modelling the effect of the spatial distribution of agricultural...

Ecological Modelling 137 (2001) 93–105

Modelling the effect of the spatial distribution ofagricultural practices on nitrogen fluxes in rural catchments

Veronique Beaujouan a,*, Patrick Durand b, Laurent Ruiz b

a Laboratoire de spatialisation Numerique, ENSAR, Rennes, 65 rue de St Brieuc, 35042 Rennes Cedex, Franceb Unite Sol et Agronomie Rennes Quimper, INRA, Rennes, 65 rue de St Brieuc, 35042 Rennes Cedex, France

Received 25 February 2000; received in revised form 4 October 2000; accepted 27 October 2000

Abstract

An integrated, hydrology and nitrogen dynamics model was developed to study the spatial interactions between soiland groundwater that can affect the nitrogen delivery to streamwater in rural catchments. The hydrological modelTNT is based on TOPMODEL hypotheses but is it fully distributed according to a regular square grid. A subsurfaceflow component was distinguished to account for the supply of groundwater and nitrate to downslope soils. The cropgrowth and nitrogen biotransformations were simulated using an existing generic crop model, STICS. Both modelsare process-based, but kept as simple as possible. The integrated model was applied to theoretical catchments toanalyze the combined effects of geomorphology and crop distribution on the whole catchment nitrogen budget. Thecatchments differed both in the slope profile and in the pattern of water pathways. The results suggest that placingcrops acting as nitrogen sinks downslope potentially polluting crops could reduce significantly the streamwatercontamination by nitrate. This effect is the highest for catchments with parallel water pathways and a wide concavebottomland. Nitrogen uptake by sink crops was quantitatively more important than denitrification to reduce nitrogenoutput. It is concluded that this model, although still in development, may prove an interesting working tool toinvestigate the effect of the landscape structure on nutrient budgets in ecosystems. © 2001 Elsevier Science B.V. Allrights reserved.

Keywords: Distributed hydrological model; Crop model; Water pollution; Geomorphology; DEM; Buffer zone

www.elsevier.com/locate/ecolmodel

1. Introduction

In intensive farmland regions, streamwater pol-lution by nitrate is due to excess organic andmineral nitrogen fertilization. Even with careful

farming practices, it is probably incompatible tomaintain an optimal and regular crop yield and tokeep nitrogen losses under about 30 kg N ha−1

year−1 (Mariotti, 1997). This is especially truewhen the proportion of organic fertilization ishigh, or in oceanic regions where mild winters andhighly organic soils promote high mineralizationrates which vary according to climate conditions.Since this level of nitrogen loss is still sufficient to

* Corresponding author. Tel.: +33-2-23485438; fax: +33-2-23485430.

E-mail address: [email protected] (V.Beaujouan).

0304-3800/01/$ - see front matter © 2001 Elsevier Science B.V. All rights reserved.

PII: S 0304 -3800 (00 )00435 -X

V. Beaujouan et al. / Ecological Modelling 137 (2001) 93–10594

affect streamwater quality, other remediationmeasures, such as landscape management, mustbe considered.

Landscape management measures to controlnitrogen contamination of surface waters includebuffer zones preservation or restoration (Haycocket al., 1996; Martin and Reddy, 1997; Merot etal., 1998), these ‘buffer zones’ being riparian wet-lands, vegetated strips or hedgerows. Commonsense and a few studies (e.g. Decroux et al., 1991)suggest that the location of the pollution ‘sourceparcels’ in the landscape, especially the distance tothe watercourse, may affect the amount of pollu-tant reaching the stream. This is empirically inte-grated in the export coefficient approach (Johnes,1996) and similar approaches (Skop andSorensen, 1998). The processes involved dependon the water pathways prevailing in the catch-ments. In temperate areas with imperviousbedrock, the groundwater table is often locatedwithin the soil or at the surface in the lower partof the hillslopes: subsurface saturated flow makesit possible for upslope contaminated water to‘irrigate’ downslope soils. This allows nitrogenuptake by crops or riparian vegetation and deni-trification in waterlogged areas. To what extentcould such processes affect nitrogen losses fromrural catchments? The experimental evidence atthe catchment scale is very hard to obtain. Itwould require either truly ‘comparable’ catch-ments with different landscape patterns or catch-ments where the land use pattern could bedrastically changed. A simpler approach is to usesimulation models to compare different scenarios.

