Agriculture and groundwater nitrate contamination in the Seinebasin. The STICS–MODCOU modelling chain
E. Ledoux a,⁎, E. Gomez a, J.M. Monget a, C. Viavattene a, P. Viennot a,A. Ducharne d, M. Benoit b, C. Mignolet b, C. Schott b, B. Mary c
a Centre de Géosciences, ENSMP, UMR Sisyphe, Fontainebleau, Franceb INRA, Station de Recherche SAD, 662 avenue Louis Buffet, 88500 Mirecourt, France
c INRA, Unité d'Agronomie Laon-Reims-Mons, Laon, Franced Laboratoire Sisyphe, CNRS/Université Pierre et Marie Curie, Paris, France
Abstract
A software package is presented here to predict the fate of nitrogen fertilizers and the transport of nitrate from the rooting zone ofagricultural areas to surface water and groundwater in the Seine basin, taking into account the long residence times of water and nitrate inthe unsaturated and aquifer systems. Information on pedological characteristics, land use and farming practices is used to determine thespatial units to be considered. These data are converted into input data for the cropmodel STICSwhich simulates the water and nitrogenbalances in the soil–plant system with a daily time-step. A spatial application of STICS has been derived at the catchment scale whichcomputes the water and nitrate fluxes at the bottom of the rooting zone. These fluxes are integrated into a surface and groundwatercoupledmodelMODCOUwhich calculates the daily water balance in the hydrological system, the flow in the rivers and the piezometricvariations in the aquifers, using standard climatic data (rainfall, PET). The transport of nitrate and the evolution of nitrate contaminationin groundwater and to rivers is computed by themodelNEWSAM.Thismodelling chain is a valuable tool to predict the evolution of cropproductivity and nitrate contamination according to various scenarios modifying farming practices and/or climatic changes.
Keywords: Seine river basin; Hydrological modelling; Water quality; Nitrates; Agriculture; Pollution; Economical evaluation
1. Introduction
Surface and groundwater of the Seine River systemhas been progressively contaminated with nitrate(Meybeck et al., 1998). The main reason for this is the
⁎ Corresponding author. Centre de Géosciences, Ecole des Mines deParis, 35 rue St Honoré, 77305 Fontainebleau cedex, France. Tel.: +331 64 69 47 02; fax: +33 1 64 69 47 03.
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
intensification of agricultural activities. Because of theinertia of the hydrological system, due to the lowdischarge of groundwater both through unsaturated zoneand aquifer, the impact of agricultural activity on waterquality has become clear over the past thirty years.
The increasing concern about protection of waterquality has led managers to encourage the development ofsimulation tools able to mimic the water and nitrogenbehaviour and transport at the regional scale (Cabon,1993; Dupuy et al., 1997; Geng, 1989; Schnebelen, 2000).
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
2 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
In this work, an integrated model has been developed tosimulate the fate of nitrogen in soil and the nitrate transportfrom arable soils to surface water and aquifers at theregional scale, accounting for the retardation due topossible very long residence times of water and nitrate invadose zones and aquifer systems. The capacity of theintegrated model is illustrated in the case of the Seine riverbasin and more particularly the Marne basin, a sub-catchment of theSeine,where data about farming practicesis more extensive (Mignolet et al., 2001). The model isused to assess the impact of future diffuse nitrogenpollution scenarios on the groundwater resources.
2. Materials and methods
The methodology of nitrogen transport simulation inthe hydrological system is based on the coupling ofexisting models. The hydrological model MODCOU(Ledoux, 1980; Ledoux et al., 1984, 1989) calculates thedaily water balance in the hydrological system, the flowin rivers and the piezometric variations in aquifers,using standard climatic data (rainfall, PET). This modelis coupled to the NEWSAM model adapted to computethe transport of nitrate in the unsaturated and saturatedzones and to the rivers. The agronomic model STICS(Brisson et al., 1998, 2002, 2003) simulates thefunctioning of the soil–plant system with a daily time-step, taking into account the growth of crops and thenitrogen and water balances.
2.1. Water fluxes
MODCOU is a regional spatially distributed model,which describes surface and groundwater flow at a dailytime step. It uses a conceptual reservoir-based approachin every cell of the surface layer to partition precipitationinto evapo-transpiration, surface runoff (resulting fromoverland flow and interflow) and infiltration. Infiltrationrecharges the groundwater that can consist of a singleor multi-layered aquifer. A delay is imposed betweensurface infiltration and aquifer recharge using a cascadeof equal linear reservoirs through the unsaturated zone(Nash and Sutcliffe, 1970). The recharge flow con-tributes to the dynamics of groundwater, given in eachaquifer layer by a finite-difference solution of the two-dimensional diffusivity equation. The resulting ground-water head is dynamically coupled to the water level insurface “river” cells. This defines the baseflow compo-nent of river-flow which also includes surface runoffand is routed through the drainage network with transfertimes that depend on topography and a basin-wideparameter, the concentration time.
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
This model has successfully predicted surface andgroundwater flow in many basins of varying scales andhydrogeological setting. In particular, it was used toassess the hydrological impact of climate change on theFrench part of the Rhône basin (87,000 km2; Etcheverset al., 2002).
