This article appeared in a journal published by Elsevier. The attachedcopy is furnished to the author for internal non-commercial researchand education use, including for instruction at the authors institution
and sharing with colleagues.
Other uses, including reproduction and distribution, or selling orlicensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of thearticle (e.g. in Word or Tex form) to their personal website orinstitutional repository. Authors requiring further information
regarding Elsevier’s archiving and manuscript policies areencouraged to visit:
http://www.elsevier.com/copyright
http://www.elsevier.com/copyright
Author's personal copy
Multi-year assessment of photochemical air quality simulation over Spain
Marta G. Vivanco a,*, Inmaculada Palomino a, Robert Vautard b, Bertrand Bessagnet c,Fernando Martı́n a, Laurent Menut d, Santiago Jiménez e
a CIEMAT, Atmospheric Pollution Unit, Avda. Complutense, 22, 28040 Madrid, Spainb LSCE/IPSL, Laboratoire CEA/CNRS/UVSQ, Gif sur Yvette, Cedex, Francec Institut National de l’ Environnement Industriel et des Risques, INERIS, Parc Technologique Alata, 60550 Verneuil en Halatte, Franced Laboratoire de Meteorologie Dynamique IPSL, Ecole Polytechnique, 91128 Palaiseau, Francee INYPSA (Informes y Proyectos, S.A.), Madrid, Spain
a r t i c l e i n f o
Article history:Received 17 September 2007Received in revised form 13 May 2008Accepted 15 May 2008Available online 1 July 2008
Keywords:OzonePhotochemical modellingAir qualityEvaluating model performanceTropospheric chemistry
a b s t r a c t
Ground-level ozone concentrations in the atmospheric boundary layer over Spain are still exceedingthresholds established in EU legislation to protect human health and prevent damage to ecosystems. Theincreasing role that air quality models play in air quality management requires comparison betweenmodel results and previous observations in order to determine the capacity of the model to reproducepast events.The CHIMERE chemistry-transport model has been used by several research groups to estimate airpollutant concentrations in different European countries. An evaluation of the model performance of theCHIMERE air quality model was carried out for the spring and summer periods of 2003–2005 in Spain,using EMEP emissions. This evaluation has demonstrated a fair agreement between observed andmodelled ozone values for background stations, with a mean normalized absolute error below 15% forrural background air quality sites. This value lays inside the range proposed in EPA’s guideline for anacceptable level of model performance. In spite of this acceptable model performance, further studiesneed to be carried out to explain some underestimation found over Madrid surroundings.
� 2008 Elsevier Ltd. All rights reserved.
1. Introduction
In spite of the efforts focused on reducing pollutant emissions,concentrations in excess of European air quality standards havebeen recorded in Spain for some air pollutants, such as troposphericozone and particulate matter (Baldasano et al., 2003; EEA, 2006).Monitoring data indicates that ozone concentrations are aboveEuropean standards at many locations, representing a potential riskto human health.
In Europe, the increasing concern for public health hasmotivated a continuous improvement of air quality models.Together with the increase of the monitoring activity, thedevelopment of regional and local air quality models has allowed,in the last decade or so, a comprehensive picture of the three-dimensional distribution of some pollutants like ozone. Computerpower has reached a sufficient power to allow long-term (one orseveral years) simulations of air quality at regional scale (Jonsonet al., 2005; Vautard et al., 2006) or city scale (Cuvelier et al., 2007),with extensive comparisons between simulated and observed
values at ground-level. Simulations of the impact of Europeanpolicies on pollutant emissions have also been carried out, withinthe Clean Air For Europe (CAFE) programme (Amann et al., 2005;Cuvelier et al., 2007).
Air quality models have also been applied over the IberianPeninsula for short time periods (Jiménez et al., 2006; Pérez et al.,2006). Long-term simulations of ozone and particulate matter havebeen carried out over a smaller area of the Iberian Peninsulacovering Portugal (Monteiro et al., 2005). Presently, a new model-ling system for high-resolution air quality forecast in Spain is beingdeveloped under the financial support of the Environment Minis-try. Several Spanish research institutions (BSC, CIEMAT, IJA–CSICand CEAM) are participating under the leadership of BarcelonaSupercomputing Center (BSC). This system is called CALIOPE and itis based on a set of models: HERMES for emissions, WRF for me-teorology, and the CHIMERE and CMAQ for air quality.
In this article, the skill of a three-dimensional chemistry-transport model for simulating background photochemical airquality over Spain is examined by means of multi-year simulationsand systematic comparisons with observations.
Jakeman et al. (2006) have proposed some guidelines for goodmodelling practice in a broad range of natural resource modellingsituations. They have outlined 10 steps in model development in
* Corresponding author. Tel.: þ34 913466711.E-mail address: [email protected] (M.G. Vivanco).
Contents lists available at ScienceDirect
Environmental Modelling & Software
journal homepage: www.elsevier .com/locate/envsoft
1364-8152/$ – see front matter � 2008 Elsevier Ltd. All rights reserved.doi:10.1016/j.envsoft.2008.05.004
Environmental Modelling & Software 24 (2009) 63–73
Author's personal copy
order to have credible models. Under this framework, the evalua-tion of the model performance presented in this paper is linked tothe last step (‘Model evaluation or testing’) proposed by Jakemanet al. The CHIMERE model has been used for some years by researchgroups in several European countries. A detailed description of themodel configuration and performances over Europe is presented inprevious studies, using surface observations (Schmidt et al., 2001;Bessagnet et al., 2004; Vautard et al., 2003; Derognat et al., 2003;Hodzic et al., 2005) or remote sensed observations (Hodzic et al.,2004). The results of the evaluation of the CHIMERE model in Spainwill allow to improve model formulations and to determine if themodel is suitable for some air quality management tasks, such asbackground air quality assessment and prediction.
