed computing environment for modelling environmental space-time data, Environmental Modelling & Software, accepted. DOI: 10.1016/j.envsoft.2009.02.009.Fassò A.,Cameletti M. & Nicolis O. (2007) Air quality monitoring using heterogeneous networks. Environmetrics, 18:245�264.Zhang, H. (2007) Maximum-likelihood estimation for multivariate spatial linear coregionalizationmodels. Environmetrics, 18, 125-139.keywords: Hierarchical spatio-temporal models, coregionalization, EM algorithm, Kalman �lter, particulate mattersCinzia D'Ariano, Via Marconi 5, 24044 Dalmine (BG), ItalyE-mail address:cinzia.dariano@unibg.it

Supporting grant: Contributed GRASPA, PRIN n.2006131039, "Statistical analysis and modelling of impact and risk forenvironmental phenomena in space and time" and Regione Piemonte project CIPE 2004 Statistical methods and spatio-temporalmodels for atmosphere

Oral Presentation

MATRIX INVERSION AND STATISTICAL DATA ANALYSIS.

Abdel El-Shaarawi1

1National Water Research Institute and McMaster University

Recent work on spatial temporal analysis of environmental data requires the inversion of large dimensionalsymmetric matrices. Although Gaussian elimination is the best-known general direct method for matrixinversion, special methods will often be useful when inverting matrices of special form. Indirect methods arechie�y iterative in nature and can be used in combination with direct methods to improve the accuracy of thecomputation. The objectives of this talk are to review these methods, suggest improvements, present someenvironmental applications, and link the computation of matrix inversion to statistical thinking.Keywords:Matrix inversion, Gaussian Elimination, Choleski decomposition, Cross ValidationAbdel El-Shaarawi, National Water Research Institute 867 Lakeshore Rd, Burlington, ON, L7R 4A6CanadaE-mail address:abdel.el-shaarawi@ec.gc.ca

Oral Presentation

COPPER REMOVAL FROM AQUEOUS SOLUTIONS BY USING PORTLAND CEMENTKILN DUST.

Mokhtar Elatrash1, Aly Okasha1, Hesham Ibrahim1

1Al-Mergheb University

Sustainable development is a development that meets the needs of humans at present without compromis-ing the ability of future generations to meet their own needs. This experimental study aims to investigatethe possible use of particulate emissions of Portland Cement Kiln Dust (PCKD) collected from Al-MerghebPortland cement factory at Al-Khoms, Libya, a low cost adsorbent to remove copper from aqueous solu-tions.The e�ects of various parameters such as adsorbent dosage, initial concentration of copper, agitationrate, contact time and solution pH level on the adsorption e�ciency were studied through batch experiments.The collected experimental data �t well with the Langmuir and Freundlich models. The statistical evaluationof the experimental data against the theoretical model is based on the total deviation error and the correla-tion coe�cient.The optimum physical conditions for copper removal from aqueous solution are identi�ed andFreundlich model o�ers the best representation of the experimental data. Furthermore, the results indicatethat the total deviation error is a realistic description of the deviation between the collected experimentaldata and the theoretical model. The study concludes that the PCKD can be utilized as an e�ective adsorbentto remove copper from aqueous solutions.keywords: Adsorption, Copper, Portland, Cement, DustMokhtar Elatrash, Department of Environmental Science, Faculty of Science, Al-Mergheb university, Al-Khoms City,Libya.E-mail address:m_elatrash2000@yahoo.com

Supporting grant: Al-Mergheb university

35

Bologna, Italy TIES 2009

Oral Presentation

CHARACTERIZING SPATIAL PATTERNS OF FIRE WEATHER USING HISTORICALDATA.

Sylvia Esterby1, Zuzana Hrdlickova2, Steve Taylor3

1University of British Columbia Okanagan , 2Brno University of Technology, 3Canadian Forest Service,Paci�c Forestry Centre

The Fire Weather Index (FWI), which provides a numerical rating of �re intensity, is a component of theCanadian Forest Fire Danger Rating System. Calculations in the FWI system are based on the consecutivedaily weather measurements accumulated into fuel moisture codes and �re behavior indices. Automatedelectronic �re weather stations have been operating in British Columbia since the 1980s and there are nowabout 200 stations with records of daily �re weather for as long as 25 years. Currently agencies such asCanadian Forest Service and BC Forest Protection post maps daily which display regions of various levelsof risk. Several methods that would allow us to summarize the change of regional risk over time have beeninvestigated. This includes methods that cluster stations on the basis of the time series for each station. Inaddition to using the complete record, we have investigated changes in the annual station maximum, 90%quantile and the number of days within a �xed �re danger class. The utility of such summaries as a basis forcomparison with current values and scenarios for the near future will be considered.Sylvia Esterby, Mathematics, Statistics and Physics, Irving K. Barber School of Arts & Sciences, University of BritishColumbia Okanagan, 3333 University Way, Kelowna, BC Canada V1V 1V7E-mail address:sylvia.esterby@ubc.ca

Supporting grant: PIMS, GEOIDE Project SII-51

Oral Presentation

A NON-STATIONARY NEYMAN-SCOTT MODEL FOR RAINFALL.

Guillaume Évin1, Anne-Catherine Favre1

1Institut National de la Recherche Scienti�que - Eau, Terre et Environnement, Université du Québec

Stochastic point processes for rainfall are known to be able to preserve the temporal variability of rainfall onseveral levels of aggregation (hourly, daily), especially thanks to the cluster approach. One major assumptionin the applications seen up to now is the stationarity of the rainfall properties in time. However, if we wishto consider a climate change in the rainfall model, this postulate must be reconsidered. Thus, we proposenew developments of Poisson-based models with clusters which consider a nonstationarity function on stormarrivals. The key is to imagine that storms will be more frequent in the future, this assumption being inaccordance with the increase of the mean annual precipitation. Thus, the basic theory is reconsidered andthe moments are derived up to the third order in the general case and with a simple function on stormarrivals. However, these developments raise deeper issues, that is how to link the theoretical moments withthe observed ones. If we consider that the statistical properties vary temporally, the computation of theobserved moments is not straightforward. In this work, a calibration method is proposed and discussed.Observed moments are computed thanks to moving window statistics and a regression method. Then, thegeneralized method of moments can be applied. This nonstationary rainfall model in time is applied to climatechange simulations of the Canadian Regional Climate Model (CRCM). These rainfall series are generatedfrom 1961 to 2100 at a 15min time-step with a spatial resolution of 45kmx45km.Guillaume Évin, Institut national de la recherche scienti�que Centre - Eau Terre Environnement 490, rue de laCouronneQuébec (Québec) G1K 9A9CANADAE-mail address:guillaume_evin@ete.inrs.ca

36

Bologna, Italy TIES 2009

Oral Presentation

SAMPLING STRATEGIES FOR THE ASSESSMENT OF ECOLOGICAL DIVERSITY.

Lorenzo Fattorini1

1Università di Siena

The problem of de�ning and measuring ecological diversity by means of well-behaved indexes is considered.Any diversity measure is a function of the species abundances in the community, which are usually unknownquantities. Accordingly, the abundances are estimated on the basis of suitable sampling strategies whichare able to handle the problems arising when working with ecological communities as well as ensuring goodstatistical properties. Subsequently, the abundance estimates are used to make inference about diversity onthe whole community. Emphasis is laid particularly on the problem of comparing diversity among severalcommunities. Finally the statistical challenge of estimating species richness is treated together with therelated topics of analysing species accumulation curves.Lorenzo Fattorini, Dipartimento di Metodi Quantitativi, P.za S. francesco 8, 53100 Siena (Italy)E-mail address:fattorini@unisi.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Poster Presentation

SPATIAL VARIABILITY OF HYDROCARBON POLLUTED SOILS: MAIN CONTRIBU-TIONS OF THE LOQUAS PROJECT.

Claire Faucheux1, Yves Benoit2, Claire Carpentier3, Chantal de Fouquet4, Bruno Fricaudet3,Jean-Christophe Gourry5, Edwige Polus-Lefebvre4

1Mines ParisTech-Centre de Géosciences-Géostatistique, 2IFP, 3Arcadis ESG, 4Mines ParisTech, 5BRGM

The context of LOQUAS (Localisation and quanti�cation of organic pollutant in soil) project is thecharacterization and the remediation of hydrocarbon polluted soils. One of the critical steps in such a kindof remediation is the spatial variability of hydrocarbon concentrations at small distances. The developmentof an on-site measurement device of hydrocarbon pollution, the Pollut-Eval®, enables to carry out anintensive sampling survey including measurements at small distances and a multi-scale sampling schemegoing from centimetric to decametric scales. After an initial e�ective phase of validation of the new gauge,made by comparison with a standard method (Gas Phase Chromatography), the variance of measurementerror is investigated. Then, a detailed study of the variability at small distances is performed: what isthe representativeness of one small sample analysed with the Pollut-Eval®? What is the horizontal andthe vertical variability of the concentrations? Then, the proportional e�ect of concentrations is underlinedand computed. Thanks to the nested sampling scheme, several variograms are calculated at di�erent scalesand adjusted together. Finally, the e�ect of the homogenization of a soil sample is examined in order toreduce the number of samples required to characterize one soil sample. The results point out the bene�tof performing an initial regular sampling scheme with some areas including small distances measurementsto obtain a multi-scale sampling. Depending on the spatial variability, the choice of an adapted samplingstrategy can be done and the feasibility of a selective remediation can be evaluated.keywords: geostatistics, hydrocarbon concentrations, soil pollution, spatial variability, Pollut-Eval®Claire Faucheux, Mines ParisTech,Centre de Géosciences-Géostatistique, 35 rue Saint-Honoré, 77300 FontainebleauE-mail address:claire.faucheux@ensmp.fr

Supporting grant: ANR (French National Research Agency)

37

Bologna, Italy TIES 2009

Oral Presentation

HOW TO BUILD AN INITIAL SAMPLING SCHEME: RECOMMENDATIONS FOR MEA-SUREMENT SURVEYS OF AIR QUALITY.

Claire Faucheux1, Chantal de Fouquet2, Giovanni Cárdenas3, Laure Malherbe3

1Mines ParisTech-Centre de Géosciences-Géostatistique, 2Mines ParisTech, 3Ineris

Air quality sampling schemes are designed to characterize at best contaminant concentrations (benzeneor nitrogen dioxide for example) and their relations with environment so as to obtain maps of pollution asprecise as possible. Two cases can be encountered: either implementation of an a priori sampling schemeusing only auxiliary variables, such as population density or land use, or modi�cation of a sampling schemeafter a �rst survey, in order to reduce the sampling costs.This study deals with the �rst case; it is assumedthat there are relations between contaminant concentrations and auxiliary variables. The most pertinentauxiliary variables linked to one given pollutant can be found in literature. Therefore the sampling schemewill depend on two di�erent spaces: the geographical space, in which samples are spread, and the auxiliaryvariables space, a �ctitious space described by the auxiliary variables scatter plot(s) which the samples tryto cover entirely.The chosen process is illustrated through an example. First, boundaries of the area to mapare de�ned. Then, a �rst regular sampling scheme is created and at the same time the auxiliary variables touse are chosen. The points de�ned in the geographical space (on the regular sampling scheme) are projectedon the auxiliary variables scatter plot(s). The sampling scheme is then completed in order to cover the wholescatter plot area and the new points are projected on the geographical space. Finally, the created samplingscheme is compared with an existing one to validate the proposed process.keywords: geostatistics, sampling scheme, air pollution, benzene, nitrogen dioxideClaire Faucheux, Mines ParisTech,Centre de Géosciences-Géostatistique, 35 rue Saint-Honoré, 77300 FontainebleauE-mail address:claire.faucheux@ensmp.fr

Supporting grant: Ineris

Oral Presentation

GENERATING FUZZY LOGIC-BASED RAINFALL-RUNOFF MODELS USING SELF OR-GANIZING MAPS.

Christophe Faust1, Peter Gemmar1, Oliver Gronz2, Markus Casper2

1University of Applied Sciences (FH) Trier , 2University Trier

Rainfall-runo� models are used to describe the hydrological behavior of catchments. Besides the amountof precipitation, the soil moisture is an important factor determining the runo� intensity in small catch-ments.Although there is a relationship between the discharge and the soil moisture, it is di�cult to describetheir functional connection. In this approach, specialized Self-Organizing Maps are used to detect similaritiesin a large set of soil moisture probes. If there are similarities between probes in the set, only a subset isneeded to get valuable information about the moisture state. A greedy algorithm is used to determine aminimum set of independent probes.As rainfall-runo� model, a Sugeno-type fuzzy system has been devel-oped. The fuzzy rules use selected soil moisture probes in the premise part and combine precipitation datain the crisp conclusion for runo� prediction. The fuzzy sets for the premise variables and the set of accordingrules are determined automatically, whereas the generic function for the conclusion was manually de�ned.The parameters of the rule base's conclusion functions were optimized using the least squares method.Thevalidation of this approach was done for a 7 km2 basin. Besides precipitation and temperature, we used15 soil moisture probes and the main gauge collected in this basin over two years. To evaluate the modele�ciency, the Nash-Sutcli�e coe�cient was calculated for di�erent test cases with respect to training andvalidation data; the best results achieved a Nash-Sutcli�e coe�cient of 0.92.keywords: Fuzzy Model , Self-Organizing Maps, Automatic Model Generation, Rainfall-Runo� ModellingChristophe Faust, faustc@fh-trier.deE-mail address:gronzo@fh-trier.de

Supporting grant: Project: Hochwasservorhersage, Programme: Wissen scha�t Zukunft, Ministry of Education, Science, Youthand Culture in Rhinland-Palatinate, Germany

38

Bologna, Italy TIES 2009

Oral Presentation

BAYESIAN INFERENCE FOR CLUSTERED EXTREMES.

Lee Fawcett1, David Walshaw1

1Newcastle University

In this talk we consider Bayesian inference for the extremes of temporally dependent wind speed dataobserved at a high altitude site in the U.K. Through a simulation study, we show that the common practiceof analysing peaks over thresholds (POT) is liable to incur serious bias in the estimation of model parameters,and in particular return levels - often used as design speci�cations when building to withstand extreme windspeeds. In fact, the direct analysis of all threshold exceedances can reduce this bias to negligible levels, whilstalso avoiding the sometimes arbitrary choice of cluster identi�cation scheme for identifying POT. Wherethere is a deeper interest in the temporal dependence itself, we develop an inference procedure based on an'automatic' cluster identi�cation scheme. Using simulated data, we implement and assess this procedure,making inferences for the extremal index and two cluster functionals. We then apply this procedure to thewind speed data, where the clusters correspond to storms and the two cluster functionals correspond tomean storm length and mean inter-storm duration. We also consider inference for long-period return levels,advocating the posterior predictive distribution as being most representative of the information required byengineers interested in design level speci�cations.keywords: Bayesian inference, Clusters, Extreme values, Wind speedsLee Fawcett, School of Mathematics & Statistics, Newcastle University,Newcastle NE1 7RU, UK.E-mail address:lee.fawcett@ncl.ac.uk

Poster Presentation

A HIERARCHICAL MODEL FOR EXTREME WIND SPEEDS.

Lee Fawcett1, David Walshaw1

1Newcastle University

A typical extreme value analysis is often carried out on the basis of simplistic inferential procedures, thoughthedata being analysed may be structurally complex. Here we develop a hierarchical model for hourly gustmaximum wind speed data, which attempts to identify site and seasonal e�ects for the marginal densitiesof hourly maxima, as well as for the serial dependence present at each location. A Gaussian model for therandom e�ects exploits the meteorological structure in the data, enabling increased precision for inferencesat individual sites and in individual seasons. The Bayesian framework adopted is also exploited to obtainpredictive return level estimates at each site, which incorporate uncertainty due to model estimation, as wellas the randomness that is inherent in the processes that are involved.keywords: Extreme Value Theory, Generalised Pareto distribution, Hierarchical model, Markov chain Monte Carlo,Wind speedsLee Fawcett, School of Mathematics & Statistics,Herschel Building, Newcastle University, Newcastle,UK NE1 7RU.E-mail address:lee.fawcett@ncl.ac.uk

Poster Presentation

MONITORING ANDMANAGING CYANOBACTERIAL RISKS IN FRESHWATER LAKES.

Claire Ferguson1, E. Marian Scott1, Laurence Carvalho2, Geo�rey A. Codd3, Andrew Tyler4

1Department of Statistics, University of Glasgow, Glasgow, UK, 2Centre for Ecology & Hydrology,Edinburgh, UK, 3University of Dundee, Dundee, UK, 4University of Stirling, Stirling, UK

Cyanobacterial toxins constitute one of the most high risk categories of waterborne toxic biological sub-stances. Mass populations of cyanobacteria are an increasingly common occurrence in inland waters ande�ective strategies for monitoring and managing cyanobacterial health risks are required to safeguard animaland human health.A multi-disciplinary study was undertaken to explore di�erent approaches for the iden-ti�cation, monitoring and management of potentially-toxic cyanobacterial populations and their associatedrisks. This included (i) using statistical and process-based models to investigate cyanobacterial bloom oc-currence; (ii) monitoring lake status using remote sensing (iii) investigating cyanobacterial toxin transfer tospray-irrigated crops; and (iv) assessing public attitudes and perceptions towards associated risks. Statisticalmodels were used to investigate which freshwater environments are most susceptible to the development oflarge populations of cyanobacteria. Phytoplankton data from 134 UK lakes were used to develop a seriesof Generalized Additive Models for predicting the presence/absence of cyanobacteria or their abundance

39

Bologna, Italy TIES 2009

(relative proportion or actual biovolume) from the knowledge of widely-available lake variables (such as alka-linity, colour, altitude, total phosphorus and retention time). These models can contribute to the assessmentof risks to public health by targeting lake monitoring and management at those lakes at highest risk ofbreaching World Health Organisation guideline levels, or national legislation/guidelines, for cyanobacteria inrecreational waters. The models indicated that, in the highest risk lakes (alkaline, low colour lakes), risks canbe lessened through management such as reducing nutrient loads and increasing �ushing during summer.keywords: generalised additive model, cyanobacteria, human healthClaire Ferguson, Department of Statistics, 15 University Gardens,University of Glasgow, G12 8QWE-mail address:c.ferguson@stats.gla.ac.uk

Supporting grant: This work was part of the project, "Strategies to manage toxic cyanobacterial blooms in lakes - remotesensing, modelling and cost bene�t analysis" funded under the NERC Environment and Human Health directed programme(Project reference NE/E009360/)

Oral Presentation

APPROXIMATED WRAPPED DISTRIBUTIONS FOR MODELING CIRCULAR DATA.

Clarissa Ferrari1, Alan E. Gelfand2, Giovanna Jona Lasinio3, Daniela Cocchi1

1University of Bologna, 2Duke University, NC, USA, 3Sapienza Univerity of Rome

Although circular data arise in many di�erent contexts, their statistical analysis is not trivial. Standardstatistical techniques cannot be used to analyze circular data because of the circular geometry of the samplespace; consequently, ad hoc approaches are needed to handle circular data.In this work, we analyze and showthe main drawbacks and advantages of the wrapping approach, through that the circular distributions areobtained wrapping the distributions on the real line onto the unit circle. Moreover, focusing on the wrappedNormal distribution, we provide a distribution approximation that turns out to be very useful to improvethe inferential results. This approximation, in fact, is directly used into the Bayesian inference procedureallowing to overcome the main disadvantage of this method given by the identi�ability problem. Overcomingthe identi�ability problem allows, substantially, to apply the standard in line models and procedures tocircular data as well.In order to appreciate the �exibility and the ease of applicability and interpretability ofthe wrapping approach a spatial application for circular data is presented.keywords: circular data, wrapping approachClarissa Ferrari, Dipartimento di Scienze Statistiche "Paolo Fortunati"Via delle Belle Arti 4140126 Bologna ItaliaE-mail address:clarissa.ferrari@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812 and2006131039) "Methods for collecting and analyzing environmental data" and "Statistical analysis and modelling of impact andrisk for environmental phenomena in space and time".

Oral Presentation

INTEGRATING SATELLITE AND GROUND LEVEL DATA FOR AIR QUALITY MONI-TORING AND DYNAMICAL MAPPING.

Francesco Finazzi1, Cinzia D'Ariano1, Alessandro Fassò1, Gianandrea Mannarini2, Orietta Nicolis1

1Dipartimento di Ingegneria dell'informazione e metodi matematici - Università degli Studi di Bergamo,2Dipartimento di Elettronica per l'Automazione - Università degli Studi di Brescia

Air quality monitoring networks have been increasingly installed around EU, mainly on a local basis. Asa result, the EU monitoring network is very expensive and appears rather heterogeneous at the continentalscale from the point of view of spatial representativeness, human risk exposure etc. (Fassò et al., 2007). Onthe other side, satellite measurements give interesting data because of homogeneity over time and space. Forexample, aerosol optical thickness (AOT) may be used to get information on airborne particulate mattersPM10 and PM2.5, see e.g. Chu et al. (2003). Since AOT are less precise than ground-level measurements ofparticulate matters, various studies, with the aim of calibrating satellite data, have reported positive temporalcorrelations between satellite data and both PM10 and PM2.5 (Koelemeijer et al. 2006, Wang & Christopher,2003). Along these lines, in this talk, we discuss spatio-temporal modelling for merging ground level data,which are local concentrations, and satellite data. In order to improve calibration capability, several �eldsare considered; e.g. boundary layer height, albedo, meteorology and land elevation. Due to the fact thatAOT availability is restricted to cloud-free conditions, the model has to cover with extensive missing data.We

40

Bologna, Italy TIES 2009

discuss a frequentist hierarchical model (Fassò & Cameletti, 2009a,b) which generalizes coregionalization ofZhang (2007) and missing data handling of Amisigo & Van de Giesen (2005). Maximum likelihood estimationis covered by the generalized EM algorithm.keywords: spatio-temporal models, coregionalization, EM algorithm, missing data, particulate mattersFrancesco Finazzi, via Marconi, 5 - 24044 - Dalmine (BG) - ITALYE-mail address:francesco.�nazzi@unibg.it

Supporting grant: Invited session, Fassò - GRASPAPRIN n.2006131039,"Statistical analysis and modelling of impact and risk forenvironmental phenomena in space and time" and Regione Piemonte project CIPE 2004 Statistical methods and spatio-temporalmodels for atmosphere

Oral Presentation

A HIERARCHICAL MIXTURE MODEL FOR ESTIMATING ZERO INFLATED CONTIN-UOUS FOREST VARIABLES.

Andrew Finley1, Sudipto Banerjee2

1Department of Forestry, Michigan State University, 2Division of Biostatistics, University of Minnesota

We are commonly interested in predicting one or more continuous forest variables (e.g., biomass, volume,age) at a �ne resolution (e.g., pixel-level) across a speci�ed domain. Given a de�nition of forest/non-forest,this prediction is typically a two step process. The �rst step determines which prediction units are forested.The second step predicts the value of the variable for only those forested units. Rarely is the forest/non-forestpredicted without error. However, the uncertainty in this prediction is typically not propagated through tothe subsequent prediction of the continuous forest variable of interest. Failure to acknowledge this errorwill result in biased and perhaps falsely precise estimates. In response to this problem, we o�er a modelingframework that will allow propagation of this uncertainty. Here, we envision two latent processes generatingthe data. The �rst is a continuous spatial process while the second is a binary spatial process. We assume thatthe processes are independent of each other. The continuous spatial process controls the spatial associationstructure of the forest variable of interest, while a binary process indicates presence of "measurable" quantityat a given location. Finally, we explore the use of a predictive process for both the continuous and binaryprocesses to reduce the dimensionality of the data and ease the computational burden. The proposed modelsare motivated using georeferenced National Forest Inventory (NFI) data and coinciding remotely sensedpredictor variables.keywords: Bayesian, geostatistical, predictive process, forest inventory, spatialAndrew Finley, Department of Forestry Michigan State University East Lansing, Michigan, USAE-mail address:�nleya@msu.edu

Supporting grant: National Science Foundation DMS-0706870; USDA Forest Service FIA and FHTET programs

Poster Presentation

THE PROBLEM OF THE COMPLEX HYDROCARBON POLLUTION IN STOCKAGESITES: THE IMPORTANCE OF GEOSTATISTICS.

Sara Focaccia1

1DICMA

Hydrocarbon pollution deriving from stockage activities is a di�used pollution, besides impacting boththe soil and the groundwater. The study and the characterization of these areas results therefore di�cultbecause it must comprise both an aspect of geological modelisation, both hydrological of the groundwater,as well as an analysis of the pollutants' distribution: in order to do that we make use of the geostatisticalinstrument. This characterization is a problem in space and time that should be coped with di�erent ap-proaches (space and time covariances; �xing one variable and using time as di�erent realization of the samealeatory function;...).Consequentially, it is multivariable because we have a pollution due to several hydro-carbons in di�erent phases. Finally, it's not stationary from the geostatistical point of view. Apart from thegeostatistical problems, there are also others components that make site analysis not simple, as, for instance,a complex geology or human operations above the area. Geostatistics is a useful tool for the characterizationof these sites because it lets us treating di�erent problems related to these zones, using the data given by

41

Bologna, Italy TIES 2009

well monitoring, security wells, sampling and surveys. We could in fact advance in the reconstruction of thegeological horizons, of the evolution of the piezometric level , exploiting di�erent geostatistical estimationmethods (kriking, cokriging, simulations,...).keywords: pollution, geostatisticsSara Focaccia, DICMA, University of BolognaE-mail address:sara.focaccia@gmail.com

Oral Presentation

A MARKOV MODEL APPROACH TO THE ANALYSIS OF VIDEO TRANSECT DATAFROM THE MARINE ENVIRONMENT.

Scott D. Foster1, Mark V. Bravington1

1CSIRO Mathematical and Information Sciences

Underwater video is a powerful and increasingly a�ordable tool for monitoring the seabed environment.However, analysis is not straightforward because the data is discrete-valued and heavily autocorrelated. Weaddress these challenges via a set of nested non-stationary Markov models that use remote-sensing data (e.g.depth, acoustic re�ectivity) as covariates. Remote sensing data can be collected cheaply for an entire regionbut is not directly informative about seabed fauna, whereas the opposite is true for video data. Our aim istherefore to combine the two data sources to predict faunal composition, and quantify its uncertainty, acrosslarge regions such as putative marine reserves. We present two diagnostic tools for assessing the adequacyof our models, and show how prediction variance can be calculated. The techniques are illustrated usingcontinental-slope data from Tasmania.keywords: Video data, Markov model, Diagnostics, Areal prediction, Marine environmentScott D. Foster, CSIRO Mathematical and Information Sciences GPO Box 1538 Hobart 7001 Tasmania AustraliaE-mail address:scott.foster@csiro.au

Poster Presentation

BIODIVERSITY ANALYSIS USING RANK ABUNDANCE DISTRIBUTIONS.

Scott D. Foster1, Piers K. Dunstan1

1CSIRO Mathematical and Information Sciences

Understanding and predicting patterns of biodiversity is a key topic for ecological research and envi-ronmental management. In the marine environment, a common form of data collected is the number ofindividuals of species at a series of locations. The set of encountered species changes between samples andthe majority of species may only be observed at a single location. Analyses based on the most ubiquitousspecies could be potentially misleading, since they occur at many locations. If species identities are discarded,Rank Abundance Distributions (RADs) can be used to provide a full description of assemblages at di�erentlocations. Samples that span an environmental gradient should contain information on how biodiversitychanges with the environment. We show that the rank abundance distribution representation of the dataprovides a convenient method for quantifying biodiversity. We outline a statistical framework for modellingRADs and allow their multivariate distribution to vary according to environmental gradients. The statisticalmethod combines elements of popular methodologies with some novel extensions that are necessary for RADrepresentation of the data. The method is motivated by, and applied to, a large scale marine survey o� thecoast of Western Australia, Australia. It provides a rich description of biodiversity and how it changes withenvironmental conditions.keywords: Biodiversity, Rank Abundance Distribution, Modi�ed Dirichlet-Multinomial, Evenness, Marine Environ-mentScott D. Foster, CSIRO Mathematical and Information Sciencesc/o Marine Laboratories GPO Box 1538 Hobart 7001Tasmania AustraliaE-mail address:scott.foster@csiro.au

42

Bologna, Italy TIES 2009

Oral Presentation

SAMPLING PROPERTIES OF SPATIAL TOTAL ESTIMATORS UNDER TESSELLATIONSTRATIFIED DESIGNS.

Sara Franceschi1

1Università di Siena

Environmental scientists are frequently interested in attributes which are scattered over a planar regionin such a way that the target parameter turns out to be the corresponding attribute spatial total. In thisframework, the design-based approach under the continuous-population paradigm may be adopted in order toprovide inference on the spatial total. The design is usually carried out by locating sample sites according tosuitable strati�ed protocols based on the tessellation of the planar region. The tessellation strati�ed designsare commonly adopted in the �eld since they allow for an even sampling coverage of the study area. In thispaper it is shown that spatial total estimators adopted under tessellation-based designs display a very elevateaccuracy and are normally distributed for large samples.keywords: design-based inference, continuous population, spatial sampling strategiesSara Franceschi, Piazza San Francesco 8, 53100, Siena, ItalyE-mail address:franceschi2@unisi.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

EXTREME EVENTS AND BUSINESS CONTINUITY PLANNING - EXPLORATION INFLOOD RISK STRATEGIES WITHIN SCOTLAND.

Maria Franco Villoria1, Marian Scott1, Trevor Hoey1, Denis Smith1, Alistair Cargill2

1University of Glasgow, 2SEPA

Recent studies report increases in both frequency and intensity of extreme events such as �oodings. Cur-rent methods, that are typically based on 50 years of data and assume constant climatic conditions, mayunderestimate �ood risk. Hence, given ongoing climate change, it is of interest to investigate the usefulnessand validity of current methods of �ood-risk estimation and provide new tools for decision making around�ood hazards in Scotland. A major cause of �ooding is extreme river �ow levels, and therefore river �ooddata are being explored in order to identify the seasonal pattern and investigate any changes that havehappened over the years. So far, traditional methods of time series analysis have been applied (sinusoidalregression, STL decomposition, periodical normal quintile transformation) to deseasonalize the data. How-ever, the resulting deseasonalized series still shows some presence of a seasonal pattern. The results of thisanalysis suggest that the assumption of stationarity, on which traditional methods are based on, does nothold. This is probably linked to climate change. Therefore, alternative methodology for dealing with this isneeded. Long-term memory analysis and wavelet analysis will be explored as possible methods for identifyingthe local behaviour of non-stationary time series.Maria Franco Villoria, Department of Statistics, University of Glasgow,University Avenue, Glasgow G12 8QQ UnitedKingdomE-mail address:mvilloria@stats.gla.ac.uk

Poster Presentation

BAYESIAN FORECASTING ALGORITHM FOR ACCOMMODATING NON-GAUSSIANAIR CONCENTRATION OBSERVATIONS.

Ali Gargoum1

1UAE University

In the Gaussian model given in Smith et al. (1995), used for predicting air concentration after a nuclearaccident, it is natural to assume that the distribution of observations conditional on their states is non-normal(e.g., Poisson or lognormal). The states are the quantities of mass under pu� and fragments of contaminationwhere pu�s are emitted stochastically from a chimney and directed by a known wind-�eld across space. TheMarkovian property of the stochastic emission process and deterministic fragmentation process, means thatthe joint distribution of mass fragments at any time is decomposable with its small clique dimension in a

43

Bologna, Italy TIES 2009

Bayesian network. In this work, I propose an approximate algorithm for quick Bayesian inference in non-Gaussian dynamic systems (the lognormal case) based on the dynamic generalized linear models when usedon junction trees. The algorithm validity can be checked numerically- for example by using the Hellingerdistance metric.keywords: Bayesian networks, Dynamic linear models, Pu� modelsAli Gargoum, Dept. of Statistics, College of Business & Econ, UAE University, Al Ain, United Arab Emirates. P. O.Box. 17555E-mail address:alig@uaeu.ac.ae

Oral Presentation

CLIMATE RECONSTRUCTED FROM POLLEN DATA USING A DYNAMIC VEGETA-TION MODEL.

Vincent Garreta1, Joël Guiot1

1CEREGE, CNRS, Aix-Marseille Universities

The objective of this talk is to present an inference method of past climate from a vegetation proxy:the pollen sampled with high time-resolution along a sediment core. The link between climate and pollenthrough the vegetation is simulated using a dynamic vegetation simulator, here, LPJ-GUESS. The model forreconstruction is hierarchical Bayesian and embeds the physical vegetation model. Inference or reconstructionof past climate is computationally challenging due to the climate space dimension considered (number of pollensamples * number of climate variables) and the need of vegetation simulations over thousand of years. Wesolve this problem using a particle �lter algorithm.Compared to classic statistical methods, this approach,by explicitly modelling the vegetation:1. allows to control external forcings of the relation climate-pollensuch as C02, 2. models -through the vegetation model- a temporal correlation and possible disequilibriumbetween changing climate and vegetation. These advantages are conditional upon the hypothesis that staticand dynamic components of the vegetation model are validated.We start the talk by stressing the importanceof taking into account the dynamics of climate and vegetation when reconstructing climate from pollen.After describing the model and the particle �lter we show an example of Holocene climate reconstruction atMeerfelder Maar (South East Germany) from a very high resolution sediment core.keywords: Paleoclimate, pollen data, vegetation simulator, Bayesian hierarchical model, particle �lterVincent Garreta, CEREGE, UMR 6635 Europôle de l'arbois 13545 Aix en ProvenceE-mail address:garreta@cerege.fr

Poster Presentation

HIERARCHICAL CLUSTERING OF SPATIALLY CORRELATED FUNCTIONAL DATA.

Ramón Giraldo1, Pedro Delicado1, Jorge Mateu2

1Universitat Politècnica de Catalunya 2Universidad Jaume I

Classi�cation problems of functional data arise naturally in many applications. Several approaches havebeen considered for solving the problem of forming groups based on functional data. Techniques for clusteringfunctional data are focused on independent functions. However, in several disciplines of applied sciencesthere exists an increasing interest for modelling correlated functional data: it is the case when samples offunctions are observed over a discrete set of time points (temporally correlated functional data) or whenthese functions are observed in di�erent sites of a region (spatially correlated functional data). In this workwe join hierarchical clustering methods for both geographically referenced data and functional data in orderto give a solution to the problem of classifying spatially correlated curves. Our methodology allows �ndingspatially homogeneous groups of sites when the observations at each sampling location consist of samplesof random functions. In univariable and multivariable geostatistics various methods of incorporating spatialinformation into the clustering analysis have been considered. Here we extend some of them to the functionalcontext in order to ful�l the task of clustering spatially correlated curves. In our approach we initially usebasis functions for smoothing the observed data and subsequently we weight the dissimilarity matrix among

44

Bologna, Italy TIES 2009

curves by either the trace-variogram or the multivariable variogram calculated with the coe�cients of thebasis functions. As an illustration the methodology is applied to a real data set corresponding to averagedaily temperatures measured at 35 Canadian weather stations.keywords: Basis functions, Functional cluster, Geostatistics, Multivariable variogram, Trace-variogramRamón Giraldo, Universidad Nacional de Colombia, Ciudad Universitaria, Departamento de Estadística, o�cina 340,Bogotá, ColombiaE-mail address:rgiraldoh@unal.edu.co

Oral Presentation

DETECTING HIGH-RISK REGIONS IN DISEASE MAPPING USING SPATIAL P-SPLINEMODELS.

Tomás Goicoa1, M.D. Ugarte1, Jaione Etxeberria1, Ana F. Militino1

1Universidad Pública de Navarra

The detection of high-risk areas in disease mapping is an important task as it allows public health agenciesto develop prevention/intervention programs. The available tools have been developed under both an empiri-cal and a fully Bayes approach. Within an empirical Bayes framework, con�dence intervals for the risks basedon an appropriate estimator of the mean squared error (MSE) of the log-risk predictor have been investigated.From a Fully Bayes approach, credibility intervals and decision rules have been used to unmask high-riskregions. All these techniques have been investigated on the basis of conditional autoregressive (CAR) models.In this work, penalized splines (P-spline) models with bases constructed from the tensor product of marginalB-splines bases are considered to smooth the risks. Moreover, a MSE estimator of the log-risk predictor isderived to build con�dence intervals for the log-relative risk predictor obtained from P-spline models. Theability of these intervals to detect high-risk areas and discard false positives created by noise are studiedthrough a simulation study. The results are compared to those obtained from the classical CAR models.keywords: Con�dence Intervals, MSE, sensitivity, speci�cityTomás Goicoa, Department of Statistics and Operations Research Public University of Navarra.31006 Pamplona,SpainE-mail address:tomas.goicoa@unavarra.es

Supporting grant: This research has been supported by the Spanish Ministryof Science and Innovation (MTM 2008-03085/MTM).