The area studied comprises the whole Seine basinincluding the hydraulic boundaries of the aquifers whichare represented by the rivers of the adjacent basins(Fig. 1). Its area is 95,560 km2 (Gomez, 2002; Gomezet al., 2002, 2003). This superficial domain is discretizedby a progressive multi-scale grid of embedded squaremeshes, the size of each cell varying from 1 to 8 km; thetotal number of cells is 35,198. The groundwater domainis modelled by a progressive multi-scale grid with amulti-layer structure which represents the three majoraquifer systems encountered in the centre of the basin(sands and limestone of Oligocene, sands and lime-stones of Eocene, and Cretaceous chalk). Deep aquifers,which outcrop at the eastern border of the basin, are notrepresented by the spatially-distributed model and aretreated by a simpler procedure with a single reservoirmodel.
The spatially-distributed structure of the hydrologicalmodel is made of 57,930 cells. It accounts for theheterogeneity of soils encountered on the whole Seinebasin. A meteorological database (precipitation andPET) with a daily time step and a spatial resolution of8 km×8 km has been derived by Météo-France on the1971–2005 record. The model MODCOU has beencalibrated over the period 1985–2004 using a set ofabout 120 hydrometric stations and more than a hundredpiezometres distributed among the three aquifersconsidered. Figs. 2 and 3 show results of simulation ofthe flow in the river Seine in Paris and of a piezometrictime series in the chalky aquifer.
2.2. Nitrogen fluxes
STICS is a plot-scale crop model that simulates theimpact of climate, soil and crop management on bothplant production and environment. Water, carbon andnitrogen budgets are calculated dynamically within thesoil–crop system, at a daily time step. The crop ischaracterised by its aerial biomass (carbon and nitro-gen), leaf area index (LAI), biomass (carbon andnitrogen) of harvested crop organs and root lengthprofile. Phenomenology is driven by the climaticconditions prevailing within the canopy (temperatureand photoperiod). Growth is driven by photosyntheti-cally active radiation, intercepted as a function of LAI,and converted into biomass increase, partly directed
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 1. Conceptualisation of the hydrological system of the Seine river basin. Extension of surface and groundwater layers (Oligocene in dark grey,Eocene in medium grey, Cretaceous chalk in light grey).
3E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
towards the harvested organs for yield calculation.Water and nitrogen stresses act as limiting factors of themain physiological functions, i.e. LAI dynamics andgrowth.
The lower limit of the system corresponds to the soil/subsoil interface, through which the water and nitrogenfluxes toward the aquifers (also called drainage andnitrogen leaching) are calculated. The soil is describedas a succession of horizontal layers, each characterisedby its content of water, mineral and organic nitrogen.The related fluxes between layers are simulated by acombination of tipping buckets and convection–disper-sion formalisations, and roots ensure the interactionsbetween the soil and the crop. In the ploughed layer,carbon and nitrogen dynamics are also controlled by thebiological transformations of the various organic matterpools (crop residues, organic fertilizers, microbial bio-mass and humus).
The upper limit of the system is the atmosphere,characterised by standard daily climatic variables (solarradiation, daily minimum and maximum temperature,precipitation, PET). The soil–plant functioning dependson crop management, through the amounts of applied
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
fertilizers and irrigation water, as well as the technicalcalendars.
This model was mainly validated at the plot scale.Regarding crop production, a detailed validation wasconducted for wheat and maize crops (Brisson et al.,2002) and the model showed good performances forother crops over various soil types (Nicoullaud et al.,2004). The performances of STICS to simulate organicmatter mineralisation and nitrate leaching were alsodemonstrated at the plot scale under various farmingpractices (Schnebelen et al., 2004), and over a largerange of soil humidity and temperature.
The nitrate leaching flux computed by STICS in eachcell is diluted by the surface runoff and the infiltrationsimulated by MODCOU. The coupled fluxes aretransmitted to the aquifer system through the unsaturat-ed zone, after a delay modulated by the depth of thewater table. The evolution of nitrate concentrations inthe groundwater and the return flow to the rivers is ruledby the convective transport equation and solved using afinite difference method in the above three aquiferlayers. The transport of nitrate in the saturated zone isassumed to be conservative — most nitrogen cycle
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 2. Comparison between observed and computed flow of the river Seine at Paris along the period of time from 1999 to 2004.
4 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
processes take place within the root-zone (Brisson et al.,2003). Denitrification can also occur close to the river inthe riparian zone; this aspect has been ignored at thattime of the study as the objective is the simulation ofnitrate concentrations in the body of the aquifer system.
The coupling procedure between STICS and MOD-COU could lead to some inconsistencies as the twomodels compute independent water fluxes to thegroundwater system from the same meteorological
Fig. 3. Comparison between simulated and observed hydraulic heads
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
data. To avoid this, the water fluxes were calibrated inMODCOU, by fitting the computed river flows andpiezometric levels to field data. Then, the water fluxessimulated by STICS was adjusted to the ones simulatedby MODCOU, by slightly tuning the water balanceparameters of STICS when necessary (Gomez, 2002).
The application of STICS model at the whole Seinebasin scale requires defining “spatial simulations units”homogeneous regarding their pedological,meteorological
for the Cretaceous chalk aquifer during the 1985–2004 period.