The evaluation presented in this study is based on simulationsfor the spring and summer seasons of 2003–2005. Section 2describes the emissions and their evolution in Spain. Details aboutthe model simulations are included in Section 3. In Section 4 theevaluation process is described and the results are given in Section5. Section 6 contains concluding remarks.
2. The evolution of emissions in Spain
NOx and non-methane volatile organic compounds (NMVOC)play a critical role in ozone formation and constitute the most
important ozone precursors. According to the Spanish EnvironmentMinistry (http://cdr.eionet.europa.eu/es) NMVOC emissions in2003 were 3 million tons/year over Spain. They have been relativelystable during the past 10 years. Natural sources represent 47.2% oftotal NMVOC emissions in 2003, with ‘‘solvent use and otherproduct use’’ (SNAP activity) contributing 18.9%. In spite of theSolvents’ Emissions Directive 1999, emissions derived from solventuse have increased significantly, from 377 kt in 1990 to 516 kt in2003.
Again according to the official Environment Ministryinformation, emissions from the other important ozone precursor,NOx, represent in 2003 1.56 million tons/year, which constitutesan increment of nearly 21% with respect to 1990. Althoughcleaner vehicles lead to a decrease in NOx transport emissions,mobile emissions are still the main contributor to total NOxemissions. Considering both ‘‘road transport’’ and ‘‘other mobilesources and machinery’’ activities, mobile emissions represent53% of the 2003 total NOx emissions. Emissions from combustionin manufacturing industry sector (stationary sources) haveincreased over the last years. These emissions, added to otherindustrial emissions (combustion in energy and transformationindustries, production processes, extraction and distribution offossil fuels and geothermal and combustion in manufacturingindustry), represent 38% of NOx emissions to the atmosphere in2003. Thus, industrial processes and mobile sources contribute90.8% to 2003 total NOx emissions.
3. Model simulations
Simulations of photochemical compounds were carried outusing a regional version of the CHIMERE chemistry-transportmodel. This version (V200603par-rc1) calculates the concentrationof 44 gaseous species and both inorganic and organic aerosols ofprimary and secondary origin, including primary particulatematter, mineral dust, sulphate, nitrate, ammonium, secondaryorganic species and water.
Six-month periods (from April 1 to September 30) weresimulated with the CHIMERE model for 2003–2005. These periodswere chosen because it is important to have air quality forecast inthese months with highest radiation, when most polluted episodestake place in Spain. Vingarzan (2004) describes the annual
Table 1Definition of the metrics used in the evaluation of the CHIMERE model performance
Mean bias
BMB ¼1N
XðMi � OiÞ ¼ M � O
Mean normalizedbias
BMNB ¼1N
X�Mi � OiOi
�¼�
1N
XMiOi� 1�
Mean absolutegross error
EMAGE ¼1N
XjMi � Oij
Mean normalizedabsolute error
EMNAE ¼1N
X�jMi � OijOi
�
Correlationcoefficient
Corr: coef : ¼PðMi �MÞðOi � OÞnP
ðMi �MÞ2PðOi � OÞ2
o12
Daily ozone profile
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24hours
micro
gr/m
3
Observations - Ruralbackground stationsObservations -Urban&suburbanbackground stations
Model - Urban&suburbanbackground stationsModel - Rural backgroundstations
Fig. 1. Daily ozone profile, calculated for the 6-month period, 2003 at rural backgroundstation and urban and suburban background stations. The model estimates a similardaily profile for rural and urban and suburban sites. Observations show different ozoneprofiles for both station types.
Table 2Analysis of CHIMERE model performance for hourly ozone
Stationtype
All available stations Common stations
BMB (mg/m3) BMNB (%) EMNAE (%) NS BMB (mg/m
3) BMNB (%) EMNAE (%) NS
2003RB �9.6 �7.6 14.3 28 �8.1 �6.3 13.5 17SB �4.2 �2.3 15.2 13 �5.5 �3.4 15.7 9UB �3.6 �1.8 16.1 17 �2.9 �1.1 15.8 10
2004RB �9.2 �7.5 13.4 25 �10.1 �8.2 13.5 17SB �3.2 �1.5 13.4 14 �4.7 �2.8 14.6 9UB 2.6 �1.1 13.8 16 �0.6 0.7 13.1 10
2005RB �10.7 �8.6 14.5 47 �11.6 �9.4 14.8 17SB �5.1 �3.4 13.6 30 �7.8 �5.6 15.2 9UB �5.1 �3.6 13.0 29 �3.6 �2.2 13.4 10
Statistics were calculated using all the available stations for every year (left side ofthe table) and using the same set of stations for the 3 years (‘‘common stations’’,right side of the table). NS: number of stations; cutoff: 80 mg/m3.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–7364
Author's personal copy
background ozone cycle over the mid-latitudes of the NorthernHemisphere being characterized by a spring maximum peakingduring the month of May. In Spain, maximum hourly values at ruralbackground sites are usually registered in August, July or June.