Many nitrogen simulation models have beendeveloped in the past three decades, and somereviews and comparisons can be found in theliterature (Addiscott and Wagenet, 1985; Vachaudet al., 1988; Kauark-Leite, 1990; Ball andTrudgill, 1995; Diekkruger et al., 1995; VanGrinsven et al., 1995; Addiscott and Mirza, 1998).A lot of models (SOILN, WAVE, LEACHN,CREAMS, SLIM, . . .) compute the nitrogenfluxes and transformations at the plot scale, as-suming one dimensional, vertical transfer. Thesemodels have sometimes been used at the catch-ment scale, simply by aggregating the plot scaleresults (ANSWERS, AGNPS, . . .). Such models

are of little use here, since the key process tosimulate is the spatial interaction between thedifferent zones of the catchment. In order to takesome spatial processes into account, Krysanova etal. (1998) have coupled plot scale models withhydrologic reservoirs, but the routing from up-slope to downslope is averaged and not simulatedexplicitly. The global conceptual models, repre-senting the catchment — or the subcatchments— as a homogenous entity, are not more ade-quate for this. In such models, the spatial interac-tions are accounted for implicitly, in the fitting ofthe model results to the whole catchment re-sponse, but the identification and the quantifica-tion of these interactions are difficult: this shouldrequire extensive data either from different catch-ments or from manipulated catchments. The ade-quate models to study these spatial interactionsmust be process-based and spatially distributed.More precisely, they must simulate fairly accu-rately subsurface lateral flow, nitrogen transfor-mations in soil and plant nitrogen uptake.Complex, mechanistic models such as SHE (Ab-bott et al., 1986) could be used as the hydrologicalmodelling basis for such models, coupled withcrop growth and nitrogen transformation models.For example, the WAVE and DAISY modelshave been used together with SHE to simulatenitrogen dynamics in catchments (Christiaens andFeyen, 1997). A simpler approach was once re-ported, coupling the SLIM model with a dis-tributed hydrological model (Cooper et al., 1994).None of these models were found readily opera-tional and, following Addiscott and Wagenet(1985) and Beven (1996), the mechanistic ap-proach was discarded because at the catchmentscale, its great complexity does not warrant betterperformance or sounder theoretical basis.

The choice was made to build up a new modelthat would better match the requirements of thestudy objectives. Since the objective was not todevelop new modelling concepts but to quicklyhave an operational tool, one decided to adaptand to put together existing modelling ap-proaches, preferably reasonably simple ones. Thispaper describes the structure of this model andpresents preliminary results, obtained from anapplication of the model to ‘virtual’ catchments,

V. Beaujouan et al. / Ecological Modelling 137 (2001) 93–105 95

aimed at giving a first order of magnitude of theeffect of spatial interactions on nitrogen losses.

2. Material and methods

2.1. Models

The simulation model is constituted of twomodels running separately but synchronized fol-lowing the producer–consumer technique (Ben-Ari, 1982) and exchanging data at each time step.The first model is a distributed hydrologicalmodel called topography-based nitrogen transfer(TNT), and the second is a crop model, STICS(Brisson et al., 1998).

2.2. From TOPMODEL to TNT and couplingprinciples

Since its creation by Beven and Kirkby (1979),TOPMODEL has been widely developed andused (Beven, 1997). It simulates flows in smallcatchments on impermeable bedrock, in whichsubsurface flow and overland flow on variablecontributing area predominate. It is assumed thatthe downslope flux can be described as a steadystate, saturated Darcian flux driven by a uniformrecharge rate to the water table. The effectivehydraulic gradient is taken as constant and equalto the local surface slope and the hydraulic trans-missivity depends on soil saturation deficit. TOP-MODEL calculates water fluxes at the outlet andthe mean saturation deficit of the whole catch-ment at each time step, and distributes the satura-tion deficit at each point of the watershed,depending on a topographic index.

To represent the different flow pathways andthe heterogeneous nitrogen inputs within thecatchment, it was necessary to abandon the hy-pothesis of uniform recharge and to replace thecalculations based on topographic index classesby an explicit cell to cell routing. Following thework from Crave and Gascuel-Odoux (1997) thehydraulic gradient was better estimated by the‘downslope gradient’, i.e. the gradient between theelevation of the cell considered and the firststream cell reached when following the flow path-

way. Conductivity was supposed to decrease ex-ponentially with the depth like in the initialTOPMODEL version, but another profile couldbe adopted following catchment properties as sug-gested by Ambroise et al. (1996).