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
5E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
and agricultural characteristics. For that purpose, threeprincipal databases have been established at the regionalscale:
A soil database. The study zone has been split intoseveral classes of soil, spatially defined as “Soil MapUnits” (SMU), on the basis of a mapping andpedological analysis available at 1/1,000,000 scale(King et al., 1995). These classes of soil arecharacterised by a distribution of soil types, called“Soil Type Units” (STU), based on their rate of areawithin a given SMU. A pedological data base isassociated with each STU.An agricultural database. Agricultural activitieshave been recorded over a 30 years period (1970–2000) using official agricultural surveys and inqui-ries (Mignolet et al., 2001; this volume). Statisticaltreatments enable “Small Agricultural Districts”(SAD) which are characterised by a homogeneousfarming production system to be defined (Thisse,1997). Each SAD has been described by a spatialdistribution of crop successions determined inaccordance with the area of each crop. Each cropsuccession can be composed of a variable number ofcrops. A data base concerning farming practices hasbeen associated to each crop composing successions.The change in farming practices over time has beentaken into account by considering varied successionsover different periods of time.A meteorological database. Geographical zones havebeen defined by Météo-France on a square grid of8×8 km2 in which the temporal evolution of meteo-rological variables is assumed to be homogeneous.
Associating these three data bases within a commonGIS interface leads to define 11,610 spatial units.Several spatial units have the same meteorological,pedological and agricultural characteristics and theseunits are grouped together to leave 7903 GeneralSimulation Units (GSU). STICS simulations were thenrun on each GSU.
2.3. Modelling nitrogen production and nitrate trans-port at the regional scale
2.3.1. Spatialised application of STICS modelThe STICS model has been designed to simulate
crops and their water and nitrogen balances at the scaleof the agricultural field. It has also been used on zonesof several hundred or thousand hectares, using GIS(Beaudoin et al., 2002; Mary et al., 1996; Mary andLaurent, 2002; Sauboua, 2001; Schnebelen, 2000). In
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
our study, STICS has not been coupled to a GIS but hasbeen integrated into a computing structure allowing itsuse in a spatial way.
The spatialization procedure is the following. For eachSmall Agricultural Districts (SAD), land use changes aredescribed by a succession of time periods with constantagricultural practices. Each time period is characterisedby an association of crop successions. This ensures atemporal continuity, during the simulation, between twotime periods. Moreover, the spatial procedure has toexplore all the pedological and agricultural combinations,because of the low resolution of databases that does notallow the localisation of these parameters inside the GSU.For example we consider the case of one SMU with twosoil type units (STU1, STU2), and one SAD with oneagricultural succession composed of two crops for the firstperiod (succP1) and three crops for the second one(succP2). A crop succession is defined as an orderedseries of n crops. Yet, at a given time, crops composing thesuccession are present simultaneously on the sameagricultural zone. To take into account this temporalvariability, the first simulated crop is alternatively one ofthe n crops of the succession. Consequently, the resultingnitrate leaching rate is the average of the n fluxessimulated for a given crop succession. This operation isapplied to all soil units (STU) encountered in the generalsimulation unit (GSU). At the end of the first period, twosituations are possible for STU1 depending whether thesoil is in the state after the first crop or in the state after thesecond. We obtain the average soil state of STU1 at theend of the first period by taking the average of these twostates. This method is also applied to STU2. These soilstates are used as initial data for simulating the secondperiod. Following this procedure, the simulation of thesecond period begins with one of the 3 crops composingthe crop succession succP2.
The advantage of this procedure is to preserve in thesimulation a differentiation among soil types, taking intoaccount their evolution in time. This method allowsinitialising automatically the agronomic state of the soilat the junction between two periods. In spite of the lowresolution of database due to the large working scale, themethod enables to keep the main spatial characteristicsby preserving the diversity of soil type and cropsuccessions in each GSU (12 km2 in average).
The simulation results, generated in parallel alongtime, were combined accounting for the proportion ofsoil types and crop successions present within a squaremesh of the superficial layer of MODCOU, andaccounting for variation in the starting crop.
The organisation chart of the spatial procedure isshown in Fig. 4. STICS is run inside a general computer
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 4. Functional diagram of the STICS–MODCOU–NEWSAM modelling instrument.
6 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
code, where it is considered as a subroutine, whichautomatically generates input data.
Databases come from several softwares (data basemanager, GIS) and are stored under text or binaryformat. Output data are distributed on the surface layergrid to allow their mapping. These output data are, onone hand, the nitrogen flux density (kg N ha−1) and thedrainage of water under the root-zone (mm). On theother hand, crop yield (t ha−1) is also obtained for eachcrop and each agricultural succession. This structureallows integration of new meteorological and agricul-tural data bases when required.
2.3.2. Nitrogen transfer towards aquifersNEWSAM uses the same grid than MODCOU and
computes the convective transport of nitrate in thesaturated zone and the variations of nitrate contamina-tion in groundwater. Due to the embedded grid system,the communication between MODCOU and NEWSAMis easy by transferring the infiltration computed byMODCOU to NEWSAM through NONSAT whichintroduces a delay using the cascade of equal linear
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
reservoirs. When a surface cell is not connected to thegroundwater system, which happens along the eastborder of the Seine Basin, nitrogen is transferred to asingle reservoir with an exponential discharge to thenearest river.