Because of the possible influence of long range pollutanttransport the model system was first used at European scale overa domain ranging from 10.5W to 22.5E and from 35N to 57.5Nwith a 0.5� horizontal resolution and 14 vertical sigma-pressurelevels extending up to 500 hPa. The focus over the IberianPeninsula was achieved by simulations over a fine-scale domainwith a 0.2� resolution, using a one-way nesting procedure wherecoarse-grid simulations force the fine-grid ones at the boundarieswithout feedback. For both domains emissions were derived fromthe annual totals of the EMEP database for 2003 (Vestreng et al.,2005). Original EMEP emissions were disaggregated taking intoaccount the land use information, in order to get higher resolu-tion emission data. The spatial emission distribution from theEMEP grid to the CHIMERE grid is performed using anintermediate fine-grid at 1 km resolution. This high-resolutionland use inventory comes from the Global Land Cover Facility(GLCF) data set (http://change.gsfc.nasa.gov/create.html). For eachSNAP activity sector, the total NMVOC emission is split intoemissions of 227 real individual NMVOC according to the AEATspeciation (Passant, 2002), and real species’ emissions areaggregated into model species’ emissions. Biogenic emissions arecomputed according to the methodology described in Simpsonet al. (1995), for alpha-pinene, NO and isoprene. First, they arecomputed for standard meteorological conditions, using vegeta-tion and soil inventories (Simpson et al., 1999); then, theseemissions are tuned according to the meteorological conditionsprevailing during the simulation period.
Boundary conditions for the coarse domain were provided frommonthly climatology from LMDz-INCA model (Hauglustaine et al.,2004) for gases’ concentrations and from GOCART model (Chinet al., 2002) for particulate species, as described in Vautard et al.(2005).
The MM5 model (Grell et al., 1995) was used to obtainmeteorological input fields. The simulations were carried out fora coarse domain and a finer one, with respective resolutions of 36and 19 km. MM5 simulations are forced by the National Centre forEnvironmental Prediction model (GFS) analyses at both scales.
4. Description of the evaluation process
In the last decades Spain has made some important progressregarding the knowledge of air quality. The number of air qualitymonitoring stations covers currently the whole territory. Quality ofthe measurements has also increased and presently many pollut-ants are being monitored routinely accordingly to the EU air qualitydirectives and EMEP requirements.
In order to evaluate the CHIMERE ozone simulations, hourly anddaily ozone concentrations were compared to observationsseparately for rural, suburban and urban background stations.Traditional metrics commonly used in model evaluation (Tescheet al., 1990; Yu et al., 2006; Chang and Hanna, 2004), such as meanbias, mean normalized bias, mean absolute gross error, meannormalized absolute error and correlation coefficient, wereestimated for the three pollutants (O3, NO2 and NO). Definitions ofthese metrics are indicated in Table 1.
Due to the different number of available stations for each year,the evaluation process was done in two ways: (1) using all theavailable observations, in order to have the most complete pictureof the quality of model predictions and (2) using a homogeneousset of stations for all the 3 years (referred as ‘‘common stations’’ inTable 2). These stations have been used by the EnvironmentMinistry to report to the European Commission about the airquality in Spain. A statistical evaluation of the model performancefor nitrogen oxides (NO2 and NOx) was also carried out, for 2003and 2004.
Bearing in mind the use of CHIMERE model for background airquality assessment, cutoff levels were set up when computingthe statistics for model performance for ozone and NOx. No cutoff
Table 3Hourly ozone statistical values for background rural stations
Station code Station name Longitude Latitude BMB (mg/m3) BMNB (%) EMNAE (%)
1016001 IZKI �2.5 42.66 �5.8 �4.4 10.91055001 VALDEREJO �3.23 42.84 �2.3 �1.2 10.56016999 BARCARROTA �6.92 38.48 �4.5 �3.7 10.48298008 BD-VIC 2.24 41.94 �9.9 �7.4 15.612099001 SANT_JORDI 0.37 40.56 �4.1 �2.5 12.212141002 ZORITA �0.17 40.73 �9.4 �7.7 12.117001002 AL-AGULL 2.84 42.39 �8.2 �6.5 1417032999 CABO_CREUS 3.32 42.32 �6.1 �4.1 14.617125001 AK-PARDIÑAS 2.22 42.31 �12.6 �10.5 14.418189999 VIZNAR �3.47 37.24 �7.0 �6.1 11.819061999 CAMPISÁBALOS �3.14 41.28 �12.7 �11.3 14.120016001 PAGOETA �2.15 43.25 �9.3 �7.8 17.225051001 AZ-BELLV 1.78 42.37 �13.7 �11.5 14.825224999 ELS_TORMS 0.72 41.39 �4.3 �2.7 10.927058999 O_SAVIÑAO �7.7 43.23 1.8 2.3 14.128027001 BUITRAGO �3.62 40.98 �16.7 �13.6 16.228051001 CHAPINERÍA �4.2 40.38 �19.3 �17.6 20.428068001 GUADARRAMA �4.1 40.68 �20.7 �17.3 2028123001 RIVAS-VACIAMADRID �3.5 40.32 1.9 2.7 11.328133001 SAN_MARTIN �4.3 40.38 �24.2 �19 20.433036999 NIEMBRO �4.85 43.45 1.2 1.8 14.339086001 LOS_TOJOS �4.25 43.15 �16.1 �13.7 17.241059001 ALJARAFE �6.04 37.34 �16.4 �13 16.843064001 AY-GANDESA 0.44 41.06 �1.7 �0.7 13.244054001 CAMARENA �0.02 41.1 �20.0 �15.1 17.545153998 RISCO_LLANO �4.35 39.52 �11.6 �9.2 12.346263999 ZARRA �1.1 39.09 �3.6 �2.5 9.449149999 PEÑAUSENDE �5.87 41.29 �13.3 �10.7 14
Cutoff: 80 mg/m3.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–73 65
Author's personal copy
was used for correlation coefficients, in order to have thecomplete picture of the temporal behaviour of the model.Regarding ozone, only statistics for moderate-to-high ozoneconcentration cases (more important for human healthprotection) were considered by selecting predicted–observedvalue pairs when hourly observations were equal to or greaterthan the cutoff of 80 mg/m3. For NO2 and NOx a cutoff value of5 mg/m3 was used. In addition, statistics parameters for daily-maximum ozone values are included in Section 5.