Oral Presentation

PREDICTION OF WATER QUALITY VARIABLES USING STATE-SPACE AND LINEARMODELS FOR RIVER NETWORK DATA.

A. Manuela Gonçalves1, Marco Costa2

1Universidade do Minho, 2Universidade de Aveiro

Abstract: Water quality monitoring networks are important tools in management and evaluation of surfacewater quality and they could be improved with precise forecasts of the surface water variables. We presenta comparative study of space-time models considering two approaches: state-space and linear models, bothassociated to clustering techniques. We identify homogeneous regions, based on similarities in the temporaldynamics of variables of water quality measured patterns, like adopted in Bengtsson and Cavanaugh (2008).For each cluster, we establish space-time models in which the goal is to model and forecast water qualityvariables. The state-space models, associated with the Kalman �lter, allow us to model the studied variableestablishing a dynamic model where the dependence structure is modelled by a latent state variable. Thelinear models contain a term for the global trend and the seasonal variation throughout the year. We discussthe quality of the predictions produced by two approaches comparing them, by the mean squared error, ina period of time. This methodology is illustrated in a case study in which the evolution of the DissolvedOxygen concentration is predicted in a River Basin located in Portugal.keywords: water quality, clustering, state-space models, linear modelsA. Manuela Gonçalves, Arminda Manuela GonçalvesDepartamento de Matemática para a Ciência e Tecnologia Uni-versidade do MinhoCampus de Azurém 4800-058 Guimarães PORTUGALE-mail address:mneves@mct.uminho.pt

Supporting grant: Centro de Matemática da Universidade do Minho Centro de Matemática e Aplicações Fundamentais daUniversidade de Lisboa

45

Bologna, Italy TIES 2009

Oral Presentation

A MULTIVARIATE CAR MODEL AND ITS APPLICATIONS.

Fedele Greco1, Daniela Cocchi1, Carlo Trivisano1

1Department of Statistics - University of Bologna

Areal data modelling has seen a considerable growth in last years following the increasing availabilityof Geographic Information System and spatial datasets. Among the various �elds of applications, diseasemapping is probably the leading one, both because mortality data are often available at area level due tocon�dentiality restrictions and because smoothed maps of relative risks are very informative in planning publichealth policies. Despite disease mapping studies have been widely performed at univariate level, contributesconcerning multivariate modelling have been only recently developed. In this work, by extending the well-known univariate CAR model, we propose a �exible multivariate areal data model that overcome somerestrictive hypotheses underlying models previously proposed in this context. In particular, a multivariatedistribution for spatial random e�ect is constructed by explicit modelling of cross-correlation that is allowedto be asymmetric. Such distribution can be adopted as a prior for regression coe�cients in a spatially varyingcoe�cients model. This is of interest since, when several covariates representing risk factors are involved inthe analysis, it might be the case that the impact of risk factors is not constant in space and that the impactof risk factors is correlated. The proposed methodology is implemented in a Bayesian framework by meansof MCMC algorithms and is illustrated via its applications in multivariate disease mapping.keywords: Multivariate CAR, Hierarchical Bayesian Models, Disease Mapping, Areal DataFedele Greco, Via delle Belle Arti, 4140126 Bologna ItalyE-mail address:fedele.greco@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

MODELLING JOINT EXTREMES: APPLICATION TO UK RIVER FLOW DATA.

Olivia Grigg1, Jonathan Tawn1

1Lancaster University

A joint probability analysis of extreme events with application to UK river �ow data will be given. Chal-lenges in this application include assessing the e�ects of seasonality and investigating the structure of condi-tional relationships between variables of interest with a view to characterising seasonal extremal dependence.Estimates of the marginal features and tail dependence structure between observed �ow and other responsevariables of interest will be presented, where the level of associated covariates is accounted for. The tail de-pendence model takes the form of the He�ernan & Tawn (2004) semiparametric model for multivariate datawhere the margins are transformed to the Laplace scale to allow for the possibility of negative dependenceand where the marginal features are described through a generalised Pareto distribution. Once obtained, themarginal model for �ow can be used to predict T-year return levels conditional on covariate level, and thedependence model can be used to estimate the probability of extreme sets in the joint domain for multivariateresponses.keywords: Extremes, Joint probability, Flooding, SemiparametricOlivia Grigg, Maths & Stats Dept Lancaster University Lancaster LA1 4YFE-mail address:o.grigg@lancaster.ac.uk

Supporting grant:NERC thematic grant NE/F001118/1

Oral Presentation

DATA ANALYSIS OF SEAWATER INTRUSION IN THE PEGO-OLIVA MARSHLAND.

Juan Grima1, Bruno Ballesteros1, José Antonio Domínguez1, Juan Antonio Luque1

1Geological Survey of Spain (IGME)

The Pego-Oliva marshland was included in the list of protected areas under the Ramsar Convention in1971. It was declared Natural Park by the Regional Government of Valencia on January 9th, 1995. Thewhole area has been investigated to identify the major causes of potential degradation. Salinization has beenidenti�ed as one of the key points for a sustainable management, and it has been found that the processes

46

Bologna, Italy TIES 2009

related with an increase of salinity are due to the existence of both natural and anthropogenic factors. Anumber of springs have been sampled and their salinity analyzed in the context of a study rather broader inscope and focussed on the sustainability of the area. It has been observed that some of them show an increasein salinity during low �ow periods, whereas some others exhibit a speci�c behaviour, with low salinity values,during the same scarce precipitation intervals. For a better understanding of the salinization mechanismin such a complex environment, a study has been conducted to analyze the interactions between the basaldetritic aquifer and the karstic lateral ones. Within the former the �ow is laminar and homogeneous, whereasin the karstic ones, the �ow is faster, through preferential channels. The relations between conductivity,precipitation and piezometric levels have been checked. Furthermore, the relation of the di�erent aquiferswith salinity values has been considered, and somewhat di�erent statistical aspects, like the detection ofstatistically signi�cant upward or downward trends and its correlation with precipitation data have beensummarizedkeywords: seawater intrusion, coastal aquifers, springs, correlation, trendsJuan Grima, Cirilo Amorós 42 entlo46004 Valencia SpainE-mail address:j.grima@igme.es

Oral Presentation

INTEGRATING SMOOTHING AND REGRESSION TREES FOR CHANGE-POINT DE-TECTION IN ENVIRONMENTAL DATA.

Anders Grimvall1, Sackmone Sirisack1

1Linkoping University

The most widespread methods for change point detection in univariate or multivariate time series ofobservational data have their roots in models for which the mean is constant between a set of level shifts.We have developed methods permitting joint estimation of smooth trends and abrupt level shifts in vectortime series. This was accomplished by employing a back-�tting algorithm alternating between estimation ofsmooth trends and detection of discontinuities. Potential change-points were identi�ed using regression treesto partition the space of time points and vector coordinates into rectangular subsets, and a post-pruningtechnique was then employed to reduce undesirable e�ects of over-�tting. Computer experiments involvingsimulated data showed that our algorithm performed well when synchronous level shifts in several of thevector coordinates played a prominent role. This was con�rmed by applying our method to long time seriesof water quality data and climate records from networks of meteorological stations.keywords: Change-point detection, Time series, Smoothing, Regression treesAnders Grimvall, Department of Computer and Information Science, Linkoping University, SE-58183 Linkoping,SwedenE-mail address:anders.grimvall@liu.se

Oral Presentation

RESTRICTING PARAMETER SPACE IN MESOSCALE WATER BALANCE MODELS US-ING HYDROLOGICAL SOIL MAPS.

Oliver Gronz1, Peter Gemmar1, Markus Casper2

1University of Applied Sciences (FH) Trier , University of Trier , 2University of Trier

In mesoscale water balance models, a catchment is usually subdivided into several model elements and thewater balance of each element is simulated individually. Representing the correct spatial variability is feasiblefor some hydrological processes, e.g. for interception by using land use maps. Deriving model parametersfor other processes, especially the soil-related ones, is di�cult. These parameters are calibrated, but dueto the size of possible parameter con�gurations, the same value is used for all model elements. Thus, thevarious model elements represent rather the mean behavior than the correct spatial distribution and thespeci�c strength of the runo� processes. To support a spatially distributed parameterization, new sourcesof information need to be incorporated. One way of incorporating additional information is the usage ofhydrological soil maps, which are available today. They indicate the potentially dominant runo� processeslike Horton overland �ow, subsurface �ow, deep percolation etc. An interdisciplinary project has startedto integrate these maps in the calibration process. The main aim is to transform the spatial distributionshown by the map into the model parameters. An initial idea is to �nd parameter prototypes for each of theruno� processes indicated by the map. In the following calibration process, these prototypes are scaled on

47

Bologna, Italy TIES 2009

a global level to adapt to a speci�c catchment. As a result, the developed parameterization strategies willpossibly make the calibration process less expensive and more transparent. Additionally, the improvementin representing the spatial distribution will enhance the model's e�ciencykeywords: Water Balance Model, Hydrological Soil MapsOliver Gronz, gronzo@fh-trier.deE-mail address:gronzo@fh-trier.de

Oral Presentation

STOCHASTIC GROUNDWATER MODELLING FOR A FRACTURATED AQUIFER INAUGUSTA AREA (SYRACUSE, ITALY).

Enrico Guastaldi1, Andrea Carloni1, Vincenzo Ferrara2, Claudio Gallo3

1Centro di GeoTecnologie, Università degli Studi di Siena, 2Dipartimento di Scienze della Terra, Universitàdegli Studi di Catania, 3CRS4 Cagliari

Simulation of groundwater �ow in fractured aquifers modelling involves problems related to heterogeneityof the medium. Therefore, natural system needs mathematical approximations. In such aquifer systems, thestudy of groundwater pollution scenarios and risk analysis are complex because the di�culty in determiningboth times and maximum distances covered by pollutants.The objective is to assess the risk of contaminationof deep groundwater in carbonate aquifer of Augusta coastal area, to evaluate possible aquifer pollution sce-narios. In order to reconstruct the complex geological framework of the study area, we utilise a geostatisticalapproach out of the usual numerical codes. We implemented a numerical model by a stochastic approach tosimulate both the groundwater �ow (MODFLOW-2000 code) and the pollutants transport (MT3DMS code).A critical step was the regionalization of hydrodynamic parameters: hydraulic conductivity requested prob-abilistic analyses to minimize uncertainty of spatial distribution of available data.We assigned the hydraulicconductivity values at the model by implementing an algorithm: we generated a stochastic distributionMonte Carlo type, based on gaussian probability density function appropriately formulated. Afterwards,we assigned to the model a discrete set of hydrodynamic conductivity values related to the main geologicaldiscontinuities.The model allowed to reproduce complex hydrogeological systems, and then to perform thecontamination risk analysis. Furthermore, the algorithm that we programmed represents a computationalelement that makes the model more �exible. In fact, this code generates di�erent con�gurations of hydrodi-namic parameters for stochastic simulation, the more precautionary way to formulate risk analysis in termsof probability.keywords: Risk analysis, Monte Carlo simulation, GeostatisticsEnrico Guastaldi, Via Vetri Vecchi 3452027 - San Giovanni Valdarno (AR)E-mail address:guastaldi@unisi.it

Oral Presentation

DETECTION OF OCEANIC INFLUENCE ON THE PRECIPITATION OF THE CENTRALVENEZUELAN COAST USING TIME-VARYING MODELS.

Lelys Guenni1, Gabriel Huerta2, Bruno Sansó3

1Universidad Simón Bolívar, Caracas, Venezuela, 2Department of Mathematics and Statistics, University ofNew Mexico, Albuquerque, USA, 3University of California, Department of Applied Mathematics andStatistics, Santa Cruz, USA

Exceptional rainfall events occurred during mid-December 1999 produced �oods and landslides along thenorth central coast of Venezuela with over 10,000 fatalities reported and economic looses estimated at over$1.8 million (Lyon, 2003). Similar events occurred also in February, 1951 and February 2005. Wieczorek etal. (2001) also reported that many of these severe events documented in the region have occurred during theperiod November-February. Common features of the combined anomalies in the Equatorial Paci�c and theNorth Tropical Atlantic sea surface temperature (SST) were found for most of the extreme rainfall events.The aim of the analysis is to detect potential changes in mean daily precipitation and monthly daily maximaduring the November-February months. Dependencies of extremes and mean daily values on the oceanicfeatures are analyzed using time varying models. To explore changes in mean daily rainfall dependence onthe SST anomalies, a normal distribution for the cubic root of mean daily rainfall with a temporal componentde�ned through a Dynamic Linear model (DLM) or state space representation was used. On another hand,

48

Bologna, Italy TIES 2009

a non-stationary Generalized Extreme Value (GEV) model with a time-varying dependence of the locationparameter on the oceanic anomalies, was used to evaluate monthly daily maxima changes with time. A moreclear signal of change is observed for the extreme values than for the mean values, which agrees with thepotential rainfall changes projected under climate change.keywords: oceanic in�uence, time-varying models, rainfallLelys Guenni, Universidad Simón Bolívar Department of Scienti�c Computing and Statistics APDO. 89.000. Caracas1080-A VenezuelaE-mail address:lbravo@cesma.usb.ve

Supporting grant: FAPESP

Oral Presentation

SIMULTANEOUS MODELLING OF SPATIAL AND TEMPORAL VARIATIONS IN AIRPOLLUTION EXPOSURES FOR HEALTH RISK ASSESSMENT AND EPIDEMIOLOGI-CAL ANALYSIS.

John Gulliver1, Marta Blangiardo2, David Briggs2, Anna Hansell2

1University of the West of Scotland, 2Imperial College

Air pollution episodes are often characterised by a dominance of long-range, secondary air pollution. Dur-ing episodes there is, however, underlying spatial variation in air pollution due to local tra�c and other sources(e.g. industrial and domestic), and locally derived air pollution may also be elevated due to meteorology.Thus, during air pollution episodes there is still potential for heterogeneity in exposures across the popula-tion. This may become signi�cant if long-range air pollution has a di�erent composition from local sources,which leads to di�erent biological properties and therefore di�erent health impacts. This paper presents anew approach, developed as part of the EU-funded GEMS project, to simultaneously account for temporaland spatial variations in daily levels of air pollution as a basis for health risk assessment and epidemiologicalstudies. Bayesian methods are used to model exposures at residential address level, combining informationfrom 1) a model of local, city-wide air pollution sources (ADMS-Urban) coupled with a GIS, and 2) estimatesof long-range air pollution from EU-scale, trans-boundary modelling. The Bayesian model was developedagainst data on monitored concentrations of PM10 from 25 sites. Comparison between model predictionsand monitored concentrations for 20 additional sites in London gave r2= 0.5-0.6. Including parameters onsite type (i.e. roadside, background), season, and day of the week was shown to improve performance. Themodel is currently being applied in an epidemiological analysis of health impacts of air pollution episodes inLondon.keywords: Exposures, Bayesian, Air pollution, Modelling, GISJohn Gulliver, University of the West of Scotland, High Street, Paisley, PA1 2BE, ScotlandE-mail address:john.gulliver@uws.ac.uk

Oral Presentation

LOOKING FOR CLIMATE CHANGE SIGNALS IN EXTREME TEMPERATURES.

Peter Guttorp1

1University of Washington and Norwegian Computing Center

In looking for climate change signals in observed data, it is often valuable to study extremes. As annualextremes by de�nition are relatively rare events one needs longt time series to evaluate trends in them. Weuse a long temperature series from Stockholm to look for expected climate signals. A related series fromUppsala is used for comparison to the �ndings from Stockholm. We use both parametric and nonparametrictools.Peter Guttorp, Box 354322 Seattle, WA 98195-4322USAE-mail address:peter@stat.washington.edu

49

Bologna, Italy TIES 2009

Poster Presentation

SPATIAL VARIATION OF FACTORS OBTAINED FROM TWOMONITORING STATIONSALONG A STREAM BY FACTOR ANALYSIS.

Serdar Göncü1, Erdem Ahmet Albek1, Ömer Güngör1

1Anadolu University

In water quality assessment studies, measurements require extensive labor force, time and cost. Studieswhich are aimed to determine water quality necessitate a large number of parameter measurements. Insteadof making numerous measurements, a smaller number of factors which represent these parameters can bemeasured so that cost and time savings can be made. This study is aimed at determining these factors. Thedatasets consist of monitoring data collected at two stations located 200 km apart from each other on thePorsuk stream which is an important watercourse in northwestern Turkey. 29 parameters are monitored inthese stations by the General Directorate of State Hydraulic Works of Turkey. The factorability of the datahas been tested using the Kaiser-Meyer-Olkin (KMO) and Barlett's test. The factors are analyzed usingprinciple component analysis. On a yearly basis, 6 potential factors have been identi�ed which explain 82%of the total variance in the �rst station and 7 potential factors have been identi�ed which explain 87% of thetotal variance in the second station. These factors express inorganic, organic and microbiologic water qualityparameters and give information about river water quality and pollution sources. The Porsuk Dam upstreamof the �rst station shows an equalization e�ect which is perceived at the �rst station factor analysis results.The second station indicates the e�ects of a major city and di�use agricultural sources which are seen in thedi�erent factors obtained in the two datasets.keywords: Factor analysis, Monitoring, Principal component analysis, Pollution, Water quality assessmentSerdar Göncü, Anadolu University Ikieylul CampusEngineering and Architecture FacultyEnvironmental Eng. Dept.26555 Eskisehir/TURKEYE-mail address:sgoncu@anadolu.edu.tr

Supporting grant: Anadolu University

Oral Presentation

EXTREME VALUE THEORY APPLIED TO THE DEFNITION OF BATHINGWATERDIS-COUNTING LIMITS.

Ruth Haggarty1, Marian Scott1, Claire Ferguson1

1University of Glasgow

The European Community Bathing Water Directive European Parliament, 2006) set compliance standardsfor bathing waters across Europe, with minimum standards for microbiological indicators to be attained atall locations by 2015. Classifcation of sites identifed as bathing waters is based on 90th and 95th percentilevalues of four-year datasets. The Directive allows up to 15% of samples a�ected by short term pollutionepisodes to be disregarded from the �gures used to classify bathing waters, provided certain managementcriteria have been met, including informing the public of short term water pollution episodes. Therefore, ascienti�cally justi�able discounting limit is required which could be used as a management tool to determinethe samples that should be removed. We have considered di�erent methods of obtaining discounting limits,with particular focus on extreme value methodology applied to data from Scottish bathing waters. Blockmaxima and threshold models were applied to data collected by the Scottish Environment Protection Agency(SEPA) between 2003 and 2007 in order to identify suitable return levels which could subsequently be appliedas discounting limits. The impact of the limits obtained on the levels of compliance achieved was used toassess he e�ectiveness of each of the methods considered. Site speci�c threshold models provided the moste�ective discounting limits, reducing the percentiles used to assess compliance.keywords: discounting, extreme, thresholdRuth Haggarty, Department of Statistics, University of Glasgow, 15 University Gardens, Glasgow, G12 8QWE-mail address:rhaggarty@stats.gla.ac.uk

50

Bologna, Italy TIES 2009

Oral Presentation

A TALE OF TWO PHASES: DESIGN AND ESTIMATION OF TREE FOLIAGE BIOMASS.

Temesgen Hailemariam1, Vicente Monleon2, Aaron Weiskittel3, Duncan Wilson1

1Oregon State University, 2USDA - PNW Station, 3The University of Maine

Conifer crowns are highly variable both within- and between-trees, particularly with respect to foliagebiomass and leaf area. A variety of sampling schemes have been utilized to estimate biomass and leafarea at the individual tree and stand scales including randomized branch sampling and importance sampling(Ecology (1995) 76: 1181-1194), two-stage systematic and ellipsoidal sampling (Can. J. For. Res. (2003): 82-95). Rarely has the e�ectiveness of these sampling schemes been compared across stands or even species. Inaddition, sample size estimates for achieving a certain level of precision have rarely been given. Using a MonteCarlo simulation study on extensive branch and tree leaf area datasets for Douglas-�r and ponderosa pine,we compared alternative sampling alternatives and examined the use of auxiliary information at the designand estimation phases. The use of auxiliary information at the design phase resulted in lower mean squareerror values than when auxiliary information was used at the estimation phase. Contrarily, using auxiliaryinformation at the estimation phase resulted in lower precision/kg than using the same at the estimationphase. For both species, systematic sampling with ratio estimation provided the highest precision/kg. InDouglas-�r, stratifying by branch type (i.e. whorl vs. interwhorl) resulted in a marginal gain in precision.On average, the root mean square error decreased by 26 and 14% when the sample size was increased from6 to 12 branches per tree and from12 to 18 branches per tree, respectively.keywords: Tree crown sampling, ratio estimation, Oregon and WashingtonTemesgen Hailemariam, hailemariam.temesgen@oregonstate.eduE-mail address:hailemariam.temesgen@oregonstate.edu

Oral Presentation

STATISTICAL METHODS IN THE RECONSTRUCTION OF PALAEOCLIMATE.

John Haslett1

1Trinity College Dublin

The understanding of climate change is enriched by study of past climate; the 2007 IPCC report devoted anentire chapter to palaeoclimates. As the instrumental record is only about 150 years long, the main challengeis uncertainty. Knowledge is indirect and rests on (a) proxy data (typically multivariate counts) from longenvironmental archives such as lake sediments and ice deposits; (b) computer models of the climate systemforced by, for example, information on past values of CO2; and latterly (c) models of climate/vegetationinteraction. Jointly and separately these permit limited forms of conditional statistical inference about theuncertain past climate, given data and models. These latter include both physical models of the energytransfers in climate and statistical models, calibrated on the observed relationships between proxy data andclimate in the modern world.Atmospheric climate - C(t,s) - is a multivariate space time process associatewith time t and location s; it is spatially and temporally coherent as it must satisfy laws of physics. Howeverancient proxy data are located at ill-de�ned times, for typically the age of a sample must be inferred fromits depth in sediment; spatial resolution is better de�ned. Ideally inference for an entire space-time region isbased on many cores covering di�erent proxies at di�erent depths at many spatial locations in that region.Thetechnical focus of this introductory talk will be on the separate marginal inference for C(t,s) correspondingto a single proxy sample.John Haslett, School of Compuet Science and Statistics, Trinity College Dublin 2. IrelandE-mail address:jhaslett@tcd.ie

Supporting grant: Science Foundation Ireland grants 04/BR/M0049, 05/RFP/MAT044, 07/RFP/MATF164

51

Bologna, Italy TIES 2009

Oral Presentation

PROJECTIONS OF FUTURE INSURANCE LOSSES FROM CLIMATE MODEL DATA.

Ola Haug1

1Norwegian Computing Center

The anticipation of substantial future climate change gives rise to increased focus on weather related risksfrom the insurance industry. The vulnerability of life as well as non-life products is a�ected, and knowledgeof future loss levels is valuable. Most apparently, risk and premium calculations may be updated accordingly,but equally cost-e�ective is perhaps the communicating of dedicated loss-preventive measures to customersand regulators. We have established statistical claims models for the coherence between externally in�ictedwater damage to private buildings in Norway and selected meteorological variables. Based on these modelsand downscaled climate model predictions with two di�erent CO2 emissions scenarios, projected loss levelsof a future scenario period (2071-2100) are compared to those of a control period (1961-1990). Inherentlyinvolved in the loss predictions are uncertainties introduced from i) imprecise climate model scenario data,ii) claims models misspeci�cation and iii) �tting the claims models to limited amounts of data. Among theseerror terms, the estimation uncertainty (iii) is the only component that is quanti�able. Disregarding the�rst two components, our analyses identify an incontestable, but moderate increase in the losses. In mostareas, the claim projections do not di�er between the emissions scenarios. As a spin-o�, spatial variations invulnerability are recognized as well.keywords: Building water damage, Weather observations, Claims models, Climate model data, PredictionOla Haug, P.O.Box 114, BlindernN-0314 OsloNorwayE-mail address:ola.haug@nr.no

Supporting grant: Statistics for Innovation, (s�)2

Oral Presentation

SPATIO-TEMPORAL DISEASE MAPPING USING INLA.

Leonhard Held1, Birgit Schrödle1

1University of Zurich

Integrated nested Laplace approximations (INLA) have been recentlyproposed for approximate Bayesianinference in latent Gaussian models(Rue, Martino and Chopin, 2009, JRSSB). The INLA approach isappli-cable to a wide range of commonly used statistical models, includingmodels for spatial and spatio-temporaldisease mapping.In this talk I will �rst review the INLA methodology and contrast itwith more establishedinference approaches such as Markov chain MonteCarlo (MCMC). In the second part of the talk I will illus-trate howparametric (Bernardinelli et al., Stats in Med, 1995) andnonparametric (Knorr-Held, Stats in Med,2000) models forspatio-temporal disease mapping can be �tted using INLA. I will alsodiscuss how the INLAapproach can be used for model assessment andmodel comparison based on leave-one-out cross-validation.Themethodology will be applied to case reporting data on BVD (bovineviral diarrhoe) and Salmonellosis incattle provided by the Swissfederal veterinary o�ce.keywords: Disease mapping, Spatio-temporal models, INLA, Leave-one-out cross-validationLeonhard Held, University of ZurichInstitute of Social and Preventive Medicine Biostatistics UnitHirschengraben848001 ZurichSwitzerlandE-mail address:held@ifspm.uzh.ch

Supporting grant: Support by the Swiss Veterinary O�ce is gratefully acknowledged.

52

Bologna, Italy TIES 2009

Oral Presentation

COMPARATIVE ANALYSIS OF MODEL BEHAVIOUR FOR FLOOD PREDICTION PUR-POSES USING SELF-ORGANIZING MAPS.

Marcus Herbst1, Markus C. Casper2, Jens Grundmann3, Oliver Buchholz4

1Economic and Social Statistic Dept. and Centre for Regional and Environmental Statistics, University ofTrier, 2Department of Physical Geography, University of Trier, Germany, 3Department of Hydrology andMeteorology, University of Dresden, Germany, 4Hydrotec GmbH, Aachen, Germany

Distributed watershed models constitute a key component in �ood forecasting systems. It is widelyrecognized that models because of their structural di�erences have varying capabilities of capturing di�erentaspects of the system behaviour equally well. Of course, this also applies to the reproduction of peakdischarges by a simulation model which is of particular interest regarding the �ood forecasting problem.Inour study we use a Self-Organizing Map (SOM; Kohonen [Self-Organizing Maps, 3rd ed., Springer, 2001,501 pp.]) in order to examine the conditions under which three di�erent distributed watershed models arecapable of reproducing �ood events present in the calibration data. The SOM helps to analyze speci�cindices calculated on model output time series which are obtained from Monte-Carlo simulations. The SOMproduces a discretized mapping of these index distributions on a two dimensional plane such that theirpattern and consequently the patterns of model behaviour can be conveyed in a comprehensive manner. Itis demonstrated how the SOM provides useful information about details of model behaviour and also helpsidentifying the model parameters that are relevant for the reproduction of peak discharges and thus for �oodprediction problems. The most prominent advantage of using SOM in the context of model analysis is that itallows to comparatively evaluating the data from two or more models. Our results highlight the individualityof the model realizations in terms of the index measures.keywords: Self-Organizing Maps, model evaluationMarcus Herbst, Universität Trier Campus II - Behringstrasse 54286 TrierE-mail address:herbstm@uni-trier.de

Oral Presentation

BAYESIAN SCALE SPACE ANALYSIS WITH APPLICATION TO REMOTE SENSINGAND CLIMATE MODELING.

Lasse Holmström1, Leena Pasanen1

1University of Oulu

The idea of a scale space has its origin in computer vision where it refers to a family of smooths of a digitalimage. No particular level of smoothing is regarded as "correct" and each smooth is thought to provide infor-mation about the object of the image at a particular scale, little smoothing revealing small details and heavysmoothing displaying only the coarsest features. Scale space analysis was only relatively recently introducedto Statistics by P. Chaudhuri and J.S. Marron in the form of SiZer methodology where the goal is to makeinferences about scale-dependent features of curves and images from noisy observations.The talk describesiBSiZer, a Bayesian version of SiZer that can be used to �nd credible, scale-dependent di�erences betweentwo digital images or two random �elds. The advantages of the Bayesian approach include straightforwardsimulation-based inference, �exible modeling and the possibility to incorporate relevant prior information inthe analyses. As examples we discuss scale space analyses of pairs of Landsat satellite images as well as atentative application to the interpretation of climate model outputs.keywords: Bayesian SiZer, satellite images, climate modelingLasse Holmström, Department of Mathematical Sciences P.O.B. 300090014 University of Oulu FINLANDE-mail address:lasse.holmstrom@oulu.�

53

Bologna, Italy TIES 2009

Oral Presentation

DOES ENVIRONMENTAL PERFORMANCE AFFECT FINANCIAL PERFORMANCE? AMETA-ANALYSIS.

Eva Horvathova1

1Charles University in Prague

What do we know about the impact of environmental regulations on �rm performance? After more thanthree decades of theoretical as well as empirical research, the results seem to be still inconclusive. Somepapers suggest that the regulations harm �rms, others claim that regulations may contribute positively andgive an impetus to innovations e.g. Palmer et al. [J.Econ.Pers. (1995):119-132], Porter [SciAm. (1991):96].Therefore, I examine the heterogeneity in �nancial environmental nexus empirically performing a meta-analysis of 57 outcomes from 36 empirical studies to uncover the underlying factors, such as econometricmethods, geographical area/legal systems, time-coverage, type of �nancial variables used that can in�uencethe observed variation in the empirical results. The results suggest that empirical method used matters forthe nexus and that the likelihood of �nding a negative link between environmental and �nancial performancesigni�cantly increases using simple correlation coe�cients instead of more advanced econometric analysis (i.e.a sign of omitted variable bias). The results indicate that the portfolio studies tend to report a negativelink between environmental and �nancial performance. This likely re�ects the omitted factors. The positivelink between environmental and �nancial performance is found more frequently in the common law countriesrather than in civil law countries. The results also point to the importance of appropriate time coverage inorder to establish a positive link between environmental and �nancial performance. This suggests that ittakes time until the environmental regulation materializes in �nancial performance.keywords: meta-analysis, environmental performance, �nancial performanceEva Horvathova, Eva Horvathova Boleslavova 18140 00 Praha 4 Czech RepublicE-mail address:ev.horvathova@gmail.com

Supporting grant: GACR 402/09/2049

Oral Presentation

RECONSTRUCTINGHOURLY PM10 GRAVIMETRICMEASUREMENTS THROUGH TRANS-FER FUNCTION MODELS.

Daniele Imparato1, Mauro Gasparini1

1Department of Mathematics - Politecnico di Torino

Several di�erent technologies to measure particulate matter PM10 have beenintroduced in the last fewyears: gravimetric(low volume and high volume), TEOM, beta and nephelometric. As of today, gravimetriclow volume measures are the only o�cial data that local administrations are legally allowed to refer to forPM10. However, because of technical reasons, only daily measures are available with gravimetric tools,whereas for other technologies, such as TEOM, hourly data of PM10 can be obtained. Because of the highcorrelation between these di�erent measures, more detailed information on PM10 gravimetric levels can bereconstructed through suitable statistical modeling. In this work the whole PM10 gravimetric measure timeseries is reconstructed at hourly frequency from TEOM hourly measures. For this purpose, we discuss transferfunction models where the input and the output time series are, respectively, PM10 TEOM and gravimetricmeasures collected by the same monitoring station. Before we can identify and estimate the model, we needto reconstruct missing values for both time series via joint use of di�erent techniques such as Kalman �lterand time series multiple regression. The �nal results are expected to increase the actual forecasting accuracyof PM10 gravimetric levels in Piemonte.keywords: PM10 gravimetric measurements, transfer function models, time series reconstructionDaniele Imparato, corso Duca degli Abruzzi, 24 - 10129 TorinoE-mail address:daniele.imparato@polito.it

Supporting grant: Regione Piemonte, Research Project CIPE 2004: �Rilevamento di inquinanti in diverse matrici ambientali´´

54

Bologna, Italy TIES 2009

Oral Presentation

APPLICATIONS OF GREY RELATIONAL METHOD TO RIVER ENVIRONMENT QUAL-ITY EVALUATION IN CHINA.

Wai Cheung Ip1, Baoqing Hu2, Heung Wong1, Jun Xia3

1The Hong Kong Polytechnic University, 2Wuhan University, China, 3Chinese Academy of Sciences

Inexact data, short sample and incomplete hydrological data in water research are very commonly encoun-tered. Basing on the grey relational method, we introduce a new relational degree formula to water qualityevaluation and to demonstrate how it can be applied to evaluate river water quality of the Wuhan segmentof Hang Jiang River, a major branch river in China, for the years 1992 to 1998. As available water qualitydata are largely vague and imprecise existing methods would produce very coarse results. Our proposedmethod is an improvement over both the fuzzy set theory and the rough set. It possesses some essential anddesirable properties which ensure that translational properties do not exist and it will give a more preciseand �ner grading of the overall water quality than existing methods. Empirical applications of our proposedmethod are seen to produce results that broadly agree to those by existing methods. The �ner grading ofthe water quality produced enables us to easier visualize year-to-year �uctuations and to make year-to-yearcomparisons of the water quality. Our numerical results reveal that the water quality of the Wuhan segmentdid not experienced signi�cant changes between 1992 and 1997 and the ample-�ow period had better waterquality than the moderate-�ow period which in turn was better than the low-�ow period. The only exceptionis perhaps in 1996 when the water quality stayed at roughly same level for all �ows.keywords: Water quality index, water quality evaluation, grey relational method, triangle relational degreeWai Cheung Ip, Department of Applied Mathematics, The Hong Kong Polytechnic University, Hung Hom, Kowloon,Hong KongE-mail address:mathipwc@polyu.edu.hk

Oral Presentation

EARTHQUAKES FROM THE INDONESIAN REGION: AN APPLICATION OF EXACTCOMPUTABLE EXPRESSIONS FOR THE ASYMPTOTIC DISTRIBUTION OF CHANGE-POINT MLE IN THE EXPONENTIAL CASE.

Venkata K. Jandhyala1, Stergios B. Fotopoulos1, Elena Khapalova1

1Washington State University

Adapting Hinkley's set up of an abrupt change, we derive explicit computable expressions for the as-ymptotic distribution of the maximum likelihood estimate of an unknown change-point in a sequence ofindependently and exponentially distributed random variables. The solution we provide ends decades ofsearch for exact computable expressions for the distribution of the change-point mle, for the case of exponen-tial sequences. The derived method can be applied equivalently to the change-point mle in a Poisson processwhere one observes the time occurrences of each Poison event. Through the application of change-pointanalysis to data on earthquakes from the Indonesian region, we �nd evidence to support the hypothesis ofdynamic triggering mechanism for seismic faults.keywords: Maximum likelihood, Random walk with negative drift, Ladder height, change-point, dynamic triggeringVenkata K. Jandhyala, Department of Statistics Washington State University Pullman WA 99164-3144USAE-mail address:jandhyala@wsu.edu

Supporting grant: United States National Science Foundation Grant Number DMS-0806133

Oral Presentation

A TEST FOR CHANGE IN MEAN OF RANDOM VECTORS WITH APPLICATION TOTEMPERATURE SERIES.

Daniela Jaruskova1

1Czech Technical University, Prague

At the previousa TIES meetings we showed how to apply change point methods to detect changes in theannual mean or annual maximal and minimal temperature. This time we would like to deal with methodsfor detecting changes in behavior of temperature during a year that may be represented by a vector of dailyor monthly means. A statistic that characterizes the di�erence in the behavior "before and after a changepoint" may be the euclidean distance of the vectors of the means of the �rst and second part of the series.