The watershed is modelled as a set of columnsas shown on Fig. 1, each column corresponding toa cell of a grid digital elevation model (DEM).The flow direction is determined following thegreatest elevation gradient between a cell and itseight neighbours. River cells are determined by adrainage area threshold: for the cells over thisthreshold the outflow is routed directly to theoutlet. TNT is based on a water balance at eachtime step for each column of the watershed. Thecolumn is divided into two stores:� the soil store, that can be seen as the retention

porosity of the root zone, where water can betaken up by plants or evaporated;

� the drainage store, that can be seen as thedrainage porosity of the soil and of the subsoil,with a fixed capacity SZMAX. The amount ofwater in this store (SZ) determines two zones:the saturated zone where the groundwaterflows laterally, and the unsaturated zone (orvadose zone) where water percolates vertically.When the drainage store is full (SZ=SZ-MAX), the vadose zone disappears and over-land flow is generated.Retention porosity under the soil is not taken

into account.In this coupled version of the model, the soil

store is controlled by STICS, which simulatescrop uptake, evapotranspiration, vertical waterfluxes, and nitrogen transfers and transformationsafter rain and fertilization. Drainage water andnitrogen leaching under the soil calculated bySTICS at each time step are summed with thesurface fluxes calculated from the upslope cells toenter in TNT’s drainage store. If the water inputis higher than the capacity of the unsaturatedzone, excess water forms overland flow at thesurface. The percolation of water from the vadosezone to the saturated zone is controlled by theamount of water in the vadose zone (SUZ) and amaximum percolation rate given by Darcy’s lawat the interface between the two zones with ahydraulic gradient equal to 1:


Qp(t)

=minimum(K0 exp((SZ(t−1)−SZMAX)/M);

SUZ(t)) (1)

where K0 is the vertical hydraulic conductivity atthe soil surface at saturation (m day−1); M theexponential decay rate of the hydraulic conductiv-ity with depth (m); SZ(t−1) the saturated watercontent at time (t−1) (m3 m−2); SZMAX themaximum water content in the saturated zone (m3

m−2); SUZ(t) the water content in the unsatu-rated zone and Qp(t) the percolation (m3 m−2

day−1).Percolation and deep groundwater flow from

the upslope cells Qbu(t) increase the saturatedzone content SZ(t−1). Lateral flow to thedownslope cell is calculated using Darcy’s law,and is separated into two different flows, Qb(t)and Qes(t), depending if the groundwater level isabove a threshold (SZT) (Robson et al., 1992):

If SZ(t)BSZT :

Qb(t)

=T0 tan b(exp((SZ(t)−SZMAX)/M)

−exp(−SZMAX/M)) (2)

Qes(t)=0 (3)

If SZ(t)\SZT :

Qb(t)

=T0 tan b(exp((SZT(t)−SZMAX)/M)

−exp(−SZMAX/M)) (4)

Qes(t)

=T0 tan b exp((SZ(t)−SZMAX)/M)−Qb (5)

where SZT is the threshold above which subsur-face flow is generated (m3 m−2); Qb(t) the deepgroundwater flow (m3 m−2 day−1); Qes(t) thesubsurface flow (m3 m−2 day−1); tan b the

Fig. 1. Diagram of the catchment representation in the integrated model.


downslope gradient (–); and T0 the lateral trans-missivity at saturation (m2 day−1).

The deep groundwater flow is routed to thedownslope cell saturated zone whereas the subsur-face flow is routed to the surface. This waterenters the soil zone controlled by STICS, andnitrates are exchanged between soil water andsubsurface water. This process permits exchangesof nitrate and water between soil and groundwa-ter, and the regions where this subsurface flowoccurs are thus called the ‘interaction area’ in thefollowing text.

Nitrate is considered as a perfect solute in thehydrological model, and the mixing in each reser-voir is instantaneous and complete.

The model parameters are the following: theeffective vertical hydraulic conductivity at the soilsurface (K0, in m day−1), the total lateral trans-missivity of the column (T0, in m2 day−1), theexponential decay factor of the hydraulic conduc-tivity with depth, supposed to be the same verti-cally and laterally (M in m), the maximumamount of water of the mobile groundwater (SZ-MAX in m), and the threshold above which sub-surface flow occurs (SZT in m). These parameterscan be different for each cell.

2.3. The crop model

Many crop models simulating nitrogen verticalfluxes and transformations in soils have been de-veloped so far (see, e.g. Ball and Trudgill, 1995).STICS (Brisson et al., 1998) has been chosenbecause: (i) it simulates explicitly the effect ofwater and nitrogen stresses on crop developmentand growth; (ii) it is generic, i.e. its structureallows to simulate different crops with the sameset of equations; (iii) it is relatively simple and thewater transfer component is compatible withTOPMODEL; and (iv) it has been well parame-terized and tested for winter wheat and maizeunder temperate climate.