Time series showing model and observed daily values (O3, NO2)at some rural stations are also included in order to illustrate thebehaviour of the model in a graphical way.
5. Results
First of all, an analysis of the ozone diurnal cycle was done foreach station type, in order to examine the hourly ozone evolution atthe different sites. Fig. 1 shows the mean diurnal profile for ozone at
2003- BIAS FACTOR AT RURAL BACKGROUND STATIONS
< - 15
-10 < Bias < -5-15 < Bias < -10
-5 < Bias < 0 0 < Bias < 5
10 < Bias < 15Bias > 15
5 < Bias < 10
EMEP sites
< - 15
-10 < Bias < -5-15 < Bias < -10
-5 < Bias < 0 0 < Bias < 5
10 < Bias < 15Bias > 15
5 < Bias < 10
2003- BIAS FACTOR AT URBAN AND SUBURBANBACKGROUND STATIONS
Fig. 2. Distribution map of bias for rural background station (top of the figure) and urban and suburban background stations (bottom of the figure).
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–7366
Author's personal copy
rural and urban and suburban background sites, considering all theobserved and modelled ozone hourly values for the 2003 6-monthperiod. Observations indicate that the diurnal cycle at suburbanand urban stations presents lower ozone values than rural back-ground stations, especially at nighttime. Nevertheless, the modeltends to show a similar profile at all station types, similar to that atrural background stations. The mean O3 profile observed at ruralsites is well reproduced by the model. Nevertheless, minimumvalues observed at urban and suburban stations are not
reproduced. This fact is likely related to model resolution. As NOxemissions are being distributed over a 0.2� � 0.2� area, it is hardlypossible to represent the ozone destruction that occurs over urbanor suburban areas with higher NOx concentrations duringnighttime.
Table 2 reports the statistical results for ozone, using hourlyozone values for the three 6-month periods. Left side of this tablereflects the statistical results when all the available stations wereused (different number of available stations in 2003–2005). Right
2003- CORRELATION COEFFICIENT AT URBAN AND SUBURBANBACKGROUND STATIONS
Corr > 0.70.5 < Corr < 0.7Corr < 0.5
2003- CORRELATION COEFFICIENT AT RURAL BACKGROUND STATIONS
Corr > 0.70.5 < Corr < 0.7Corr < 0.5EMEP sites
Fig. 3. Distribution map of correlation coefficients for rural background station (top map) and urban and suburban background stations (bottom map).
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–73 67
Author's personal copy
side of the table shows the statistical results when the same set ofstations is used for the three 6-month periods (‘‘common sta-tions’’). Statistical values for hourly ozone at the individual ruralbackground stations are shown in Table 3.
The US EPA (1991) suggested a range of�5–15% as an acceptableguideline of model performance for normalized bias. When back-ground rural stations in 2003 are considered (Table 3), just threestations, all of them located in Madrid area, do not meet the criteriaof acceptable model performance with respect to this guideline.
Palacios et al. (2004) have shown the influence of Madrid onregions located 100 km away from the city, which could suggestthat these stations are not rigorously measuring background levels.When averaged overall sites (Table 2), however, mean normalizedbias for rural background stations in 2003 is �8%, well within theEPA’s criterion. This value indicates that the model tends to under-predict hourly observed values higher than 80 mg/m3. Also, meannormalized bias presents acceptable values for the suburban andurban stations, and for the 3 years considered in this study. The
43064001
0
30
60
90
120
150
180
21020030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
27058999
0
30
60
90
120
150
180
210
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
28123001
0
30
60
90
120
150
180
210
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
O3 µµg
/m
3
O3 µg
/m
3
O3 µg
/m
3
O3 µg
/m
3
O3 µg
/m
3
O3 µg
/m
318189999
0
30
60
90
120
150
180
210
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
28133001
0
30
60
90
120
150
180
210
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
17032999
0
30
60
90
120
150
180
210
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
Model Observations
Model Observations
Model Observations Model Observations
Model Observations
Model Observations
Fig. 4. Time series showing daily averaged modelled and observed ozone for rural background stations in the period April–September 2003.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–7368
Author's personal copy
EPA’s guideline for mean normalized absolute error is 30–35% foran acceptable level of model performance. Compared with thisguideline, all the rural background stations in 2003 meet thecriteria (Table 3). When averaged overall monitoring sites, thecriteria are achieved for all the station types and for all the years, asit can be observed in Table 2. The errors in 2005 for ruralbackground stations are higher than those in 2004 and 2003, whenwe compare error values for ‘‘common stations’’ (right side of Table2). This small increase in error values could be related to the factthat emission data correspond to the 2003 EMEP emissioninventory.