55

Bologna, Italy TIES 2009

For testing whether a change occured the maximum of these statistics computed for all possible splittingmay serve as a test statistic. We study its limit distribution taking into account the dependence betweencoordinates of observed random vectors. We also apply the procedure to chosen temperature series.keywords: change point methods, random vectors with dependent components, max-type test statistic, limit distri-butionDaniela Jaruskova, Dept. of Mathematics, Faculty of Civil Engineering, Czech Technical University, Thakutova 7, CZ166 29 Prague 6, Czech RepublicE-mail address:jarus@mat.fsv.cvut.cz

Oral Presentation

SPATIO TEMPORAL DATA MODELING IN ENVIRONMENTAL SCIENCES A REVIEW..

Giovanna Jona Lasinio1

1DSPSA University of Rome La Sapienza

In the last decade, an ongoing change in the perception of the role of data and models in environmental andecological statistics is becoming manifest. At the heart of the matter is how we study complex environmentalprocesses. Though the traditional empirical strategy in these �elds made use of designed experiments, a hugespread of di�erent kinds of data was seen in the last ten years. Away from a disaggregated examinationof process components, there is an increasing shift in thinking to an integrated understanding of processbehavior. A larger scale view of the process, introducing all of our knowledge -theoretical, physical, chemicaland empirical- is becoming both more appealing and necessary. The way forward seems to be throughhierarchical and other complex modeling approaches pooling information from various sources and includingthe de�nition of complex variability structures, multi-step procedures involving exploratory tools, missingvalue imputation and data modeling. The most challenging framework where all this is taking place is theinvestigation of processes that have spatial and/or spatio-temporal structure that are typical of this inter-science topic. In this talk I'll try to outline the main results and proposals presented in the recent literatureand the GRASPA members contributes to the development of this lines of thoughts in terms of both ideasand applications to speci�c problems.keywords: space-time models, environmental sciencesGiovanna Jona Lasinio, DSPSA University of Rome La Sapienza, P.le Aldo Moro 5, 00185 Rome, ItalyE-mail address:giovanna.jonalasinio@uniroma1.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

CHANGEPOINT DETECTION IN AUTOCORRELATED TIME SERIES.

Laimonis Kavalieris1

1University of Otago

The existence of changepoints in climatological and econometric time series has received extensive attentionin the literature. Of particular interest is the possibility of confusing long memory with structural changessuch as abrupt shifts in the mean or slope, Diebold and Inoue [J. Econometrics 105 (2001): 131-159], Mills[J. Roy. Statist. Soc.(A) 170 (2007): 83-94]. Alternatively focus may be on the detection and estimationof undocumented changepoints in historical time series of temperature and precipitation, Lund, et. al. [J.Climate 20 (2007): 5178-5190].The approach is often based on a sequence of hypotheses tests.In this paper weare concerned with the estimation of the number of structural breaks in a time series using penalty functioncriteria. Such criteria include the familiar AIC and BIC, as well as procedures based on the MinimumDescription Length principle due to Rissanen [Ann. Statist. 14 (1986) 1080�1100]. We will present somelarge sample theory in order to justify the use of penalty function criteria. Application of our approach isillustrated using examples from dendrochronology, historical instrumental records of climate, and long termreconstructions of global temperatures.Laimonis Kavalieris, Department of Mathematics and Statistics, University of Otago,Dunedin,New ZealandE-mail address:lkavalieris@maths.otago.ac.nz

56

Bologna, Italy TIES 2009

Oral Presentation

EFFECT OF TROPOSPHERIC OZONE ON TOBACCO PLANTS IN VARIOUS EXPO-SURE SERIES DURING GROWING SEASONS.

Dariusz Kayzer1, Klaudia Borowiak2, Anna Budka1, Janina Zbierska2

1Poznan University of Life Sciences, Department of Mathematical and Statistical Methods, 2PoznanUniversity of Life Sciences, Department of Ecology and Environmental Protection

Tropospheric ozone is created during photochemical reaction of primary air pollutants emitted mainlyfrom mobile sources. The number of cars has been increasing in recent times, so far all scenarios predictincrease of tropospheric ozone concentration during next 50 years. Ground level ozone is very phytotoxic airpollutant, which could cause visible injuries and losses of yield crop and other plants. Tobacco plants and itsBel W3 cultivar has been selected as a very sensitive for tropospheric ozone. It has been used in biomonitoringprograms all over the world. At our experiment we used this cultivar to bioindication experiment at nine sitesin Poznan city and one at rural area. Plants were exposed to ambient air during nine exposure series (fromthe end of May to the end of September) in three growing seasons. Visible leaf injury (necrosis) were assessedafter each exposure series. The results obtained have been presented with the use multivariate analysis ofvariance [1]. The aim of this paper was to investigate di�erences of leaf injury levels between expositionseries. Some regression method for studying the series*environment interactions were considered.[1] CalinskiT., Czajka S., Kaczmarek Z. (1987); A model for the analysis of a series of experiments repeated at severalplaces over a period of years. Biuletyn Oceny Odmian Cultivar Testing Bull. 12, Nos. 17-18, 7-71.keywords: tropospheric ozone, bioindication, tobacco plants, multivariate analysis of varianceDariusz Kayzer, Poznan University of Life Sciences, Department of Mathematical and Statistical Methods, WojskaPolskiego 28, 60-637 Poznan, PolandE-mail address:dkayzer@up.poznan.pl

Oral Presentation

DESIGN OF ANTARCTIC CIRCUMPOLAR WHALE RESEARCH CRUISES: A MODEL-BASED APPROACH.

Natalie Kelly1, David Peel1, Mark Bravington1, Sharon Hedley2

1CSIRO, 2Consultant Statistician, St Andrews, Scotland

The usual overarching aim in designing any experiment is to minimise expected uncertainties in estimatesof model parameters. Model-based methods of designing experiments use models of the underlying systemto guide placement of e�ort or samples, s, so as to minimise the expected prediction error conditional ons. We propose a simple framework to test various survey designs for estimating animal abundance wherethe underlying system is modelled using generalized additive models (GAMs). Our case study is a decade-long series of Antarctic circumpolar line-transect surveys undertaken to estimate Antarctic minke whaleabundance. From these data, we produced two linked spatial GAMs: one for mean school size and onefor school density. Although the GAMs are much simpler than would be used for a proper model-basedabundance estimate, they do capture the main sources of uncertainty, and are straightforward to work with.We used them to estimate how the precision of estimates for future surveys would vary in response to surveydesign features such as track length and con�guration, latitudinal distribution of e�ort, total amount oflongitude traversed in a single summer and whale spotting methods. This exercise identi�ed some of theimportant sources of uncertainty in model-based estimates of total minke whale abundance which will aid inproducing more e�cient survey designs in the future.keywords: model-based design, line transect surveys, generalized additive models (GAMs), Antarctic minke whales,animal abundanceNatalie Kelly, C/- Australian Antarctic Division, 203 Channel Highway, Kingston,Tasmania, 7050, AustraliaE-mail address:natalie.kelly@csiro.au

57

Bologna, Italy TIES 2009

Oral Presentation

RECORDS METHOD FOR THE NATURAL DISASTERS APPLICATION TO THE STORMEVENTS.

Zaher Khraibani1, Hussein M. Badran2, Hussein Khraibani3

1Université Paris 10, 2Lebanese University, Department of Probability and Statistics, 3Laboratoire Centraldes Ponts et Chaussées, Nantes

The goal of this study is to introduce a new approach for testing if the �rst observed events are sporadic (theinterarrival times between two successive events are independent identically distributed). For that we use anonparametric test based on the record process in particularly the number of record among n variables, thosevariables are represented by the waiting time between two successive cases. The distribution of the numberof records is particularly robust since it is independent of the cumulative distribution of the observations andit is exactly calculated for each value of n. We illustrate this statistic by testing on the �rst observations ofa simulation then we apply in real data of natural disasters as storms in France.keywords: Statistical inference, Number of records, Sporadic, StormZaher Khraibani, Université Paris 10, equipe ModalX, 200 avenue de la République, 92001 Nanterre Cedex, FranceE-mail address:zaher.khraibani@gmail.com

Oral Presentation

COMPARING PIECEWISE LINEAR TRENDS.

Hyune-Ju Kim1, Jun Luo2, Michael Barrett2, Eric Feuer3

1Syracuse University, 2Information Management Services, Inc., 3National Cancer Institute

A Joinpoint regression model, a piecewise linear regression model with unknown change-points, has beensuccessfully applied to cancer trend data and Joinpoint software available athttp://srab.cancer.gov/joinpoint/index.html helps one to select the number of change-points and to esti-mate the model parameters.Our interest in this talk is in comparing such trend data. We �rst present apermutation test to compare two groups of trend data to determine if the two groups share the same slopesor if the mean functions are identical. We then propose two methods to divide groups of trend data intoseveral clusters with common characteristics. The �rst method uses the p-values from pairwise comparabilitytests and incorporates an adjustment proposed in literature to improve the overall power of the multiplecomparison. The second method is based on an exhaustive clustering, for which we discuss how to deter-mine initial conditions and suggest ways to improve computational e�ciency. Finally we will illustrate thesemethods with cancer trends and water quality trends.Hyune-Ju Kim, Department of Mathematics 215 Carnegie Building Syracuse University Syracuse, New York 13244,U.S.A.E-mail address:hjkim@syr.edu

Poster Presentation

IMPACT OF SEASONAL CHANGES OF THE RIVER WATER ON DRINKING WATERQUALITY.

Ivan Klozyatnyk1, Natalia Klymenko1

1Institute of Colloid Chemistry and Chemistry of Water, National Academy of Sciences of Ukraine

The cascade of the Dniper storeges features a high content of natural organic matter (NOM) and itssigni�cant �uctuations depending on seasonal phenomena.It is shown that data on the origin of NOM inwater can be obtained by comparing the ratio of color and oxidizability. A higher value of this ratio indicatesa predominant content in the water of stable humic substances of marsh origin. The ratio of humic compoundsof marsh or plankton origin depends on seasonal or annual climatic conditions. Humic substances of planktonorigin prevail in the period of �oods, while humic substances of plankton-arid periods of the year. For example,the lowest value of color-to-oxidizability ratio of 2.5-3 was registered in the dry 2002. This fact indicated apredominant content in the water of humic matter of plankton origin. A sharp rise in the ratio during certainperiods of 2001 (up to 5.2) and 2003 (up to 5.0) is the evidence of the growing content of humic matter of

58

Bologna, Italy TIES 2009

marsh origin. This statement is corroborated by correlation of these indicators with the rise of ammoniacontent and the reduction of oxygen content in the water. It is related to the intensive summer rains duringthis period of the year in the upstream region of the Dnieper River.keywords: Water storage, seasonal changes, drinking waterIvan Klozyatnyk, 42 Vernadsky Avenue, Kyiv 03680, UkraineE-mail address:koziatnik@ukr.net

Oral Presentation

DEALING WITH UNCERTAINTIES IN ENVIRONMENTAL BURDEN OF DISEASE AS-SESSMENT.

Anne Knol1, Arthur Petersen2, Jeroen van der Sluijs3

1RIVM, 2PBL, 3Utrecht University

Disability Adjusted Life Years (DALYs) combine the number of people a�ected by disease or mortality ina population and the duration and severity of their condition into one number. The environmental burdenof disease is the number of DALYs that can be attributed to environmental factors. Environmental burdenof disease estimates enable policy makers to evaluate, compare and prioritize dissimilar environmental healthproblems or interventions. These estimates often have various uncertainties and assumptions which arenot always made explicit. Besides statistical uncertainty in input data and parameters which is commonlyaddressed a variety of other types of uncertainties may substantially in�uence the assessments results. Wehave reviewed how di�erent types of uncertainties a�ect environmental burden of disease assessments, and wegive suggestions as to how researchers could address these uncertainties. We propose the use of an uncertaintytypology to identify and characterize uncertainties. Finally, we argue that uncertainties need to be identi�ed,assessed, reported and interpreted in order for assessment results to adequately support decision making.keywords: Burden of disease, uncertainty, DALY, environment, healthAnne Knol, anne.knol@rivm.nlE-mail address:anne.knol@rivm.nl

Oral Presentation

SPATIAL MODEL OF PERSISTENT ORGANIC POLLUTANTS VOLATILIZATION FROMSOIL: CONNECTION OF DETERMINISTIC AND STOCHASTIC APPROACH.

Jiri Komprda1, Klara Kubosova1, Milan Sanka1, Ondrej Hajek2, Ivan Holoubek1

1RECETOX (Research Centre for Environmental Chemistry and Toxicology), Masaryk University,2Department of Botany and Zoology, Faculty of Science, Masaryk University

Persistent organic pollutants (POPs) have been produced for various agricultural and industrial purposesduring last decades. These compounds are still present in the environment, particularly in soil, in considerableamount. One of the important POPs properties is semivolatility. When atmospheric concentrations of POPsdecreased to low level as a result of restrictions, volatilization from soil has become their signi�cant or evendominant emission source to the environment. The goal of this study was to calculate volatilization �uxes ofselected POPs (HCB, PCB) from the area of the Czech Republic during years 2006-7. Stochastic model basedon statistical analysis (regression trees) of available data from monitoring in combination with environmentalparameters (organic carbon content, land cover, land use...) was used for prediction of the POPs concentrationmap in the area of the CZ. Main sources of monitoring data were Basal monitoring of agricultural soilsconducted by CISTA (Central Institute for Supervising and Testing in Agriculture), Basal monitoring ofsoils in protected areas conducted by ANCLP (Agency for Nature Conservation and Landscape Protection),projects conducted by RECETOX (Research Centre for Environmental Chemistry and Ecotoxicology) invarious spatial scales and the Czech Hydrometeorological Institute as a source of temperature data. Spatiallyresolved deterministic model based on fugacity approach was created for calculation of volatilization �uxesand total amount of pollutants being volatilized was predicted. E�ects of environmental temperature changesand variation of organic carbon content in soil were included in the model.keywords: POPs volatilization , soil, regression treeJiri Komprda, Faculty of Science MU, Kotlarska 2, Brno 60200, Czech RepublicE-mail address:komprda@iba.muni.cz

Supporting grant: This research was supported by the Czech Ministry of Education, Youth and Sport (MSMT 0021622412).

59

Bologna, Italy TIES 2009

Oral Presentation

THE CAUSES OF CLIMATE CHANGE.

Milena Ková°ová1

1University of South Bohemia,Institute of Physical Biology

Drought and �oods, low predictability of weather, extremes in temperature and precipitation, destructivewinds in formerly calm areas have been frequent reported at present time. Global climate change, and inparticularly global warming is mostly explained as a consequence of an increased concentration of greenhousegases. A new view to the cause of climate change will be presented.Milena Kovárová, University of South Bohemia, Institute of Physical Biology, Zámek 136, 373 33 Nové Hrady, CzechRepublicE-mail address:kovarova@greentech.cz

Poster Presentation

SOME BLOCK DESIGNS WITH NESTED ROWS AND COLUMNS FOR RESEARCH ONPESTICIDE DOSE LIMITATION.

Maria Kozªowska1, Agnieszka �acka1, Roman Krawczyk2, Radosªaw J. Kozªowski3

1Department of Mathematical and Statistical Methods, Poznan University of Life Sciences, 2Institute ofPlant Protection-NRI, 3Institute of Agricultural Engineering, Department of Applied IT, Poznan Universityof Life Sciences

In 1979, Singh and Day de�ned a block design with nested rows and columns (NRC in short). In this paperwe take the view, common among statisticians, that principles for planning of factorial experiments and near-factorial experiments in NRC designs are not the same. A group divisible NRC design for factorial experimentis considered. Some constructions of the designs are given. There are necessary and su�cient conditionsformulated and proven for group divisible NRC design to be a C-design. For near-factorial experiments,when there are a levels of experimental factor A and b levels of experimental factor B and there is onecontrol treatment added, we proposed a method for constructing NRC designs. Plant protection experimenton limitation of pesticide dose is given to show how the obtained results can be applied.keywords: block design with nested rows and columns, group divisible design, C-design, near-factorial experiments,pesticide doseMaria Kozlowska, Department of Mathematical and Statistical Methods, Poznan University of Life Sciences, WojskaPolskiego 28, 60-637 Poznan, PolandE-mail address:markoz@up.poznan.pl

Poster Presentation

ANALYSIS OF STREAM MACROINVERTEBRATES RESPONSE TO ENVIRONMENTALCONDITIONS: RESEARCH SUPPORT OF THE WATER FRAMEWORK DIRECTIVE IM-PLEMENTATION IN THE CZECH REPUBLIC.

Klara Kubosova1, Jiri Jarkovsky1, Karel Brabec1, Svetlana Zahradkova2, Jindriska Bojkova2, PavelBartusek2

1RECETOX (Research Centre for Environmental Chemistry and Toxicology), Masaryk University,2Department of Botany and Zoology, Faculty of Science, Masaryk University

A suitably designed monitoring network and a reasonable selection of indicative parameters are essentialcomponents of the standardized assessment of the surface waters ecological status. In order to achieve this,it is necessary to assess the ecological characteristics and indicative power of taxa that might be used for theneeds of the ecological conditions assessment. Macroinvertebrate communities have wide indicative potentialfor assessment of various impairments in �uvial ecosystems. A standardized database of macroinvertebratesamples from the Czech Republic was used for this purpose. Suitable organisms that meet certain criteria(su�cient occurrence, consistent relationship to environmental conditions, and sensitivity to stressors) wereselected. Analytical methods designed for such topics represent a relatively wide range of approaches fromsimple logistic regression (and other models of the GLM - Generalized linear models) through the GAM(Generalized additive models), HOF (Huisman-Ol�-Fresco models), ISA (Indicator species analysis) or IR(Index of representation), to nonparametric techniques such as decision trees and forests. We used thesemethods in relation to the distribution characteristics of individual variables. We also aimed to comparestandard and innovative techniques for the analysis of taxa preference and tolerance to the environmental

60

Bologna, Italy TIES 2009

parameters, particularly in terms of their statistical assumptions and complexity. The outputs of theseanalyses are: (1) Ecological characteristics and indicative power of relevant taxa usable in the monitoring ofthe surface waters ecological status, (2) Comparison of environmental preference pro�les of macroinvertebratetaxa at pristine and degraded sites.keywords: macroinvertebrates, environmental preference, impact responseKlara Kubosova, Kotlarska 2, Brno 60200, Czech RepublicE-mail address:kubosova@iba.muni.cz

Supporting grant: This research was supported by the Czech Ministry of Environment (VaV MZP SP2e75008) and the CzechMinistry of Education, Youth and Sport (MSMT 0021622412).

Oral Presentation

INCORPORATING UNCERTAINTY INTO GULLY EROSION CALCULATIONS: A RAN-DOM FOREST APPROACH.

Petra Kuhnert1, Anne Kinsey-Henderson2, Rebecca Bartley2, Alexander Herr3

1CSIRO Mathematical and Information Sciences, 2CSIRO Land and Water, 3CSIRO SustainableEcosystems

In Australia, studies performed on various catchments (e.g. Burdekin) that drain directly into the GreatBarrier Reef (GBR) lagoon have indicated that gully erosion is a major contributor to sediment loss. Gullyerosion represents a major input into deterministic sediment budget models and is computed from gullydensity measurements, where gully density represents the total length of gullies per square kilometre in aregion. Despite calculations of gully erosion and associated sediment loss being reported for the Burdekin,there has been a lack of emphasis on the quanti�cation of the prediction uncertainty around these calculations.Furthermore, since the calculations are based on a very small proportion of mapped gully sites (0.88%) it isparamount that an investigation into the uncertainty around the gully erosion calculations takes place.Weuse Random Forests to predict gully density from mapped gullies using a suite of environmental variables.This model builds an ensemble of regression trees using bootstrap samples of the data, from which gullydensity predictions can be obtained by averaging predictions from the bootstrap trees. Using the bootstrappredictions, variances and covariances from this model, we calculate gully erosion and its correspondinguncertainty. Results show that predictions from the Random Forest model provided greater accuracy thanprevious approaches in terms of the mean square error and spearman rank correlation. We found that wherethe uncertainty was high the quality of the mapped data was generally poor but also found areas with "good"gully mapping that still had high prediction uncertainty.keywords: Bootstrap, Dependence, Ensemble methods, Gully Density, Prediction UncertaintyPetra Kuhnert, CMIS, CSIRO Laboratories, 233 Middle Street, Cleveland QLD 4163 AustraliaE-mail address:Petra.Kuhnert@csiro.au

Supporting grant: This work was undertaken through the Water for a Healthy Country Flagship as part of the Reef CatchmentFutures program.

Poster Presentation

ESTIMATING ABUNDANCE OF PELAGIC FISHES USING GILLNET CATCH DATA INDATA-LIMITED FISHERIES: A BAYESIAN APPROACH.

Petra Kuhnert1, Shane Gri�ths2, William Venables1, Stephen Blaber2

1CSIRO Mathematical and Information Sciences, 2CSIRO Marine and Atmospheric Research

We describe a Bayesian modelling approach to estimate abundance, and biomass, of pelagic �shes fromgillnet catches in data-limited situations. By making a number of simple assumptions, we use �sh sustainedswimming speed to calculate the e�ective area �shed by a gillnet in a speci�ed soak time in order to estimateabundance (�sh km-2) from the number of �sh caught. We used catch data from various sampling methodsin northern Australia and elicited anecdotal information from experts to build a size distribution of the truepopulation in order to compensate for size classes that were unlikely to be represented in the catch due tosize selectivity of the gear. Our �nal abundance estimates for various sized tunas and mackerels (0.04 to 4.17�sh km-2) and bill�shes (0.004 to 0.005 �sh per km-2) were similar to what has been estimated for tropicalecosystems elsewhere in data-rich situations. The model is particularly useful in data-limited situations whereabundance or biomass estimates are required for pelagic �sh species of low economic importance. These data

61

Bologna, Italy TIES 2009

are often required for ecosystem models (e.g. Ecopath) which are increasingly being used as tools worldwidefor ecosystem-based �sheries management. This work has been recently published in the Canadian Journalof Fisheries and Aquatic Sciences.keywords: Elicitation, Expert Opinion, Prior Distribution, SelectivityPetra Kuhnert, CMIS, CSIRO Laboratories, 233 Middle Street, Cleveland QLD 4163 AustraliaE-mail address:Petra.Kuhnert@csiro.au

Oral Presentation

MULTIPLE IMPUTATION OF INCOMPLETE OCEANOGRAPHIC LINEAR-CIRCULARDATA USING MULTIVARIATE MIXTURE MODELS.

Francesco Lagona1, Marco Picone1

1DIPES - University of Roma Tre

Oceanographic databases include observations from linear and circular variables such as signi�cant waveheight and direction. Typically provided by buoys networks, these data are often missing because of discon-tinuous device functioning or broadcasting interruption. We propose latent class analysis for the multipleimputation of these data, by using a battery of nonparametric bootstrap samples, drawn from the incompletedatabase. Each sample is modelled by a mixture of multivariate, parametric circular-linear distributions.A conditional independence assumption allows each multivariate mixture component to be speci�ed by theproduct of univariate parametric distributions. For each bootstrap sample, we hence obtain a predictivedistribution of the missing values given the observed data. Being based on bootstrap samples, imputationsobtained by these predictive distributions re�ect the uncertainty about parameter values.The methodology isillustrated on missing values of signi�cant wave height, wave direction, wind speed and wind direction in theItalian Wave Metric Network Database, by using a mixture of multivariate distributions that are speci�edby the product of two Von Mises and two Gamma distributions.keywords: linear-circular data, mixture models, multiple imputation, nonparametric bootstrapFrancesco Lagona, DIPES - University Roma TreVia G. Chiabrera, 19900145E-mail address:lagona@uniroma3.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

A CRITICAL LOOK AT THE FULFILLMENT OF BASIC ASSUMPTIONS FOR THE AP-PLICATION OF TWO COMMON RECEPTOR MODELS CMB AND PMF FOR SOURCEAPPORTIONMENT OF PM10.

Bo R. Larsen1

1EU Joint Research Centre, Institute for Health & Consumer Protection

Episodes of high ambient PM10 concentrations occur regularly during winter over Europe. The combi-nation of on one hand strong sources such as emissions from the tra�c sector (exhaust and re-suspension),home heating, industrial and non-industrial combustion, agricultural activities, and power generation andon the other hand the speci�c atmospheric conditions characterised by low wind speeds and temperatureinversions are responsible for these pollution events. With the aim o� setting up an approach in supportfor the identi�cation of e�cient PM10 pollution abatement strategies during 2005-2010 the EU Joint Re-search Centre in collaboration with European Research Institutions has embarked on major integrated airquality projects focused on PM10 in highly polluted areas with source apportionment using an array of re-ceptor modelling techniques as a key activity. For the present presentation a critical look is taken at theful�lment of the basic assumption for the applied multivariate statistical models and based upon two realdata sets (Lombardy Region, Italy and Krakow, Poland carried out outdoor as well as indoor during winterepisodes) sensitivity calculations have been performed by yard-stick variation of the input parameters.Themajor sources contributing to the PM10 pollution in both studies were secondary aerosols (ammonium nitrateand ammonium sulphate), residential heating and tra�c exhaust plus re-suspension. An overview is given

62

Bologna, Italy TIES 2009

of the source apportionment results and the results on the sensitivity of the Source Contribution Estimateoutput to the simulated variation of the input pro�les.keywords: Source-apportionment, PM10, Lombardy, Krakow, Indoor-outdoorBo R. Larsen, Institute for Health & Consumer ProtectionEU Joint Research Centre, Ispra.Via E. Fermi 2749, ED-28F,21020, Ispra (VA), ItalyE-mail address:bo.larsen@jrc.ec.europa.eu

Poster Presentation

NON-STATIONARY FREQUENCYANALYSIS OF EXTREME PRECIPITATION IN SOUTHKOREA.

YoungSaeng Lee1, SangHoo Yoon1, Jeong Soo Park1

1Chonnam National University

Annual maxima of daily rainfall data, dating from 1901 to 2007 are modeled for 28 locations in SouthKorea. The generalized extreme value distribution (GEVD) is �tted to the data for each of the location.In the course of climatic change, thelocation parameter of GEVD is formulated as a function of time toexplore temporaltrends of the maximum precipitation and to predict their future behaviors. We�nd evidenceof non-stationarity in the form of increasing trends for six of the 28 locations considered. Seventeen locationswere �tted well to the stationary Gumbel distribution. We quanti�ed the changes in extreme rainfall for eachlocation and provided return levels for the years of 2010, 2020, 2050 and 2100. This paper deals with theusefulness of the non-stationary GEVD in modeling extreme rainfall.keywords: Climate change, Generalized extreme value distribution, Gumbel distribution, Return level, Non-stationaryYoungSaeng Lee, (500-757)300 Yongbong-dong, Chonnam National University, Gwang-Ju, Korea.E-mail address:hellight@naver.com

Supporting grant: This work was funded by the Korea Meteorological Administration Research and Development Program underGrant CATER 2009-4507.

Oral Presentation

FIRE RISK ASSESSMENT IN MUSKOKA, ONTARIO.

Jonathan Lee1, W. John Braun1, Bruce Jones1, Doug Woolford2, Mike Wotton3

1University of Western Ontario, 2Simon Fraser University, 3University of Toronto-Canadian Forest Service

Fire risk in the wildland-urban interface is an important modelling problem in the presence of climatechange and increased human habitation. In this talk, we describe a work in progress in which �re riskto recreational properties in the Province of Ontario is being assessed using the Prometheus-Burn-P3 �respread model. Data collection issues are discussed, and a proposal for a weather data simulator for inputinto Burn-P3 is presented.Jonathan Lee, Dept. of Stat and Act. Sci. University of Western Ontario London, Ontario N6A 5B7 CANADAE-mail address:jlee253@uwo.ca

Oral Presentation

SPATIALMODELLING AND ECOLOGICAL BIASWITHIN AIR POLLUTION ANDHEALTHSTUDIES.

Duncan Lee1, Gavin Shaddick2

1University of Glasgow, 2University of Bath

In studies estimating the short-term e�ects of air pollution on health, daily counts of mortality from a studypopulation are regressed against ambient air pollution concentrations and other covariate risk factors. Suchstudies are at the ecological rather than individual level, which raises a number of statistical problems. Firstly,the pollution-health association estimated from these studies may not be the same as the desired individuallevel relationship, due to the possible presence of ecological bias. Secondly, the pollution and health dataare spatially misaligned, as the former is measured at a number of monitoring locations, while the latter isonly available as a single summary for the entire study population. The �rst of these problems is typicallyignored, while the second is overcome by averaging the monitored values to produce a single representativepollution concentration. We present a simulation study that quanti�es the likely biases induced by theseproblems, and propose alternative modelling approaches that produced less biased results. Average pollutionconcentrations are estimated using a spatio-temporal model, while the health model is re�ned to re�ect the

63

Bologna, Italy TIES 2009

aggregation from the individual to the population level. We then apply our methods to a case study basedin Greater London. The studies show that while there can be considerable variation in estimates of averagepollution levels, the potential for errors in the health e�ects are small. However, the uncertainty intervalsfrom the standard approach may be too narrow as they do not incorporate the inherent variability in theprocess.keywords: Air pollution and health, Space-time modelling, ecological biasDuncan Lee, Department of Statistics, 15 University Gardens, University of Glasgow, Glasgow,G12 8QQ, UnitedKingdomE-mail address:duncan@stats.gla.ac.uk

Poster Presentation

USING QUANTILE REGRESSION IN ENVIRONMENTAL EPIDEMIOLOGY.

Duncan Lee1, Tereza Neocleous1

1University of Glasgow

The long-term health e�ects associated with air pollution exposure can be estimated by cohort and eco-logical studies, with the latter based on observational data from contiguous small areas. Such studies regressyearly counts of mortality or morbidity events, Yt, against pollution concentrations and other covariates,using Poisson log-linear models and allowing for spatial correlation. These models estimate the relationshipbetween air pollution and the conditional mean of the health responses, E(Yt), but ignore its e�ects onother parts of the response distribution, such as the quartiles. This provides an incomplete picture of thepotentially complex relationship between air pollution and health, just as the sample mean is an incompletesummary of a data set. This poster presents a quantile regression approach that investigates how air pol-lution a�ects the entire distribution of the health responses. Only Machado and Santos Silva [American.Statistical. Association (2005): 1226-1237] have proposed quantile regression methods for count data, andtheir model is implemented within a frequentist setting and assumes the responses are independent. Here weadopt a Bayesian approach and additionally incorporate random e�ects to model the spatial correlation in theresponses. We illustrate our methods by investigating the relationship between average NO2 concentrationsover 2002 to 2004, and respiratory hospital admissions in 2005 in Scotland. For low quantiles of the responsedistribution the relative risk is around 1.06 (for an 8 micron increase in NO2), which decreases and levels o�at 1.02 as the quantile approaches one.keywords: Air pollution and health, quantile regression, spatial modellingDuncan Lee, Department of Statistics, 15 University Gardens, University of Glasgow, Glasgow, G12 8QQ,UKE-mail address:duncan@stats.gla.ac.uk

Oral Presentation

USING ESTIMATED PERSONAL EXPOSURES IN STUDIES OF THE EFFECTS OF AIRPOLLUTION ON HEALTH.

Duncan Lee1, Gavin Shaddick2, Ruth Salway2, Jim Zidek3

1University of Glasgow, 2University of Bath, 3University of British Columbia

This presentation describes how a probabilistic model can be used to estimate personal exposures toairborne pollutants. A computer model is used to simulate the exposures experienced by individuals in anurban area, using data on ambient concentrations and temperature, whilst incorporating the mechanismsthat might determine exposures. The output from the model comprises a set of daily exposures for asample of individuals from the population of interest. These daily exposures are then approximated byparametric distributions, so that the predictive exposure distribution of a randomly selected individual can begenerated. These distributions are then incorporated into a hierarchical Bayesian framework (with inferenceusing Markov Chain Monte Carlo simulation) in order to examine the relationship between short-term changesin exposures and health outcomes, whilst making allowance for long-term trends, seasonality, the e�ect ofpotential confounders and the possibility of ecological bias.This approach is applied to a case study comprisingdata on particulate pollution (PM10) and respiratory mortality counts. Of particular interest is the di�erencein using such exposures in comparison to ambient concentrations in a health model in terms of the resultingrelative risks and measures of uncertainty.keywords: air pollution and health, personal exposures, Bayesian hierarchical modelsDuncan Lee, Department of Statistics, 15 University Gardens, University of Glasgow, G12 8QQE-mail address:duncan@stats.gla.ac.uk

64

Bologna, Italy TIES 2009

Oral Presentation

SOLID WASTE LANDFILLS MONITORING BY AERIAL INFRARED THERMOGRA-PHY.

Massimiliano Lega1, Luca d'Antonio1, Rodolfo M.A. Napoli1

1DiSAm-Dipartimento di Scienze per l'Ambiente-University of Naples PARTHENOPE (Italy)

The soil matrix anthropogenic pollution phenomena are many and of varied nature, but all share thedanger to human health derived by the direct impact or by the resulting contamination of other matrices.Afamous and critical soil pollution source is the solid waste land�ll and the reason of this is the potentialproduction of undesiderable and/or toxic compounds that �ow directly or through multiple patterns.Theconventional land�lls monitoring techniques have shown, in recent years, to not be able either to highlightwhen and where pollution takes place or prevent it through the identi�cation of anomalies in the land�llbody.The aerial infrared thermography is a powerful innovative monitoring technique, suitable in the envi-ronmental sector. This technique lets to investigate the thermal phenomena both qualitatively and quantita-tively producing, respectively, a false colours image, revealing the thermal anomalies, and radiometric dataset.Measurement campaigns carried out on disused or working land�lls, using an integrated system (infraredcamera + "manned" and/or "unmanned" aerial system), have identi�ed areas where the heat was due tolocal fermentation or, even more, starting points of �re. Then, it was possible to detect unexpected biogasleakage and covering or waterproo�ng material joins not performed in accordance with best practice.Theoperating capabilities of the proposed system could be very useful also to �ght Waste Dumps (WD) ad ille-gal discharges.The measurement campaigns results will be presented in this contribution, showing how theradiometric processed images can easily and quickly prevent environmental pollution.keywords: SOLID WASTE LANDFILLS, AERIAL INFRARED THERMOGRAPHY, THERMOGRAPHY, IRCAMERA, ENVIRONMENTAL MONITORINGMassimiliano Lega, Centro Direzionale di Napoli-Isola C4, 80143 Naples, ItalyE-mail address:lega@uniparthenope.it

Poster Presentation

SOIL MONITORING BY AERIAL INFRARED THERMOGRAPHY.

Massimiliano Lega1, Claudia Ferrara1, Patrizia Manganiello1, Vincenzo Severino1

1DiSAm-Dipartimento di Scienze per l'Ambiente - University of Naples PARTHENOPE (Italy)

This paper introduces the �rst results of an innovative soil monitoring technique: a LTA and/or ULMplatform with onboard an IR camera and a speci�c payload.The digital IR camera is able to measure thethermal energy radiating from surfaces generating a false colors image, revealing the thermal anomalies, and aradiometric data set. The use of a speci�c software permits the analysis of data-set, a visible augmentation ofthe anomalies and the data-fusion between visible and IR shots. The e�ectiveness of IR thermography could bealso increased using an aerial platform; in fact, changing the point of view (POV) of thermal sensor, increasesthe Field Of View (FOV) and we can compare more elements in the same scene, maintaining a good resolutionand accuracy. Nowadays, it's possible to monitor at low altitude (proximal sensing) using LTA and/or ULMplatforms, this approach give us the opportunity to analyze small-scale phenomena. Our research grouphave classi�ed several environmental problems using IR signatures, an �nding a related thermal pattern.Ifwe introduce in the scene a warmer or cooler body, following the signature of the temperature "smoothing",we could trace the thermal-�ux using thermal signature (thermal tracking).The real cases studied for thepurposes of this contribution, validated aerial infrared thermography as optimal tool to support the SoilMonitoring.In the paper it will be introduced a report about the "discovery" of the soil problems with the�rst IR/visible shots.keywords: Soil Monitoring, aerial infrared thermography, thermography, Infrared pictures, environmental monitoringMassimiliano Lega, Centro Direzionale di Napoli-Isola C4, 80143 Naples, ItalyE-mail address:lega@uniparthenope.it

65

Bologna, Italy TIES 2009

Oral Presentation

ELIMINATING THE PRACTICAL BOUNDARY BETWEENMARKOVANDOTHERGAUSS-IAN RANDOM FIELDS.