The STICS–MODCOU–NEWSAM model chainhas been developed to simulate the impact ongroundwater resources of nitrogen diffuse pollutionoriginating from agricultural practices extending overthe whole Seine river basin. Following the EuropeanWater Framework Directive (WFD) guidelines and inconcordance with recommendations of the AESN basinAgency (“Agence de l'Eau Seine Normandie”, inFrench), the evolution of the overall quality of ground-water resources has been assessed for an agriculture andland use baseline scenario up to the 2015 horizon. Forthe definition of the spatial variability of agronomicinput, the watershed has been divided into 12 areas
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
7E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
known for their homogeneity in farm practices (Fig. 5).This segmentation results from the overall aggregationof 150 initial regional units (SAD) available in theagricultural data base.
A calibration period of the model has been firstconsidered from 1970 to 1990. Then a simulation exer-cise was carried out, based on the repetition over severaldecades of the agriculture practices observed during the1990–2000 decade and the climate conditions of the1970–1990 decades.
3.1. Model calibration
Model calibration has been conducted over the1970–2000 decades using existing nitrogen concentra-tion measurements. It is based on an innovativestatistical approach using the Chi2 distribution proba-bility distance (Monget et al., 2004) performed on theprobability distribution of measured and computedvalues.
Fig. 5. Homogeneous agricultural zones of the Seine River basin used forpedological data results in 12 areas identified by levels of grey on the right
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
3.1.1. The trends in groundwater nitrate concentrationdata in the Seine basin
The French Ministry of Environment created in 1984a national repository for groundwater quality monitoringcalled the “Observatoire National de la Qualité des EauxSouterraines” (ONQES). This database has been main-tained by the “Service Géologique National” (NationalGeological Survey) inside the “Bureau de RecherchesGéologiques et Minières” (BRGM). It is a composite ofdifferent data sources which were set up either locally,regionally or at the scale of whole river basins. It isessentially recording quality measurements made fromfresh groundwater wells. Management basin organiza-tions like the AESN have followed the same scheme fortheir own surveillance network database.
Over the geographic area covered by the Seine riverbasin, data from 6500 water boreholes concerning theOligocene, Eocene and Chalk aquifers, are stored in theONQES database for a surveying period ranging from1972 to 1995. It was later augmented by the SISE–
the spatial application of STICS. The aggregation of agricultural andpart of the figure.
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
8 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
EAUX (DDASS) and ADES data bases for the period1996–2003. For the ONQES database alone, themedian, first and third quartiles of the nitrate concen-tration statistical distribution were computed for a subsetof 4442 wells for which uninterrupted yearly series wereavailable from 1975 until 1988 (Fig. 6). These curvesdisplay a strikingly linear joint increase for the threeindices. This sustained increase rate is estimated to0.64 mg NO3/L/year. The standard deviation of theannual distribution is about 10 mg NO3/L and is alsostable over the period analyzed.
This overall statistical evolution is not necessarilyrepresentative of the evolution of nitrate concentrationin a specific water borehole. Several particular freshwater wells (Chalk aquifer: Strabel et al. (1989); Brielimestone: Roberts and Marsh (1987); Champignylimestone: Poitevin (1997)) display much higherincrease rates than the ones computed from the medianevolution.
In 2004 it was possible to complete the ONQES datawith those coming from SISE–EAUX (DDASS) andADES data bases for the period 1996–2003.
3.1.2. Statistical calibration of computed nitrateconcentrations
In order to reduce the effect of erratic naturalvariations of the median of measured concentrations ineach aquifer, a mean square linear adjustment of 3297wells selected from the total data base because their
Fig. 6. Observed evolution of nitrate concentration in water wel
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
measurements series spanned the 1972–2003 timeinterval (300 in Oligocene, 1118 in Eocene and 1879in chalk). The adjusted series have been normalized overthe 1970–2000 period. The median computed fromthese adjusted values displays smoother time behaviourthan that obtained from the original measurements thusbeing more adapted to a direct comparison with themedian coming from simulated values produced by themodel. Thus, calibration of the model was done usingthe evolution of the median of the least-square adjustedlinear nitrate concentration series.
One of the major problems with the calibration of themodel is that the initial state of concentrations inaquifers is unknown because the simulation had to startin 1970, the oldest date for which it was possible toacquire the required data about nitrogen inputs andagricultural practices. To avoid this problem, the medianchronology was adjusted over the 30 years for themeasured concentrations within each aquifer using atechnique of mean square over sliding intervals, in orderto determine the simulated period of 30 years showingthe most similar data among a computed chronology of200 years chosen to definitely overlap the observedsituation. For each aquifer, once the best fit obtained, theadjustment was refined by varying the nitrogen fluxdelivered by STICS with a multiplicative coefficient.The final result (shown for the Oligocene aquifer onFig. 7) is acceptable, as the standard deviation of errorsdoes not exceed 1 mg NO3/L. However, the method of
ls over the whole aquifer system of the Seine River basin.
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 7. Temporal adjustment of STICS–MODCOU results on observed nitrate concentration values for Oligocene aquifer.
9E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
adjustment allows a very good estimation of theevolution of median concentrations for each aquiferwithout any guarantee on its reliability at the local scaleof a given bore-hole. The correction factors are appliedto concentrations at the basis of the root-zone beforethey are routed to the groundwater system. In that way,they act as a corrector of the nitrogen input which israther uncertain. To be rigorous, this correction shouldhave been applied to the inputs of STICS, but thathappened not possible considering the very long com-puter time needed by the spatial procedure with STICSat that state of the study. The correction factors weredivided by 1.7 for the Oligocene and 2 for the Eocenewith multiplication by 1.3 for the chalk.