Maps in Fig. 2 show the spatial distribution of bias in 2003for background stations. Bias maps reveal that highest ozoneunderestimations are located in Madrid area. In addition, weevaluated the correlation coefficient in 2003 for rural, urban andsuburban stations, as it is presented in Fig. 3. These maps
indicate that a high number of rural stations are presentinga very good correlation, with a correlation coefficient higherthan 0.7, generally located in less mountainous areas. Never-theless, there are also some urban and suburban backgroundstations presenting a lower correlation, including those locatedSouthern and Southeastern Spain. This area presents a complextopography and the model resolution is not adequate to satis-factorily represent the dominant atmospheric circulations andtransport patterns. Palau et al. (2005) have shown that ina complex-terrain coastal site, such as Mediterranean area, usingan inadequate scale to solve the meteorology can result in a verybig gap in the simulation of lower-layer pollutant behaviour aturban scales.
Time series plots for some representative sites, demonstratingpatterns of modelled and observed ozone concentrations over theentire 6-month period for 2003, are shown in Fig. 4. These pictures
Table 4Daily-maximum ozone statistical values for background rural stations
Station type Corr. Coeff. BMB (mg/m3) BMNB (%) EMNAE (%)
Ozone peaks 2003 (common stations)RB 0.67 �10.7 �6.3 14.7SB 0.74 �15.9 �10.7 15.5UB 0.65 �4.2 0.1 15.9
Ozone peaks 2004 (common stations)RB 0.60 �11.6 �6.1 16.7SB 0.59 �12.6 �7.9 16.1UB 0.55 2.0 6.9 18.2
Ozone peaks 2005 (common stations)RB 0.63 �13.5 �8.6 16.3SB 0.58 �15.1 �6.1 22.0UB 0.60 �2.5 1.8 15.9
43064001
0
40
80
120
160
200
24020030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928O
3 µµg
/m
3)
O3 µg
/m
3)
O3 µg
/m
3)
O3 µg
/m
3)
39086001
0
40
80
120
160
200
240
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
28123001
0
40
80
120
160
200
240
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
28133001
0
40
80
120
160
200
240
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
Model ObservationsModel Observations
Model Observations Model Observations
Fig. 5. Time series showing daily peaks (maximum hourly ozone value/day).
Table 5Analysis of CHIMERE model performance for NOx and NO2
Station type BMB(mg/m3)
BMNB(%)
EMNAE(%)
NS BMB(mg/m3)
BMNB(%)
EMNAE(%)
NS
NOx – 2003 NO2 – 2003RB �3.8 �15.8 52.8 20 �1.6 �3.3 52.8 19SB �21.3 �51.8 64.1 16 �11.8 �36.4 62.4 16UB �30.3 �47.6 66.6 15 �15.3 �34.9 65.3 16
NOx – 2004 NO2 – 2004RB �4.4 �24.1 53.2 20 �2.1 �11.5 60.3 19SB �24.5 �54.9 66.6 16 �13.1 �41.0 62.8 16UB �25.2 �49.5 69.5 15 �13.6 �35.0 68.4 16
NS: number of stations.Statistics were calculated for hourly values. The same set of stations was used for the2 years.Cutoff: 5 mg/m3.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–73 69
Author's personal copy
show the daily averaged ozone values for some rural backgroundstations. These stations represent an area similar to that repre-sented by the model grid point. Graphics indicate that ozone valuespredicted by the CHIMERE model are in a fair agreement withobserved values at all rural background sites. Presently, the airquality modelling group in CIEMAT is carrying out deeper studies inorder to determine the areas where the model is providing the bestand worst results and to identify the reason for the discrepancies.
An evaluation of daily ozone maximum was also carried out.Table 4 shows statistics calculated for daily ozone peaks forbackground stations (urban, suburban and rural station types).Statistics for the 3 years present acceptable values, with mean
normalized bias values between �10.7 and 6.9%. In spite of theseresults, a general underestimation is observed, especially for ruraland suburban background stations, presenting higher ozoneobserved values. Fig. 5 includes time series of observed andmodelled hourly maximum ozone values for rural backgroundstations. This figure illustrates the fact that the model tends toreproduce time variations, although highest observed values(higher than 160 mg/m3) are clearly underestimated by the model.Time series for ‘‘280133001’’ site (a rural background stationlocated in Madrid surroundings) in Fig. 5 illustrate this behaviour.Further studies are required to clarify the reasons for thisunderestimation, especially over Madrid area, but it seems to be
2003: 6-monthly mean of NOx concentrations.
Urban background stations
0
20
40
60
80
100
120
140
3014006
4902001
9018001
18140001
19130001
28006004
28092005
33044032
40194001
41004006
46131002
23050003
41038001
41091019
42173001
Station number
NO
x ( µg
/m
3)
ModelMeasurements
2003: 6-monthly mean of NOx concentrations.
Suburban background stations
0
20
40
60
80
100
120
140
2003001
19046001
11008003
21041017
4049001
28079024
28080003
37274009
39059001
39087003
45168001
46250043
44216001
48055002
29069001
37274010
Station number
NO
x ( µg
/m
3)
2003: 6-monthly mean of NOx concentrations.