Finn Lindgren1, Håvard Rue2, Johan Lindström1, David Bolin1

1Lund University, 2Norwegian University of Science and Technology

Gaussian random �eld models are used extensively in spatial andspatio-temporal statistics. Traditionally,two largely separateapproaches have been used; covariance function speci�cations andgrid-based Markovrandom �elds. The former method is appealing inits directness, but computationally costly, whereas thelatter isappealing for its computational bene�ts. The two approaches havecoexisted without much directlinks between the speci�cations.In this talk I will explain how to construct a direct speci�cationof Markovrandom �elds approximating the Matérn family of covariancemodels, through the use of stochastic partialdi�erential equations.As a simple side-e�ect, this model class can be expanded to �elds oncurved surfaces,such as a globe, as well as anisotropic, oscillating,and non-stationary �elds, as illustrated with geo-statisticaldata.The approach also provides a link to other popular random �eld models,such as convolution �elds andspectral representations.Finn Lindgren, Centre for Mathematical Sciences, Lund University,Box 118, SE-22100 Lund,SwedenE-mail address:�nn@maths.lth.se

Oral Presentation

POSSIBLE CLIMATE CHANGE EFFECTS ON MARINE SAFETY.

Georg Lindgren1

1Lund University

Climate change can a�ect marine safety in a number of ways. In the talk I will summarize some ofthe �ndings obtained by the partners in the EU Marie Curie RTN Seamocs, Applied Stochastic Modelsfor Ocean Engineering, Climate, and Safe Transportation:http://www.maths.lth.se/seamocs/ Waves, winds,and currents are the main factors that a�ect marine and o�shore safety, including the coastal zone changes.Changes in global circulation patterns will obviously change the wind climate on the oceans, which in turnwill a�ect the wave climate, but these changes tend to be regional, depending very much on local conditions.The mechanisms and methods to infer wave properties from wind are fairly well understood and thereforereliable simulation of wind �elds are possible within the di�erent ensemble studies that are performed in theclimate change scenarios.Another safety issue, partly related to climate change, but also to increased safetyrequirements, is that related to the detailed wave properties in di�erent sea states. Changing wind patternscan a�ect complex spectral properties of the sea state, for example, related to multidirectional seas. Thesespectral properties have shown to be important for the formation of extreme waves. As a �nal example,possible e�ects of climate change on coastal regions, sand drift, etc, will be dealt with.keywords: wave climate, wind climate, coastal changeGeorg Lindgren, Mathematical statistics, Lund university Box 118SE-22100 Luind SwedenE-mail address:georg@maths.lth.se

Supporting grant:SEAMOCS: Contract MRTN-CT-2005-019374

Oral Presentation

FAST ESTIMATION OF NON-STATIONARY GAUSSIAN MARKOV RANDOM FIELDS.

Johan Lindström1, Finn Lindgren2, Peter Jonsson2, David Bolin2, Håvard Rue3

1University of Washington, Lund University, 2Lund University, 3Norwegian University of Science andTechnology

Modelling spatially varying data is an important statistical problem, with a wide array of possible applica-tions. Spatial data is often non-stationary, and new measurement techniques has lead to increases in the sizeof the datasetswe wish to model. Thus there is a need for models that are able to handle large, non-stationarydatasets.It has been demonstrated that Gaussian Markov Random Fields (GMRFs) can be constructed todirectly approximate spatial �elds with Matérn covariance (Lindgren and Rue, Tech. report, MathematicalSciences, Lund University, 2007:12).Here we extend the stationary model to construct a non-stationary �eldwhere the variance and range depend on location. Given Gaussian observations of the underlying �eld itis possible to explicitly calculate derivatives of the log-likelihood allowing for very fast estimation of the

66

Bologna, Italy TIES 2009

non-stationary �eld. The resulting model is applied to two di�erent datasets: Point measurements of seadepth where the model is used to reconstruct the topography of the seabed, and reconstruction of globalsea level pressure from measurements at weather stations. The last application demonstrates the ability tomodel non-stationary �elds on a sphere, an important application for global data.Johan Lindström, Box 354322 University of Washington Seattle WA, 98195-4322USAE-mail address:johanl@maths.lth.se

Supporting grant: STINT (The Swedish Foundation for International Cooperation in Research and Higher Education) GrantIG2005-2047

Oral Presentation

TEMPORAL AND SPATIAL ANALYSIS OF CLIMATIC CYCLES IN A DETRITIC AQUIFER:BEHAVIOUR OF RECHARGE.

Juan Antonio Luque-Espinar1, Eulogio Pardo-Igúzquiza2, Mario Chica-Olmo2, María José García-Soldado2

1Geological Survey of Spain (IGME), 2University of Granada

We have studied the temporal and spatial behaviour of the climatic cycles in a detritic aquifer. In previouswork, we determined the existence of four climatic cycles by spectral methods. We have identi�ed: decadal,ENSO, annual and semiannual cycles. In each climatic cycle, we di�erentiated �ve levels of trust betweennon-detectable (0), < 90% (1), 90-95% (2), 95-99% (3) and > 99% (4). The spatial analysis has beendivided into two phases. By mean ordinary kriging (KO), we have estimated the spatial variability of the�ve categories to analyze the behaviour of the aquifer at each climatic cycle. We have completed the studywith indicator kriging (KI). This dual approach, estimation of the spatial detection of the four cycles andthe probabilistic study, has allowed us to study the spatial distribution of recharge and determine the mostsigni�cant areas where the water enters the aquifer. This analysis has revealed that the rivers are the mainaxis of recharge and that the provision of water varies signi�cantly from one area to another. The behaviourdetected would indicate that the rainfall runo� they generate shows an important variability in space. Theresults on recharge analysis re�ect the unequal role of the rivers or the storm waters. Ultimately, the recharge,except the annual recharge, was concentrated in a few sectors in relationship with the network drainage anddepends on the cycle analysis.keywords: climatic cycles, geostatistic, recharge modellingJuan Antonio Luque-Espinar, ja.luque@igme.esE-mail address:ja.luque@igme.es

Oral Presentation

ANALYSIS OF THE PIEZOMETRIC SPATIAL DISTRIBUTION BASED ON THE ESTI-MATION OF PIEZOMETRIC DIFFERENCES.

Juan Antonio Luque-Espinar1, Mario Chica-Olmo2, Eulogio Pardo-Igúzquiza2, María José García-Soldado2,Juan Grima-Olmedo1

1Geological Survey of Spain (IGME), 2University of Granada

The experimental variograms have usually drift. This phenomenon is due to the gradient of the spatialpiezometric data. When we calculate spatial piezometric data using the geoestatistical methods, the exper-imental variograms drift too. Because of this behaviour, we normally use more complicated geostatisticalmethods to take into account the presence of piezometric drift. Moreover, the piezometric data varies littlein time. For this reason, the variogram models adjusted closely in time are identical. This means that theestimates are equal and only di�er in the estimation error. In this paper we show the results of a piezometricstudy of a detritic aquifer. We have selected piezometric data taken in di�erent months of the hydrologicalyear. We have compared the results of di�erent months, we then estimated piezometric variations betweenone month and another. We compared both estimationsWith this approach we have achieved several things:- The experimental piezometric variograms do not have any drift, thus simplifying the geoestatistical estimateof the piezometry. - We can do better monitoring of the annual evolution of piezometry and perhaps a betteridentify of the areas where there is a further exploitation. - The behaviour of �ow in the aquifer can bede�ned in more detail.keywords: piezometry, geostatistic, space-time modellingJuan Antonio Luque-Espinar, Urb. Alcázar del Genil, 4. ed. Zulema bajo. 18006 Granada (Spain)E-mail address:ja.luque@igme.es

67

Bologna, Italy TIES 2009

Oral Presentation

POISSON NONLINEAR MIXED MODELS FOR ENVIRONMENTAL DATA.

Renjun Ma1

1University of New Brunswick, Fredericton, Canada

Correlated count data with intrinsically nonlinear nature frequently appear inenvironmental and geneticstudies; however, little has been done in the analysis ofcorrelated count data with intrinsically nonlinearlink despite great amount researchIn the areas of generalized linear mixed models, nonlinear mixed modelsand generalizednonlinear models. In this talk, we discuss some data examples and the correspondinganalysisapproaches.keywords: clustered data, generalized linear models, nonlinear mixed models, overdispersion, random e�ectsRenjun Ma, Department of Mathematics and Statistics, University of New Brunswick, Fredericton, Canada E3B 5A3E-mail address:renjun@unb.ca

Supporting grant: Natural Sciences and Engineering Research Council of Canada

Oral Presentation

MONITORING USING CHANGEPOINTS.

Ian MacNeill1, Krishna Jandhyala2, Elena Naumova3

1Department of Stats and Act Ac, U of Western Ontario, 2Department of Statistics, Washington StateUniversity, Pullman, WA., 3Department of Family Medicine and Community Health, Tufts University,Boston, MA

The changepoint (CP) statistic introduced by Cherno� and Zacks [Ann. Math. Statist. (1964): 994-1018] isthe focus of this discussion, with particular emphasis on its power properties and on itsuse as amonitoring tool. A discussion is given regarding the factorsa�ecting the ability of the CP-statistic to detectthe presence ofchangepoints in a time series. A part of this discussion is concerned with thedistributionof the weights assigned to the possible changepoints, particularlyas it pertains to focusing emphasis on theportion of the time series just priorto the evolving end of the time series; this is of interest when the statisticis used formonitoring. The distribution of the statistic, applied to time series ofnormal variates, is obtainedunder null and alternative hypotheses for essentiallyarbitrary alternatives and weighting distributions. Powercomputations are carriedout for a collection of relevant variables. Costs and bene�ts are discussed foraselection of parameters and are illustrated using power surfaces. Discussion is givenof when central limittheory is applicable and when it is not.Discussion of models of outbreaks of infections and sudden increases involatility ofdata are discussed and it is shown that changepoint modelswith weightings that emphasize changesat the evolving end of a time series havepower properties that compare favorably with other monitoringmethods.keywords: Monitoring, Changepoints, Power surfaces, Time seriesIan MacNeill, Department of Statistical and Actuarial Sciences, University of Western Ontario, London, Ontario, N6A5B7, CANADAE-mail address:macneill@uwo.ca

Oral Presentation

A BOOTSTRAP VARIANCE ESTIMATOR FOR THE OBSERVED SPECIES RICHNESSIN QUADRAT SAMPLING FROM FINITE POPULATIONS.

Steen Magnussen1, Ron McRoberts2

1Canadian Forest Service, 2US Forest Service, MN

In ecological quadrat sampling more than one species can be observed in a single quadrat. When speciesco-occurrence in quadrats is non-random, the variance of the number of observed species in a sample will belarger than the variance under an assumed model of independence. Species sampling by quadrats is akin tosampling with a random cluster size. A design-based estimator of variance requires information about theprevalence and covariance of all species in a population and is therefore impractical. Instead, a bias-correctedtwo-step bootstrap version of the design-based variance estimator is proposed. Hot-deck imputations aremade for non-sampled units (N) in step one. In step two, a random sample of size (n) is taken from the Npseudo records. After B replications of the two steps, a design-based variance is estimated. A �nal correctioncompensates for the variance of species lost in a bootstrap sample. The proposed estimator is compared

68

Bologna, Italy TIES 2009

to two alternatives in simulated sampling from seven large forest inventory samples from Georgia (GA),Minnesota (MN), Utah (UT), and Wisconsin (ASP212). The bootstrap estimator was best in GA and inUT when sample sizes were ≤60. Otherwise an estimator exploiting ideas in balanced repeated replicationswas best. A variance approximation proposed by Haas, Liu, and Stokes (2006, Biometrics 62: 135-162) wasunreliable.Steen Magnussen, 506 West Burnside Road, Victoria, BC V8Z 1M5, CanadaE-mail address:steen.magnussen@nrcan.gc.ca

Poster Presentation

SEVERE WEATHER UNDER A CHANGING CLIMATE: LARGE SCALE INDICATORSOF EXTREME EVENTS.

Elizabeth Mannshardt-Shamseldin1, Eric Gilleland2, Harold Brooks3

1Duke University and the Statistical and Applied Mathematical Sciences Institute (SAMSI), 2NationalCenter for Atmospheric Research, 3National Oceanic and Atmospheric Administration

One of the more critical issues with a changing climate is the behavior of extreme weather events, asthese can cause loss of life, and have huge economic impacts. It is generally thought that such events wouldincrease under a changing climate. However, climate models are currently at too coarse of a resolutionto capture the very �ne scale extreme events such tornadoes or hurricanes. One approach is to look atthe behavior of large scale indicators of severe weather. Here several factors are considered as large scaleindicators of severe weather, including convective available potential energy and wind shear. This presentssome interesting statistical issues. Numerous approaches, including the use of the generalized extreme valuedistribution for annual maxima, the generalized Pareto distribution for threshold excesses, a point processapproach, and a Bayesian framework, are examined. Each approach is critiqued and compared for goodnessof �t, model robustness, and predictive attributes on both re-analysis data and climate model output data.For the univariate case, it is relatively straightforward to analyze such data though numerous issues mustbe resolved. These issues include appropriate techniques for threshold selection and prior speci�cation. Abivariate approach can also be considered. In addition, when analyzing weather extremes, one is faced witha spatial �eld. Predicting extreme weather events is an important, growing area of research and there remainmany avenues for further exploration. Acknowledgements to Patrick Marsh and Matt Pocernich.keywords: severe weather, extreme events, climate change, large scale indicators, �ne scale predictionElizabeth Mannshardt-Shamseldin, SAMSIP.O. Box 14006Research Triangle Park, NC 27709-4006E-mail address:elizabeth@stat.duke.edu

Supporting grant: This material was based upon work partially supported by the National Science Foundation under GrantDMS-0635449 to the Statistical and Applied Mathematical Sciences Institute.

Oral Presentation

A BAYESIAN APPROACH TODETERMINE THE RAINFALL THRESHOLDS FOR LAND-SLIDES TRIGGERING.

Mario Lloyd Virgilio Martina1, Sara Pignone2, Ezio Todini1

1University of Bologna, 2ARPA - SIM, Bologna

The rainfall thresholds are considered by several authors the critical values for one or more precipitationfeatures such as intensity, duration, antecedent volume, to trigger surface landslides. Rainfall thresholds,for instance, may represent the minimum intensity and duration of rainfall which induce a landslide. In theliterature two types of rainfall thresholds are established: (1) empirical thresholds based on historic analysisof relationship rainfall/landslide occurrence; (2)physical thresholds based on numerical models that takeinto account the relationship between rainfall pore pressure and slope stability by coupling hydrological andstability models. In this work we stress the idea of de�ning the rainfall threshold on the landslides hazard, i.e.on the probability of occurrence. The Bayes' theorem it is used to estimate the probability of a landslide giventwo rainfall features selected among �ve considered to play a physical role in the triggering mechanism. Onthe basis of a historical database for the Emilia-Romagna regions set up with more than 2000 landslides from1951 till today, we show the method use for the computation of the marginal and conditional distributions ofthe rainfall variables and we apply the Bayes' theorem to determine the rainfall thresholds corresponding to

69

Bologna, Italy TIES 2009

a landslides occurrence probability. The results are physical meaningful and show the di�erent signi�cativityof the rainfall variables for predicting landslides.keywords: rainfall thresholds, landslides, prediction, BayesMario Lloyd Virgilio Martina, Department of Earth Sciences, University of Bolognavia Zamboni, 6740126 - BolognaE-mail address:mario.martina@unibo.it

Oral Presentation

A HURDLE MARKOV MODEL FOR POLLUTANTS CONCENTRATIONS.

Antonello Maruotti1, Francesco Lagona1

1Dipartimento di Istituzioni Pubbliche, Economia e Società - Università di Roma Tre

A common feature of ecological data sets is their tendency to contain many zero values. We discuss amethod for analyzing data that are positively skewed and contain a substantialproportion of zeros, focusing ontime series of pollutants exceedances that are typically a�ected by latent heterogeneity and by the persistenceof zeros that challenge the identi�cation of typical patterns of exceedances probabilities.Statistical inferencebased on such data are likely to be ine�cient or wrong unless carefulthought is given to how these zerosarose and how best to model them. In this paper, wepropose a framework for understanding how zero-in�ated data sets originate anddeciding how best to model them.We propose to model these data throughregression models for count data that allow for overdispersion and zero-in�ation, in a Hidden Markov modelframework. We account for unobserved heterogeneity in the data mixing two di�erent approaches. Onthe one hand, we present an extension of the Poisson model for count data based on a semiparametricmodel for unobserved heterogeneity in the conditional mean. Further, a new zero-in�ated model, in whichoverdispersion is assumed to be caused by an excessive number of zeros, is discussed by extending the well-known hurdle model. Heterogeneity sources that in�uence the �rst process (i.e. the hurdle step) are assumedto in�uence also the (truncated) distribution of the positive outcomes. Estimation is carried out through theforward-backward algorithm without any parametric assumptions on heterogeneity sources distributions.keywords: pollutants exceedances , hidden markov model, hurdle model, random e�ects model, zero-in�ated countdataAntonello Maruotti, Dipartimento di Istituzioni Pubbliche, Economia e Società - Università di Roma Tre - Via G.Chiabrera 199 - 00145 RomaE-mail address:antonello.maruotti@uniroma3.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

WAVELET-BASED ANALYSIS OF THE SPATIAL STRUCTURE OF POINT PATTERNS.

Jorge Mateu1, Orietta Nicolis2

1University Jaume I, 2University of Bergamo

We analyze and detect structures in spatial point patterns by using a multiresolution wavelet approach.The idea is to consider a"modi�ed" �rst-order intensity function where peaks are given by abrupt changesand then use wavelet techniques for edge detection toextract information about clusters (shape, position andscale), and about theconditions of homogeneity and isotropy. In particular, we use the detail coe�cients ofwavelet transforms and maxima modula for identyi�ng discontinuites. A simulation study has been carriedout to evaluate the method using di�erent type of wavelets (decimated and non-decimated, discrete andcontinue wavelets, etc.). Finally, an application to real environmental data has been considered.keywords: Spatial structures, Multiresolution wavelets, Intensity functionJorge Mateu, University Jaume I, Department of Mathematics, Campus Riu Sec, E-12071 Castellon, SpainE-mail address:mateu@mat.uji.es

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time", and by grant MTM2007-62923 from the Spanish Ministry ofScience and Education

70

Bologna, Italy TIES 2009

Poster Presentation

NATIONAL FRESHWATER QUALITY INDICATOR-A CANADIAN ENVIRONMENTALSUSTAINABILITY INDICATOR.

Beverly McNaughton1

1Science and Technology Branch, Department of the Environment

The health and social and economic well-being of Canadians are closely linked to the quality of theirenvironment. In 2004, the Government of Canada committed to establishing a national indicator to assessfreshwater quality. The goal of this new indicator is to provide Canadians with more regular and reliableinformation on water quality and how it is linked with human activities. The Freshwater Quality Indicator,represented statistically by the Water Quality Index, is a tool that allows experts to translate large amounts ofcomplex water quality data into a more easily understood format for the general public. The Index compareswater quality monitoring data against water quality guidelines, which are values that de�ne water qualityconditions beyond which aquatic life may be adversely a�ected. The formula for the Water Quality Indextakes into account the number of water quality parameters that do not meet guidelines, how frequently thisoccurs and by how much. The Water Quality Index then ranks waterbodies into a simple overall rating of'excellent, good, fair, marginal or poor'. The National Water Quality Monitoring and Surveillance Division ofthe Department of the Environment together with provincial and territorial government partners calculate theWater Quality Index annually for the Canadian Environmental Sustainability Indicators Initiative Report(www.environmentandresources.ca). The report provides the public with the status and trends of waterquality throughout Canada with the goal of supporting sustainable water management decisions. This posterwill highlight examples of the indicator calculations and application in western Canada.keywords: water quality indicator, water quality index, water quality assesment, water qualityBeverly McNaughton, 201-401 Burrard Street Vancouver, British Columbia Canada V6C 3S5E-mail address:beverly.mcnaughton@ec.gc.ca

Oral Presentation

SATELLITE IMAGE-BASED MAPS: SCIENTIFIC INFERENCE OR JUST PRETTY PIC-TURES?.

Ronald E. McRoberts1

1U.S. Forest Service

The scienti�c method has been characterized as having two distinct components, Discovery and Justi�ca-tion. Discovery emphasizes ideas and creativity, focuses on conceiving hypotheses and constructing models,and is generally regarded as lacking a formal logic. Justi�cation begins with the hypotheses and models andends with a valid scienti�c inference. Unlike Discovery, Justi�cation has a formal logic whose rules must berigorously followed to produce valid scienti�c inferences. In particular, when inferences are based on sampledata, the rules of the logic of Justi�cation require valid assessments of unbiasedness and precision. Thus,satellite image-based maps that lack such assessments may be of little utility for scienti�c inference; essen-tially, they may be just pretty pictures. Probability- and model-based approaches to inference are explained,illustrated, and compared for producing inferences for maps depicting three land cover classes: non-forest,coniferous forest, and deciduous forest. Emphasis is placed on methods for assessing unbiasedness and pre-cision of maps constructed using logistic regression models with forest inventory data and Landsat satelliteimagery.keywords: forest inventory, bias, precisionRonald E. McRoberts, 1992 Folwell Avenue Saint Paul, Minnesota 55108E-mail address:rmcroberts@fs.fed.us

Poster Presentation

FORESTS ON THE EDGE.

Ronald E. McRoberts1, Susan M. Stein1, Lisa G. Mahal2

1U.S. Forest Service, 2University of Nevada, Las Vegas

The Forests on the Edge project uses geographic information systems (GIS) techniques to assess thecontributions of and threats to private forest land in the USA. Contributions and threats are discussed anddescribed in the context of the Montréal Process criteria and indicators; contributions include timber supply,wildlife habitat, and water quality, while threats include urbanization, wild�re, pollution, and insects and

71

Bologna, Italy TIES 2009

disease. Nationally consistent digital layers depicting the contributions and threats have been acquired orconstructed to support the project. Watersheds have been selected as the basic analytical unit to emphasizethe dependency of forest contributions on water-related features. Results of the analyses indicate that mostprivate forest land in the USA is in the East, as are both their greatest contributions and the greatest threats tothose contributions. Watersheds where urbanization most threatens private forest land and their contributionsto water quality and timber supply are located in New England, while watersheds most threatened bypollution, insects, and disease are distributed throughout the country but with concentrations in the MiddleAtlantic States. The publication reporting Phase I of Forests has had more than 10,000 requests and hasgone through multiple printings. Results from the Forests on the Edge project have been quoted in a widevariety of news outlets from trade journals to the Washington Post.Ronald E. McRoberts, 1992 Folwell Avenue Saint Paul, Minnesota 55108 USAE-mail address:rmcroberts@fs.fed.us

Oral Presentation

SMOOTHINGANDCHANGE POINT ESTIMATION FOR IRREGULARLY SPACED PALAEOPROXY TIME SERIES.

Patricia Menéndez1, Sucharita Ghosh2, Hans Rudolf Künsch3, Willy Tinner4

1Biometris, Wageningen University & Swiss Federal Research Institute WSL, Switzerland, 2Swiss FederalResearch Institute WSL, Birmensdorf, 3Seminar for Statistics, ETH, 4Institute of Plant Sciences,University of Bern

Long-term records such as fossil pollen and stable oxygen isotopes reveal major �uctuations in the past en-vironmental conditions, including abrupt climate changes and long term temperature shifts. Oxygen isotopes,used to reconstruct past temperatures, show that the Holocene (approximately the last 11,500 years) has beenmuch warmer than the Younger Dryas (ca. 11,500-12,700 years before present) whereas the transition fromthe Younger Dryas to the Holocene saw abrupt changes in temperature. Pollen records from sedimentaryarchives show that temperature changes had strong e�ects on vegetation.Due to the nature of the sedimen-tation and dating process, palaeo time series such as fossil pollen or oxygen isotope records are irregularlyspaced in time and depth. Therefore, given the scenarios of arbitrary changes in the underlying processes,it seems reasonable to use nonparametric smoothing methods for irregularly spaced time series to assessdi�erent aspects of these proxy records. As for the model, of special interest are processes with time varyingdistribution functions, such as a time dependent one-dimensional transformation of a stationary Gaussianprocess. For the analysis of such palaeo data, in addition to estimating trend functions, their derivatives,and probability distribution functions over time, one important problem is to detect 'rapid change points',i.e. where rapid or abrupt changes in the earth's climate have taken place.The methods will be illustratedvia application to oxygen isotopes and pollen records.keywords: Palaeo research, Irregularly spaced time series, Smoothing, Long-memory, Change point estimationPatricia Menéndez, Biometris Droevendaalsesteeg 1, Building 107 6708 PB WageningenE-mail address:patricia.menendez@wur.nl

Supporting grant: Biometris

Poster Presentation

DATA ANALYSIS ON ENVIRONMENTAL MONITORING NETWORKS.

Raquel Menezes1, Ana Cristina Fernandes1

1Minho University, Portugal

In environmental monitoring networks, preferential sampling happens if there is a natural inclination toplace more monitors in areas classi�ed as high risk for pollution. The traditional geostatistical methods relyon the fundamental assumption that the sampling locations X have been chosen independently of the values ofthe spatial variable S under study; however, preferential sampling does not allow this stochastic independence.Diggle, Menezes and Su (2009) propose a model-based approach, which considers log-Gaussian Cox processesto model the stochastic dependence of X on S. Moreover, one possible solution for the preferability issue is toseek explanatory variables which eliminate, or at least reduce, the resulting adverse e�ects. The underlyingidea is that when X and S are unconditionally dependent, they may nevertheless be conditionally independent,or approximately so, given suitable covariate information. In this work, we study the performance of these

72

Bologna, Italy TIES 2009

approaches when applied to a real dataset of pollution considering the ground elevation as a covariate. Thedata were collected from Galicia region in Spain, where air quality was monitored through biomonitors, basedon the high bioconcentration of heavy metals in land mosses (Fernández et al., 2000).Diggle P.J., Menezes R.and Su T. (2009). "Geostatistical Inference Under Preferential Sampling". Under revision.Fernández J.A.,Rey A. and Carballeira A. (2000). "An extended study of heavy metal deposition in Galicia (NW Spain)based on moss analysis". Sci Total Environ, 254, pp. 31-44.keywords: geostatistics, air quality monitoring, preferential samplingRaquel Menezes, Dep Matemática para a Ciência e Tecnologia Universidade do Minho Campus de Azurém 4800-058Guimarães PortugalE-mail address:rmenezes@mct.uminho.pt

Poster Presentation

DATA HANDLING BASED ON AUTOCOVARIANCE FUNCTION FOR DECODING COM-PLEX SIGNALS FROM ENVIRONMENTAL MONITORING: IDENTIFICATION OF OR-GANIC TRACERS IN ATMOSPHERIC AEROSOL.

Mattia Mercuriali1, Maria Grazia Perrone2, Luca Ferrero2, Ezio Bolzacchini2, Maria Chiara Pietrogrande1

1Department of Chemistry, University of Ferrara, 2Department of Environmental Sciences, University ofMilano Bicocca

In the present paper, a signal processing procedure based on the AutoCovariance Function (ACVFtot)is applied to the signals obtained from Gas Chromatography-Mass Spectrometry (GC-MS) analysis of at-mospheric aerosols. Mathematical equations have been derived and a computation algorithm implementedto extract information on the homologous series of n-alkanes as molecular tracers of speci�c input sources.In fact, these compounds help to di�erentiate aerosols of anthropogenic origin from those of biogenic origin(Oliveira et al. Atm. Environ., 41 (2007), 5555-5570).The key parameters useful for a good source appor-tionment are the number of terms (nmax) and the carbon preference index (CPI). These parameters can bedirectly computed from the AutoCoVariance Function (ACVFtot) computed on the acquired chromatogram(Pietrogrande et al., Anal. Chim. Acta, 594 (2007) 128-138, Pietrogrande et al., Analyst, 134 (2009) 671-680). The method was applied to GC-MS chromatograms of di�erent atmospheric aerosols (PM10, PM2.5and PM1 samples) collected in rural and urban areas close to Milan (Italy) in di�erent seasons (winterand summer).The obtained results are compared with the data obtained by the conventional time and la-bor consuming procedure based on peak integration from the acquired chromatogram.The procedure makesit possible to extract information relevant to trace the origin and the fate of atmospheric aerosols and todi�erentiate type and abundance of the di�erent input sources reducing the labour and time required andincreasing the quality and objectivity of the results.keywords: AutoCovariance, n-alkanes, organic tracers, PM, source apportionmentMattia Mercuriali, Via L.Borsari 46, 44100 FerraraE-mail address:mrcmtt@unife.it

Oral Presentation

BAYESIAN SPATIAL PANEL DATA.

Maura Mezzetti1, Samantha Leorato1

1Facoltà Economia, Università "Tor Vergata", Roma

A hierarchical Bayesian model for spatial panel data is proposed. The idea behind the method is to analyzepanel data taking into account a possible dependence within observations, besides temporal pattern, due toa geographical structure. The underlying idea is the introduction of a separable covariance matrix for theobservationvector: a Kronecker product of a matrix indicating spatial covariance and a component represent-ing the temporal covariance.Our proposal has the advantage of allowing forpurely non-informative patternof the covariance matrices, whereas most of the proposals in literature, stems from the strict assumption ofautoregressive pattern of order one for both spatial and temporal components.The patterns of the spatial andtemporal components of the covariance matrices can in fact be arbitrary, the only constraint being the multi-plicative structure of the covariance ofthe errors. A natural choice for the spatial covariance pattern is oftengiven by some decreasing function of a measure of distancebetween regions.Analysis of spatial panel data isof great importance and interest in many �elds as epidemiology, environmetrics and spatial econometrics.We consider di�erent applications in econometrics and envirometrics. Through uninformative prior, on one

73

Bologna, Italy TIES 2009

hand, we investigate presence of correlation between observations, beside temporal one, on the other hand wecan investigate weather this dependence is due to geographical distance or to other similarities. Simulateddata allows the study of the implemented MCMC algorithm, and the sensitivity to the prior distributionsassumed. Finally we compare our model with existing spatial correlated data.keywords: spatial panel data, separable covariances, Kronecker product, Gibbs samplingMaura Mezzetti, università tor vergata, via columbia 2, 20131 romaE-mail address:maura.mezzetti@uniroma2.it

Oral Presentation

FLEXIBLE STATISTICAL MODELS IN THE STUDY OF VULNERABILITY TO ENVI-RONMENTAL EXPOSURE.

Rossella Miglio1, Francesca Bruno1

1Department of Statistics, University of Bologna

Identi�cation of factors that confer susceptibility to environmental exposures has become an importantissue in the scienti�c community. Several studies have investigated the potential association between exposureto air pollution and adverse health e�ects and the relationship between high temperature and mortality.Somestudies have reported a greater susceptibility for the elderly, for those with a lower socio-economic status andwith previous chronic clinical conditions. For example, a "J" shaped relationship between daily temperatureand all-causes mortality has been found, with an immediate time lag (same day or previous day at the most) ofthe heat e�ect. The interest in this situation is devoted to modelling a non linear functional relationship. Thisproblem could be even more complicated when the relationship di�ers among subgroups and robust analyticalmethods that can disentangle complex e�ects are required. The semi-varying coe�cient models constitute arecent alternative to the well-known non parametric techniques as proposed by Fan and Zhang [StatisticalInterface (2008): 179-195]. An important feature of these models is to allow the coe�cients to vary smoothlyover groups and hence permit to include in their formulation non-linear interaction between variables. Wewill compare the results obtained with di�erent approaches in order to investigate the short term relationshipbetween high summertime temperature and all-cause mortality with the aim to evaluate e�ect modi�er ofthis association in population subgroups de�ned by demographic characteristics, socio-economic status andpre-existing clinical conditions in the city of Bologna.keywords: semivarying coe�cient models, conditional logistic regression, environmental epidemiologic studies, marsRossella Miglio, Department of Statistics, University of Bologna via Belle Arti, 4140126 Bologna, ITALYE-mail address:rossella.miglio@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

SPACE-TIMEMODELS FORMOVING FIELDS. APPLICATION TO SIGNIFICANTWAVEHEIGHT.

Valérie Monbet1, Pierre Ailliot2, Anne Cuzol1, Nicolas Raillard2

1Université de Bretagne Sud, 2Université de Bretagne Occidentale

Recent technological advances such as satellite scanning result in increasingly complex environmental dataover large spatial domains and long periods of time. In parallel, there is a growing demand for statisticalmethods appropriate to the handling of such space-time data.We will focus on the signi�cant wave height(Hs) a parameter related to the energy of the sea-state, which is an important parameter for many marineapplications. Stochastic space-time models for Hs are generally used for spatio-temporal interpolation, andthen used to produce climatology or real-time analysis of Hs �elds.An important feature of Hs is that the �eldsare moving (propagation of the waves and displacement of the air masses). Using Lagrangian reference frame,instead of a �xed (Eulerian) reference frame, appears to be natural for modelling such moving processes. Inthis study, we propose to use outputs of a numerical weather forecast system to estimate the motion of thelarge scale structures of Hs �elds. Although these numerical models are sometimes inaccurate, we will showthat they provide enough information about the state of the atmosphere and the ocean in order to get arough estimate of the motion of the sea-state. We will then propose an appropriate space-time covariance

74

Bologna, Italy TIES 2009

model which takes into account the motion and show that the proposed model improves the accuracy of theinterpolated �elds obtained using satellite data.keywords: State space model, Space time interpolation, Motion estimation, Signi�cant wave heightValérie Monbet, Université de Bretagne Sud, Lab-STICC,BP 573, F-56000 Vannes cedexE-mail address:valerie.monbet@univ-ubs.fr

Oral Presentation

FUNCTIONAL KRIGING OF OCEAN PROFILE DATA.

Pascal Monestiez1, David Nerini2

1BioSP, INRA, 2LMGEM - UMR 6117 CNRS, COM

In environmental sciences, many spatial data are curves or can be considered as curves. Examples aregranulometry distributions in soils or vertical pro�les of meteorological variables recorded by radiosonds.Surveys in oceanography also provide vertical pro�les of temperature, salinity or biological variables that arespatially dependent and sampled along transects. In geostatistics, there are two usual way to analyse suchdata; the �rst one is to consider the pro�le as a third dimension which is often problematic due to strong andcomplex anisotropy and to non-stationarity along the vertical dimension. The second way is to discretize thecurves and to model them in multivariate geostatistics. We propose here a general functional kriging whichgeneralizes multivariate geostatistics on discretized curves and formally accounts for the functional nature ofthe data. A general formulation of the solution is given. In practice, each data curve is decomposed on anorthogonal functional basis and we show that performing a kriging on functional data is equivalent to a fullisotopic cokriging on the function-basis coe�cients. We applied it to interpolation of vertical pro�les in thesouthern Antarctic Ocean where elephant seals, equipped by Argos tracking device and data logger, sampledand recorded pro�les of temperature, salinity and �uorescence (chlorophyll a).keywords: functional data analysis, coregionalization, curve cokriging, RKHSPascal MONESTIEZ, Unité Biostatistique et Processus Spatiaux, INRA Avignon, Domaine Saint-Paul, Site Agroparc84914 AVIGNON cedex 9, FranceE-mail address:monestiez@avignon.inra.fr

Supporting grant: ANR Project IPSOS-SEAL, CNES-TOSCA

Oral Presentation

ZERO-INFLATEDGENERALISED POISSON REGRESSIONMODEL TODESCRIBE PSEUDO-NITZSCHIA CONCENTRATION IN LISBON BAY.