3.1.3. Evaluation of errors after calibration of STICS–MODCOU
The objective of the temporal adjustment of themodel is to make available a prediction tool for nitrateconcentrations in aquifers, taking account the incertitudeassociated to the data. It is therefore important toevaluate with more accuracy the statistics of differencesbetween observed and simulated values. These devia-tions are calculated for each well between the yearlyaverage of measured concentrations and the simulatedaverage as it comes from MODCOU–STICS at the cellcontaining the well. The histogram of deviations for thebore-holes belonging to Oligocene shows a nearlygaussian distribution (without bias estimator) with astandard deviation of 28 mg/L.
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
3.2. Outputs of the model, example of the Marne river
The outputs of the STICS–MODCOU model chainare nitrate flux and concentrations at the basis of the root-zone, nitrate concentrations in the aquifers and nitrateflux coming to the rivers. The complete evaluation ofSTICSmodel is difficult due to both lack of experimentaldata and to the different scales of measurement andsimulation. Observations made in soils at the 1 m2 levelcannot be easily compared to simulations made at 1 km2
level. For this reason, the model has been mainly eval-uated on the basis of crop yields at the scale of smallagricultural districts (SAD). Reasonable agreement wasobtained between simulated and observed yields(expressed in t/ha) for the main crops encountered inthe Marne basin, where agricultural data were the morecompletely available at this time of the study: winterbarley (simulated: 5.2, observed: 5.5), winter wheat(sim: 7.3, obs: 6.4), oilseed rape (sim: 4.1, obs: 3), maize(sim:7.9, obs: 7.7) and spring pea (sim: 4.6, obs: 5.2).
The mean N fluxes calculated in the hydrologicalcompartments of the Marne basin during the last30 years are given in Fig. 8. Nitrate concentration inthe hydrological system has been initialised at 0. At thebeginning of the simulation, a time delay of about3 years is observed between the nitrogen flux below therooting zone and below the non-saturated zone.Nitrogen fluxes then increased until 1982–1983. Thatmeans that the non-saturated zone of the Marne basinstored nitrogen during the first eleven years. The major
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 8. Evolution of nitrogen fluxes in the Marne basin: fluxes below the rooting zone, fluxes below the non-saturated zone, fluxes exchanged with theriver (drainage) and stored in aquifers (storage).
10 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
part of the Marne basin is therefore occupied by a non-saturated zone in which the time delay is smaller than orequal to 11 years. The nitrogen flux drained by riversfollowed the evolution of the output flux of the non-saturated zone. Finally, it appears that aquifers havebeen storing nitrogen during the 1971–1999 period.
4. Assessing the impact of future nitrogen diffusedpollution scenarios on the groundwater resources ofthe Seine river basin
4.1. Groundwater and the EU WFD
Within the European WFD guidelines, baselinescenarios are meant as “business as usual” hypotheses(Commission of the European Community, 2003) usedin a forward extrapolation of present pressures on theenvironment as well as management practices. Theirreal purpose is to conduct a risk analysis of non-compliance and evaluate the probability of economicalternatives to reach the “good water status” forparticular water bodies. In this context, the productionof concentration maps at given time horizons (here,2015) is seldom the appropriate end product for decisionmakers. In risk management studies it is often preferableto manipulate entities to which probabilities can beattached and eventually combined with monetaryconcepts such the Net Present Value (NPV) (Pollio,1999) or the various economic values attached to theenvironment (Pearson, 2000). In order to fulfill these
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
demands, uncertainty probability modeling of the typementioned in this paper enables to produce thresholdprobability maps displaying the probability that a certainregulatory index level be exceeded. Fig. 9 provides suchan example, where STICS–MODCOU results at the2015 horizon have been used in order to evaluate theprobability of exceeding a 50 mg NO3/L regulatorylevel in the water wells drawing water out of theOligocene, Eocene and Chalk aquifers of the Seine riverbasin.
4.2. Evaluation of basin wide nitrate reductionmeasures and economic impact
The WFD requires the achievement of good ground-water status and to that end provides for the monitoringof groundwater bodies as well as measures to protect andrestore groundwater quality. But does it pay for society toreduce groundwater nitrate concentrations by investingin programs that result in increased adoption of bestmanagement practices (Yadav and Wall, 1998)? In orderto answer this question the STICS–MODCOU modelchain was used to compare different policy alternativesfrom an economic perspective.
4.2.1. Scenarios including groundwater protectionmeasure
In France, groundwater protection measures aremainly based on the control of fertilizers input and theuse of catch crops (intermediary culture acting as a
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 9. Per municipality probability map of exceeding the level of 50 mg NO3/L in groundwater wells for the aquifers of the Seine river basin at the2015 horizon.
11E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
nitrate trap, Culture Intermédiaire Piège à Nitrates(CIPAN). New agricultural scenarios including these twopractices have been simulated with the STICS model asan input to the MODCOU–NEWSAMmodel in order toevaluate their effect at the regional level. The “businessas usual” hypothesis (no changes in agriculturalpractices, scenario A) has been compared with threechanges in agricultural practices applied after 2005— a
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
reduction of 20% of N-input and use of “CIPAN”(scenario B); a reduction of 20% of N-input (scenario C);sole use of “CIPAN” (scenario D). A fifth hypothesis(scenario E) consists in a complete ban of any agri-cultural activities after 2005 which in fact consists of astop in new N-inputs.