Rural background stations
0
20
40
60
80
100
120
140
1055001
6016999
10182001
12099001
12141002
17032999
18189999
19061999
20016001
25224999
27058999
11008009
33036999
39086001
41059001
44054001
45153998
46263999
49149999
41088001
Station number
NO
x ( µg
/m
3)
ModelMeasurements
ModelMeasurements
Fig. 6. 2003 Six-monthly mean of NOx concentrations for all the background station types. NOx levels for urban background stations present the highest values.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–7370
Author's personal copy
related to the need of a higher resolution emission inventory. Inaddition, parameterizations for dry deposition and biogenic emis-sions for high temperature must be revised. Also, other aspects,such as contribution from forest fires in Portugal, temperatureprediction or model performance for upper levels, need to beevaluated in order to identify the model limitations. Other modelsrunning with a higher resolution (5� 5 km2) were able to fit wellthe atmospheric circulations and the photochemical pollution inthe Greater Madrid Area for 1992 and 1995 (Martin et al., 2001;Palacios et al., 2002).
Table 5 presents statistical results for hourly NOx and NO2concentrations when the observed value was higher than 5 mg/m3. The same set of stations was used for both 2003 and 2004.Chemiluminescence NO/NOx/NO2 analyzers can respond toother nitrogen containing compounds, such as peroxyacetylnitrate (PAN), which might be reduced to NO in the thermalconverter (Winer et al., 1974). Although atmospheric concen-trations of these potential interferences are generally lowrelative to NO2 (especially in urban areas) we added modelledPAN and HNO3 compounds to modelled NO2, in order tocompare model results to NO2 and NOx observations. Statisticalvalues in Table 5 indicate an underestimation of NOx and NO2,especially for urban and suburban stations. Lower errors arefound for rural background stations. According to Carter (1994),the availability of NOx in the environment is the single mostimportant factor affecting reactivity rankings. Since the reactionbetween OH radicals and NO2 is an important radical termina-tion process, NOx inhibits ozone under high NOx conditions.Fig. 6 illustrates the fact that urban and suburban backgroundstations register higher NOx levels than rural stations due to theproximity to NOx sources. This figure shows the 6-month meanof NOx concentration for 2003 and for each station. NOx levelsare significantly higher for suburban and urban stations. Thesimulations’ resolution dilutes emissions in large grid cells,resulting in an underestimation of NOx close to the sources. It isclear that higher resolution simulations must be carried out. InFig. 7, comparisons between observed and predicted NO2concentrations for some monitoring sites are presented. Modelpredictions tend to reproduce the daily NO2 variation at manyrural background stations. In order to understand why negativebias is observed for both ozone and NO2 close to Madrid city,a deeper analysis was carried out. Fig. 8 shows hourly observedand predicted concentrations for both NO2 and ozone at theurban background ‘‘28092005 Móstoles’’ site on July 30–31,2004. This station is situated about 24 km southwest of Madrid.Time series in the figure indicate that ozone underestimationoccurs during daytime, when a good agreement is foundbetween modelled and observed NO2 concentrations. Never-theless, underestimation of NO2 takes place between 4:00 and8:00 in the morning and during the evening. A deeper study isbeing conducted in order to determine the reasons for thatbehaviour, such as resolution, errors in emission temporaldistribution or problems with MM5 predicted temperature.Regarding meteorological variables, an underestimation of veryhigh temperatures is found over Madrid. Fig. 9 contains themodelled temperature at Getafe station, located 13 km south ofMadrid for July 30–31, 2004.
6. Conclusions
One of the main interests of air quality modelling lies in thepossibility of applying models for forecasting or for scenarioanalysis. Predicting air quality has important applications, such asinforming and preventing standards exceedances, protectinghuman health or designing strategy plans to reduce air qualitylevels.
To be used as a forecasting tool, a system-model must be capableof reproducing observed values in order to demonstrate that itdescribed dispersion and chemical processes in the atmosphereadequately. In this paper we evaluated the performance of theCHIMERE model, using all the modelled and observed hourly valuesavailable at all the background stations in Spain in 2003–2005, forthe period between April 1 and September 30.
Using a resolution of 0.2� � 0.2� and regridded EMEPemissions, the CHIMERE model provides ozone results in a fairagreement with observations for all the background stations and
19061999
0
5
10
15
20
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
39086001
0
5
10
15
20
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
46263999
0
5
10
15
20
20030401
20030416
20030501
20030516
20030531
20030615
20030630
20030715
20030730
20030814
20030829
20030913
20030928
NO
2 µµg
/m
3N
O2 µg
/m
3N
O2 µg
/m
3
Model Observations
Model Observations
Model Observations
Fig. 7. Time series showing daily averaged modelled and observed NO2 for ruralbackground stations in the period April–September 2003.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–73 71
Author's personal copy
for the 3 years considered and presents statistical results insidethe suggested EPA range (Tesche et al., 1990) for mean bias, meannormalized bias and mean normalized absolute error. For ruralbackground stations, best reflecting areas similar to modelresolution, statistics values in 2003 are better than those in 2004and 2005, possibly due to the use of the 2003 emissioninventory. This study reveals a significant underestimation ofozone levels over Madrid surrounding areas. This behaviourindicates that precursor transport to these areas is not being wellrepresented by the model. As emissions were derived from the50� 50 km2 resolution EMEP database, simulations which willrequire higher resolution emission data are needed to improvemodel predictions in the vicinity of a large city as Madrid. Otherpotential causes of the underestimation, such as pollutanttransport from forest fires in Portugal, dry deposition andbiogenic emissions’ parameterizations for high temperature,temperature underestimation, emissions temporal distribution ormodel prediction at upper layers, must be analysed.