The main processes simulated are the growth,the development of the crop and the water andnitrogen balance of the soil–crop system, orga-nized in seven modules: development, shootgrowth, yield components, root growth, waterbalance, thermal environment and nitrogen bal-

ance. STICS water input is rainfall at the soilsurface and the model simulates evapotranspira-tion, water and nitrogen uptake by the plants,nitrogen transformations in the soil, and there-after percolation and nitrogen leaching throughthe soil considered as a succession of horizontallayers as in Burns’ model (Burns, 1974).

This model has been slightly adapted to takeinteractions between cells into account in additionto the existing transfers. Water of the subsurfaceflow coming from upslope cells, calculated byTNT, is mixed to the existing soil water in eachlayer that is not saturated, beginning from thebottom soil layer. If the layer is at field capacity,only nitrates are exchanged by homogenizing theconcentrations between the subsurface flow waterand the soil water. If the water content of thelayer is under field capacity, it is filled up to thefield capacity, and the remaining water moves tothe next layer upwards.

When the organic layer is saturated by thegroundwater, the mineralization rate of the soilorganic matter is halved (Van Der Linden et al.,1987), and denitrification is activated. The deni-trification rate is calculated using the NEMISmodel (Henault, 1995) adapted to the saturatedsoils and the local context:

VED=VPD · fT · fN (6)

where VED is the actual denitrification rate (kg Nha−1 day−1); VPD the potential denitrificationrate (kg N ha−1 day−1), at 20°C under anoxicconditions and with excess of nitrate (here takenas 30 kg N ha−1 day−1 according to Durand etal., 1998); fT, a temperature function following anArrhenius law; and fN, a function of nitrate con-centration following a Michaelis–Menten law.

The code of the model has been modified toallow multiple soil column and multiple crop sim-ulation, and the model is run for each cell at eachtime step.

2.4. Simulation protocols

Simulations are conducted on square virtualcatchments of 400 square cells 40 m wide, i.e. atotal area of 0.64 km2 (Fig. 2). The elevationrange is 26 m. Two slope profiles were chosen:


Fig. 2. Aspect of the six types of virtual catchments.

one features a wide valley bottom, called ‘con-cave’, the other features a narrow valley bottomand an extended plateau, called ‘convex’. Theseslope profiles were used to generate a catchment,either by rotation around the lowest point to forma headwater catchment, called ‘convergent’, or bytranslation to obtain a regular hillslope, called‘parallel’. An ‘intermediate’ type was constructedby introducing higher spots in the middle of theparallel catchment. This constitutes six ge-omorphological types of catchments, called in thefollowing text: Cc, Cv, Pc, Pv, Xc, Xv with Cstanding for convergent, P for parallel, X forintermediate, c for concave, v for convex. Thesecatchments can be seen as stereotypes of the reliefelements of a temperate area on crystallinebedrock.

The soil of the catchments was considered asbeing homogeneous and having the mean proper-ties of the soil of the experimental catchment ofNaizin (Britanny, France): depth of 80 cm, fieldcapacity of 180 mm, a plough layer 20 cm thick,17% clay and 2.4% organic matter (Walter andCurmi, 1998).

The catchments were divided into 16 squareparcels of 25 cells (i.e. 0.04 km2) each. In half theparcels, agricultural practices resulting in a strong

excess of nitrogen (‘source’ parcels) have beensimulated. In the other parcels, the simulatedapplied nitrogen was lower than the crop require-ments (‘sink’ parcels). The simulation exerciseconsisted in comparing the nitrogen fluxes in thesix constructed catchments with the source parcelseither upslope or downslope. In the parallel case,a chequered’ distribution of the source and sinkparcel has also been tested.

The parameters used for the source and sinkcrops are those established by the authors of theSTICS model for maize and wheat, respectively.Overfertilized maize is actually considered as animportant source of nitrate because of a shortcycle and a poor nitrate absorption capacity. Onthe other hand, under-fertilized winter wheat canbe considered as a potential sink of nitrate be-cause of a relatively high uptake capacity and along cycle. The agricultural practices have beenadapted to these crops (Table 1). Both grain andstraw are harvested for the two crops (as silagefor maize) and the roots only are left in the field.