The Fig. 10 shows the evolution of the calculatedmedian of nitrates concentration under the different
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
Fig. 10. Evolution of the median of STICS–MODCOU simulated nitrates concentration in the Oligocene aquifer of the Seine river basin for variousscenarios of modification of agricultural practices. A: pursuit of the present practices; B: reduction by 20% of N fertilization rate and introduction ofcatch crops; C: reduction by 20% of N fertilization rate; D: introduction of catch crops; E: complete stop of agricultural activity.
12 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
scenarios for the Oligocene aquifer of the Seine riverbasin. On a short-time horizon, effects of preventivemeasures have little impact on the nitrates concentrationevolution at the basin scale. This is especially the casefor deeper parts of the Chalk aquifer for which no trendreversal can be observed but merely a stabilization of thepresent level. Of all preventive measures, the use ofcatch crops seems to be a more effective measure thanto reduce N-inputs.
4.2.2. How to estimate the benefits that the differentpolicy will bring to different stakeholders?
A direct approach to measure the economic value ofgroundwater quality is to consider the costs that usershave to bear if groundwater quality deteriorates. The
Fig. 11. Distribution of the probability of exceeding 50 mg NO3/L for grounriver basin in 2015 for two scenario of future farming practices. A: pursuit of thof catch crops (CIPAN).
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
underlying idea is that these costs would no longer haveto be paid if groundwater quality could be restored. Inthis sense, the benefits of groundwater protection takethe form of avoided damage costs (Görlach andInterwies, 2003).
Groundwater contamination generates different typesof damage such as the costs for water users who have toundertake averting or corrective actions or the cost ofecological damage and loss of recreational value, whengroundwater contamination has an impact on surfaceecosystems (rivers, wetland, coastal zones) (Rinaudoet al., 2004). Our study focuses on the estimation of thedamage associated with domestic drinking water supply.In this case the estimated benefits can be interpreted aslower bound estimates of the value of groundwater
dwater of municipalities overlaying the Oligocene aquifer of the Seinee present agricultural practices, (Business as usual) and D: introduction
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
13E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
protection. This approach has already been applied inthe case of contamination of aquifer by toxic materials(Raucher, 1983). The author explained that the expectedsocial benefits of protection are defined by the change inthe expected damage E(D) associated with the contam-ination. In the case of the contamination by nitrates ofwells for domestic water supply, the concept of Rauchercan be adapted in a simple way so that the expecteddamage will be expressed by:
EðDÞ ¼ p⁎Cr ð1Þ
where p is the probability that contamination will occur(0≤p≤1) and Cr is the expense of the most econom-ically efficient response to the contamination incident(Cr≥0). The p value is computed from the results of ourmodel after an aggregation at municipality's level. Itrepresents the risk of exceeding the 50 mg NO3/Lconcentration limit in 2015. Fig. 11 shows the distribu-tion of probabilities calculated per municipalities for thescenarios “business as usual” and CIPAN. For Oligoceneand Eocene aquifers, probabilities decrease with thenumber of municipalities, most of them being under the0.5 threshold. For the Chalk aquifer the distribution ismore contrasted with higher values assigned to highprobability classes (0.9–1). This strong risk of exceedingthe 50 mg NO3/L concentration concerns mostly theNorth-East part of the Seine basin. It may be due to thecombination of two effects at local scale; the diffusenitrogen pollution is very high and the chalk aquifer issuperficial with a relative thin non-saturated zone,making it particularly vulnerable to this type of pollution.The different scenarios still have little impact atmunicipal scale for the 2015 horizon.
In order to compute the expected damage for eachlocal groundwater managing unit (also called “Unité deGestion et d'Exploitation or UGE”, in France) theassumption was made that, for each borehole of theONQES database, a treatment would be set up in case ofnitrates concentration exceeding 50 mg/L. For eachborehole the corresponding Cr term (see Eq. (1), above)is equal to:
Cr ¼ Cost of treatment plant
⁎ number of water consumers
⁎ water consumption per habitant:
The model is based on the following values: a cost ofdrinking water treatment of 0.20 €/m3; water consump-tion per inhabitant of 150 L per person per day; anumber of water consumers in 2015 based on a
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
population estimated by an extrapolation of the 1998-population census (population of the each municipalitieslinked to the particular “UGE”) using a populationgrowth (source: INSEE census) calculated at the “bas-sin de vie” scale for the period 1990–1999 (Source:INSEE). The “bassin de vie” is a group of municipal-ities representing the smallest territory where peoplehave access to equipment and employment. The Seineriver basin counts 341 “bassin de vie” for 8118municipalities. The model does not however take inaccount that treated water can be mixed with raw waterand the loss of water in the network is neglected.
The following expected damages have been com-puted following the previous method for each of 1684“UGE” of the Seine river basin (6.5 Millions inhabi-tants) outside Paris Urban Area. In the case of the“business as usual” scenario the total cost is estimated to21 M€ with an average cost of 5.7 k€ for one UGE.These costs are reduced to respectively 20.3 and 5.5when considering a “catch-crop” scenario.