The evaluating exercise for NOx using rural background stationsindicates that the model is able to reflect background processes. Forsuburban and urban background stations, higher errors arerevealed, related to the higher levels of NOx around thesemonitoring sites that cannot be represented by the model at thisresolution. Despite the limitations, the results presented in thisstudy for ozone and NO2/NOx suggest that the CHIMERE modelwith a resolution of 0.2� � 0.2� and EMEP emissions in Spain can beused to predict or reproduce rural background levels, but thata higher resolution is needed to improve the predictions forsuburban or urban background station.
References
Amann, M., Bertok, I., Cofala, J., Gyarfas, F., Heyes, C., Klimont, Z., Schöpp, W.,Winiwater, W., 2005. Clean Air For Europe (CAFE) Programme Final Report,Laxenburg, Austria.
Bessagnet, B., Hodzic, A., Vautard, R., Beekmann, M., Rouil, L., Rosset, R., 2004.Aerosol modelling with CHIMERE – first evaluation at continental scale. Atmos.Environ. 38, 2803–2817.
Baldasano, J.M., Valera, E., Jiménez, P., 2003. Air quality data from large cities. Sci.Total Environ. 307, 141–165.
Carter, W.P.L., 1994. Development of ozone reactivity scales for volatile organiccompounds. Journal Air and Waste Management 44, 881–899.
Chang, Hanna, 2004. Air quality model performance evaluation. Meteorol. Atmos.Phys. 87, 167–196.
Chin, M., Ginoux, P., Kinne, S., Holben, B.N., Duncan, B.N., Martin, R.V., Logan, J.A.,Higurashi, A., Nakajima, T., 2002. Tropospheric aerosol optical thickness fromthe GOCART model and comparisons with satellite and sunphotometermeasurements. J. Atmos. Sci. 59, 461–483.
Cuvelier, C., Thunis, P., Vautard, R., Amann, M., Bessagnet, B., Bedogni, M.,Berkowicz, R., Brandt, J., Brocheton, F., Builtjes, P., Coppalle, A., Denby, B.,Douros, G., Graf, A., Hellmuth, O., Honoré, C., Hodzic, A., Jonson, J.,Kerschbaumer, A., de Leeuw, F., Minguzzi, E., Moussiopoulos, N., Pertot, C.,Pirovano, G., Rouil, L., Schaap, M., Stern, R., Tarrason, L., Vignati, E., Volta, M.,White, L., Wind, P., Zuber, A., 2007. CityDelta: a model intercomparison toexplore the impact of emission reductions in 2010 in European cities in 2010.Atmos. Environ. 41, 189–207.
Derognat, C., Beekmann, M., Bäumle, Martin, D., 2003. Effect of biogenic volatileorganic compound emissions on tropospheric chemistry during the atmo-spheric pollution over the Paris area (ESQUIF) campaign in the Ile-de-Franceregion. J. Geophys. Res. 108 (D14), 4409–4423.
EEA, 2006. Air Pollution by Ozone in Europe in summer 2005. Overview ofExceedances of EC Ozone Threshold Values for April–September 2005. EEATechnical Report No. 3/2006.
EPA, 1991. Guideline for Regulatory Applications of the Urban Airshed Model, USEnvironmental Protection Agency Report EPA-450/4-91-013. EPA, Office of AirQuality Planning and Standards, Research Triangle Park, NC, 89 pp.
Grell, G.A., Dudhia, J., Stauffer, D.R., 1995. A Description of the Fifth-Generation PennState/NCAR Mesoscale Model (MM5). NCAR/TN-398þ STR. NCAR TechnicalNote.
Hauglustaine, D.A., Hourdin, F., Jourdain, L., Filiberti, M.-A., Walters, S.,Lamarque, J.-F., Holland, E.A., 2004. Interactive chemistry in the Laboratoirede Météorologie Dynamique general circulation model: description andbackground tropospheric chemistry evaluation. J. Geophys. Res. 109, doi:10.1029/2003JD003957.
Hodzic, A., Chepfer, H., Vautard, R., Chazette, P., Beekmann, M., Bessagnet, B.,Chatenet, B., Cuesta, J., Drobinski, P., Goloub, P., Haeffelin, M., Morille, Y., 2004.Comparisons of aerosol chemistry-transport model simulations with lidar andsun photometer observations at a site near Paris. J. Geophys. Res., doi:10.1029/2004JD004735.
Hodzic, A., Vautard, R., Bessagnet, B., Lattuati, M., Moreto, F., 2005. On the quality oflong-term urban particulate matter simulation with the CHIMERE model.Atmos. Environ. 39, 5851–5864.
Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in developmentand evaluation of environmental models. Environ. Model. Softw. 21, 602–614.
Jiménez, P., Lelieveld, J., Baldasano, J.M., 2006. Multiscale modelling of air pollutantsdynamics in the northwestern Mediterranean basin during a typical summer-time episode. J. Geophys. Res. 111, D18306, doi:10.1029/2005JD006516.
Jonson, J.E., Simpson, D., Fagerli, H., Solberg, S., 2005. Can we explain the trends inEuropean ozone levels. Atmos. Chem. Phys. 6, 51–66.