Mean nitrogen fertilization on the whole catch-ment is of 250 kg N ha−1 year−1, 80 kg asmineral N and 170 kg as cattle slurry. The figureof 170 kg N ha−1 year−1 is the EU standard formaximum organic N load to crops. Mineral fertil-


izer is applied on the whole catchment whilemanure is applied on the source parcels only. It issupposed that 30% of the nitrogen of the slurry isvolatilized during spreading; the remaining part(i.e. 158 kg N ha−1 year−1) is considered as beingspread on the source parcels at two dates, andtreated by the model as mineral N.

Another series of simulations has been con-ducted to test the effect of nitrogen deficit of thesink parcels. Different amounts of mineral fertil-izer (0, 40, 80, 120, 160, 200, 320 and 440 kg Nha−1 year−1) are applied on the sink parcels, onequarter on 15 February and the other part on 31March. Source parcels fertilization is kept at 250kg N ha−1 year−1. It is important to note that inthese simulations, the total nitrogen input to thecatchments varies between the simulations.

Climate data were obtained from measurementson the Kervidy catchment (Brittany, France) dur-ing 1994, 1995 and 1996. These 3 years showcontrasted features, both in amount and in distri-bution of the rainfall, which represent quite wellthe climate variations of Brittany. Three succes-sive cycles of these 3 years were simulated, thefirst two being used for initializing the system. Asteady state is established between the second andthe third cycle.

The results presented here are the means of thethird cycle of 3 years. Parameters used for thehydrological model were obtained by a step bystep calibration in the Kervidy catchment forthese 3 years.

3. Results and discussion

3.1. General obser6ations

The first set of results presented was obtainedwith the six geomorphologic types, crossed withthe two types of spatial distribution of the crops(source parcels upslope or downslope) and onlyone level of mineral fertilization, i.e. 12simulations.

The mean annual specific discharge varied onlybetween 416 and 429 mm for a mean annualrainfall of 850 mm for the 3 years studied. Anexample of specific discharge simulated on theparallel concave catchment is shown in Fig. 3.Comparison with the specific discharge of theKervidy catchment showed that the model simu-lates the discharge dynamics of a catchment inBrittany fairly well. The mean actual evapotran-spiration was 430 mm, 187 mm of which beingevaporated and 243 mm transpired.

The mean nitrogen export by crops was of148918 kg N ha−1 year−1. The mean export bythe ‘source’ crop was fairly steady (14995) whilethat of the ‘sink’ crop which was more variable(147935).

The spatial patterns obtained are presented inFig. 4. The maps show discontinuities that arelinked to artefacts in the DEM construction oranalysis. In particular, mono-directional drainagenetwork derivation causes non-realistic patternsof saturation in the convergent catchments (for afull discussion on mono-directionnal and multi-di-rectionnal drainage network derivation methods,see Wolock and McCabe, 1995; Beaujouan et al.,2000). The treatment of the river cells and thecomputation of downslope gradient calculationalso lead to minor irregularities in the parallelcatchments. However, these artefacts do not affectsignificantly the results.

The maps of the mean duration of the interac-tion period (Fig. 4(i)) show that the size of theinteraction area varied with catchment morphol-ogy. An interaction index was constructed foreach catchment to represent the spatial and tem-poral extent of the interaction area: this index isequal to the mean daily proportion of the catch-ment area where subsurface flow occurs. Fig. 5

Table 1Cultural practices

‘Sink’ crop (wheat)October 24 SowingFebruary 15 20 kg N ha−1 (mineral fertilizer)

60 kg N ha−1 (mineral fertilizer)March 31July 24 Harvest

‘Source’ crop (maize)170 kg N ha−1 (slurry)March 1

April 15 170 kg N ha−1 (slurry)SowingMay 780 kg N ha−1 (mineral fertilizer)May 7

September 30 Harvest


Fig. 3. Simulated specific discharge at the outlet of the ‘parallel concave’ catchment (black line) compared to the specific dischargeobserved at the outlet of the catchment of Kervidy (Brittany, France) (grey line). On the secondary Y-axis are plotted the dailyrainfall (grey bars) and evapotranspiration (black line).

Fig. 4. Maps of the mean values of different variables for the last 3 simulation years: (i) mean duration of the interaction periods(expressed in days year−1) on three catchment types: Cc: convergent concave; Pc: Parallel concave; Pv: Parallel convex. (ii) Meandenitrification (kg N ha−1 year−1) on the parallel concave catchment depending on the location of the crops. (iii) Mean nitrogenstress index (INN) during the reproductive phase of the crop on the Parallel concave catchment depending on the location of thecrops. ‘s’ letter indicates the position of source crop parcels. Outlet of each catchment is on the bottom right.


shows two major trends: convex catchments havea smaller index than concave ones; convergentcatchments have a smaller index than the catch-ments with a parallel structure, namely the paral-lel and intermediate ones.