5. Conclusion
The purpose of this research was to show the abilityto assess nitrate evolution of groundwater at the basinwide scale. The results of simulation compared toobserved data, available over a period of time of nearlythirty years, indicate hopefully that the approachcoupling a spatially distributed hydrological model toa plot scale crop-model is suitable. One of the interestsof the integrated model comes from the fact that eachcomponent has been tested independently which conferssome validity to their assembly. Moreover, the spatialdiscretization allows an evolution of the modelling asthe data base becomes enriched or modified. This bringssome flexibility to the model which makes it well suitedfor the integration of future date and for prospectivestudies. In this paper, it has been used to quantify theevolution of nitrate contamination as a consequence ofmodifications of farming practices. In another paper(Ducharne et al., 2006-this issue) it has also been used toevaluate the effect of climatic changes on hydrologicalconditions and nitrate concentration.
One must bear in mind that this kind of spatialisedmodel requires a large amount of information. Therepresentativeness of databases established at such alarge spatial scale is questionable. As a consequence, alarge uncertainty is attached to the simulation results; itis the reason why we have made an attempt to quantify itby calibrating the model on a statistical way and byexpressing the results in terms of probability oftransgression of concentration thresholds.
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
14 E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
Taking into account this uncertainty, one of the majorconclusions of the prospective studies exploring variousscenarios of changes in agricultural practices, is to pointout the low sensibility of nitrate contamination inaquifers at the 2015 horizon, due the long transfer timethrough unsaturated zone and aquifers.
Acknowledgements
This work was supported by the programme PIREN-Seine and by the programme Gestion et Impact duChangement Climatique of the Ministère de l'Ecologieet du Développement Durable. We are grateful for thedata provided by Météo-France (actual meteorologicaldata from the SAFRAN database), by the Agence del'Eau Seine-Normandie and by the Direction Régionalede l'Environnement Ile-de-France regarding pointsource pollution.
References
Beaudoin N, Coquet Y, Mary B. Estimation des pertes de nitrates et depesticides en zone de grande culture. Etude à l'échelle du bassinhydrologique de Bruyères-et-Montberault Compte rendu finalADEME n° 9701001; 2002. 68 pages.
Brisson N, Mary B, Ripoche D, Jeuffroy M, Ruget F, Nicoullaud B,et al. STICS: a generic model for the simulation of crops and theirwater and nitrogen balances. 1-theory and parametrizationapplied to wheat and corn. Agronomie 1998;18: 311–46.
Brisson N, Ruget F, Gate P, Lorgeou J, Nicoullaud B, Tayot X, et al.STICS: a generic model for simulating crops and their water andnitrogen balance. II. Model validation for wheat and maize.Agronomie 2002;22: 69–92.
Brisson N, Gary C, Justes E, Roche R, Mary B, Ripoche D, et al. Anoverview of the crop model STICS. Eur J Agron 2003;18: 309–32.
Cabon, F., Modélisation du cycle de l'azote dans le système sol-eau-plante du lysimètre au bassin hydrologique. Ph.D. thesis,Université Pierre et Marie Curie, 1993.
Ducharne A, Baubion C, Beaudoin N, Benoit M, Billen G,Brisson N, et al. Long term prospective of the Seine Riversystem: Confronting climatic and direct anthropogenic changes.2006-this issue. doi:10.1016/j.scitotenv.2006.12.011.
Dupuy A, Razack M, Banton O. Contamination nitratée deseaux souterraines d'un basin versant agricole hétérogène. 1:evaluation des apports à la nappe (modèle Agriflux). Rev Sci Eau1997;1: 23–40.
Etchevers P, Golaz C, Habets F, Noilhan J. Impact of a climate changeon the Rhone River catchment hydrology. J Geophys Res 2002;107(D16): ACL6 1-ACL6 18.
Geng, Q.Z., Modélisation conjointe du cycle de l'eau et du transfertdes nitrates dans un système hydrologique. Ph.D. thesis, Ecole desMines, 1989.
Gomez, E. Modélisation intégrée du transfert de nitrate à l'échellerégionale dans un système hydrologique. Application au bassin dela Seine. Thèse de Doctorat de l'Ecole Nationale Supérieure desMines de Paris, France, 2002, 287 pp.
Gomez E, Mignolet C, Mary B, Schott C, Brunstein D, Bornerand C,et al. Dynamique agricole et pollution nitrique diffuse: Mod-
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
élisation intégrée du transfert des nitrates sur le bassin de laSeine. Rapport de synthèse du Programme PIREN-Seine 1998–2001, UMR CNRS 7619 Sisyphe Paris Jussieu; 2002.
Gomez E, Ledoux E, Viennot P, Mignolet C, Benoit M, Bornerand C,et al. Un outil de modélisation intégrée du transfert des nitrates surun système hydrologique: application au bassin de la Seine. LaHouille Blanche; 2003 . p. 38–45. n°3/2003.
Görlach B, Interwies E. Economic assessment of groundwaterprotection: a survey of the literature. Berlin: Ecologic; 2003. 64 pp.
King D, Le Bas C, Jamagne M, Daroussin HR, J. Base de donnéesgéographique des sols de France à l'échelle du 1/1000000. Noticegénérale d'utitlisation. Rapport technique, INRA. Service d'étudedes sols et de la carte pédologique de France; 1995.
Ledoux E, Girard G, Villeneuve JP. Proposition d'un modèle couplépour la simulation conjointe des écoulements de surface et desécoulements souterrains sur un bassin hydrologique. La HouilleBlanche; 1984 . p. 101–10.