Martin, F., Crespı́, S.N., Palacios, M., 2001. Simulations of mesoscale circulations inthe center of the Iberian Peninsula for thermal low conditions. Part I: evaluationof the TVM model. J. Appl. Meteorol. 40, 880–904.
MÓSTOLES - 28092005
0
40
80
120
160
200
2004073000
2004073004
2004073008
2004073012
2004073016
2004073020
2004073100
2004073104
2004073108
2004073112
2004073116
2004073120Time
µg/m
3
g/m
3
OBSERVED OZONE
MODELLED OZONEOBSERVED NOx
MODELLED NOx
Fig. 8. Time series showing O3 and NOx hourly values at an urban background site during July 30–31, 2004.
0
5
10
15
20
25
30
35
40
0:00
3:00
6:00
9:00
12:00
15:00
18:00
21:00 0:0
03:0
06:0
09:0
012
:0015
:0018
:0021
:00
Hour
Tem
peratu
re (°C
)
MM5OBS
Fig. 9. Predicted and observed temperature at Getafe meteorological station for theperiod July 30–31, 2004.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–7372
Author's personal copy
Monteiro, A., Vautard, R., Borrego, C., Miranda, A.I., 2005. Long-term simulations ofphoto oxidant pollution over Portugal using the CHIMERE model. Atmos.Environ. 39, 3089–3101.
Palacios, M., Martin, F., Martilli, A., Kirchner, F., Clappier, A., Rodriguez, M.E., 2002.Photochemical modelling of summer ozone episodes in the Greater MadridArea. Int. J. Environ. Pollut. 18, 243–259.
Palacios, M., Martı́n, F., Aceña, B., 2004. Estimate of potentially high ozoneconcentration area in the centre of the Iberian Penı́nsula. Int. J. Environ.Pollut. 24 (1/2/3/4).
Palau, J.L., Ṕerez-Landa, G., Diéguez, J.J., Monter, C., Millán, M.M., 2005. Theimportance of meteorological scales to forecast air pollution scenarios oncoastal complex terrain. Atmos. Chem. Phys. 5, 2771–2785.
Passant, N.R., 2002. Speciation of UK Emissions of Non-methane Volatile OrganicCompounds. AEAT/ENV/R/0545, issue 1.
Pérez, C., Nickovic, S., Baldasano, J.M., Sicard, M., Rocadenbosh, F., Cachorro, V.E.,2006. A long Saharan dust event over the western Mediterranean: Lidar, sunphotometer observations and regional dust modelling. J. Geophys. Res. 111,D15214, doi:10.1029/2005JD006579.
Schmidt, H., Derognat, C., Vautard, R., Beekmann, M., 2001. A comparison ofsimulated and observed ozone mixing ratios for the summer of 1998 in westernEurope. Atmos. Environ. 35, 6277–6297.
Simpson, D., Guenther, A., Hewitt, C.N., Steinbrecher, R., 1995. Biogenic emis-sions in Europe 1. Estimates and uncertainties. J. Geophys. Res. 100 (D11),22875–22890.
Simpson, D., Winiwarter, W., Borjesson, G., et al., 1999. Inventorying emissions fromnature in Europe. J. Geophys. Res. 104, 8113–8152.
Tesche, T.W., Georgopoulos, P., Seinfeld, J.H., Cass, G., Lurmann, F.W., Roth, P.M.,1990. Improvement of Procedures for Evaluating Photochemical Models.Draft Final Report Prepared for California Air Resources Board. Radian,Sacramento, CA.
Vautard, R., Martin, D., Beekmann, M., Drobinski, P., Friedrich, R., Jaubertie, A.,Kley, D., Lattuati, M., Moral, P., Neininger, B., Theloke, J., 2003. Paris emissioninventory diagnostics from ESQUIF airborne measurements and a chemistrytransport model. J. Geophys. Res. 108 (D17), 8564–8585.
Vautard, R., Bessagnet, B., Chin, M., Menut, L., 2005. On the contribution of naturalAeolian sources to particulate matter concentrations in Europe: testinghypotheses with a modelling approach. Atmos. Environ. 39, 3291–3303.
Vautard, R., Szopa, S., Beekmann, M., Menut, L., Hauglustaine, D.A., Rouil, L.,Roemer, M., 2006. Are decadal anthropogenic emission changes in Europeconsistent with surface ozone observations? Geophys. Res. Lett. 33, L13810, doi:10.1029/2006GL026080.
Vestreng, V., Breivik, K., Adams, M., Wagener, A., Goodwin, J., Rozovskkaya, O.,Pacyna, J.M., 2005. Inventory Review 2005, Emission Data Reported to LRTAPConvention and NEC Directive, Initial review of HMs and POPs, Technical ReportMSC-W 1/2005, ISSN: 0804-2446.
Vingarzan, R., 2004. A review of surface ozone background levels and trends.Atmos. Environ. 38, 3431–3442.
Winer, A.M., Peters, J.W., Smith, J.P., Pitts, J.N., 1974. Response of commercialchemiluminescent NO–NO2 analyzers to other nitrogen-containing compounds.Environ. Sci. Technol. 8, 1118–1121.
Yu, Eder, Dennis, Chu, Schwartz, 2006. New unbiased symmetric metrics forevaluation of air quality models. Atmos. Sci. Lett. 7, 26–34.
M.G. Vivanco et al. / Environmental Modelling & Software 24 (2009) 63–73 73