Helena Mouriño1, So�a Palma2, Maria Teresa Moita2, Maria Isabel Barão1

1Universidade de Lisboa, Portugal, 2Instituto Nacional de Recursos Biológicos, Portugal

Located in the Iberian Peninsula, the West Coast of Portugal is the northern limit of the North Atlanticupwelling system. Along this area, upwelling is identi�ed as the major source of seasonal and spatial variabilityof phytoplankton biomass and composition.Phytoplankton assemblage in estuarine and coastal ecosystemsconsists of a wide variety of algae. In particular, species of the diatoms genus Pseudo-nitzschia are commonin the Portuguese Coast. There is a lack of knowledge about the dynamics that rule Pseudo-nitzschiablooms. This work aims at shedding some light into the intricacies of this physical phenomenon. We builda statistical model that describes Pseudo-nitzschia concentration in Lisbon Bay as a function of the SeaSurface Temperature and the upwelling index. Wind data were obtained at the meteorological station ofCabo Carvoeiro.Due to the count nature of Pseudo-nitzschia time series, the statistical model is constructedwithin the framework of the Poisson Regression. The data set under study is not only characterised byexcess of zeros but also the non-zero part of the data is over-dispersed. As a consequence, the Zero-In�atedGeneralised Poisson Regression Model is used to model Pseudo-nitzschia concentration.The mathematicalmodel that describes the amount of Pseudo-nitzschia in Lisbon Bay shows a lag of four to six days betweenthe upwelling events and the presence of Pseudo-nitzschia in the monitoring station. Also, there is a signi�cantpositive relation between Pseudo-nitzschia concentration and Sea Surface Temperature.keywords: Zero-In�ated Generalised Poisson distribution, Time Series Analysis, Upwelling, Pseudo-nitzschiaHelena Mouriño, Departamento de Estatística e Investigação Operacional,Faculdade de Ciências-Universidade deLisboa, Edifício C6, Campo Grande1749-016 Lisboa-PORTUGALE-mail address:mhnunes@fc.ul.pt

75

Bologna, Italy TIES 2009

Oral Presentation

SCORE AND QUASI-SCORE INFERENCE FOR CHANGE-POINTS, WITH APPLICA-TIONS IN MARINE ECOLOGY AND GROUNDWATER MONITORING.

Vito Muggeo1, Gianfranco Lovison1

1Dipartimento di Scienze Statistiche e Matematiche "S.Vianelli", University of Palermo

Segmented regression represents a useful framework for modelling the e�ect of an explanatory variable,when interest lies in estimating one or more threshold values where such e�ect changes. Inference on theseunknown breakpoints is complicated by several non-regularities. In this paper we discuss some possibleapproaches based on the Score statistic and some of its modi�cations and extensions; in particular, we focuson con�dence intervals for the breakpoint. Unlike the log-likelihood, interestingly the score and pseudo-scorefunctions are less a�ected by violations of usual regularity conditions, and therefore permit to make inferencein classical linear models as well as generalized models with non-Gaussian response and non-identity linkfunction. We present two applications in environmetrics, where segmented regression is employed and score-based or quasi-score-based con�dence intervals for the breakpoint provide reliable results: the �rst dealswith monitoring groundwater levels over time; the second with the e�ect of age on the growth of Posidoniaoceanica, a marine plant which represents a biological marker of the quality of coastal marine waters.keywords: changepoint, segmented regression, score inferenceVito Muggeo, Dipartimento di Scienze Statistiche e Matematiche "S.Vianelli", viale delle Scienze, ed. 13, 90128Palermo - ITALYE-mail address:vmuggeo@dssm.unipa.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time" and "Fondi di Ateneo ORPA044431 2004".

Poster Presentation

MODELLING SPATIO-TEMPORAL FOREST HEALTH DATA.

Monica Musio1, Nicole, H. Augustin2, Klaus von Wilpert3, Edgar Kublin3, Simon Wood2, MartinSchumacher4

1University of Cagliari, 2University of Bath, 3Forest Research Centre Baden-Württemberg, 4University ofFreiburg

Forest health monitoring schemes were set up across Europe in the 1980's in response to concern about airpollution related forest die back (Waldsterben) and have continued since then. The main in�uencing factorsfor forest health are the deposition of pollutants through the air and rainfall; weather (extreme heat anddroughts - climate change) and biotic in�uences (e.g. biological pests). We model yearly data on tree crowndefoliation, an indicator of tree health, from a monitoring survey carried out in Baden- Württemberg. On achanging irregular grid, defoliation and other site speci�c variables are recorded. Our objective is to providea statistical tool for monitoring in order to replace the traditional trend estimation method which does nottake spatial and temporal correlation into account. The tool should produce estimates of average defoliationover time and space with con�dence bands, hence allowing assessment of changes in trends of defoliation. InBaden-Württemberg the temporal trend of defoliation di�ers between areas because of site characteristicsand pollution levels, making it necessary to allow for space-time interaction in the model. For this purposewe use generalized additive mixed models (GAMMs) incorporating scale invariant tensor product smooths ofthe space-time dimensions. The main reference for this work is (Augustin et al. (2009)).Augustin, N. Musio,M. von Wilpert, K. E.Kublin, E. S.Wood, S. and Schmacher, M. Modelling spatio-temporal trends of foresthealth monitoring data, Journal of the American Statistical Association, 2009, to appearkeywords: generalized additive mixed models, forest damage, tree defoliation , spatio-temporal model, tensor productsmoothsMonica Musio, Dipartimento di Matematica ed Informatica, via Ospedale, 72, - 09124 Cagliari, ItaliaE-mail address:mmusio@unica.it

76

Bologna, Italy TIES 2009

Oral Presentation

MODELLING SPACE-TIME VARIATION OF CANCER INCIDENCE DATA: A CASE STUDY.

Monica Musio1, Erik Sauleau2, Nicole, H. Augustin3

1University of Cagliari, 2University of Strasbourg, 3University of Bath

Cancer incidence data are typically available as rates or counts for contiguous geographical regions and arecollected over time. Recent methodological developments have moved in the direction of univariate space-timemodeling of incidence data especially in a Bayesian context. Based on an example of data on cancer incidencecollected between 1988 and 2005 in a speci�c area of France, this work describes an approach to analyze thespace-time evolution of the disease taking into account also of possible non linear e�ects of other covariates.For this purpose, we consider Generalized Additive Mixed Models (GAMMs) with a Poisson response, usingthe methodology presented in (Wood 04, Wood 06). The proposed method allows to incorporate a widerange of correlation structures. Besides one dimensional smooth functions accounting for non-linear e�ectsof covariates, the space-time interaction can be modeled using scale invariant tensor product smooths, wherethe smoothness parameter is estimated and does not depend on the di�erent scales of the covariate axes.Another possibility investigated to account for space-time dependency is to use varying coe�cient models. Insuch case, to explore spatio-temporal patterns, analyzes focused on six time periods, each 3 years in length,between 1988 and 2005. For model implementations we use the R package mgcv.WOOD,S. 2004. Stable ande�cient multiple smoothing parameter estimation for generalized additive models. JASA, 99,467, 673-686.WOOD,S. 2006. Low-rank scale-invariant tensor product smooths for generalized additive mixed models.Biometrics, 62,4, 1025-1036.keywords: Cancer incidence data, Generalized additive mixed models, spatio-temporal interactions, Disease mappingMonica Musio, Dipartimento di Matematica ed Informatica, via Ospedale, 72- 09124 Cagliari, ItalyE-mail address:mmusio@unica.it

Oral Presentation

A UNIFIED FRAMEWORK OF FRACTAL AND WAVELET RANDOM FIELDS.

Orietta Nicolis1, George Christakos2

1Dept. of Information Technology and Mathematical Methods, 2Department of Geography, San Diego StateUniversity

The mathematical apparatus of modern spatiotemporal geostatistics includes a variety of theories andmodels. Among them, a powerful uni�ed framework is provided by the generalized space time random �eldstheory. In this work we analyze a special case of generalized space/time random �elds (GS/TRF): thewavelet random �eld model (WRF) . Starting from the work of Christakos (1991) we consider the waveletsas test functions and we analyse their main properties such as the second order moments, self-similarity andanisotropy. We also show how the wavelet random �eld depends on the properties of wavelets (regularity,spatial orientation, ecc.). Some examples are provide using discrete and continue wavelets.keywords: Directional wavelets, Generalized space/time random �eld, Self-similarity, Wavelet random �eldsOrietta Nicolis, University of Bergamo,Via Marconi 5, 24044 Dalmine BG I, ItalyE-mail address:orietta.nicolis@unibg.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

SPATIAL STATISTICS, COMPUTER MODELS AND REGIONAL CLIMATE CHANGE.

Douglas Nychka1, Stephen Sain1, Linda Mearns1, Cari Kaufman2

1National Center for Atmospheric Research, 2University of California/Berkeley

What weather can I expect in my city 50 years from now? As attentionshifts from broad global summariesof climate change to more speci�cregional impacts there is a need for statistics to quantify theuncertaintyin regional predictions. This talk will provide an overviewon interpreting regional climate experiments. Aregional climate model(RCM) is a computer code based on physics that simulates the detailed�ow of theatmosphere in a particular region based on large scaleinformation from a global climate model. One intent isto comparesimulations under current climate to future scenarios to infer the kinds of climate change expectedat a location. Clearly this an areawhere statistical science has an important role to play. The RCM outputis

77

Bologna, Italy TIES 2009

large and requires summarization simply to be interpreted. Thenumerical experiments typically simulate 20-30 years and so statisticalanalysis is needed to separate the underlying climate of the model fromyear to yearvariations from weather. Finally, it is important to attachmeasures of uncertainty to the model predictionsin order to make thesegeophysical results relevant for forming local policy and makingeconomic decisions.To address these problems requires some new andchallenging applications of spatial statistics with respect toRCMoutput. These include strategies to handle large spatial �elds,considering extremes in spatial �elds andANOVA decompositions ofspatial �elds into di�erent sources of variation.Douglas Nychka, National Center for Atmospheric Research PO Box 3000 Boulder CO 80307-3000E-mail address:nychka@ucar.edu

Supporting grant: Supported by the National Science Foundation

Poster Presentation

A SPATIO-TEMPORAL MODEL BASED ON THE SVD TO ANALYZELARGE SPATIO-TEMPORAL DATASETS.

Rossella Onorati1, Paul D. Sampson1, Peter Guttorp1

1University of Washington

A common problem in the analysis of space-time data is to compress a large dataset in order to extractthe underlying trends. Empirical orthogonal function (EOF) analysis isa useful tool for examining boththe temporal and the spatial variation in atmospherical and physical process and a convenient method ofperforming this is the Singular ValueDecomposition (SVD). Many spatio-temporal models for measurementsZ(s, t) at location s at time t, can be written as a sum of a systematic component and a residual component:Z= M+E, where Z, M and E are all TxN matrices. Our approach permits modelling of incomplete data matricesusing an "EM-like" iterative algorithm for the SVD. We model the trend, M, by linear combinations of smoothtemporal basis functions derived from left (temporal) singular vectors of the SDV of Z with dimension ofthe model chosen by cross-validation. We further decompose by SVD the spatio-temporal residual matrixEcomputed as residuals from regressions at each site (column) of the observations on the smoothed temporalbasis functions. Finally we �t an autoregressive model to the columns(time series) of residuals from the SVDof E. Our aim is to illustrate a simple model to characterize trends and model the variability in large spatio-temporal data matrices. Themethodology is demostrated with a spatiotemporal dataset.keywords: Spatio-temporal processes, SVDRossella Onorati, Via dei Dori, 691100 Trapani ItalyE-mail address:onoratir@u.washington.edu

Oral Presentation

MODELLING OF SPATIAL EXTREMES: A REVIEW.

Simone Padoan1, Anthony Davison1, Mathieu Ribatet1

1Ecole Polytechnique Fédérale de Lausanne

The areal modelling of the extremes of a natural process such as rainfall or temperatureobserved at pointsin space is important in environmental statistics; for example,understanding extremal spatial rainfall is crucialin �ood protection. This article reviewsapproaches to the statistical modelling of spatial extremes, startingwith a discussion ofgeostatistical models and of mathematical models for rare events, based on the notionofmax-stability. The main types of models, based on Gaussian anamorphosis, on latentvariables, and on the�tting of spatial max-stable processes through composite likelihoodmethodology, are described and comparedon a rainfall datasetkeywords: Composite likelihood, Copula, Environmental data analysis, Max-stable process, Statistics of extremesSimone Padoan, Route de Chavannes 17 Lausanne 1007E-mail address:simone.padoan@ep�.ch

Supporting grant: Extremes project, http://www.cces.ethz.ch/projects/hazri/EXTREMES

78

Bologna, Italy TIES 2009

Poster Presentation

MODELLING SORPTION OF TRICYCLAZOLE ON RICE PADDY SEDIMENTS BY STA-TISTICAL ANALYSIS OF LAB-SORPTION DATA.

Ottorino-Luca Pantani1, Irene Lozzi1, Luca Calamai1, Marinella Bosetto1, Ettore Capri2

1Dip di Scienza del Suolo e Nutrizione della Pianta, Univ. di Firenze, 2Ist. di Chimica agraria edambientale,Univ. Cattolica del Sacro Cuore

Tricyclazole sorption in paddy sediments representative of rice-growing regions of Northern Italy wasmodelled by statistical analysis of distribution coe�cient (Kd) obtained in lab-scale experiments conductedeither in distilled water or in 0.01 and 1.0 M NaCl-containing solutions.The data analyses aimed to highlight:a) which were the chemical-physical characteristics of the sediments that in�uenced the tricyclazole sorptionin water; b) the e�ect of the ionic strength on sorption. The strategies used for data analysis in a or bwere:a) A multi-linear model was �tted to the data where the mean Kd values, calculated for each sedimentat EL = 0, were regressed over the chemical-physical parameters of the sediments.b) a mixed-e�ects modelwas built with Kd as the response, and the EL as experimental factor within each sediment. Also pH andelectrolyte level were analysed by mixed-e�ects model.The data analysis showed two distinct clusters in Kdvalues separated by a 5% threshold in the organic matter (OM) content of the sediments. At OM > 5%, theKd values depended upon both OM and clay content. NaCl signi�cantly increased the sorption and decreasedthe bulk pH that, however, was always above the pKa of the fungicide (1.6) and did not a�ected Kd. Sincethe Kd was signi�cantly increased in presence of both NaCl 1 M and OM > 5%, the hypothesized sorptionmechanism for tricyclazole were hydrophobic interactions.keywords: Sediments, Rice paddy �elds, Tricyclazole, Lab experiments, SorptionOttorino-Luca Pantani, olpantani@uni�.itE-mail address:olpantani@uni�.it

Supporting grant: supported by the Italian Ministry of Education, University and Research within the national project MIUR-COFIN 2002

Oral Presentation

QSAR MODELLING AND MULTIVARIATE ANALYSIS OF THE ENVIRONMENTAL BE-HAVIOUR OF ORGANIC POLLUTANTS.

Ester Papa1, Paola Gramatica1

1DBSF - Università dell'Insubria

The limited availability of experimental data which are necessary for risk assessment of chemicals, and thederived general lack of knowledge of properties and activities of existing substances, increased, in the lastdecades, the interest on development and validation of alternative predictive methods. QSARs (QuantitativeStructure-Activity Relationships) are based on the assumption that the structure of a molecule contains thefeatures responsible for its physical, chemical, and biological properties, which can be described by numericaldescriptors. By applying QSAR models, based on di�erent chemometric methods, the activity (property,or reactivity, etc.) of a new or untested chemical can be predicted from the molecular structure of similarcompounds whose activities are known. The application of these models produces predicted data whichare useful to �ll data gaps and for the creation of priority lists through chemical screening approaches.Persistent, Bioaccumulative and Toxic (PBTs) as well as Endocrine Disruptors (EDs) chemicals are amongthe compounds of higher concern due to the harm they pose to human and environment, and they requireAutorization in the new European regulation of chemicals REACH. This presentation shows di�erent QSAR-based and multivariate approaches, which include externally validated regression and classi�cation modelsthat allow for the prediction of the potential behaviour of PBTs and EDs. The application of these QSARscould be particularly useful for screening and prioritization purposes, also a priori (before chemical synthesis),and in the design of new, safer alternatives to existing dangerous products.keywords: QSAR, chemometric methods, PBT, Endocrine Disruptors, predictivityEster Papa, Dipartimento di Biologia Strutturale e Funzionalevia J.H. Dunant 321100 VareseE-mail address:ester.papa@uninsubria.it

79

Bologna, Italy TIES 2009

Oral Presentation

BAYESIAN METHODS FOR RECONSTRUCTING PAST CLIMATE HISTORIES.

Andrew Parnell1, John Haslett2, Michael Salter-Townshend1

1University College Dublin, 2Trinity College Dublin

Statistical palaeoclimate reconstruction involves making inference on the past climate c corresponding to asingle sample at a given depth in a core of, for example, sediment. Such a sample is, typically, a multivariatecount vector y; here the data are pollen counts, changes in which re�ect changes in past vegetation, driven(largely) by climate change. We regard reconstruction as making statistical inference on c, given y; that isthe study of p(c|y). For a single core, this may be thought of as reconstructing the climate history for theentire period of time corresponding to the entire core. This presentation will focus on the joint aspects ofsuch inference; that is on p(c(T) | y1,..., yn) where c(T) denotes c(t) for all points t in a given time periodT - a "history". Joint inference exploits temporal smoothness in c(t); 'mostly' the climates at times t1 andt2 are similar if t1 is close to t2. Such smoothness permits borrowing-of-strength; several observations y cancontribute to inference on a every c(ti). One technical issue is the modeling of such stochastic smoothness; weadopt a random walk model. A challenge arises however, from the fact that, although depths (in the sediment)are known for every sample, information on the associated time is uncertain; we discuss the inference. Wepresent an e�cient Monte Carlo algorithm for sampling such histories given all the associated uncertainties.keywords: Bayes, Monotone Processes, mixtures of Gaussians, long-tailed random walk, normal inverse GaussiandistributionAndrew Parnell, Room 551 Library Building School of Mathematical Sciences University College Dublin Dublin 4IrelandE-mail address:andrew.parnell@ucd.ie

Oral Presentation

A CAUSAL MODELLING APPROACH TO SPATIAL AND TEMPORAL CONFOUNDINGIN ENVIRONMENTAL IMPACT STUDIES.

Warren Paul1

1Department of Environmental Management and Ecology, La Trobe University

Deciding whether an anthropogenic disturbance caused a change in an ecosystem is problematic owing tospatial and temporal confounding, and the generally accepted approach to the problem has been to use aBefore-After Control-Impact (BACI) type of design. However, advances in graph theory and causal mod-elling now provide a formal basis for selecting those covariates that need to be observed in order to controlconfounding bias [Pearl, J., Biometrika (1995), 82, 669-710], and the solution provided by this approach issimpler than that provided by any of the BACI designs, including BACI, BACIP, Beyond-BACI, or MBACIdesigns. Based on an explicit description of the nature of spatial and temporal confounding in causal mod-els for two hypothetical environmental impact studies, it is argued that confounding can be controlled byadjusting directly for spatial or temporal location in a Before-After (BA) or Control-Impact (CI) study. Itis further argued that there is no advantage in combining these designs in a BACI-type study, both from acausal modelling perspective and from the perspective of the assumptions implicit in BACI designs. Issuesfor the analysis of data from BA and CI designs are discussed and two applications involving multispeciesdata are given: the 1978 Amoco Cadiz oil spill o� the coast of Brittany, France [Dauvin, J-C, Estuarine,Coastal and Shelf Science (1982), 14, 517-531], and an example of 'no disturbance' from the Rose River inVictoria, Australia [Shackleton, M., Honours Thesis, La Trobe University (2006)].keywords: causal modelling, spatial and temporal confounding, environmental monitoring, before-after and control-impact designs, pseudoreplicationWarren Paul, La Trobe University PO Box 821 Wodonga, Victoria, Australia, 3689.E-mail address:w.paul@latrobe.edu.au

Supporting grant: Travel Grant from The Ian Potter Foundation

80

Bologna, Italy TIES 2009

Oral Presentation

HARP-A SOFTWARE TOOL FOR DECISION SUPPORT DURING NUCLEAR EMER-GENCIES.

Petr Pecha1, Radek Hofman1, Emily Pechova2

1Institute of Information Theory and Automation, 2Institute of Nuclear Research, division EGP

The contribution presents software system HARP designed for fast assessment of radiological consequencesof accidental releases of radionuclides into the living environment. A special segmented Gaussian plume modelis introduced, which can take into account both short-term meteorological forecast and release dynamics ofdischarged admixtures. The system o�ers various alternative options of input parameters de�nition of therelease scenarios in their atmospheric, deposition, ingestion and dose parts. For that reason the softwareproduct can serve as a training tool enabling responsible sta� to improve their knowledge and perception ofthe problem details.A special emphasis is laid on proper treatment of types of input parameter �uctuationsin sense of di�erentiation between variability and uncertainty. Some model errors arising from the conceptuallimitations can be roughly estimated on the basis of computations with alternative submodel options (atmo-spheric dispersion formulae for smooth/rough terrain, e�ect of near-standing buildings, in�uence of size ofaerosol particles on dry deposition velocity, variability in Julian day of radioactive fallout, alternative semi-empirical expressions for time evolution of deposited radioactivity on terrain and some others). The optionscan be entered interactively from the screen and provide fast response for comparison. The presentationdemonstrates the initialisation of computation runs consisting in interactive de�nitions of accidental releasescenario, atmospheric and deposition parameters and dynamic food chain model data. Visualisation of theradiological outputs and various countermeasure actions can be done on the proper map backgrounds of theNPPs operating in the Czech Republic.keywords: radioactive pollution, atmospheric propagation, radiological consequencePetr Pecha, Institute of Information Theory and Automation, Pod Vodarenskou vezi 4, 182 00 Prague 8, CzechRepublicE-mail address:pecha@utia.cas.cz

Supporting grant: Project No. 102/07/1596 (2007-2009) of the Grant Agency of the Czech Republic:ASSIMILATION METH-ODS OF MATHEMATICAL MODEL OF HARMFUL SUBSTANCES PROPAGATION WITH REAL OBSERVATIONS DES-IGNATING FOR EMERGENCY MANAGEMENT SUPPORT

Oral Presentation

FORECAST OF OZONE POLLUTANT UP TO 5 DAYS IN ADVANCE IN ROME URBANAREA BY MEANS NEURAL NET.

Armando Pelliccioni1, Stefano Lucidi2, Vittorio La Torre2, Fabrizio Pungì3

1Ispesl, 2University of Rome "La Sapienza"- DIS, 3Ispesl-Dipia

Air quality problems produced by high levels of ozone regards e�ects on human health and is relatedwith respiratory problems. Ozone is a reactive gas and presents concentration which are dependent bothfrom the meteorological conditions and seasonal e�ects. The prediction of Ozone levels is very complex toobtain as described in di�erent studies. For Ozone models one of the most di�cult problems to deal with,is the simulation of the chemical reactions, linked to the long range transport and to the incoming solarradiation and turbulence conditions. Among the complex systems, an important tool in order to forecast airpollution data is the neural network (NN), that be used in assessing the dynamics of such systems. In ourwork, NN methods have been developed to forecast hourly ozone levels using data from one day before up tothe �fth day in advance. We analyze one year data of urban Rome city. As input variables we consider twosimulations: �rst using only pollutants and meteorological measurements and second one by using, other thanthe conventional variables, hour of day and the day itself. For every simulation, we calculate a correlationdecay curve for the ozone levels and we �nd good correlations for both simulations. The use of the hourof days as input variables improves in signi�cant way the result of simulations and suggest a way how tooptimize the environmental simulation by Neural net approaches.keywords: Neural Net, Ozone forecast, Optimization methods, Urban siteArmando Pelliccioni, Ispesl-Dipia, Via Fontana Candida 1, 00040, Monteporzio Catone (RM), ItalyE-mail address:armando.pelliccioni@ispesl.it

81

Bologna, Italy TIES 2009

Oral Presentation

ENVIRONMENT MONITORING PERFORMED BY ADVANCED HYBRID AIRSHIP ATLOW ALTITUDE.

Giuseppe Persechino1, Pasquale Schiano1, Massimiliano Lega2, Rodolfo Napoli2

1CIRA-Italian Aerospace Research Centre, 2DiSAm-Dipartimento di Scienze per l'Ambiente-University ofNaples PARTHENOPE

This paper introduces last results about the use of an Integrated System (Advanced Hybrid Airship, BI-Lift,and a speci�c "payload") developed by us and specialized to the environmental monitoring at low altitude.The low altitude concept, emphasized by the capability to move through a 0 to 300 metres range, allowsthe measurement of pollution through a 3d air matrix close to ground and the sampling, at high accuracy,of the data on ground with advanced technology (i.e.infrared thermography).The develop of the airship hasbeen done by verifying the particular correspondence of it to the requisites of a mission and, at the sametime, appraising the same one like e�cient "ampli�er" of what is measurable from the sensors positionedon the ground in "pure" con�guration.Particularly, the IS will allow the exploration of three-dimensionalspaces without altering the measurement parameters, returning geo-referenced data, guaranteeing minimalinvasion and maximal safety for the operational context. The "payload" is an array of sensors �nalized to themeasurement of environmental parameters and it is specialized to be hosted on board of this speci�c aircraft.The results will be useful for both the domains: the aerospace and the environmental �eld. On the aerospacedomain it will be important to validate the system behaviour in terms of Flight mission and capability tocomplement the pre-existing acquisition resources. On the environmental domain we will have an innovativepoint of view to observe several matrixes and to grab data at di�erent altitudes.keywords: Environment Monitoring, aerial platform , airship, low altitudeGiuseppe Persechino, CIRA-Italian Aerospace Research Centre-via Maiorise Capua (CE) 81043 ItalyE-mail address:g.persechino@cira.it

Oral Presentation

BAYESIANNONPARAMETRICMIXTURES FOR LOCAL CLUSTERING OF FUNCTIONALDATA.

Sonia Petrone1, Michele Guindani2, Alan E. Gelfand3

1Bocconi University, 2University of New Mexico, 3Duke University

I discuss a Bayesian mixture model with local random e�ects for dimension reduction and inference withfunctional data. In many applications, one has noisy observations of a sample of curves, or surfaces, thatpresent a similar pattern, except perhaps for a few regions that provide heterogeneus behavior (e.g., damaged,or polluted, areas on an otherwise smooth surface). For data of this nature, one wants both a global clusteringof similar patterns, as well as local clustering of sub-regions of the curves. To this aim, we model theindividual curves as realizations of a Gaussian process, with individual-speci�c, local random e�ects. In aBayesian nonparametric approach, global and local clustering is induced by an appropriate choice of the prioron the latent distribution of the random e�ects, that generalizes the popular Dirichlet priors. The dimensionreduction obtained with the proposed model is illustrated in an example with image data. Extensions tospatio-temporal, environmental data will be discussed.Sonia Petrone, DEC - Università BocconiVia Roentgen 120136 MilanoE-mail address:sonia.petrone@unibocconi.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

82

Bologna, Italy TIES 2009

Oral Presentation

A CHEMOMETRIC APPROACH BASED ON THE AUTOCOVARIANCE FUNCTION FORHANDLING COMPLEX SIGNALS FROM ENVIRONMENTAL MONITORING.

Maria Chiara Pietrogrande1, Mattia Mercuriali1, Nicola Marchetti1, Luisa Pasti1, Dimitri Bacco1, GaetanoZanghirati2, Francesco Dondi1

1Department of Chemistry, University of Ferrara, 2Department of Mathematics, Math4Tech Center

Atmospheric aerosols consist of a complex mixture of hundreds of compounds belonging to many di�erentclasses: despite this complexity, in environmental monitoring and assessment studies, the more useful analy-sis is limited to selected compounds, representing a chemical �ngerprint of the possible input sources. Thismotivates the need for computer-assisted signal processing procedures to extract usable information from thecomplex signals obtained from analytical techniques. This is particularly true for hyphenated techniques suchas Gas chromatography-Mass Spectrometry (GC-MS) which generate data sets. In this work a chemometricapproach based on the AutoCovariance Function (ACVFtot) is described as a very powerful tool for inter-preting complex chromatograms (Pietrogrande et al., Anal. Chim. Acta, 594 (2007) 128-138). Attention isfocused on the terms of homologous series, n-alkanes and n-alkanoic acids, since they contain very relevantinformation to trace the origin and fate of atmospheric aerosols. The recognition of the presence of the termsof the series and information on their distribution pattern, i.e., odd/even predominance, are diagnostic forassessment of natural versus anthropogenic origin. Two main parameters can be directly extracted from theACVFtot computed on the acquired chromatographic signal: the number of terms of the series and the carbonpreference index (CPI) (Pietrogrande et al., Analyst, 134 (2009) 671-680).The ACVFtot approach was alsoextended from one dimensional chromatograms to two dimensional data obtained from GC-MS hyphenatedtechnique or bi-dimensional separations (Marchetti et al., Anal. Chem., 76 (2004) 3055-3068)keywords: signal processing , complex chromatograms, GC-MS, organic tracersMaria Chiara Pietrogrande, Via L.Borsari 46, 44100 Ferrara, ItalyE-mail address:mpc@unife.it

Oral Presentation

OPERATIONAL OCEANOGRAPHY: THE SCIENCE BASED APPROACH TO MARINEMANAGEMENT PROBLEMS.

Nadia Pinardi1, Srdjan Dobricic2, Ralph Milli�3

1University of Bologna and INGV, 2Centro Euro-Mediterraneo per i Cambianeti Climatici, 3Coloradoresearch Associates, NWRA

Operational oceanography nowadays embraces the frontier research for ocean monitoring and forecasting,as operational meteorology did in the �fties and it is continuing to do today. From the nineties, the availabilityof reliable and real time satellite and in situ data together with advanced hydrodynamics numerical modelshas allowed the development of a system that monitors in an integrated way the ocean variability and forecastsinto the future. A Mediterranean ocean Forecasting System (MFS) has been developed in the past twentyyears to deliver generic services based upon common-denominator ocean state variables and products thatare required to help meet the needs for information of those responsible for marine environment managementand protection, civil and military security at sea, monitoring of climate variability and change. It is thescience based approach to the needs for ocean information from our society, organised as a meteorologicalo�ce for the marine environment. The key element of this service is the set of basic statistical tools thatmerge deterministic numerical model output with observational data, so-called data assimilation techniques,and the Montecarlo methods to understand and quantify uncertainties in the forecasts. The presentation willdiscuss in particular the data assimilation approach developed for the MFS for both satellite and in situ dataand the Bayesian Hierarchical Model developed for surface winds in order to produce ensemble forecasts ofocean currents.Nadia Pinardi, Gruppo Nazionale di Oceanogra�a OperativaIstituto Nazionale di Geo�sica e Vulcanologia Via AldoMoro 44 40127 BolognaE-mail address:n.pinardi@sincem.unibo.it

83

Bologna, Italy TIES 2009

Oral Presentation

DARWINIAN THEORY BASED TECHNIQUE TO PREDICT AIR POLLUTANT CONCEN-TRATIONS.

José Pires1, F.G. Martins1, M.C.M. Alvim-Ferraz1, M.C. Pereira1

1LEPAE, Faculdade de Engenharia da Universidade do Porto

Tropospheric ozone (O3) is a secondary air pollutant and it is one of the most important due to theseveral negative impacts on human health, vegetation, climate and materials. The high complexity thatcharacterizes its formation led the constant use of statistical models to predict the concentrations of thisair pollutant, aiming to inform the population when high concentrations are expected. The model structureis not a degree of freedom when the model is determined and this constitutes an important limitation forstochastic case studies. Genetic programming (GP) is a procedure that uses the Darwin principles to createa model, optimizing its structure and parameters, simultaneously. The models are encoded in tree structures(individuals) that are modi�ed in an iterative process. An improved technique, called multi-gene GP, uses thesame procedure. However, it considers several tree structures for each individual and the linear combinationof the correspondent outputs gives the output of the model. This study aims to predict the next day hourlyaverage O3 concentrations using multi-gene GP algorithm. The study period was two weeks of May 2004,where high O3 concentrations were measured. The last seven days were used for the test period and thecorresponded O3 data of each one was predicted using the data of the seven days before. The results showedthat multi-gene GP achieved good performance in the prediction of O3 concentrations, which means thatthis is a promising technique to model stochastic problems.keywords: Multi-gene genetic programming, tropospheric ozone, air pollution modellingJosé Pires, Faculdade de Engenharia da Universidade do Porto LEPAERua Dr. Roberto Frias, s/n4200-465 Porto,PortugalE-mail address:dce05005@fe.up.pt

Oral Presentation

ANAGGREGATE AIR QUALITY INDEX CONSIDERING INTERACTIONS AMONG POL-LUTANTS.

Antonella Plaia1, Mariantonietta Ruggieri1, Anna Lisa Bondì1

1DSSM - University of Palermo

Several countries provide an Air Quality Index (AQI) to communicate air pollution, but there is not aunique and internationally accepted methodology for constructing it. The most of the proposed indices arebased on the USA AQI by EPA and are de�ned by the value of the pollutant with the highest concentration.For each pollutant, a sub-index is computed by linear interpolation according to the grid in a table, butthe breakpoints of such a table may di�er from one country to another, as well as the descriptors of eachcategory, the air quality standards, the functions chosen as daily synthesis to aggregate hourly values at eachsite for each pollutant, and so on. Anyway the main drawback is that such indices do not take into accountthe combined e�ects of all the considered pollutants, giving little emphasis to e�ects occurring over long timeperiods, such as chronic health e�ects, damages on vegetation, animals, monuments. With the purpose toaccount for multiple pollutant exposure, some attempts have already been made in literature. In this paperwe propose an aggregate AQI which tries to overcome the problems listed above. The proposed index canbe easily computed and interpreted, so allowing comparisons among di�erent situations, both in time andspace. Moreover, the association of an appropriate measure of dispersion, that can also be considered as ameasure of concentration, adds very important information.keywords: Air quality index, Multiple pollutant exposureAntonella Plaia, viale delle Scienze ed. 1390128 PalermoE-mail address:plaia@unipa.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

84

Bologna, Italy TIES 2009

Oral Presentation

THREE NONLINEAR STATISTICAL METHODS TO ANALYZE PM10 POLLUTION INROUEN AREA.

Jean-Michel Poggi 1, Francois-Xavier Jollois2, Bruno Portier3

1Univ. Paris Descartes & Univ. Orsay, 2Univ. Paris Descartes, 3INSA Rouen

The problem is to analyze PM10 pollution during 2004-2006 in Haute-Normandie area (France) using sixdi�erent monitoring sites and to quantify the e�ects of variables of di�erent types, mainly meteorologicalversus other pollutants measurements. This work takes place in a scienti�c collaboration between Air Nor-mand (the local observatory of air quality) from the applied side and Paris-Descartes University and INSAof Rouen from the academic side.The talk focus on two aspects: pollution modeling and quanti�cation of alocal part and a regional part of PM10 pollution. We will introduce and motivate the three modern nonpara-metric statistical methods used to handle the problem: �rst, random forests focusing on relative importanceof variables and variable selection issues as well as marginal e�ects of variables; second, partially nonlinearadditive model using two original climatic variables to partition data and model each cluster, and �nally,clusterwise linear modeling. Next, we will focus on an attempt of quanti�cation of what we call in a broadsense a local part and a regional part of PM10 pollution.So the talk will give some methodological statisticalinsights together with some conclusions useful from the environmental perspective.keywords: Pollution, PM10, Random Forests, Regression, Classi�cationJean-Michel Poggi , University Paris Sud-Orsay, Lab. de Mathematiquesbat. 425, 91405 Orsay, FranceE-mail address:Jean-Michel.Poggi@math.u-psud.fr

Supporting grant: GRASPA invited/contributed session

Oral Presentation

A CRITICAL REVIEW OF SOME STATISTICAL ISSUES IMPLIED BY THE USE OFMULTIVARIATE RECEPTOR MODELS.

Alessio Pollice1

1Dipartimento di Scienze Statistiche - Università degli Studi di Bari

In pollution source apportionment studies, multivariate receptor models heavily rely on statistical factoranalytic techniques to estimate the source-speci�c contributions from a large number of observed chemicalconcentrations. The scope of this talk is to o�er a review of some recent statistical literature in order todescribe the main features and recent advances of this �eld, advice on the possible "statistical risks" in usingstandard methods and existing software and �nally show how some theoretical and practical failures of thecommonly used methodologies can be addressed by proper statistical modeling and estimation tools. Thetopics addressed include: the estimation of the number of sources, model identi�ability issues, the considera-tion of the temporal dependence in the data and systematic e�ects of physical factors such as meteorologicalconditions, possible extensions to spatial data collected by multiple receptors and the assessment of sourcespeci�c health e�ects.keywords: Multivariate receptor models, Pollution source apportionment, Factor analysisAlessio Pollice, Dipartimento di Scienze Statistiche - Università degli Studi di Bari, Via C. Rosalba n.53, 70124 BariE-mail address:apollice@dss.uniba.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

85

Bologna, Italy TIES 2009

Oral Presentation

FUNCTIONAL ESTIMATION OF GAUSSIAN DAGUMRANDOMFILEDS ANDRELATEDMODELS.