3.2. Effect of spatial arrangement of crops onstream pollution

The mean nitrate concentration in the riverover the 3 years of simulations for each catchmentand for the different crop distributions is given inthe Table 2. This concentration was almost thesame for the six catchments when the source cropswere located on a downslope. It was lower andmore variable when the source parcels were lo-cated on an upslope. The difference in concentra-tion due to the change in crop locationrepresented 12–34% of the maximum concentra-

tion: spatial distribution of the crops may wellsignificantly affect water quality by mean of ex-changes between soil and groundwater.

This effect varied with catchment morphology.Fig. 6 shows that, in the cases studied, the effectof crop location is strongly linked to the interac-tion index. This interaction index accurately syn-thesizes the environmental characteristicsfavouring the effect of a change in the crop spatialdistribution.

3.3. Origin of this effect

The general trend observed while changing thedistribution of crops were the same for all thecatchments. The following analysis focuses on thecatchment for which the most important effectwas obtained, i.e. the parallel concave catchment.

Fig. 5. Interaction indexes calculated for the different catchment types. The interaction index is defined as the mean daily proportionof the catchment area where subsurface flow occurred.

Table 2Mean streamwater nitrate concentration (mg NO3 l−1) for each catchment type, depending on crop location

XvXcPvPcCvCc

161 171Source crops downslope 157 167 163 166Source crops upslope 139127 113 136151 103

3421 17 31 1812Difference (% of max. concentration)


Fig. 6. Difference in mean nitrate concentration at the outlet of each catchment induced by the change in crop location(concentration when source parcels are located downslope minus concentration when parcels are located upslope), as a function ofthe interaction index.

Three different simulations were made by chang-ing the location of the source parcels: upslope,downslope or chequered.

The gross input of nitrogen (i.e. by rainfall,mineralization and fertilization) was roughly thesame for the three simulations, namely 294 kg Nha−1 year−1. The amount of nitrogen stored inthe soil and groundwater was also almost thesame for all three simulations after the 3 years.Therefore, the variations in nitrogen discharge inthe river were due to denitrification and exportsby the crops only.

The mean simulated denitrification on thewhole catchment was 9, 12.5, and 17 kg N ha−1

year−1 when sources were upslope, chequered ordownslope, respectively. The maximum denitrifi-cation rate simulated for a cell was 51 kg N ha−1

year−1. These values are low in comparison to thepotential denitrification rate chosen (30 kg Nha−1 day−1, i.e. 10 950 kg N ha−1 year−1,Durand et al., 1998). The spatial distribution ofdenitrification is shown in Fig. 4(ii). Denitrifica-tion is obviously located in the potentially satu-rated areas, but it is sometimes low even for the

cells with lasting saturation. This result suggeststhat denitrification was essentially limited by ni-trate availability, which has also been observed inthe field (Durand et al., 1998). Moreover, highdenitrification rates are located in the source par-cels. In these simulations, denitrification was moreefficient if nitrogen was applied right into themost humid areas: nitrate exchanges betweengroundwater and soil water were of secondaryimportance for denitrification.

Mean nitrogen export by the crops was 127,155, and 180 kg N ha−1 year−1 when source

Table 3Amount of nitrogen exported by crops (kg N ha−1 year−1) forthe parallel concave (Pc) catchment as a function of croplocation

Location of source crops

Downslope Chequered Upslope

116Sink crops 167 204155138 144Source crops180127 155Mean exports


parcels were upslope, chequered or downslope,respectively. This variation was essentially due toexport by sink crops, as shown in Table 3. Exportby sink crops depended to a great extent on theirlocation with respect to source crops: the highestexport took place in sink parcels downslope fromthe source parcels. The increase in export waslarger when the sink parcels were located immedi-ately downhill the source parcel, but it still existedeven far away from the source parcels. The mapsof nitrogen stress indexes during the reproductivestage of the crop (Fig. 4(iii)) show that the im-provement in nitrogen nutrition of the sink cropsis significant even for the cells remote from thesource parcels. As opposed to the case of denitrifi-cation, nitrogen exchanges between groundwater

and soils may play an important part in thenitrogen nutrition of crops.