Ledoux, E., Modélisation intégrée des écoulements de surface et desécoulements souterrains sur un bassin hydrologique. Ph.D. thesis,Ecole des Mines, 1980.
Ledoux E, Girard G, deMarsily G, Villeneuve JP, Deschenes J. Spatiallydistributed modeling: conceptual approach, coupling surface waterand groundwater. In: Morel-Seytoux HJ, editor. Unsaturated flow inhydrologic modeling — theory and practice. NATO ASI Ser.CNorwell, Massachussett: Kluwer Academic; 1989 . p. 435–54.
Mary B, Beaudoin N, Benoit M. Prévention de la pollution nitrique àl'échelle du bassin d'alimentation en eau. In: Lemaire G,Nicolardot N, editors. Maîtrise de l'azote dans les agrosystèmes,Reims, 19–20/10/1996. Paris, INRA Editions, vol. 83. Collectionles colloques; 1997 . p. 289–312.
Mary B, Laurent F. La gestion durable de la fertilisation azotée.Colloque IIRB. Bruxelles; 2002 . p. 59–67.
Meybeck M, De Marsily G, Fustec E. La Seine en son bassin.Fonctionnement écologique d'un système fluvial anthropisé. Paris:Elsevier; 1998. 749 pages.
Mignolet C, Bornerand C, Benoit M. Dynamique spatiale et temporellede l'activité agricole dans le basin de la Seine au cours des trentedernières années. C R Acad Agri 2001;87: 99-109.
Monget, J.-M., C., Viavattene, P., Viennot., (2004) Simulation aumoyen du logiciel STICS–MODCOU des pollutions azotées surl'Oligocène du Bassin Parisien. Mise en oeuvre et confrontationavec les données de terrain. Rapport d'activité de l'année 2003.Programme Piren-Seine. (Unpublished report).
Nash JE, Sutcliffe JV. River flow forecasting through conceptualmodels, 1. A discussion of principles. J Hydrol 1970;10: 282–90.
Nicoullaud B, Couturier A, Beaudoin N, Parnaudeau V, Mary B,Coutadeur C, et al. Modélisation spatiale à l'échelle parcellaire deseffets de la variabilité des sols et des pratiques culturales sur lapollution nitrique agricole. In: Monestiez P, Lardon S, Seguin B,editors. Organisation spatiale des activités agricoles et processusenvironnementaux, INRA Editions; 2004 . p. 143–61.
Pearson CS. Economics and the global environment. CambridgeUniversity Press; 2000. 583 pp.
Poitevin J. Les contrats de nappes : une nouvelle approche de lagestion des eaux souterraines pour un développement durable.Paris: Institut d'Aménagement et d'Urbanisme de la Région Ile-de-France; 1997.
Pollio G. International project analysis and financing. Mac MillanPress; 1999. 235 pp.
Proposal for aDirective of theEuropeanParliament and of theCouncil onthe protection of groundwater against pollution. Draft, COM (2003),Brussels, Commission of the European Communities, 21 pp.
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
15E. Ledoux et al. / Science of the Total Environment xx (2007) xxx–xxx
ARTICLE IN PRESS
Raucher RL. A conceptual framework for measuring the benefits ofgroundwater protection. Water Resour Res 1983;19: 320–6 [n°2].
Rinaudo J-D, Arnal C, Blanchin R, Elsass P, Meilhac A, Loubier S. Thecost of groundwater diffuse pollution in the Upper Rhine valley.Conference on nutrient management — aquatec; 2004. 11 pp.
Roberts G, Marsh T. The effects of agricultural practices on the nitrateconcentrations in the surface water domestic supply sources ofWestern Europe. Institute for Agronomical and HydrologicalStudies Publication, 164; 1987 . p. 365–80.
Sauboua E., Modélisation stochastique fonctionnelle du transfert d'eauet d'azote sous culture de maïs. Application à l'évaluation del'impact des pratiques agricoles en plaine de Bièvre, Ph.D. thesis,Université de Grenoble 1, 2001.
Schnebelen N., Analyse et modélisation de l'impact de la maîtrise despratiques agricoles sur la pollution diffuse par les nitrates.
Please cite this article as: Ledoux E et al. Agriculture and groundwatermodelling chain. Sci Total Environ (2007), doi:10.1016/j.scitotenv.2006.1
Application à l'aquifère des calcaires de Beauce (site deVilamblain), Ph.D. thesis, Université d'Orléans, 2000.
Strabel O, Duynisveld WHM, Bottcher J. Nitrate pollution ingroundwater in Western Europe. Agric Ecosyst Environ 1989;26:189–214.
Schnebelen N, Nicoullaud B, Bourennane H, Couturier A, VerbequeB, Revalier C, et al. The STICS model to predict nitrate leachingfollowing agricultural practices. Agronomie 2004;24: 423–35.
Thisse J-F. De l'indétermination des régions et de quelquesinconvénients qui en résultent. L'Espace Géographique 1997;2:135–48.
Yadav SN, Wall David B. Benefit-cost analysis of best managementpractices implemented to control nitrate contamination ofgroundwater. Water Resour Res 1998;34: 497–504 [n°3].
nitrate contamination in the Seine basin. The STICS–MODCOU2.002