Emilio Porcu1, Maria Dolores Ruiz Medina2, Rosaura Fernàndez Pascual3

1University of Göttingen, 2University of Granada, 3University of Jaen

In this paper, fractal and long-range dependence Gaussian random �eld models are introduced. Speci�cally,we consider the Dagum class of covariance functions and other classes related to them. We discuss theproperties of their associated Gaussian random �elds in terms of regularity, fractality and long memory. Wealso discuss their pseudodi�erential representation and characterisation in terms of Sobolev spaces.keywords: Fractals, Long Memory, Sobolev SpacesEmilio Porcu, Universitat Göttingen, Goldschmidtstrasse 737077 Göttingen GermanyE-mail address:porcu@mat.uji.es

Poster Presentation

BIO-DIVERSITY IN VINEYARDS WITH CONVENTIONAL, BIOLOGICAL AND INTE-GRATED TREATMENT.

Zdenek Pospisil1, Jitka Kuhnova1

1Masaryk University, Brno, Czech Republic

The contribution presents some �ndings on impact of agricultural technology to bio-diversity.Twelve South-Moravian vineyards were observed; three of them were protected by conventional chemical treatment, threeby introduction of natural enemies of pests, three by bio-agents supplied by mild and restricted chemicalintervention and three vineyards were abandoned to natural succession. Abundances of butter�ies, beetlesand birds were watched and relevant indeces of diversity and equitability were evaluated. Surprisingly,the greatest diversity was found on areas treated by integrated way not on the biologically treated ones.One possible conclusion is that the bio-diversity is in�uenced rather by the distance of agricultural areafrom natural surroundings than by an agriculture applied. Some data and analyses supported the idea willbe presented.The second investigation was concerned with bio-diversity recovery (or decay) after massivechemical intervention. Data having reference to an evolution of insect abundances on two agricultural areas(predominantly treated by conventional and integrated means) were collected and analyzed.keywords: bio-diversity, pest controlZdenek Pospisil, Masaryk University, Department of Mathematics and Statistics, Kotlarska 2,617 37 Brno, CzechRepublicE-mail address:pospisil@math.muni.cz

Oral Presentation

OPTIONS FOR THE DESIGN OF A SOIL MONITORING SCHEME.

Jacqueline Potts1

1Biomathematics and Statistics Scotland

A recent project examined options for the design of a UK soil monitoring scheme. Both model-based anddesign-based methods were considered. Any new scheme will be more e�cient for estimating changes in soilquality if it incorporates at least some data from existing monitoring networks. These existing monitoringnetworks tend to use grid-based sampling designs. Grid-based schemes have the advantage that they are oftenmore e�cient than other designs because spatial coverage is optimized. They are also easier to implementand more �exible than strati�ed random sampling schemes, because it is not necessary to have accurate mapsof strata in advance and it may be easier in the future to provide estimates for reporting classes that were notincluded in the initial strati�cation. However, the main disadvantage is that there is no design-based unbiasedestimator of the variance of the sample mean.A simulation study for several variogram models will be used toexamine the bias in the estimation of the variance of the sample mean if the systematic sample is treated asthough it were a simple random sample and if pseudo-strati�cation is used. Although treating the systematic

86

Bologna, Italy TIES 2009

sample as a simple random sample can result in considerable bias, the bias under pseudo-strati�cation isgenerally fairly small.keywords: soil monitoring, sampling designJacqueline Potts, Biomathematics and Statistics Scotland, Macaulay Land Use Research Institute, Craigiebuckler,Aberdeen, AB15 8QH, UKE-mail address:j.potts@bioss.ac.uk

Poster Presentation

DO OZONE CONCENTRATIONS AFFECT RESPIRATORY HEALTH IN SCOTLAND?.

Helen Powell1, Duncan Lee1, Adrian Bowman1

1University of Glasgow

Short term exposure to air pollution can cause and aggravate a number of respiratory conditions. Ouraim is to develop a model which investigates whether ozone concentrations are contributing to the numberof people who are admitted to hospital with a respiratory disease. For the city of Glasgow, in Scotland, weregressed daily counts of respiratory admissions against daily mean ozone concentrations, and other covariaterisk factors using Poisson generalized linear models. It was not necessary to use a distribution suitablefor dealing with overdispersion as we found no evidence of this. The other covariates included minimumtemperature levels and a day of the week indicator as well as a substantial non-linear trend. We modelledthe trend using a natural cubic spline of day of the day study because it is numerically stable and left littlestructure or correlation in the residuals. We �tted the model with ozone lagged by 1, 2 and 3 days afterinitial exposure, and calculated the relative risk for an increase of 17 microns in ozone. We found convincingevidence for a relationship between admissions and ozone concentrations lagged by 1 day, with a relative riskof 1.7%. There was also some evidence of 'admissions displacement', as the relative risk at lag 3 was lessthan one, suggesting that all those who were a�ected by the rise in ozone would still be in hospital form lag1.keywords: air polltuion, respiratory admissions, generalised linear modelHelen Powell, Department of Statistics, 15 University Gardens, University of Glasgow, G12 8QQE-mail address:h.powell@stats.gla.ac.uk

Oral Presentation

OPTIMAL CONSTRAINED CONFIDENCE ESTIMATION VIA TAIL FUNCTIONS WITHAPPLICATIONS TO ENVIRONMENTAL DATA.

Borek Puza1, Terence O'Neill1

1Australian National University

It is often the case that the target parameter in a statistical model is constrained within some speci�edregion. This presents a problem if the data lead to a con�dence interval (CI) which lies entirely outside thatregion. For example, in an environmental study a regression parameter representing the e�ects of pollutionon human mortality might be known to be positive, whereas the data by chance lead to a 95% CI that isentirely negative, and hence to an empty CI after excluding all inadmissible values. This paper presents away of dealing with this problem via the method of tail functions, as recently developed by Puza and O'Neill[Canadian Journal of Statistics (2006): 299-310]. If prior information is available, the method can be used toengineer an optimal CI in terms of prior expected length, and in that case it provides an alternative to theBayesian approach, which does not lead to a proper frequentist CI. The method is compared with the uni�edapproach of Feldman and Cousins [Physical Reviews D (1998): 3873-3889], which does not permit the use ofprior information. The usefulness of the tail functions approach is illustrated by application to inference onthe normal mean and binomial proportion in models involving environmental data.keywords: Con�dence interval, Constrained estimation, Optimal inference, Prior information, Tail functionBorek Puza, School of Finance and Applied Statistics,ANU College of Business and Economics, Australian NationalUniversity,Canberra, ACT 0200, AustraliaE-mail address:borek.puza@anu.edu.au

87

Bologna, Italy TIES 2009

Oral Presentation

SUSTAINABLE ECONOMICS: THE CONTRIBUTION OF OFFICIAL STATISTICS.

Walter J. Radermacher1

1Eurostat

Sustainable development is about improving human well-being. The quick proliferation of measurementconcepts during the last years has, however, created a need for harmonisation and comparability. A joint UN-ECE/OECD/Eurostat Working Group on Statistics for Sustainable Development (WGSSD) has submitteda framework for the design of sustainable development indicator sets by national governments and interna-tional organisations.With the forthcoming Climate Change summit in Copenhagen, attention focuses on theresponses of politicians on meeting the climate change targets, and the societal responses on the questionsof the relation between globalisation and sustainability. Environmental e�ects of globalisation follow thoseof economic and social globalisation. O�cial statistics are needed to monitor economic developments andtheir environmental impacts to allow policy makers to steer the process of using the environment in thedirection of a sustainable situation and towards an ecological modernisation of the economy. A developmentthat even has strengthened with the recent enormous investment programs for stimulating a fast economicrevival among others by stimulating the green economy. In the presentation the role of o�cial statistics inenvironmental statistics, sustainability measurement and statistics for climate change will be discussed. Thepresentation will show the urgent need for improving the availability of basic environment statistics, in par-ticular regarding quality and timeliness and the continuation of work on sustainable development indicators,the full operation of the Environmental Data Centres and bringing Environmental Accounting to maturity.keywords: Sustainable Development, O�cial Statistics, Climate Change, Globalisation, Green EconomyWalter J. Radermacher, Eurostat Joseph Bech building 5 Rue Alphonse Weicker L-2721 LuxembourgE-mail address:walter.radermacher@ec.europa.eu

Poster Presentation

TREND TESTING: SEARCHING FOR A BETTER ESTIMATION OF THE VARIANCEOF THE TEST STATISTIC.

M. do Rosário Ramos1

1Universidade Aberta(PT) and CMAF, University of Lisbon

The purposes of a monitoring program include the identi�cation of changes or trends in water quality vari-ables over time. Frequently is need to check if a new environmental policy giving new standards, introducedat a certain time, is really having results. The problems with standard statistical tests like the t test for theslope of the linear regression model arises when we are in the presence of one or more of the following: serialcorrelation, short or moderate time series, missing values not randomly distributed and a skewed distributionin the sample. In this work we study two trend tests: The parametric test to the least squares estimator ofthe slope and the nonparametric test based on Kendall's Tau. In previous work by the author and in papersfrom other researchers we veri�ed that the poor estimation of a strong or moderate correlation is possibly themost important cause to the weak performance of both type of tests, even when the e�ect of correlation wastaken into account in the variance of the test statistic, as we review here. Now, assuming a linear regressionmodel with AR(1) errors, we try another approach to estimate de variance, using Bootstrap methods fordependent data and an imputation method to some missing values. Through a simulation study, we comparethe performance of both tests under di�erent assumptions. Finally we present an application of the methodsin a series of water quality variables from Portuguese waters.keywords: Trend testing, Least squares, Mann-Kendall, Bootstrap, Water qualityM. do Rosário Ramos, CMAF-Centro de Matemática e Aplicações Fundamentais Complexo Interdisciplinar da Uni-versidade de Lisboa Av. Prof. Gama Pinto 2, 1649-003 Lisboa, PortugalE-mail address:ramos@cii.fc.ul.pt

88

Bologna, Italy TIES 2009

Poster Presentation

TRI-PARAMETER LOGNORMAL DIFFUSION PROCESSWITH EXOGENOUS FACTORS:INFERENCE BASED IN SIMULATED DATA.

Eva Mª Ramos-Ábalos1, Ramón Gutiérrez-Jaimez1, Ramón Gutiérrez-Sánchez1, Ahmed Na�di1

1Universidad de Granada (SPAIN)

The principal objective of the present work is to study a new stochastic lognormal di�usion process withexogenous factors, one with threshold parameter, which can be considered as an extension of the homogeneouslognormal process with the addition of a threshold parameter. This original process can be used pro�tablyin a variety of disciplines including Biology, Geology, Agriculture, Environmetrics and Population Dynamics.From the corresponding Ito's stochastic di�erential equation and the Kolmogorov equations, we obtain theanalytical expression of this process, its transition probability density function and its trend function. Thestatistical inference of the parameter is studied by considering discrete sampling of the sample paths of themodel and then using the maximum likelihood method. The estimation of the threshold parameter requiresthe solution of a nonlinear equation. To do so, we propose the Newton-Raphson method. Finally, thismethodology is applied to an example with simulated data corresponding to the process.keywords: Lognormal di�usion process, Threshold parameter, Discreet sampling, Stochastic di�erential equation,SimulationEva Mª Ramos-Ábalos, Departamento de Estadística e I.O. Facultad de Ciencias. Universidad de Granada CampusFuentenueva, s/n 18071 Granada (Spain)E-mail address:ramosa@ugr.es

Supporting grant: MTM2008-05785P06-FQM-02271

Oral Presentation

LIKELIHOOD-BASED INFERENCE FOR MAX-STABLE PROCESSES.

Mathieu Ribatet1, Simone A. Padoan2, Scott A. Sisson3

1Institute of Mathematics, EPFL, 2Laboratory of Environmental Fluid Mechanics and Hydrology, EPFL,3School of Mathematics & Statistics, University of New South Wales

The last decade has seen max-stable processes emerge as a common tool for the statistical modeling ofspatial extremes. However, their application is complicated due to the unavailability of the multivariate den-sity function, and so likelihood-based methods remain far from providing a complete and �exible frameworkfor inference. In this presentation we develop inferentially practical, likelihood-based methods for �ttingmax-stable processes derived from a composite-likelihood approach. The procedure is su�ciently reliableand versatile to permit the simultaneous modeling of marginal and dependence parameters in the spatialcontext at a moderate computational cost. The utility of this methodology is examined via simulation, andillustrated by the analysis of U.S. precipitation extremes.keywords: Composite likelihood, Extreme value theory, Max-stable processes, Spatial ExtremesMathieu Ribatet, Institute of Mathematics Ecole Polytechnique Fédérale de Lausanne STAT-IMA-FSB-EPFL, Station8CH-1015 Lausanne SwitzerlandE-mail address:mathieu.ribatet@ep�.ch

Supporting grant: GRASPA invited/contributed session

Oral Presentation

GENERALIZED PROBABILITYWEIGHTEDMOMENTSMETHODS IN EXTREMEVALUETHEORY.

Pierre Ribereau1, Armelle Guillou2, Philippe Naveau3

1Université de Montpellier 2, 2Université de Strasbourg, 3LSCE/CNRS

In 1985 Hosking et al. estimated with the so-called Probability-Weighted Moments (PWM) methodthe parameters of the Generalized Extreme Value (GEV) distribution.Their approach is still very popularin hydrology and climatology because of its conceptual simplicity, its easy implementation and its goodperformance for most distributions encountered in geosciences. Its main drawback resides in its limitationswhen applied to strong heavy-tailed densities. Whenever the GEV shape parameter is larger than 0.5, theasymptotic properties of the PWMs cannot be derived and consequently, asymptotic con�dence intervals

89

Bologna, Italy TIES 2009

cannot be obtained.To broaden the validity domain of the PWM approach, we take advantage of a recentextension of PWM to a larger class of moments, called Generalized PWM (GPWM). This allows us to derivethe asymptotic properties of our estimators for larger values of the shape parameter.The performance of ourapproach is illustrated by studying simulations of small, medium and large GEV samples. Comparisons withother GEV estimation techniques used in hydrology and climatology are performed.Pierre Ribereau, I3M, CC051, Place Eugène Bataillon 34095 MontpellierE-mail address:pribere@math.univ-montp2.fr

Supporting grant: GRASPA invited/contributed session

Poster Presentation

GOMPERTZ-LOGNORMAL DIFFUSION PROCESS FOR MODELLING THE ACCUMU-LATED NUTRIENTS DISSOLVED IN A CULTIVATION OF CAPISCUM ANNUUM.

Nuria Rico1, Desiree Romero1, Francisco Torres1, Patricia Román1

1Universidad de Granada, Spain

With the aim to minimize the pollution of soils and underground aquiferous, there have been realizeddiverse studies focused in the analysis of the productivity and evaluation of the dissolution of nutrients. Thesestudies provide data showing a mixed behaviour between Gompertz and exponential growth; concretely, thepattern growth is associated with a curve that shows a change-point where the growth turns from Gompertzto exponential.In order to study this type of behaviour pattern from a dynamical point of view, in this paperwe propose a new non-homogeneous di�usion process. This process is built (by di�erent methods) in sucha way that its mean function is a mixture between Gompertz and exponential curves, so it can be usefulto study data with behaviour as the mentioned above.In addition, a wide study of the same one is realized,including its main characteristics. An inferential study is carried out based on discrete sampling including analternative approach for solving the likelihood equation, taking into account the di�culties encountered withstandard procedures. Two applications have been developed: the former, based on simulated data, validatesthe procedure followed, and the second considers an application to real data showing its usefulness to modelthe accumulated nutrients dissolved in a cultivation of Capiscum Annuum along time.keywords: Lognormal di�usion process, Gompertz di�usion process, growth modellingNuria Rico, Departamento de Estadística e Investigación Operativa. Faculta de Ciencias, Universidad de Granada.Avenida Fuentenueva, sn. 18071. Granada. SpainE-mail address:nrico@ugr.es

Supporting grant: This work was supported in part by the Ministerio de Educación y Ciencia, Spain, under Grants MTM2008-05785/MTM and HI2007-0034, and by the Consejería de Innovación, Ciencia y Empresa (Junta de Andalucía), Spain, underGrant P06-FQM-02271.

Oral Presentation

DOES IGNORING MODEL SELECTION EFFECTS IN ASSESSING THE EFFECT OF PMON MORTALITY MAKE US TOO VIGILANT?.

Steven Roberts1, Michael Martin1

1Australian National University

This study investigates the extent of "overcon�dence" that can result if model selection e�ects are notaccounted for in time series studies of the association between particulate matter air pollution (PM) andmortality. The results of our study show that not accounting for model selection e�ects can result in thestandard errors estimated for PM-mortality e�ect estimates being approximately 40% too small and theobserved size of tests of no association between PM and mortality being approximately �ve times the nominalsigni�cance level. These results illustrate that studies that fail to account properly for the e�ect of modelselection risk lowering the accepted burden of proof for concluding a statistically signi�cant associationbetween PM and mortality.keywords: mortality, time series, air pollution, model selectionSteven Roberts, School of Finance and Applied statistics,College of Business and Economics, Australian NationalUniversity, Canberra ACT 0200,Australia. E-mail address: steven.roberts@anu.edu.au

Supporting grant: ARC Discovery grant

90

Bologna, Italy TIES 2009

Oral Presentation

A SPATIO-TEMPORAL MODEL FOR POINT-SOURCE MODELLING IN CRIMINOL-OGY: A CCTV APPLICATION.

Alexandre Rodrigues1, Peter Diggle1

1Lancaster University

Closed Circuit Television Cameras (CCTV) have become a common feature of public life. In an attempt tocontrol the increase in the rate of crimes, many local authorities have installedCCTV cameras in their towncentre streets. We propose a spatio-temporal model framework to, �rst, quantify the e�ect of the CCTVcameras in the spatial distribution of crimes and, mainly, to look into the disappearance, if it occurs, of thise�ect over time. We apply our methodology in a complex CCTV intervention problem in Belo Horizonte,Brazil.keywords: CCTV, spatio-temporal modellingAlexandre Rodrigues, 29, Portland Street, Lancaster, UKPost code: LA1 1SYE-mail address:a.rodrigues@lancaster.ac.uk

Poster Presentation

MULTILEVEL ZERO-INFLATED REGRESSION FORMODELLING SPECIES ABUNDANCEIN RELATION TO HABITAT: A BAYESIAN APPROACH.

Marco A. Rodríguez1, Clarice G.B. Demétrio2, Silvio S. Zocchi2, Roseli A. Leandro2, Julie Deschênes1

1Université du Québec à Trois-Rivières, Trois-Rivières, Canada, 2ESALQ/Universidade de São Paulo,Piracicaba, Brasil

Ecological data often have clustered or nested structure, in which observations are made on units grouped atdi�erent hierarchical levels. We examined the variation in counts of the slimy sculpin, a �sh species commonlyfound in North American streams, in relation to habitat descriptors from sites nested hierarchically withina river basin in eastern Canada. The sampling scheme comprised three levels: 600 sites distributed among120 reaches and 22 streams of the Cascapedia River, Québec. The distribution of counts was highly over-dispersed and had excess zeros, with nearly 10% of the sampled sites having zero counts. Zero-in�ated Poisson(ZIP) regressions, as well as multilevel extensions of this model incorporating random e�ects, were used toaccount for overdispersion and potential intra-group correlations arising from the nested sampling scheme.Eight environmental variables were considered as predictors. Parameter estimates were obtained by MCMC;comparisons between models of di�ering complexity were based on DIC and posterior predictive checks. Theinclusion of random e�ects allowed for improved assessment of the environmental in�uences and the spatialstructure of unexplained variation. The zero-in�ated regressions were useful in distinguishing structuralfrom sampling zeros and identifying the main environmental determinants of incidence (presence/absence)separately from those of abundance (number of individuals), two key objectives in studies of habitat quality.Heterogeneity in count data is common in ecological studies and is probably best viewed as a potentially richsource of information rather than as a nuisance.keywords: clustering, count data, overdispersion, random e�ects, slimy sculpinMarco A. Rodríguez, Département de chimie-biologie, Université du Québec à Trois-Rivières,3351 boul. des Forges,Trois-Rivières (Québec),G9A 5H7, CanadaE-mail address:marco.rodriguez@uqtr.ca

Supporting grant: Natural Sciences and Engineering Research Council of Canada Discovery Grant

Oral Presentation

BAYESIAN ESTIMATION OF THE CONDITIONAL INTENSITY FUNCTION IN SELF-CORRECTING POINT PROCESSES APPLIED TO THE SEISMIC ACTIVITY OF ITAL-IAN TECTONIC REGIONS.

Renata Rotondi1, Elisa Varini1

1CNR - IMATI

To enrich the point processes analysed in statistical seismology with recently developedgeologic and ge-odetic measurements, we propose a new version of self-correcting-type processwhich compares the expectedand the observed displacement obtained respectively by the unknownseismic slip rate and by the sum of the

91

Bologna, Italy TIES 2009

slips caused by every earthquake, calculated fromthe moment magnitude.Assuming that the larger the di�er-ence D(t) between these displacements is, the largerthe occurrence probability is because there is a slip de�citwhich could be covered by aforthcoming earthquake, we suppose that D(t) is the level of the physical processcontrollingthe rate of a stochastic point process similar to a stress release process withstepwise exponentiallyincreasing intensity function.Estimation is performed through MCMC methods. The initial displacement oftheregion is not statistically identi�able; we have therefore chosen to determine it byminimizing the areabetween the two curves representing the expected and theobserved displacement. In this way the periods inwhich there is a slipde�cit, that is when the expected slip is larger than the observed one, canbe consideredas alarm periods and we say that the events that occurred in those intervalscorrespond to right backwardpredictions, while the others would not be correctly forecast.Finally the model is validated on the basis ofthe Bayes factor by comparing it with the stressrelease model and the homogeneous Poisson process.keywords: Markov chain Monte Cralo methods, stress release model, Bayes factor, hazard functionRenata Rotondi, CNR - IMATI Via Bassini 15 20133 MilanoE-mail address:reni@mi.imati.cnr.it

Oral Presentation

ON THE CONCEPT OF STRUCTURAL COMPONENTS WITH AN APPLICATION TOWEATHER FUNCTIONAL DATA.

Valentin Rousson1, Juhyun Park2, Theo Gasser3

1University Hospital Center and University of Lausanne, 2University of Lancaster, 3University of Zurich

Weather characteristics, such as temperature, are often measured along time, leading to the analysis ofa functional data set. Functional principal components is then a useful tool to reduce the dimension andto summarize the important aspects of the data. Interpretation of principal components is however notalways obvious and alternative methods, such as varimax rotation, are often considered to provide a moreinformative summary of the data. In this talk, we introduce the concept of structural components whicho�ers an alternative approach to de�ne interpretable components. In particular, we distinguish between twotypes of components: block-components and di�erence-components. The former are positive on the wholetime period considered and are therefore easier to interpret. In practice, our approach is shown to o�er auseful compromise between functional principal components and varimax, as illustrated on weather data.keywords: functional data, functional principal componentsValentin Rousson, Rue du Bugnon 171005 Lausanne SwitzerlandE-mail address:valentin.rousson@chuv.ch

Oral Presentation

SPATIAL ANALYSIS OF REGIONAL CLIMATE MODEL ENSEMBLES.

Steve Sain1

1NCAR

The North American Regional Climate Change Assessment Program (NARCCAP) seeks to examine theuncertainty in the output of regional climate models and projections of future climate and climate change.At the heart of the program is an ambitious experiment that seeks to use a number of regional climate models(RCMs) with boundary conditions supplied by di�erent atmosphere-ocean general circulation models (GCMs)to produce a wide range of model output over North America. Our goal within this program is to developstatistical methodology to analyze this model output and assess and quantify the sources of uncertainty.To that end, we are developing a Bayesian hierarchical framework that is based upon a multivariate spatialmodel. This allows us to capture the complex distribution of the spatial �elds produced by these regionalclimate models. Case studies will be presented based on an ensemble of regional climate model output overthe western United States as well as an initial analysis of NCEP-driven regional model output associatedwith NARCCAP.Steve Sain, Geophysical Statistics Project Institute for Mathematics Applied to Geosciences National Center forAtmospheric Research P.O. Box 3000 Boulder, CO 80307-3000USAE-mail address:ssain@ucar.edu

92

Bologna, Italy TIES 2009

Oral Presentation

THE INLA METHOD FOR MULTIVARIATE COUNTS DATA.

Michael Salter-Townshend1, John Haslett2

1University College Dublin, 2Trinity College Dublin

Bayesian statistical methods often involve computationally intensive inference procedures. Sampling meth-ods may be computationally intensive and su�er from long run times and high potential sampling error. TheIntegrated Nested Laplace Approximation [Roy. Statist. Soc. Ser. B (2009): 319-392] constructs closed-formapproximations to the posterior and o�ers a fast and accurate alternative. Essentially, a Gaussian MarkovRandom Field is used to approximate the posterior of the likelihood parameters.The motivating dataset forthis work consists of pollen counts data observed at known climate-space locations [Roy. Statist. Soc. Ser. A(2006): 395-438]. A Bayesian hierarchical model is built such that the counts are i.i.d given a latent, spatial�eld de�ned across the climate space. The concern here is with the "inverse problem"; making inference onunobserved climates given (fossil) pollen data. The �rst step is to compute the posterior for the spatial �eldgiven the climate + count pairings. The INLA method is particularly well suited to this task.If there aremultiple spatial processes giving rise to multivariate counts vectors at each climate-space location then theINLA method may prove unsuitable [Salter-Townshend, Ph.D. thesis 2009]. This is due to an assumptionthat data are conditionally independent given the spatial �eld and a computationally imposed restriction onthe number of hyperparameters the hierarchical model may have. An assumption of conditional independenceallows INLA to be applied to each taxon in turn. Procedures for assessing the accuracy of this assumptionvia cross-validation are presented.keywords: Bayesian, Integrated Nested Laplace Approximation, cross-validationMichael Salter-Townshend, 552 James Joyce Library UCD Bel�eld Dublin 4 IrelandE-mail address:michael.salter-townshend@ucd.ie

Oral Presentation

SPATIAL REGRESSION MODELINGWITH NONSTATIONARY SPATIAL COVARIANCESTRUCTURE FOR AIR QUALITY EXPOSURE FROM COMPLEX SPATIO-TEMPORALMONITORING AND GIS-BASED COVARIATES.

Paul D Sampson1, Adam Szpiro1, Lianne Sheppard1, Johan Lindstrom2

1University of Washington, 2Lund University

Statistical analyses of health e�ects of air pollution have increasingly used GIS-based covariates in "landuse regression" models for prediction of air quality. More recently these spatial regression models haveaccounted for spatial correlation structure in combining monitoring data with covariates. We build on thisframework to address the spatio-temporal prediction of ambient PM2.5 concentrations using a hierarchicalmodel that decomposes the space-time �eld into a "mean" representing spatially varying seasonal and long-term trends and a "residual" that accounts for spatially correlated deviations from the mean model. Themodel accommodates complex spatio-temporal patterns by characterizing site speci�c temporal trends asa linear combination of empirically derived temporal basis functions, and embedding spatial �elds of basisfunction coe�cients in linear regression models with spatially correlated residuals (universal kriging). Thecomputations provide a pragmatic approach that can take full advantage of regulatory and other supplementalmonitoring data which jointly de�ne a complex spatio-temporal monitoring design. The work described inthis paper is motivated by the Multi-Ethnic Study of Atherosclerosis and Air Pollution (MESA Air), fundedby the U.S. Environmental Protection Agency (EPA) to predict individual ambient source exposures in orderto assess the relationship between chronic exposure and sub-clinical cardiovascular disease in six major UScities. We present empirical analyses of PM2.5 data from U.S. EPA regulatory AQS monitors and fromsupplemental monitoring campaigns designed to better characterize local e�ects of tra�c related pollution.keywords: air quality, land use regression, spatio-temporal model, universal krigingPaul D Sampson, Department of Statistics University of Washington Box 354322 Seattle, WA 98195-4322 USAE-mail address:pds@stat.washington.edu

Supporting grant: Research supported by U.S. EPA grant for the Multi-Ethnic Study of Atherosclerosis and Air Pollution andby STINT: Swedish Foundation for International Cooperation in Research and Higher Education grant.

93

Bologna, Italy TIES 2009

Poster Presentation

CHANGING OF BIODEGRADABLE ORGANIC CARBON IN SOME PROCESSES OFDRINKING WATER PREPARATION.

Olena Samsoni-Todorova1, Natalia Klymenko2, Ivan Kozyatnyk2

1National Technical University of Ukraine Kiev Polytechnic Institute, 2Institute of Colloid Chemistry andChemistry of Water, Ukrainian National Academy of Sciences

Natural organic matters (NOM) determine technology of drinking water preparation and its quality. It isvery important to determine that part of NOM, which can be degraded by microorganisms. Bioavailabilityof NOM is very signi�cant factor, which in�uences on water quality in distribution system.The main goalof present work was to establish biodegradable organic carbon (BDOC) content in di�erent types of water(Dniper river basin, artesian, tap and model solution of fulvic acids (FA)) and its change during treatmentby di�erent methods. In consideration of the fact that BDOC content in surface water sources dependson natural and climate condition, this information must be extremely useful for the development of watertreatment technology and improvement drinking water quality.In the �rst series of experiment BDOC wasdetermined in initial Dniper river water, in this water after coagulation, sedimentation and �ltration throughAC; in tap water and in this water after through AC and disinfection by UVBDOC content in river water israther low (11% with respect to DOC). BDOC content increases up to 20% after coagulation. It is conditionedby the removal of resistant biodegradable components mainly, which are more hydrophobic in comparisonwith biodegradable organic matters. Nor Dniper river water after coagulation- sedimentation-�ltration nortap water does not correspond with criteria of the biologic stability. BDOC was not determined only aftertreatment of water on AC �lter and disinfection by UV. BDOC also was not determined in artesian water.keywords: biodegradable organic carbon, drinking water qualityOlena Samsoni-Todorova, 42 Vernadsky Avenue, Kyiv 03680, UkraineE-mail address:samsoni@online.ua

Poster Presentation

SPECTRAL ANALYSIS OF IRREGULARLY SPACED PALEOCLIMATIC TIME SERIESUSING EMPIRICAL MODE DECOMPOSITION AND WAVELET LIFTING.

Jean Sanderson1, Michel Cruci�x2, Piotr Fryzlewicz3, Jonathan Rougier1

1 University of Bristol, 2Université Catholique de Louvain, 3London School of Economics

The majority of standard time series analysis tools require that the signal is composed of observationswhich are evenly spaced over time. Since paleoclimatic series taken from ice cores and marine sediment recordsexhibit a naturally irregular time sampling, it is not possible to apply these techniques without modi�cationof the data or method.For the detection of long term oscillations it is su�cient to use interpolation to transferthe data to a regularly spaced grid. However in doing this, it is possible that information at higher frequenciesis being lost. There is particular interest in high frequencies as these can clarify the role of orbital forcingin the development of glacial cycles, and so it is preferable to use methods which deal directly with theirregularly observed data values.Making direct use of the irregular observations, we demonstrate ongoingwork based on empirical mode decomposition (EMD) and wavelet lifting. Using EMD the original series canbe decomposed into a sum of Intrinsic Mode Functions (IMF) related to di�erent periods of variation withinthe data. Wavelet lifting is then used to estimate the spectral components of these irregularly spaced series.Our results are demonstrated with application to a composite carbon dioxide record dating back nearly 1million years.keywords: time series, irregular observations, spectral estimation, wavelet lifting, empirical mode decompositionJean Sanderson, Department of Mathematics,University of Bristol,University Walk, BristolBS8 1TWE-mail address:j.sanderson@bristol.ac.uk

94

Bologna, Italy TIES 2009

Oral Presentation

SPATIO-TEMPORAL MODELS FOR OCEANIC VARIABLES.

Bruno Sanso1, Ricardo Lemos2

1University of California Santa Cruz, 2Universidade de Lisboa and Maretec, Universidade Tecnica de Lisboa

We develop statistical spatio-temporal models for historical records of ocean temperatures and salinitycollected sparsely in space over a period of thirty years. Our models seek to estimate ocean climatologies forthe variables of interest, together with quanti�cations of the anomalies, the trends and the seasonal cycles.The components of the model are allowed to vary with time and location. While ocean temperature dataare relatively numerous, salinity measurements are scarce. Our model considers both variables jointly andimposes density stability conditions to avoid unstable water columns. The spatio-temporal evolution of thedata is captured using convolutions with compactly supported kernels. Particular attention is given to theproblem of handling large datasets. This is achieved with an e�cient parallelization of the Markov chainMonte Carlo method used in the estimation of the model parameters. We use data from the Northern AtlanticOcean to illustrate our method.Bruno Sanso, University of California, Applied Math and Statistics, 1156 High Street, MS: SOE, Santa Cruz, CA95064, USAE-mail address:bruno@ams.ucsc.edu

Oral Presentation

SOLAR PHOTOVOLTAIC IN ITALY: A STATISTICAL SPATIAL ANALYSIS FOR ACCU-RATE ENERGY PLANNING.

Annalina Sarra1, Gianfranco Iurisci1

1Dipartimento di Metodi Quantitativi e Teoria Economica Università "G.d'Annunzio" di Pescara

In recent years, the rapid growth of renewable energy sources (photovoltaic, biomass, geothermal, wind andhydroelectricity) constitutes a feasible solution for environmental problems created by the present production-consumption energy model. Photovoltaic is one of the most promising, renewable energy sources with greatpotential for development. In this paper, we rely on the recent spatial cluster detection method for case eventdata of Demattei et al. (2007) in order to understand the spatial distribution pattern of the photovoltaicpower (PV) systems in Italy. This study considers the spatial locations of photovoltaic plants in 103 Italianprovinces, distinguished according to their power. Data have been derived from Conto Energia (GSE report,2008). The reviewed statistical approach, based on the distance regression on selection order, exhibited�exibility and ability in detecting several arbitrarily shaped clusters. It can provide useful insight in explainingwhy certain areas are more suitable for the installation of PV power generation systems. Indeed, the processof �nding locations for new facilities requires the assessment of some discrimination factors: environmental,legal, orographic and climatological issues are to be taken into account. The evidence showed that in thedetected clusters there were many northern Italian provinces (e.g. Sondrio, Biella, Bolzano) with unfavourableclimate conditions (low global irradiance level, low annual temperatures) which have rapidly taken advantageof incentives for solar energy installations. Hence, the main �ndings suggest that legal regulations play a keyrole on fomenting the development of solar energy in Italy.keywords: renewable energy sources, photovoltaic power system, detection cluster method, distance regression onselection orderAnnalina Sarra, Dipartimento di Metodi Quantitativi e Teoria Economica, Viale Pindaro, 42 65127 PescaraE-mail address:asarra@dmqte.unich.it

95

Bologna, Italy TIES 2009

Oral Presentation

RISK STANDARDISATION AND POISSON REGRESSION IN ENVIRONMENTAL CAN-CER INCIDENCE STUDIES.