Fig. 7 presents the relative amount of the differ-ent nitrogen sinks for each spatial distribution ofcrops. Denitrification accounted for only 5% ofthe total output. In this study, denitrification didnot play a significant role in streamwater riverpollution. Variations in nitrogen export by thesink crops explained most of variations in nitrateconcentrations.

3.4. Effect of the sink crop fertilization

The effect of the sink crop fertilization rate wastested by additional simulation runs. The analysiswas done on a third set of results obtained on the

Fig. 7. Ratio of the different nitrogen outputs from the catchment as a function of the location of the source crops.

Table 4Effect of the fertilization rate of the sink crops on the differences induced by the crop location in (i) nitrate concentration of theriver (in % of the concentration with sink crops downslope) and (ii) amount of nitrogen exported by crops (amount with sourcecrops downslope minus amount with source crops upslope, in kg N ha−1 year−1)

Sink crop fertilization (kg N ha−1 year−1)

80 120 160 200 320 4400 40

Difference in nitrate concentration of the river (%) 5374 42 34 27 11 363Difference in amount of N exported by crops (kg N ha−1 year−1) 59 53 47 41 35 20 966


parallel concave catchment, with two spatial dis-tribution types (sink crops upslope or downslope),and eight nitrogen fertilization rates, i.e. 16 runs.

Results given in Table 4 indicate that the effectof spatial distribution of crops decreased when thefertilization rate of the sink crop increased. How-ever, this effect still existed even when the fertil-ization rate was much higher than the ratesactually used by farmers. This means that themodel allows very high ‘luxury consumption’ bythe wheat. However, the model does not accountfor the negative effects of excess nitrogen andwaterlogging on crop development. For thehighest fertilization rates, the model equationsmay not be applicable anymore. Still, these resultssuggest that downslope parcels, even when wellfertilized, can still uptake a significant part of thenitrate transported by the groundwater.

4. Conclusion

The combined hydrology and crop nitrogenmodel presented here was a prototype aimed atassessing the interest of this modelling approachand giving a first order of magnitude of the croplocation effect on nitrogen losses in rural catch-ments. By keeping the hydrological approach verysimple, and capacity-based, it has been possible tobuild up a robust model, using few parameters,able to simulate the main hydrological processesobserved in the real world catchments of north-western France. The co-operation between thetwo programs was operational, although not opti-mal both from the computing and conceptualviewpoints: it was relatively slow and did notallow simulation of a fully lateral subsurface flow(the simulated flow was first vertical in STICS andthen lateral in TNT). STICS is currently beingsimplified and rewritten to fit in the modular codearchitecture of TNT. The preliminary results givean incentive to develop both modelling and exper-imental investigations on the effects of the spatialdistribution of crops on the nitrogen losses ofrural catchments. The model still needs thoroughtesting and validation before it can be used topredict the response of real world catchments.

Acknowledgements

This work was supported, in part, by the INRA‘Ecospace’ program. First author was receiving aPhD grant from the french Brittany region. Theauthors acknowledge Pierre Aurousseau, HerveSquividant, Salim Combo, Tiphaine Cancouetand Gerald Diquelou for their help concerningthe computer modelling.

References

Abbott, M.B., Bathurst, J.C., Cunge, J.A., O’Connell, P.E.,Rasmussen, J., 1986. An introduction to the Europeanhydrological system-Systeme Hydrologique Europeen,‘SHE’, 1: history and philosophy of a physically-based,distributed modelling system. J. Hydrol. 87, 45–59.

Addiscott, T.M., Mirza, N.A., 1998. Modelling contaminanttransport at the catchment or regional scale. Agric. Ecosys-tems Environ. 67, 211–221.

Addiscott, T.M., Wagenet, R.J., 1985. Concepts of soluteleaching in soils: a review of modelling approaches. J. SoilSci. 36, 411–424.

Ambroise, B., Beven, K., Freer, J., 1996. Towards a generali-sation of the TOPMODEL concepts: topographic indicesof hydrological similarity. Water Resour. Res. 32 (7),2135–2145.

Ball, J., Trudgill, S.T., 1995. Overview of solute modelling. In:Trudgill, S.T. (Ed.), Solute Modelling in Catchment Sys-tems. Wiley, Chichester, pp. 3–56.

Beaujouan, V., Aurousseau, P., Durand, P., Ruiz, L., Squivi-dant, H., 2000. Comparaison de methodes de definition deschemins hydrauliques pour la modelisation hydrologique al’echelle du bassin versant. Revue internationale de Ge-omatique 10 (1), 39–60.