Erik-A. Sauleau1, Silvia Columbu2, Monica Musio2

1Faculte de Medecine, Universite de Strasbourg, France, 2Dipartimento di Matematica ed Informatica,Universita di Cagliari, Italy

In incidence cancer epidemiology, geographical unit of residence is used as environmental exposure proxy.The standardized incidence ratio is estimated by the number of observed cases divided by an expected numberof cases, obtained as the product between a population at risk and an estimate of the risk to which thispopulation is exposed. This risk, global or adjusted on variable(s), for example age, is calculated by indirectstandardization. An unexplored question is the in�uence of standardization on the estimates of variablesintroduced in Poisson regression when taking into account other e�ects. The example of lung cancer is taken.The model, using fully Bayesian inference, is a Poisson regression in which period of diagnosis and gender aremodeled as �xed e�ects (�at priors) and residence is modeled using a conditional autoregressive prior. Wecompare three models: one with standardisation on a global risk including, in addition to the three previouse�ects, age as a P-spline (M1), one using the same global risk but with age as a �xed e�ect (M2) and onewith age-standardized risk without age e�ect in the regression (M3). Nearly only the intercept estimationdi�ers between M1 and M3. Comparison of DICs promotes slightly M1. In M2, estimations of e�ects arecompletely di�erent and DIC is the worst.In standardization, an estimated risk is independent from eachother. This is equivalent to a model using a global risk but a �exible modeling of the e�ects. Inappropriatemodeling of e�ects deteriorates sharply the model.keywords: Environmental human health, Cancer incidence, Risk standardisation, Poisson regression, Bayesian in-ferenceErik-A. Sauleau, Université de Strasbourg Laboratoire de Biostatistiques Faculté de Médecine 4, rue Kirschleger F -67085 StrasbourgE-mail address:erik-andre.sauleau@medecine.u-strasbg.fr

Oral Presentation

CORRECTION OF EDGE EFFECT IN SPATIAL GENERALIZED ADDITIVEMIXEDMOD-ELS.

Erik-A. Sauleau1, Agnes Fromont2, Laurence Clerc3, Audrey Bellisario4, Claire Bonithon-Kopp 4, ThibaultMoreau2, Christine Binquet4

1Faculty of medicine, University of Strasbourg, Strasbourg, France, 2University Hospital, Dijon, France,3CNAMTS, DRSM Bourgogne Franche-Comte, Dijon, France, 4INSERM, CIE1, Dijon, France

In spatial analyses on aggregated data, edge e�ect is important when using methods borrowing informationfrom neighboring geographical units. Strategies have been developed to take this e�ect into account (mappingthe surface onto a torus, weighting, using a bu�er zone). The creation of an external bu�er has signi�cantadvantages on other methods.We use a dataset on multiple sclerosis, provided by the "Caisse Nationaled'Assurance maladie", �rst health insurance system in France and covering 84% of the French population.The objectives of this study is to estimate multiple sclerosis prevalence according to age, sex and geographicareas (France is divides in 96 administratives "departements") and to identify factors that could explaindi�erences in multiple sclerosis distribution such as socioeconomic, medical or environmental data.We buildmodels in the Bayesian framework of generalized additive mixed models, for Poisson data. A �rst series ofmodels include sex as �xed e�ect (�at prior), age (P-spline prior) and spatial e�ect adding to a normal priorfor heterogeneity either a conditional autoregressive prior or a bidimensional geospline for autocorrelation.We correct maps by adding an external bu�er zone such that each geographical unit on boundary (foreigncountries or seas) has the same number of neighbors than the mean of all other geographical units.It appearsthat geospline are more robust to the edge e�ect than conditional autoregressive prior as the posterior of thestructured spatial e�ect is less modi�ed with the �rst prior than with the second when using the bu�er.keywords: Environmental human health, Multiple sclerosis, Edge e�ect, Generalized additive mixed model, BayesianinferenceErik-A. Sauleau, Laboratoire de Biostatistiques Faculté de Médecine 4 rue Kirschleger 67085 STRASBOURG CedexFranceE-mail address:erik-andre.sauleau@medecine.u-strasbg.fr

96

Bologna, Italy TIES 2009

Oral Presentation

GAUGE IMPRECISION EFFECT ON ENVIRONMENTALMONITORINGALGORITHMS.

Michele Scagliarini1, Daniela Cocchi1

1University of Bologna

One of the principal objectives in environmental monitoring is the determination of compliance withstandards. In such surveillance activities, statistical monitoring algorithms are often used. Their statisticalproperties are known to be highly sensitive to measurement errors. In statistical quality control literature, themodel relating the measured value to the true, albeit not observable, value of the variable of interest, is usuallyGaussian and additive. However, situations arise where the measurement error is not normally distributed orwhere a di�erent error model needs to be considered. In environmental monitoring, experimental evidence hasshown that situations occur where two types of measurement errors ought to be considered: a measurementerror that is constant over a range of concentrations close to zero, while, at higher concentrations, it isproportional to the concentration of the substance in question. In this work, we re-express the additiveGaussian error model, which is traditionally used, in a more general way so as to include the structure ofthe two-component error model. We study the e�ects of the proposed error-model on the performance of themean control charts. Furthermore, we show that, with a su�cient amount of data from preliminary samples,it is possible to design the monitoring algorithm through a nonparametric approach. This methodologyreduces the e�ects of measurement errors, and allows for the evaluation of the in-control and out-of-controlperformances of the chart under a realistic error model which includes the Gaussian additive model as aspecial case.keywords: Environmental monitoring, measurement errors, control chartsMichele Scagliarini, Dipartimento di Scienze Statistiche, via Belle Arti 41, 40126 Bologna, ItalyE-mail address:michele.scagliarini@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

RADIATION-INDUCEDGENETIC EFFECTS AND ECOLOGICAL DOSE-RESPONSE ANAL-YSES.

Hagen Scherb1, Kristina Voigt1

1Institute of Biomathematics and Biometry Helmholtz Zentrum München-German Research Center forEnvironmental Health

The e�ects of ionizing radiation at doses below 10 mSv are still little understood. Recent epidemiologicevidence suggests that there is harm in the dose range of a few mSv or even below 1 mSv. The disaster atthe Nuclear Power Plant in Chernobyl in April 1986 resulted in the exposure of a large number of people inEurope to ionizing radiation that varied substantially, creating a new situation for epidemiology. In manydata sets from central, eastern, and northern parts of Europe, absolute or relative increases of birth defects,stillbirths, and the human sex ratio at birth after 1986 were observed. Marked jumps in the temporal trends ofunfortunate pregnancy outcome indicators are supported by results from analytical ecological dose-responseanalyses involving a spatial dimension represented by region-(district)-speci�c exposure data. Long-term dosedependent detrimental impacts of radioactive fallout after Chernobyl on pertinent genetic health indicators(stillbirths, birth defects, and the human sex ratio at birth) have been found. For example, from nearlyall published data concerning Down's syndrome in Europe, from 1981 to 1992, long-term increases afterChernobyl may be seen. Typical ecological relative risks for stillbirth and certain types of birth defects(Down syndrome, malformations of the heart, deformities) are in the range 1.2-2.5 per mSv/a. The sex oddsratio per mSv/a has been found in the range of 1.01-1.04, depending on whether one considers long-term orshort-term e�ects.keywords: Change-point analysis, ecological study, ecological dose-response relation, radiation-induced genetic ef-fects, spatial-temporal logistic regressionHagen Scherb, Ingolstädter Landstr. 1D-85764 Neuherberg, GermanyE-mail address:scherb@helmholtz-muenchen.de

97

Bologna, Italy TIES 2009

Oral Presentation

LOCAL APPROACHES FOR INTERPOLATING AIR POLLUTION PROCESSES.

Wolfgang Schmid1, Olha Bodnar1

1European University

In this talk we consider the problem of interpolating a spatio-temporal environmental process. Bodnar andSchmid (2009) proposed a local kriging procedure based on the observations of the neighboring measurementstations. Our aim is to compare local approaches with more global methods and, especially, we want to dealwith the question how local (global) a procedure should be. Moreover, an extension of the results of Bodnarand Schmid (2009) to multivariate environmental processes is provided. In an empirical study the mostimportant air pollutants of the Berlin-Brandenburg region of Germany are considered and it is described howour approaches can be applied.keywords: Nonlinear predictor, LOESS method, Non-stationary spatio-temporal process, Environmental statisticsWolfgang Schmid, European University, Department of Statistics,Groÿe Scharrnstr. 59,15230 Frankfurt (Oder), Ger-manyE-mail address:schmid@euv-frankfurt-o.de

Oral Presentation

MODELLING MULTIPLE SERIES OF RUNOFF: THE CASE OF RIO GRANDE BASIN.

Alexandra Schmidt1, Romy Ravines2, Helio Migon1

1IM-UFRJ, 2Bayes Forecast

This paper proposes a joint model for the rainfall and multiple series of runo� at a basin, two of themost important hydrological processes. The proposed model takes into account the di�erent spatial units inwhich these variables are measured, and as a natural bene�t its parameters have physical interpretations.Also, we propose to model runo� and rainfall in their original scales, making no use of any transformationto reach normality of the data. More speci�cally, our proposal follows Bayesian dynamic nonlinear modelsthrough the use of transfer function models. The resultant posterior distribution has no analytical solutionand stochastic simulation methods are needed to obtain samples from the target distribution. In particular,as the parameters of the dynamic model are highly correlated, we make use of the Conjugate UpdatingBackward Sampling recently proposed by Ravines, Migon and Schmidt (2007), in order to e�ciently explorethe space of the parameters.We analyze a sample from a basin located in the Northeast of Brazil, the RioGrande Basin. The data consist of monthly recorded series from January 1984 to September 2004, at threeruno� stations and nine rainfall monitoring stations,irregularly located in an area of drainage of 37 522.48km2. Model assessment, spatial interpolation and temporal predictions are part of our analysis. Resultsshow that our approach is apromising tool for the runo�-rainfall analysis. This is joint work with Romy R.Ravines and Helio S. Migon.keywords: Dynamic Models, Transfer functions, Change of SupportAlexandra Schmidt, Instituto de Matemática - UFRJ Caixa Postal 68530 Rio de Janeiro - RJ CEP:21.945-970 BrasilE-mail address:alex@im.ufrj.br

Oral Presentation

SEPARABLE CONDITIONAL INTENSITY ESTIMATES FOR SPACE-TIME POINT PRO-CESSES WITH APPLICATION TO LOS ANGELES COUNTY WILDFIRES.

Frederic Schoenberg1

1UCLA Statistics

Wild�res pose an extremely serious threat to California each year, damaging vast areas of public andprivate property and often threatening human lives. Wild�re risk assessments are presently made primarilyusing the Burning Index (BI), a numerical rating issued by the USDA Department of Forestry. Unfortunately,the BI appears to be a very poor predictor of wild�res. Because so many variables are positively associatedwith wild�re risk, the construction of models to compete with the BI is quite a di�cult task. The problemis essentially the familiar "curse of dimensionality," and is greatly alleviated when di�erent components in

98

Bologna, Italy TIES 2009

the process may be estimated separately. Some results will be presented concerning situations where suchseparability is permitted, and their use in forecasting wild�res in Los Angeles County will be explored.keywords: point processes, separability, intensity estimation, wild�res, kernel smoothingFrederic Schoenberg, 8125 Math-Science Building, UCLA Statistics, Los Angeles, CA 90095-1554, USAE-mail address:frederic@stat.ucla.edu

Oral Presentation

SIMPLE METRICS, COMPLEX ENVIRONMENTAL SYSTEMS.

Marian Scott1, Duncan Lee1, Claire Ferguson1, Ron Smith2

1University of Glasgow, 2Centre for Ecology and Hydrology

Environmental indicators are widely used and are especially favoured by environment agencies, regulatorsand policy makers. They relate directly to policy, are simple to present visually, and often form the basisof reporting changes in the state of the environment. In the science arena, a hierarchy of indices which mayrelate directly or indirectly to the policy indicator can be constructed, starting with the simplest: a singlevariable (such as nitrate concentration) in a single water body. Further levels of complexity are added bythe requirement to present a spatial summary, so the index then represents an aggregated (typically average)value of the variable of interest over space (at river basin, city or country scale), and by weighting overdi�erent variables of interest (e.g. over di�erent air pollutants). The �nal level generates a composite indexas a summary of the results of a complex and sophisticated multivariate statistical model (e.g. a biodiversitysummary of breeding bird population size or a measure of sustainability). The index, once generated, is usedas a summary statistic, though rarely accompanied by an estimate of its uncertainty and the data whichform the basis of the index, and hence the index itself, may not be representative (spatially or temporally).Fundamentally the index is a univariate summary of a complex multivariate environmental system, thus ourability to model e�ects of drivers (such as climate change or management practice) is compromised.Suchstatistical issues concerning environmental indices will be discussed with recommendations o�ered.keywords: indices, multvariate analysisMarian Scott, Dept of Statistics, University of Glasgow, Glasgow G12 8QWE-mail address:marian@stats.gla.ac.uk

Oral Presentation

BAYESIAN LATENT VARIABLE MODELLING IN STUDIES OF AIR POLLUTION ANDHEALTH.

Gavin Shaddick1, Duncan Lee2, Ruth Salway1, Stephen Walker3

1University of Bath, 2University of Glasgow, 3University of Kent

This paper describes the use of Bayesian latent variable models in the context of studies investigating theshort-term e�ects of air pollution on health. Traditional Poisson or quasi-likelihood regression models usedin this area assume consecutive outcomes are independent (although they allow for overdispersion), whichin many studies may be implausible as temporal correlation is to be expected. We compare this traditionalapproach with a series of Bayesian latent process models, which incorporate the possibility of auto-correlation.These include an autoregressive model that has been used in existing studies, as well as implementing analternative based on a moving average structure. A simulation study is presented to assess the e�ciencyof these models when there are di�erent forms of auto-correlation in the data, and the results show thatmodels that consider separate components for modelling trend and correlation do better than those thatattempt to combine the two. The models are then applied to a new epidemiological study investigating thee�ects of short-term exposure to air pollution on respiratory mortality in the elderly (aged 65 and above) inLondon, England, between 1997 and 2003. The results from this study show signi�cant relationships betweenrespiratory health and a number of pollutants, notably particulate matter and ozone, which do not appearto be sensitive to the approach adopted for modelling the trend and correlation.keywords: Air pollution and health, Bayesian modelling, latent processes, residual correlationGavin Shaddick, Mathematical Sciences University of Bath Bath BA2 7AY United KingdomE-mail address:g.shaddick@bath.ac.uk

99

Bologna, Italy TIES 2009

Poster Presentation

UNCERTAINTY ANALYSIS WITHIN THE HEIMTSA (HEALTH AND ENVIRONMENTINTEGRATEDMETHODOLOGYAND TOOLBOX FOR SCENARIO ASSESSMENT) PROJECT.

Gavin Shaddick1, Marta Blangiardo2, Ruth Salway1, Alex Zenie3, Bruce Denby4, Edzer Pebesma5

1University of Bath, 2Imperial College, 3European Council Joint Research Centre (JRC), 4NorwegianInstitute for Air Research (NILU), 5University of Munster

HEIMTSA is an Integrated Project funded under the EU Sixth Framework Programme-Priority 6.3 GlobalChange and Ecosystems. It aims to develop and apply new integrated approaches to the assessment of envi-ronmental health risks and their consequences to European policy in areas of transport, energy, agriculture,industry, household and waste treatment and disposal. Within the project, one work package (WP1.2) isdedicated to uncertainty analysis. HEIMTSA is committed to considering the e�ects of uncertainty within a'full chain' approach, where the e�ects of uncertainty at each stage in a process are considered, both qualita-tive and quantitative, where possible using Monte Carlo simulation techniques and Bayesian methods. Thetechniques developed have been used in four case studies, designed to facilitate the analysis of the full chainapproach in a practical context. These include the Health Impact Assessments of (i) outdoor air pollutants;(ii) indoor air contaminants; (iii) of complex pollutants with multipathway exposure and, (iv) of tra�c noise.The method of uncertainty analysis may di�er between the case studies in order to adapt to the di�erentchallenges they present, for example a qualitative characterisation was used for the noise case study, whilea Monte Carlo approach was implemented in the complex pollutants case study. This presentation aims topresent details of the HEIMTSA uncertainty framework, to give recommendations and describe some lessonslearned from its four case studies and to discuss possible future developments in uncertainty analysis.keywords: Uncertainty, Health e�ects, Environmental modellingGavin Shaddick, Mathematical Sciences University of Bath Bath BA2 7AY United KingdomE-mail address:g.shaddick@bath.ac.uk

Supporting grant: Funded under the EU Sixth Framework Programme for Research- Thematic area "Sustainable Development,Global Change and Ecosystems". Contract number: GOCE-CT-2006-036913-2

Oral Presentation

LINKING STATISTICS TO ECOSYSTEM SERVICES FRAMEWORKS.

Ron Smith1, Jan Dick1

1Centre for Ecology and Hydrology

The ecosystem services concept connects ideas on the structure and function of a whole ecosystem tothe context of assessing and valuing the services the ecosystem provides to humans (clean air, drinkingwater, recreation etc). Di�erent frameworks are used for the analysis of ecosystem services valuation studiesre�ecting the complexity, size, spatial and temporal scales of the di�erent studies. While some studies attemptto be all-encompassing, others focus on the parts of the ecosystem for which su�cient data is available toprovide a monetary valuation of the ecosystem. Non-monetary valuation, although recognised as important,can be di�cult to quantify and is often downgraded in the �nal analysis. In this paper we will look at someof the issues in quantifying the value of an ecosystem within statistical frameworks which may provide analternative view of the importance of some ecosystem components.keywords: Ecosystem services, Bayesian belief networks, MultivariateRon Smith, CEH Edinburgh Bush Estate Penicuik Midlithian EH26 0QBE-mail address:ris@ceh.ac.uk

100

Bologna, Italy TIES 2009

Poster Presentation

137CS, 40K, 238PU, 239+240PU AND 90SR IN ENVIRONMENTAL MATERIAL ORIGI-NATIG FROM KING GEORGE ISLAND (SOUTH SHETLANDS, ANTARCTICA).

Katarzyna Sobiech-Matura1, Maria A. Olech1, Jerzy W. Mietelski2, Anna Ma±niak1

1Institute of Botany, Department of Biology and Earth Science, Jagiellonian University, Kraków, 2TheHenryk Niewodniczanski Institute of Nuclear Physics, Polish Academy of Science, Kraków

The aim of the research was to estimate the activity levels of radionuclides: 137Cs, 40K, 238Pu, 239+240Pu,and 90Sr in plant, animal and soil material originated from Antarctica. Samples were collected at King GeorgeIsland (South Shetlands) during two expeditions to Henryk Arctowski Polish Antarctic Station in 2005/2006and 2006/2007 seasons. The research is a continuation of radioactive contamination monitoring started in1998 by J.W. Mietelski, P. Gaca, and M.A. Olech [J. of Radioanal. and Nucl. Chem. (2000): 527] andby J.W. Mietelski, M.A. Olech, K. Sobiech-Matura, et al. [Pol. Biol. (2008): 1081-1089].Among terrestrialsamples were two �owering plants, moss, algae, two lichen species, and 10 soil pro�les, two of which werelocated at the fresh outcrop. Marine environment was represented by several animal species whole bodies,and by macroalgae. Two penguin's species egg shells and fur of the south elephant seal was also examined.Samples were cleaned and homogenised. Depending on the type of the sample they were dried in 105°C(plant, lichen and soil material) or ashed in 400°C (animal material). Then activity ratios of 137Cs and40K were measured using Y-spectrometer with HPGe detector. After those measurements samples wereashed in 600°C and radiochemical procedure was applied to estimate activity ratios of 238, 239+240Pu, and90Sr. Measurements were carried out on low-background α- and β-spectrometer (respectively). The qualityof analyses was assured by reference material and blanc samples measurement.keywords: Antarctica, radioactive contamination, natural radioactivity, terrestrial environment, marine environmentKatarzyna Sobiech-Matura, ul. Kopernika 27 31-501 Kraków POLANDE-mail address:katarzyna.sobiech@uj.edu.pl

Supporting grant: The project was supported by Polish Ministry of Science and Higher Education (Grant nr 1290/P01/2007/32).

Oral Presentation

PARTIAL LEAST SQUARE (PLS) MODEL IN AIR QUALITY MODELLING.

Tarana A. Solaiman1

1Civil and Environmental Engineering, The University of Western Ontario, Canada

Since past decades, many urban areas in North America are su�ering from ozone and particulate matterpollution due to the increase of auto and industrial emissions. Air quality predictions, thus are useful foradapting appropriate environmental and health risk management strategies in advance. This study aimsat assessing the applicability of Multivariate Partial Least Square (PLS) model in modeling air qualityby investigating the e�ect of air pollution in the "Los Angeles South Coast Air Basin". Daily ozone andparticulate matters data have been collected from �ve non-attainment areas located in the cities of Azusa,Burbank, Los Angeles, San Bernardino and Rubidoux during 1995 and 2005. PLS models have been developed(i) to understand the relation between the response variables; and (ii) to assess the need for data pre-treatment. The pre-treated data has been used to model the temporal dependency of the process using(i) Finite Impulse Response model and (ii) Auto Regressive with Exogenous Input model. The best suitedmodels have been used to predict the future outputs. Analysis shows that PLS model can be used as apowerful tool for investigating the complex behaviour of the pollutants. The concentrations of ozone havefound to remain more active longer than the particulate matter (7 days vs. 21 days) in aiding pollution,indicating the necessity of combining chemical compositions of both pollutants. The model performanceswere in between 75%-85% which re�ects excellent performance of PLS model.keywords: Partial least square model, ozone pollution, particulate matter pollutionTarana A. Solaiman, 608-251 Platts Lane, London, Ontario, Canada, N6H4P4E-mail address:tsolaima@uwo.ca

101

Bologna, Italy TIES 2009

Oral Presentation

CHEMOMETRICS AND ENVIRONMETRICS: TALL SHOULDERS, ACCOMPLISHMENTS,AND FUTURE DIRECTIONS.

Cli�ord H. Spiegelman1, Abdel El-Shaarawi2

1Texas A&M University, 2National Water Research Institute, Canada

Science is the product of the accumulated human intellect and may be viewed as a tree with branchesrepresenting various �elds. As new problems arise that have a common theme new �elds emerge. Theemergence of chemometrics and environmetrics at that time was the result of current societal needs toaddress environmental problems. This talk highlights the work of pioneers who help establish chemometricsand environmetrics as important �elds. It then describes current accomplishments of these �elds. The �nalportion of the talk discusses how the �elds might position themselves for even greater success in the middleof the 21st century.keywords: History, Future DirectionsCli�ord H. Spiegelman, Cli� Spiegelman Texas A&M University Department of Statistics 447 Blocker Building 3143TAMU College Station, TX 77843-3143E-mail address:cli�@stat.tamu.edu

Oral Presentation

USING IMPUTATION TO ESTIMATE TREND AND ABUNDANCE IN COHO SALMONNUMBERS USING A MULTI-PERIOD ROTATING PANEL SAMPLING DESIGN.

Don Stevens 1

1Oregon State University

The Oregon Department of Fish and Wildlife implemented a spatially balanced rotating panel samplingdesign for Oregon coastal coho salmon in 1998. The panel structure of the design is tied to the 3-year lifecycle of coho salmon, with a panel visited every year, 3 panels visited on a 3-year repeat cycle, 9 panelsvisited on 9-year repeat cycle, and 27 panels visited on a 27-year cycle. Each year, the annual panel, andone panel from each rotation are visited, for a total of four panels per year. In this talk, I will discuss somedi�erent ways of viewing regional trend, and show how the rotating panel design can be used to increasethe e�ective sample size for both trend detection and status estimation. The technique imputes current yearvalues for trend or abundance for panels not visited in the current year using prior year information. Multipleimputation is used to estimate variance.keywords: imputation, rotating panel, trend detectionDon Stevens , Oregon State Univeristy, Department of Statistics, 44 Kidder Hall, Corvallis, Oregon 97331 USAE-mail address:stevens@stat.oregonstate.edu

Supporting grant: Oregon Watershed Enhancement Board

Poster Presentation

MULTIVARIATE NON-PARAMETRIC REGRESSION VIA GAUSSIAN MARKOV RAN-DOM FIELDS.

James Sweeney1, John Haslett1

1Trinity College Dublin

Nonparametric regression models are utilised in cases where the functional form of the relationship betweenthe response and a number of predictors is not assigned a pre-de�ned structure. This is particularly useful incases where it is di�cult to �nd parametric regression techniques which can re�ect the data structure. Thereare a number of nonparametric regression techniques: this poster is concerned with proceeding via the use ofGaussian Markov Random Fields (GMRF) and Bayesian inference. A GMRF is a simple construct: a �nitedimensional random vector following a multivariate normal distribution de�ned on a graph. Following thework of Rue et al(2007), this approach involves the creation of large numbers of latent variables, prespecifyingcovariance structures, large sparse matrices and is potentially computationally very demanding. Given dataY, the approach generally involves specifying a latent GMRF prior for the "true" surface/curve(X) conditionalon a number of hyperparameters (P) and then proceeding to make inference on the hyperparameters. Theapplication proposed is motivated by palaeoclimate reconstruction, where the relationship linking the climaticresponse of vegetation to a number of climate dimensions is modelled. Here the responses are multivariate

102

Bologna, Italy TIES 2009

zero in�ated counts. Rue's approach is extended to multivariate nonparametric regression. The approachwill involve extending the univariate nonparametric model to a multivariate case.keywords: Gmrfs, Quadratic Approximation, Non Parametric RegressionJames Sweeney, Room 118, Statistics Department, Lloyd Institute, Trinity College, Dublin 2,IrelandE-mail address:sweeneja@tcd.ie

Supporting grant: This work was funded by Science Foundation Ireland and is part of anongoing Palaeoclimate reconstructionproject headed by Prof. JohnHaslett, Dept. of Statistics, Trinity College Dublin.

Oral Presentation

ACCOUNTING FOR EXPOSURE MEASUREMENT ERROR IN ENVIRONMENTAL EPI-DEMIOLOGY.

Adam Szpiro1, Lianne Sheppard1, Thomas Lumley1

1University of Washington

Chronic air pollution epidemiology studies are improved by accurate prediction of intra-urban variationin individual pollutant concentrations. However, treating the predicted concentrations as known results ine�ect estimate bias and incorrect standard errors. The measurement error is neither Berkson nor classical,and in particular the resulting bias need not be toward the null. Therefore, standard regression calibrationtechniques are not appropriate. One approach to addressing the measurement error in this scenario is to usea double parametric bootstrap that corrects the bias and the standard errors. We show in a simulation studythat this technique is e�ective, although it is computationally intensive. As an alternative to the parametricbootstrap, we introduce a new technique termed "parameter simulation". Parameter simulation performswell at a signi�cantly reduced computational burden. It may also be more robust to model misspeci�cationsince the calculations do not require a fully speci�ed parametric model.Adam Szpiro, Department of Biostatistics University of Washington Health Sciences Building, H669E Box 357232Seattle, WA 98195-7232E-mail address:aszpiro@u.washington.edu

Poster Presentation

ROBUST QUANTILE REGRESSION ESTIMATION FOR SPATIAL PROCESSES.

Baba Thiam1, Dabo-Niang Sophie1

1University Lille 3

Spatial quantile estimation is an interesting and crucial problem instatistical inference for a number ofapplications where the in�uence of a vector of covariates on some responsevariable is to be studied in a contextof spatial dependence. We are interested in conditional quantile estimation which is an important �eld instatisticswhich dates back to Stone (1977) and has been widely studied in thenon-spatial case. It is useful in alldomain of statisticssuch as in environmental science where modelling e�ects of high order quantilesrather thanmean e�ects is more important. In our knowledge, although potential applications of quantile regression tospatial data are without number, only thepapers of Koencker and Mizera (2004), Hallin et al.(2007), have paidattention to study nonparametric quantile regression forrandom �elds.Not many theoretical works have beendevoted so farto kernel nonparametric quantile regression estimation and prediction for spatialprocesses. Wepresent a statistical framework for modeling conditional quantiles of spatialprocesses assumed to be stronglymixing in space. We are mainly concerned inthis work with L1 consistency as well as asymptotic normality ofthe kernel conditional quantile estimator in the case of random �elds.We illustrate the proposed methodologywith some simulations.keywords: quantile, spatial process, non-parametricBaba Thiam, University Lille 3, laboratory EQUIPPE, BP60149 Villeneuve d'ascq cedex, FranceE-mail address:baba.thiam@univ-lille3.fr

103

Bologna, Italy TIES 2009

Oral Presentation

PREDICTIVE UNCERTAINTY IN HYDROLOGICAL FORECASTING.

Ezio Todini1

1University of Bologna

This work aims at discussing the role and the relevance of "predictive uncertainty" in �ood forecastingand water resources management.Predictive uncertainty, is here de�ned as the probability of occurrence of afuture value of a predictand (such as water level, discharge or water volume) conditional on prior observationsand knowledge as well as on all the information we can obtain on that speci�c future value, which is typicallyembodied in one or more hydrological/hydraulic model forecasts.The aim of this work is also to clarifyquestions such as: What is the conceptual di�erence between "emulation uncertainty" (commonly usedwhen dealing with model veri�cation) from the predictive uncertainty (which is used when forecasting intothe future)? What is the di�erence between models, parameters, input output measurement errors, initialand boundary conditions, etc. uncertainty and predictive uncertainty? How one can incorporate all theseuncertainties into the predictive uncertainty and, most of all, is it really necessary?The presently availableuncertainty processors, continuous or binary depending on the scope for which they are developed and thetype of decision one must take, are then introduced and compared on the basis of their relative performancesusing operational �ood forecasting systems. Finally, the bene�ts of incorporating predictive uncertainty intothe decision making process will be compared, on actual real world derived examples, to the ones obtainablewhen using deterministic forecasts, as currently donein practice.keywords: Predictive Uncertainty, Emulation Uncertainty, Hydrological Forecasting, Bayesian Model Averaging,Model Conditional ProcessorEzio Todini, Dipartimento di Scienze della Terra e Geologico Ambientali, Università di Bologna. Via Zamboni, 67.Bologna. ItalyE-mail address:ezio.todini@unibo.it

Oral Presentation

PREDICTING TREE LEVEL VARIABLES USING AIR-BORNE LIDAR DATA AND FIELDOBSERVATIONS.

Erkki Tomppo1, Mari Myllymäki2, Antti Penttinen2

1Finnish Forest Research Institute, 2University of Jyväskylä, Department of Mathematics and Statistics

The use of LiDAR (Light Detection and Ranging) data is under intensive method development currently.Anadvantage is capability to measure the height of trees relatively accurately.Height mesurements are oftenused to predict mean or total volume of growing stock for a land element.However, interesting forestrycharacteristics are the volumes of the growing stock by tree species and also separately for pulp wood andsaw timber.Although LiDAR data can be used for accurateheight measurements, prediction of volumes oftimber assortment by tree speciesis far from trivial. Suppressed trees are not easily detectable. To distinguishtree speciesis also challenging.The overall aim of our project is to develop methods to integrate LiDAR-based predictions to the satellite image aided multi-source national forest inventory of Finland (MS-NFI)Thetwo ways to use the the predictions area) to detect the big changes on permanent �eld plots,i.e, fellingsor natural loss, b) to employthe lidar predictions for areasoutside of the permanent plots as additionaltraining datain MS-NFI.We present the �rst results aiming to predict tree levelvolumes by tree species andtimber assortmentsusing high pulse LiDAR data and �eld measurements.Hidden trees are predicted usingprobability models derived from the observed dominant and other detectable trees using LiDAR as wellmarked point processes.These data are also used to predict tree height and canopy dimensionsand, andfurther tree diameters and other characteristics neededfor volume prediction.keywords: LiDAR, Multi-source forest inventory, Marked point processes, k-NN estimationErkki Tomppo, PO Box 18 (Jokiniemenkuja 1) FI-01301 VANTAA, FINLANDE-mail address:erkki.tomppo@metla.�

104

Bologna, Italy TIES 2009

Oral Presentation

CLUSTERING OF MONITORING STATIONS IN VENICE LAGOON.

Stefano Tonellato1, Stefano Ciavatta 1, Andrea Pastore1, Roberto Pastres1

1Università Ca' Foscari Venezia

The design of monitoring plans for both fresh water and coastal water bodies represents a crucial issueboth in the EU and USA: in particular, EU member states are required to implement the European UnionWater Framework Directive 2000/60 (WFD). The main target of the WFD is the achievement of a "good"quality status in all European waters by 2015. In order to achieve this goal, member states are asked tocarry out monitoring activities, aimed at assessing the water status of their water bodies.A cost-e�ectiveapplication of the WFD should be based on an accurately designed monitoring strategy. Such a strategy iseven more crucial for transitional environments, as estuaries and coastal lagoons, which are characterized by amarked spatial and temporal variability of biological communities, chemical parameters and physico-chemicalconditions. To this regard, the identi�cation of homogeneus areas within transitional water bodies is a keystep towards the design of a science-based and cost-e�ective monitoring plan.The main purpose of this paperis to provide a classi�cation, in terms of di�erent trophic variables, of the sites of a water quality monitoringnetwork located in Venice Lagoon. In order to preserve the information about the temporal dynamics ofthe variables of interest and pursue a model based classi�cation, we follow a method based on functionaldata analysis, introduced by James and Sugar (JASA 2003: 397-408). A comparison with the most popularmethods used in the common practice will be provided.keywords: model-based clustering, functional data analysisStefano Tonellato, Università Ca' Foscari Venezia Dipartimento di Statistica San Giobbe, Cannaregio 873 30121Venezia, ITALIAE-mail address:stone@unive.it

Supporting grant: The work was partly funded by the grant n.2006131039, PRIN 2006, "Statistical analysis and modelling ofimpact and risk for environmental phenomena in space and time".

Oral Presentation

STATE-SPACE MODELS IN EXTREME VALUE THEORY.

Gwladys Toulemonde1, Armelle Guillou2, Philippe Naveau3, Mathieu Vrac3, Frédéric Chevallier3

1Université Montpellier 2, 2Université de Strasbourg, 3LSCE - IPSL

Recordings of daily, weekly or yearly maxima in environmental time series are usually �tted by the general-ized extreme value (GEV) distribution that originates from the well-established extreme value theory (EVT).One special case of such GEV distribution is the Gumbel family which corresponds to the modeling of max-ima stemming from light-tailed distributions.To capture temporal dependencies, linear autoregressive (AR)processes o�er a simple and elegant framework. It turns out that the AR process under consideration can beobserved either directly or indirectly i.e. via a second process.In this talk, we �rst propose extended linearAR models in the case of directly observed AR processes. Their main theoretical properties are derived insuch a way that they handle Gumbel distributed maxima. From the atmospheric science viewpoint, this linkbetween linear AR processes and EVT allows the statistical treatment of extreme environmental recordingsin which temporal dependencies are present. As an example, our model is �tted to daily maxima of methanemeasured in Gif-sur-Yvette (France). Simulation results are also presented in order to assess the quality ofour parameter estimations in the case of �nite samples.Finally, when the AR process of interest is indirectlyobserved, we suggest a Gumbel state-space model based on the proposed Gumbel AR process. This couldlead to important applications in �ltering schemes like data assimilation, the latter being routinely used bygeoscientists.keywords: Extreme Value Theory, Gumbel distribution, Autoregressive processes, State-space models, AtmosphericchemistryGwladys Toulemonde, Université Montpellier 2I3M - UMR CNRS 5149 - CC 051 Place Eugène Bataillon34095 Mont-pellier Cedex 5E-mail address:gwladys.toulemonde@math.univ-montp2.fr

Supporting grant: GRASPA invited/contributed session

105

Bologna, Italy TIES 2009

Poster Presentation

A SIMPLE FEEDFORWARDNEURAL NETWORK FOR THE PM10FORECASTING: COM-PARISON WITH A RADIAL BASIS FUNCTIONNETWORK AND A MULTIVARIATELINEAR REGRESSION MODEL.

Livia Trizio1, Maurizio Caselli1, Gianluigi de Gennaro1, Pierina Ielpo1

1University of Bari

The problem of air pollution is a frequently recurring situation and its management has social and economicconsiderable e�ects. Given the interaction of the numerous factors involved in the raising of the atmosphericpollution rates, it should be considered that the relation between the intensity of emission produced by thepolluting source and the resulting pollution is not immediate. The aim of this study was to realise and tocompare two support decision system (neural networks and multivariate regression model) that, correlatingthe air quality data with the meteorological information, are able to predict the critical pollution events. Thedevelopment of a back-propagation neural network is presented to predict the daily PM10 concentration 1, 2and 3 days early. The measurements obtained by the territorial monitoring stations are one of the primarydata sources; the forecasting of the major weather parameters available on the website and the forecastingof the Saharan dust obtained by the "Centro Nacional de Supercomputaciòn" website, satellite images andback trajectories analysis are used for the weather input data. The results obtained with the neural networkwere compared with those obtained by a multivariate linear regression model for 1 and 2 days forecasting.keywords: forecasting, neural network, PM10Livia Trizio, liviatrizio@lenviros.itE-mail address:liviatrizio@lenviros.it

Poster Presentation

A GEOGRAPHICALLY WEIGHTED REGRESSION-BASED GENERALIZED REGRES-SION ESTIMATOR FOR SPATIAL DATA.

Alessandro Vagheggini1, Francesca Bruno1

1Dipartimento di Scienze Statistiche, Università di Bologna

Finite population inference on spatial data is usually investigated according to the model-based viewpoint.The usual critics of arbitrariness to this framework can therefore be formulated. Purely design-based esti-mators are at present still under development. In this work, generalized regression (GREG) estimators areproposed as an unifying instrument. These estimators are known as a powerful tool for estimating a totalor a mean in �nite population since, through an e�cient use of the auxiliary information in population,they improve traditional results. GREG estimators introduce a regression model suggested by the auxiliaryinformation, and in spatial sampling the locations, where the observation are taken, belong to the set ofauxiliary information. In parallel, in quantitative geography, the geographically weighted regression allows toconsider this information: in particular it is able to stress the relationship between close and far locations bymeans of the distances between them. Di�erent weights can be assigned according to the closeness betweenobservations. We propose geographically weighted regression as the model for obtaining a GREG estimatorfor spatial data. In this way, a design-based model-assisted estimator is constructed, able to keep the spatialrelationship between locations into account. Applications with simulated and real data are here developed.keywords: Generalized Regression Estimator, Spatial Sampling, Geographically Weighted RegressionAlessandro Vagheggini, via Lauro de Bosis,153100 Siena (SI), ItalyE-mail address:avagheggini@gmail.com

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

106

Bologna, Italy TIES 2009

Oral Presentation

MULTIPLE TESTING ON STANDARDIZED MORTALITY RATIOS: A BAYESIAN HIER-ARCHICAL MODEL FOR FALSE DISCOVERY RATE ESTIMATION.

Massimo Ventrucci1

1Dipartimento di Scienze Statistiche "Paolo Fortunati"

The analysis of large datasets of Standardized Mortality Ratios (SMR) obtained by collecting observedand expected counts of disease in a map of regions is useful in descriptive epidemiology to highlight potentialenvironmental risk factors. The work focuses on small areas and spatially correlated risks case studies. Theseissues are usually addressed in the literature by means of disease mapping models providing maps of smoothedrisk estimates that are assessed by visual inspection. We propose a decision oriented approach that performshypothesis testing over all areas by controlling the False Discovery Rate (FDR). Traditional p-value basedmethods fail in FDR controlling because of the over-dispersion of small area counts. A Bayesian hierarchicalmodel including spatial random e�ects to make allowance of extra-Poisson variability is implemented to esti-mate the posterior probabilities that the null hypothesis of absence of risk is true. By means of such posteriorprobabilities, an estimate of the expected FDR conditional on data can be computed. The availability of thisestimate allows the practitioner to determine non arbitrary FDR based selection rules to identify high-riskareas given a preset FDR level. A conservative estimation of the realized FDR is needed to achieve the FDRcontrol. The goodness of FDR estimation is checked by simulating many spatial scenarios: FDR is accuratelyestimated in small areas and low risk level scenarios and, consequently, rules based on FDR= 0.05 can besuggested. The model also provides a map of posterior risk estimates.keywords: FDR, Bayesian multiple testing, Disease mappingMassimo Ventrucci, Via Belle Arti 41,40126, BolognaE-mail address:massimo.ventrucci@unibo.it

Supporting grant: work partially supported by Italian Relevant National Research Projects (PRIN-MIUR-2006139812) "Methodsfor collecting and analyzing environmental data".

Oral Presentation

DATA-DRIVEN NEIGHBORHOOD SELECTION OF A GAUSSIAN FIELD.

Nicolas Verzelen1

1Université Paris-Sud 11

Let X be a stationary Gaussian �eld observed on a regular lattice. In this talk, I will introduce a non-parametric estimation procedure for the distribution of X. This method relies on the estimation of the covari-ance in the space of Gaussian Markov random �elds (GMRF). The challenge is to select a good neighborhoodfor the GMRFs. Indeed, a too large neighborhood yields an estimator with large variance. If one choosesa small neighborhood, then the estimator has a small variance but it may not approximate well the truedistribution. This is why we introduce a data-driven model selection strategy for choosing a neighborhoodthat achieves a trade-o� between the approximation error and the variance. The so-de�ned procedure appliese�ciently to large spatial data sets. From a theoretical point of view, it is shown to achieve minimax optimalrates of convergence. Finally, simulation experiments suggest that the procedure is often a good alternativeto variogram estimation.keywords: Spatial statistics, Gaussian Markov random �elds, Model selectionNicolas Verzelen, 1 rue du Pourtalet 34380 Saint-Martin de Londres FRANCEE-mail address:nicolas.verzelen@math.u-psud.fr

Oral Presentation

ANALYSIS OF FUZZY ENVIRONMENTAL DATA.

Reinhard VIERTL1

1TU Wien

All measurement data of continuous quantities are more less non-precise, also called fuzzy. This kind ofdata uncertainty is not stochastic in nature and also di�erent from errors, called imprecision. Especiallyenvironmental data (for example CO2 emissions of a country in one year) are remarkably non-precise. Thebest quantitative description of such non-precise data is by so-called fuzzy numbers. These fuzzy numbers

107

Bologna, Italy TIES 2009

are generalizations of real numbers and intervals, de�ned by a generalization of indicator functions, so-called characterizing functions. Statistical analysis of this kind of data make generalized statistical inferenceprocedures necessary. Such generalized statistical methods will be explained in the contribution.keywords: fuzzy data, fuzzy numbers, non-precise data, statistical inferenceReinhard VIERTL, Institute of Statistics and Probability Theory, Vienna University of Technology, 1040 Wien,AUSTRIAE-mail address:r.viertl@tuwien.ac.at

Oral Presentation

DENDROCLIMATOLOGICAL RECONSTRUCTIONS OF PAST CLIMATES: IS THE PRESENTTHE KEY TO THE PAST?.

Hans Visser1

1Netherlands Environmental Assessment Agency

For dendroclimatological reconstructions of past climates it is critically important that the relation be-tween indicators of tree growth and monthly to yearly temperature variables is stable over time (the unitarianprinciple). However, recently, a vast growing number of studies have been published where instable relation-ships have been found between tree growth and climate variables. Studies report a loss of sensitivity occurringin recent years and this unwanted result has been denoted as "the divergence problem". A number of expla-nations have been brought up for this divergence, ranging from inhomogeneous temperature data to risinglevels of air pollution and from global dimming to methodological �aws. Also the non-linear reaction of treesto climate variables has been suggested. In this presentation I will show that the current dendroclimato-logical methods are not appropriate anymore for the new situation where a range of non-stationary signalshamper the process of standardization, in�uencing the subsequent estimation response functions, and this incombination with arising non-linearities in the response function itself. I will show that the application ofstructural time series models (STMs) will solve most (but not all) of these divergence-related problems. Byapplication of STMs the traditional multiple regression model is extended with a time-varying intercept andtime-varying regression coe�cients, leading to so-called stochastic response functions. The approach will beillustrated by a re-analysis of two examples from the recent literature.Hans Visser, P.O. Box 3033720 AH Bilthoventhe NetherlandsE-mail address:hans.visser@pbl.nl

Oral Presentation

STEERING FACTORS AND THE ECOLOGICAL QUALITY OF REGIONAL SURFACEWATERS:A SUCCESSFUL APPLICATION OF REGRESSION-TREE ANALYSIS.

Hans Visser1

1Netherlands Environmental Assessment Agency

In the EU Water Framework Directive (WFD), the ecological quality of regional surface waters is expressedin terms of ratios which lie between 0 and 1 (from very poor to excellent ecological quality). These ecologicalquality ratios (EQRs) are available for the following groups of species: macrofauna, �sh, macrophytes, phy-tobenthos and phytoplankton. With regard to policy, it is important to know the way in which EQR valuesdepend on steering factors. These steering factors are partly of a physical-chemical nature (concentrationsof the total of phosphorus, the total of nitrogen, oxygen) and partly hydromorphological (meandering, waterlevel dynamics, shading, river bank design, damming, maintenance). For the analysis of such datasets theapplication of regression trees (part of so-called CART models) appears to be more e�cient than multipleregression due to the many non-linear relations between variables. For the calibration and validation ofregression trees I have used the RandomForest software within S-PLUS. Seven datasets (lakes, slow and fast�owing brooks, canals, ditches and the group of brackish waters) have analyzed, where each dataset containsEQRs for macrofauna, �sh, macrophytes and phytoplankton. In the presentation my experiences with theregression-tree approach will shown. Furthermore, I will compare the prediction strength of regression-treemodels with those gained by another approach, that of neural networks, applied to the same data.Hans Visser, P.O. Box 3033720 AH Bilthoventhe NetherlandsE-mail address:hans.visser@pbl.nl

108

Bologna, Italy TIES 2009

Oral Presentation

THE INTERPLAYOF ENVIRONMETRICSWITH OTHER ENVIRONMENTALLY-ORIENTEDSOCIETIES.

Kristina Voigt1, Scherb Hagen1

1Helmholtz Zentrum Muenchen

It becomes extremely urgent to consider environmental protection as one important aspect in connectionwith sustainability and sustainable development. Environmental statistics, other mathematical approaches,environmental modelling, and environmental information systems play a major role in this respect. Thedisciplines of Environmetrics, Environmental Informatics, Environmental Modelling and Software as well asthe Partial Order in Environmental Sciences and Chemistry were established many years ago with little orno knowledge about each other. An overview on the activities working on similar environmental topics willbe given focussing on the sustainability (green) issues of the groups. Humanity has exceeded the carryingcapacity of the global environment. The only real choices for the future are to bring the throughputs thatsupport human activities down to sustainable levels through human choice, human technology, and humanorganization, or to let nature force the decision through lack of food, energy, or materials, or through anincreasingly unhealthy environment with respect to chemical contamination and physical conditions like,e.g., global warming and changing background radiation levels, as well as the increase in use of pesticides.One small but important aspect is that the groups and societies, which are already working in the �eld ofenvironmental protection and sustainability, increase their knowledge about complementary activities andfocus their e�orts by improving and intensifying their collaboration in the future. A sustainable world cannever be fully realised until this goal is widely envisioned. [Meadows D. et al. Limits to Growth: The 30-yearupdate, 2004].keywords: Environmetrics, Environmental Informatics, Environmental Modelling, Partial Order in EnvironmentalSciences, SustainabilityKristina Voigt, Helmholtz Zentrum Muenchen, Institute of Biomathematics and Biometry, Ingolstaedter Landstr. 1,85764 Neuherberg, GermanyE-mail address:kvoigt@helmholtz-muenchen.de

Oral Presentation

MIXED-EFFECTS MODEL SELECTION.

Ronghui Xu1, Michael Donohue2, Florin Vaida2, Rosanna Haut2

1University of California, San Diego, 2UC San Diego

Mixed-e�ects models are useful for analyzing environmental data. The random e�ects in the models canbe quantities of interest to estimate; for example, environmental e�ect at a particular location. As in anystatistical modeling, model selection is an important step. In the presentce of random e�ects, we discussthe marginal and the conditional focus. The marginal focus is on the �xed e�ects and variance componentsparameters, while the conditional focus is on the �xed as well as the random e�ects. For both focuses weconsider the Akaike information: for the former it is the classical Akaike information for parametric models; forthe latter a conditional Akaike information was very recently de�ned, and a corresponding Akaike informationcriteria was developed for the linear mixed model (Vaida and Blanchard, Biometrika 2005, p.). We furtherdevelop the conditional AIC for the generalized linear mixed models. Time permitting we will also discussthe handling of nuisance parameters, and the conditional AIC for the proportional hazards mixed (sometimesalso called frailty) models. The methods are illustrated on a skin cancer data set.keywords: conditional AIC, GLMM, PHMM, frailty modelsRonghui Xu, 9500 Gilman Drive, MC 0112,La Jolla, CA 92093-0112, USAE-mail address:rxu@ucsd.edu

109

Bologna, Italy TIES 2009

Oral Presentation

ANALYSIS OF CLUSTERED ENVIRONMENTAL MULTINOMIAL DATA WITH RAN-DOM CLUSTER SIZES.

Guohua Yan1, Renjun Ma1

1University of New Brunswick

Environmental contaminants inadvertantly �nd their way into the human food supply and cause potentialhealth and economic harm. We are interested in the odds of incidence due to di�erent categories of contami-nants, such as pesticides, mercury, polychlorinated biphenyls (PCBs) and polybrominated biphenyls (PBBs)etc. We regard the data for food contamination incidents as clustered counts with random cluster sizes. Tra-ditional approaches assume �xed cluster sizes. We propose a Poisson mixed model with multilevel randome�ects to take into account both intra-cluster correlation and extra-cluster variation. Our orthodox bestlinear unbiased predictor approach to this model depends only on the �rst and second moment assumptionsof unobserved random e�ects and is computationally e�cient.keywords: Food contamination, best linear unbiased predictor, random e�ects, multinomial data, overdispersionGuohua Yan, Department of Mathematics and Statistics, University of New Brunswick, Fredericton, Canada E3B5A3E-mail address:gyan@unb.ca

Supporting grant: This research is partially supported by grants from the Natural Sciences and Engineering ResearchCouncil ofCanada.

Poster Presentation

SURFACE WATER QUALITY ASSESSMENT USING STATISTICAL TECHNIQUES.

Suheyla Yerel1, Huseyin Ankara2

1Bilecik University, 2Eskisehir Osmangazi University

This study, aims to investigate the degree of the in�uence of contaminant sources on surface water in theSakarya River (Turkey). The Sakarya River is situated in the northwest Anatolian region of Turkey and itstotal length is nearly 810 km. the headwater of the Sakarya River is about 3 km southwest to the townCifteler, which is within the Eskisehir province. In this study, surface water quality dataset collected twodi�erent years from observation stations along Sakarya River are statistically examined. Descriptive statisticsfor chlorite, nitrate, nitrite, ammonium, sodium, sulfate, temperature, dissolve oxygen, biochemical oxygendemand, and chemical oxygen demand parameters are calculated. Time series analysis and Mann-Whitney Utest to compare groups also applied. Comparison of surface water quality parameters stated that; parametersare statistically di�erent from the others. These statistical techniques are believed to assist decision makersin determined priorities to improve surface water quality that has identi�ed due to various land uses.keywords: Time series analysis, Mann-Whitney U test, Sakarya River, water qualitySuheyla Yerel, Bilecik University Bozuyuk Vocational Schoo lBozuyuk/Bilecik TurkeyE-mail address:syerel@gmail.com

Poster Presentation

ON THE PARTITIONING UNCERTAINTY IN GLOBAL CLIMATE PREDICTIONS.

Stan Yip1

1National Centre for Atmospheric Science and University of Exeter

Uncertainty in climate predictions arises from several di�erent sources. Some sources of uncertainty canbe reduced by the new investment in computing, measurement facilities and the advancement of the climatescience. This paper analyses data of global annual mean of projected surface air temperature from a set ofclimate models under three emission scenarios for 20th and 21th century. Following Hawkins and Sutton(2009), we introduce some methods to improve our understanding on how di�erent sources of uncertaintyare partitioned. Recasting the problem into a time-series problem, a Bayesian dynamic linear model isproposed which allows a more �exible analysis to the problem and provides a potential to extend to a space-time model. Hawkins, E and Sutton, R. (2009) The potential to narrow uncertainty in regional climatepredictions. Bulletin of the American Meteorological Society. In Press.keywords: climate uncertainty, hierarchical model, Makrov chain Monte CarloStan Yip, University of Exeter, School of Engineering, Computing, Mathematics, Harrison Building, North ParkRoad, Exeter, EX4 4QJ,UK. E-mail address:c.y.yip@ex.ac.uk

110

Bologna, Italy TIES 2009

Oral Presentation

DEALING WITH LARGE COVARIANCE MATRICES FOR SPATIAL DATA.

Hao Zhang1

1Purdue University

When a spatial sample of size n is observed, an n by n matrix will need to be inverted for spatial predictionor inferences on parameters. When the sample size is tens of thousands or larger, innovative methods arenecessary for computing the inverse. Sparsity and the lower rank structure are the two major techniquesdealing with large matrices. This talk will review some recent advances in theory, algorithm and applications.Hao Zhang, Department of Statistics Purdue University West Lafayette, IN 47907, USAE-mail address:zhanghao@purdue.edu

Supporting grant:NSF DMS-0706835

Oral Presentation

MODELLING BACTERIAL DENSITY COUNT DATA WITH VARIOUS OVERDISPER-SION AND TAIL HEAVINESS.

Rong Zhu1, Abdel H. El-Shaarawi2, Harry Joe3

1McMaster University, Canada, 2National Water Research Institute, Environment Canada, 3University ofBritish Columbia, Canada

The distribution of organism in the environment deviates frequently from the randomness due to nat-ural cycles, availability of resources and avoidance of harm. As a result, observed data can show variousover-dispersion, skewness and heavy tail. Thus, end users face the problem to determine an appropriatedistribution for observations from many possible models. We proposes a three-parameter discrete distribu-tion family which uni�es some widely used distributions such as Poisson, negative binomial, Poisson-inverseGaussian, Polya-Aeppli, Poisson-Pascal, Neyman Type A and discrete stable. This family is called the uni�edoverdispersed Poisson (UODP), and its subfamilies have features such as equal-dispersion, over-dispersion,heavy tail or even extremely heavy tail (i.e., in�nite expectation), thus, lending the greater �exibility inmodel selection for end users. Recursive formulas have been developed for computation of the probabilitymass function, as well as statistical inferences based on likelihood. These developments are applied to abacterial density study with grouped data sets of coliform counts in Lake Erie for water quality mornitoring.keywords: count data, heavy-tailed, over-dispersed, Poisson-inverse Gaussian, discrete stableRong Zhu, Department of Mathematics and Statistics, McMaster University, 1280 Main Street West, Hamilton,Ontario,Canada L8S 4K1E-mail address:rzhu@math.mcmaster.ca

Supporting grant:NSERC

Poster Presentation

APPLICATION OF ROBUST PRINCIPAL COMPONENT ANALYSIS FOR ASSESSMENTOF CS-137 TRANSPORT IN FOREST SOIL.

Zbigniew Ziembik1, Agnieszka Doªha«czuk-�ródka1, Maria Wacªawek1

1Opole University

The work focuses on estimation of mutual relationships between Cs-137 activity in di�erent horizonsof forest soil. Cs-137 was introduced to the environment as radioactive fallout in �fties and sixties of thetwentieth century as the result of nuclear weapon tests. During the Chernobyl nuclear power plant breakdownin 1986, a large amount of radioactive matter was exhausted to the atmosphere, among others Cs-137. Thisisotope has started circulation between various alive and abiotic components of natural environment. Abig number of factors a�ecting translocation of this element in woodland ecosystems signi�cantly hampersmodeling of this process. An insight in nature of this processes might be gained by analysis of the resultsreceived from statistical methods applied to the data obtained for di�erent environmental components. Themain aim of our computations was assessment of relationships between Cs-137 activity in forest soil horizonsand subhorizons. For multidimensional data analysis the Principal Component Analysis method was chosen.To diminish the e�ect of unequal initial soil contamination the relative Cs-137 activities were calculated ineach soil pro�le. To make data distributions similar to the normal one, they were transformed using Box-Cox

111

Bologna, Italy TIES 2009

formula. The transformed data were standardized and then PCA method was applied to assess existence ofrelationships between particular variables. It was expected that in our data the certain outlying points couldbe found. To overcome this problem the robust PCA method was used. Substantial di�erences betweenresults obtained from both methods was a�rmed.keywords: 137-Cs, forest soil, circulation, robust PCAZbigniew Ziembik, Opole University,ul. kard. B.Kominka 4,45-032 Opole,PolandE-mail address:ziembik@uni.opole.pl

Oral Presentation

TESTING ISOTROPY OF SPACE COVARIANCE FUNCTIONS.

Alessandro Zini1

1Università degli studi di Milano-Bicocca

In this work, two kinds of tests of isotropy in three di�erent parametric frameworks (isotropy vs anisotropy,geometrical isotropy vs zonal isotropy, isotropy vs anisotropy, again) are proposed. In order to implementthese procedures, weighted composite likelihood (WCL) and quasi arithmetic means are needed. Theoreticalresults about WCL guarantee the consistency of the tests, which turn out asymptotically equivalent, also.Their statistical properties are evaluated, considering some well-known classes of spatial covariance functions,such as the Generalised Cauchy Family.keywords: Geometrical isotropy, Space covariance functions, Weighted composite likelihood, Zonal isotropyAlessandro Zini, via Bicocca degli Arcimboldi, 8 20126 Milano (MI)E-mail address:alessandro.zini@unimib.it

112

Index

Adel�o Giada, 16Agrimi Mariagrazia, 28Agro' Gianna, 33Ailliot Pierre, 16, 74Albek Erdem, 17Albek Erdem Ahmet, 50Albek Mine, 17Albert-Green Alisha, 18Alvim-Ferraz M.C.M., 84Amiri Esmail, 18Andriani Eleonora, 18Angulo José M., 19, 25Ankara Huseyin, 19, 110Augustin Nicole, H., 76, 77

Bacco Dimitri, 83Badran Hussein M., 58Ba�etta Federica, 28Bagliani Marco, 21Ballesteros Bruno, 46Bande Stefano, 27Banerjee Sudipto, 20, 41Barão Maria Isabel, 75Barbati Anna, 28Barrett Michael, 58Bartley Rebecca, 61Bartoletti Silvia, 29Bartusek Pavel, 60Beaulieu Claudie, 20Bell Michelle, 34Bellisario Audrey, 96Benoit Yves, 37Berrocal Veronica, 21Bertaccini Pancrazio, 21, 22Betrò Bruno, 23Biggeri Annibale, 27Binquet Christine, 96Blaber Stephen, 61Blangiardo Marta, 22, 49, 100Bodini Antonella, 23Bodnar Olha, 98Bojkova Jindriska, 60Bolin David, 23, 66Bolzacchini Ezio, 33, 73Bondì Anna Lisa, 84Bonithon-Kopp Claire, 96Borowiak Klaudia, 57Bosetto Marinella, 79Bowman Adrian, 87Brabec Karel, 60Braun W. John, 18, 63Bravington Mark, 57Bravington Mark V., 42Breysse Patrick, 31Briggs David, 49Brooks Harold, 69Bruno Francesca, 23, 24, 74, 106Brus Dick, 24Buchholz Oliver, 53Buckley Timothy, 31Budka Anna, 57Bueso María C., 25

Cárdenas Giovanni, 38Cacciari Alessandra, 33Cafarelli Barbara, 25Calamai Luca, 79Calculli Crescenza, 26Calvete H.I., 26Cameletti Michela, 27Capri Ettore, 79Cargill Alistair, 43Carloni Andrea, 48Carpentier Claire, 37Carrión J.A., 26Carvalho Laurence, 39Caselli Maurizio, 18, 106Casper Markus, 38, 47Casper Markus C., 53Castrignano' Annamaria, 25Catelan Dolores, 27Chevallier Frédéric, 105Chica-Olmo Mario, 67Chiodi Marcello, 16Chiu Grace, 28Christakos George, 77Christensen William, 28Ciavatta Stefano, 105Clerc Laurence, 96Cocchi Daniela, 23, 40, 46, 97Codd Geo�rey A., 39Colecchia Salvatore, 25Columbu Silvia, 96Corona Piermaria, 28Cosmi Carmelina, 32Cossu Q. Antonio, 23Costa Marco, 45Cotroneo Rossana, 29Cruci�x Michel, 29, 94Curci Gabriele, 33Cuzol Anne, 74

d'Antonio Luca, 65D'Ariano Cinzia, 34, 40Dabo-Niang Sophie, 30Daoud Maged, 30Davison Anthony, 78de Fouquet Chantal, 37, 38de Gennaro Gianluigi, 18, 106de Gruijter Jaap, 24Decastro Rey, 31Delicado Pedro, 44Demétrio Clarice G.B., 91Deman Dr Suresh, 31Denby Bruce, 100Deng Dianliang, 32Deschênes Julie, 91Di Leo Senatro, 32Di Nicolantonio Walter, 33Di Salvo Francesca, 33Dick Jan, 100Diggle Peter, 91Dobricic Srdjan, 83Dolhaczuk-Srodka Agnieszka, 111Domínguez José Antonio, 46Dominici Francesca, 34

113

Bologna, Italy TIES 2009

Dondi Francesco, 83Donohue Michael, 109Dubin Joel, 28Dukic Vanja, 22Dunson David, 34Dunstan Piers K., 42

El-Shaarawi Abdel, 35, 102El-Shaarawi Abdel H., 111Elatrash Mokhtar, 35Esterby Sylvia, 36Etxeberria Jaione, 45Evin Guillaume, 36

Fassò Alessandro, 34, 40Fattorini Lorenzo, 28, 37Faucheux Claire, 37, 38Faust Christophe, 38Favre Anne-Catherine, 36Fawcett Lee, 39Ferguson Claire, 39, 50, 99Fernàndez Pascual Rosaura, 86Fernandes Ana Cristina, 72Ferrara Claudia, 65Ferrara Vincenzo, 48Ferrari Clarissa, 40Ferrero Luca, 33, 73Feuer Eric, 58Finazzi Francesco, 34, 40Finley Andrew, 41Focaccia Sara, 41Foster Scott D., 42Fotopoulos Stergios B., 55Franceschi Sara, 43Franco Villoria Maria, 43Fricaudet Bruno, 37Fromont Agnes, 96Fryzlewicz Piotr, 94

Göncü Serdar, 50Güngör Ömer, 50Galé C., 26Gallo Claudio, 48García E., 26García-Soldado María José, 67Gargoum Ali, 43Garreta Vincent, 44Gasparini Mauro, 54Gasser Theo, 92Gelfand Alan, 21Gelfand Alan E., 40, 82Gemmar Peter, 38, 47Geyh Alison, 31Ghosh Sucharita, 72Gilleland Eric, 69Giraldo Ramón, 44Goicoa Tomás, 45Gonçalves A. Manuela, 45Gourry Jean-Christophe, 37Gramatica Paola, 79Greco Fedele, 46Gri�ths Shane, 61Grigg Olivia, 46Grima Juan, 46Grima-Olmedo Juan, 67Grimvall Anders, 47

Gronz Oliver, 38, 47Grundmann Jens, 53Guastaldi Enrico, 48Guenni Lelys, 48Guillou Armelle, 89, 105Guindani Michele, 82Guiot Joël, 44Gulliver John, 49Gutiérrez-Jaimez Ramón, 89Gutiérrez-Sánchez Ramón, 89Guttorp Peter, 49, 78

Hagen Scherb, 109Haggarty Ruth, 50Hailemariam Temesgen, 51Hajek Ondrej, 59Hansell Anna, 49Haslett John, 51, 80, 93, 102Haug Ola, 52Haut Rosanna, 109Hedley Sharon, 57Held Leonhard, 52Herbst Marcus, 53Herr Alexander, 61Hodges James, 20Hoey Trevor, 43Hofman Radek, 81Holmström Lasse, 53Holoubek Ivan, 59Horvathova Eva, 54Hrdlickova Zuzana, 36Hu Baoqing, 55Huerta Gabriel, 48

Ibrahim Hesham, 35Ielpo Pierina, 106Ignaccolo Rosaria, 22, 27Imparato Daniele, 54Ip Wai Cheung, 55Iurisci Gianfranco, 95

Jandhyala Krishna, 68Jandhyala Venkata K., 55Jarkovsky Jiri, 60Jaruskova Daniela, 55Joe Harry, 111Jollois Francois-Xavier, 85Jona Lasinio Giovanna, 40, 56Jonathan Rougier, 29Jones Bruce, 63Jonsson Peter, 66

Künsch Hans Rudolf, 72Kaufman Cari, 77Kavalieris Laimonis, 56Kayzer Dariusz, 57Kelly Natalie, 57Khan Shahedul Ahsan, 28Khapalova Elena, 55Khraibani Hussein, 58Khraibani Zaher, 58Kim Hyune-Ju, 58Kinsey-Henderson Anne, 61Klozyatnyk Ivan, 58Klymenko Natalia, 58, 94Knol Anne, 59

114

Bologna, Italy TIES 2009

Komprda Jiri, 59Kovárová Milena, 60Kozlowska Maria, 60Kozlowski Radoslaw J., 60Kozyatnyk Ivan, 94Krawczyk Roman, 60Kublin Edgar, 76Kubosova Klara, 59, 60Kuhnert Petra, 61Kuhnova Jitka, 86

La Torre Vittorio, 81Lacka Agnieszka, 60Lagazio Corrado, 27Lagona Francesco, 62, 70Larsen Bo R., 62Leandro Roseli A., 91Lee Duncan, 63, 64, 87, 99Lee Jonathan, 63Lee YoungSaeng, 63Lega Massimiliano, 65, 82Lemos Ricardo, 95Leorato Samantha, 73Lindgren Finn, 23, 66Lindgren Georg, 66Lindström Johan, 66Lindstrom Johan, 93Lovison Gianfranco, 76Lozzi Irene, 79Lucidi Stefano, 81Lumley Thomas, 103Luo Jun, 58Luque Juan Antonio, 46Luque-Espinar Juan Antonio, 67

Ma Renjun, 68, 110Macchiato Maria, 32MacNeill Ian, 68Madrid Ana E., 19Magnussen Steen, 68Magram Saleh, 30Mahal Lisa G., 71Malherbe Laure, 38Manganiello Patrizia, 65Mannarini Gianandrea, 40Mannshardt-Shamseldin Elizabeth, 69Marchetti Nicola, 83Martell David L., 18Martin Michael, 90Martina Mario Lloyd Virgilio, 69Martins F.G., 84Maruotti Antonello, 70Masniak Anna, 101Mateu Jorge, 44, 70Mattioli Walter, 28McNaughton Beverly, 71McRoberts Ron, 68McRoberts Ronald E., 71Mearns Linda, 77Menéndez Patricia, 72Menezes Raquel, 72Mercuriali Mattia, 73, 83Mezzetti Maura, 73Mietelski Jerzy W., 101Miglio Rossella, 74

Migon Helio, 98Mihalic Jana, 31Militino Ana F., 45Milli� Ralph, 83Moita Maria Teresa, 75Monbet Valérie, 74Monestiez Pascal, 75Monleon Vicente, 51Moreau Thibault, 96Mouriño Helena, 75Muggeo Vito, 76Musio Monica, 76, 77, 96Myllymäki Mari, 104

Núñez-Lagos R., 26Na�di Ahmed, 89Napoli Rodolfo, 82Napoli Rodolfo M.A., 65Naumova Elena, 68Naveau Philippe, 89, 105Neocleous Tereza, 64Nerini David, 75Nicolis Orietta, 40, 70, 77Nychka Douglas, 77

O'Neill Terence, 87Okasha Aly, 35Olech Maria A., 101Onorati Rossella, 78Ouarda Taha B.M.J., 20

Pérez C., 26Padoan Simone, 78Padoan Simone, A., 89Palma So�a, 75Pantani Ottorino-Luca, 79Papa Ester, 79Pardo-Igúzquiza Eulogio, 67Park Jeong Soo, 63Park Juhyun, 92Parnell Andrew, 80Pasanen Leena, 53Pasti Luisa, 83Pastore Andrea, 105Pastres Roberto, 105Paul Warren, 80Pebesma Edzer, 100Pecha Petr, 81Pechova Emily, 81Peel David, 57Pelliccioni Armando, 29, 81Peng Roger, 34Penttinen Antti, 104Pereira M.C., 84Perrone Maria Grazia, 73Persechino Giuseppe, 82Petersen Arthur, 59Petrone Sonia, 82Picone Marco, 62Pietrogrande Maria Chiara, 73, 83Pignone Sara, 69Pinardi Nadia, 83Pires José, 84Plaia Antonella, 33, 84Poggi Jean-Michel, 85Pollice Alessio, 26, 85

115

Bologna, Italy TIES 2009

Polus-Lefebvre Edwige, 37Pompei Enrico, 28Porcu Emilio, 86Portier Bruno, 85Pospisil Zdenek, 86Potts Jacqueline, 86Powell Helen, 87Puimedón J., 26Pungì Fabrizio, 81Puza Borek, 87

Rachdi Mustapha, 30Radermacher Walter J., 88Ragosta Maria, 32Raillard Nicolas, 74Ramos M. do Rosário, 88Ramos-Ábalos Eva Mª, 89Ravines Romy, 98Reese Shane, 28Ribatet Mathieu, 78, 89Ribereau Pierre, 89Richardson Sylvia, 22Rico Nuria, 90Roberts Steven, 90Rodríguez Marco A., 91Rodríguez S., 26Rodrigues Alexandre, 91Román Patricia, 90Romero Desiree, 90Rosa Rodolfo, 24Rotondi Renata, 91Rougier Jonathan, 94Rousson Valentin, 92Rue Håvard, 66Ruggieri Mariantonietta, 33, 84Ruiz Medina Maria Dolores, 86Ruiz-Medina María D., 25

Sain Stephen, 77Sain Steve, 92Salter-Townshend Michael, 80, 93Salway Ruth, 64, 99, 100Sampson Paul D, 93Sampson Paul D., 78Samsoni-Todorova Olena, 94Sanderson Jean, 94Sanka Milan, 59Sansó Bruno, 48Sanso Bruno, 95Sarra Annalina, 95Sarsa M.L., 26Sauleau Erik, 77Sauleau Erik-A., 96Scagliarini Michele, 97Scherb Hagen, 97Schiano Pasquale, 82Schmid Wolfgang, 98Schmidt Alexandra, 98Schoenberg Frederic, 98Schrödle Birgit, 52Schumacher Martin, 76Scott E. Marian, 39Scott Marian, 43, 50, 99Seidou Ousmane, 20Severino Vincenzo, 65

Shaddick Gavin, 63, 64, 99, 100Sheppard Lianne, 93, 103Sirisack Sackmone, 47Sisson Scott, A., 89Smith Denis, 43Smith Ron, 99, 100Sobiech-Matura Katarzyna, 101Solaiman Tarana A., 101Sophie Dabo-Niang, 103Spiegelman Cli�ord H., 102Stein Susan M., 71Stevens Don, 102Stocchi Paolo, 33Sweeney James, 102Szpiro Adam, 93, 103

Talenti Luca, 24Tawn Jonathan, 46Taylor Steve, 36Thiam Baba, 103Thompson Craig, 16Thomson Peter, 16Tinner Willy, 72Todini Ezio, 69, 104Tomasi Claudio, 33Tomppo Erkki, 104Tonellato Stefano, 105Torres Francisco, 90Toulemonde Gwladys, 105Trivisano Carlo, 46Trizio Livia, 106Troccoli Antonio, 25Tyler Andrew, 39

Ugarte M.D., 45

Vagheggini Alessandro, 106Vaida Florin, 109van der Sluijs Jeroen, 59Varini Elisa, 91Venables William, 61Ventrucci Massimo, 107Verzelen Nicolas, 107VIERTL Reinhard, 107Villar J.A., 26Visser Hans, 108Voigt Kristina, 97, 109von Wilpert Klaus, 76Vrac Mathieu, 105

Waclawek Maria, 111Walker Stephen, 99Walshaw David, 39Wang Lu, 31Weiskittel Aaron, 51Williams Basil, 28Wilson Duncan, 51Wong Heung, 55Wood Simon, 76Woolford Doug, 63Woolford Douglas G., 18Wotton Mike, 63

Xia Jun, 55Xu Ronghui, 109

Yan Guohua, 110

116

Bologna, Italy TIES 2009

Yao Anne-Françoise, 30Yerel Suheyla, 19, 110Yip Stan, 110Yoon SangHoo, 63

Zahradkova Svetlana, 60Zanghirati Gaetano, 83Zbierska Janina, 57Zenie Alex, 100Zhang Hao, 111Zhang Yufen, 20Zhu Rong, 111Zidek Jim, 64Ziembik Zbigniew, 111Zini Alessandro, 112Zocchi Silvio S., 91

117