Ben-Ari, M., 1982. Principles of Concurrent Programming.Prentice Hall, Englewood Cliffs, NJ, p. 172.

Beven, K.J., 1996. A discussion of distributed modelling. In:Refsgaard, J.-C., Abbott, M.B. (Eds.), Distributed Hydro-logical Modelling. Kluwer Academic, Dordrecht, pp. 255–278.

Beven, K.J. (Ed.), 1997. Distributed modelling in hydrology:applications of the TOPMODEL Concept. Advances inHydrological Processes. Wiley, Chichester, p. 350.

Beven, K.J., Kirkby, M.J., 1979. A physically based variablecontributing area model of basin hydrology. Hydrol. Sci.Bull. 24 (1), 43–69.

Brisson, N., Mary, B., Ripoche, D., Jeuffroy, M.H., Ruget, F.,Nicoullaud, B., Gate, P., Devienne-Barret, F., Antonio-letti, R., Durr, C., Richard, G., Beaudoin, N., Recous, S.,Tayot, X., Plenet, D., Cellier, P., Machet, J.-M., Meynard,J.-M., Delecolle, R., 1998. STICS: a generic model for thesimulation of crops and their water and nitrogen balances.I. Theory and parameterization applied to wheat and corn.Agronomie 18, 311–346.


Burns, I.G., 1974. A model for predicting the redistribution ofsalts applied to fallow soils after excess rainfall or evapora-tion. J. Soil Sci. 25, 165–178.

Christiaens, K. and Feyen, J., 1997. The integrated WAVE-MIKE SHE model as an instrument for nitrogen loadmodelling on a catchment scale, 2nd DHI Software UserConference, Helsingor, Denmark.

Cooper, D.M., Ragab, R., Lewis, D.R., Whitehead, P.G., 1994.Modelling Nitrate Leaching to Surface Waters. Institute ofHydrology Final Report to MAFF, Wallingford, UK.

Crave, A., Gascuel-Odoux, C., 1997. The influence of topogra-phy on time and space distribution of soil surface watercontent. Hydrol. Process. 11 (2), 43–69.

Decroux, J., Ignazi, J.-C., Puginier, M., 1991. Pour que vivel’eau, p. 25.

Diekkruger, B., Sondgerath, D., Kersebaum, K.C., McVoy,C.W., 1995. Validity of agroecosystem models. A compari-son of results of different models applied to the same dataset. Ecol. Model. 81, 3–29.

Durand, P., Henault, C., Bidois, J., Trolard, F., 1998. Ladenitrification en zone humide. In: Cheverry, C. (Ed.),Agriculture intensive et qualite des eaux, Science Update, pp.157–168.

Haycock, N.E., Burt, T., Goulding, K.W.T. and Pinay, G.,1996. Buffer Zones: Their Processes and Potential in waterprotection, Quest Environmental, Environment Agency, p.326.

Henault, C., 1995. Quantification de la denitrification dans lessols a l’echelle de la parcelle cultivee, a l’aide d’un modeleprevisionnel. PhD Thesis, ENSA, Montpellier, p. 132.

Johnes, P.J., 1996. Evaluation and management of the impactof land use change on the nitrogen and phosphorus loaddelivered to surface waters: the export coefficient modellingapproach. J. Hydrol. 183, 323–349.

Kauark-Leite, L.A., 1990. Reflexions sur l’utilite des modelesmathematiques dans la gestion de la pollution diffuseagricole. PhD Thesis, Ecole Nationale des Ponts et Chaus-sees, Paris, p. 342.

Krysanova, V., Muller-Wohlfeil, D.-I., Becker, A., 1998. Devel-opment and test of a spatially distributed hydrological/water

quality model for mesoscale watersheds. Ecol. Model. 106,261–289.

Mariotti, A., 1997. Quelques reflexions sur le cycle bio-geochimique de l’azote dans les agrosystemes. In: Lemaire,G., Nicolardot, B. (Eds.), Maıtrise de l’azote dans lesagrosystemes. Les Colloques, Reims, France, pp. 9–22.

Martin, J.F., Reddy, K.R., 1997. Interaction and spatial distri-bution of wetland nitrogen processes. Ecol. Model. 105,1–21.

Merot, P., Gascuel-Odoux, C., Durand, P., 1998. Approche desprocessus de transfert dans les petits bassins versants ruraux.In: SHF (Ed.), Colloque d’hydrotechnique. SHF, Paris, pp.59–68.

Robson, A., Beven, K.J., Neal, C., 1992. Towards identifyingsources of subsurface flow: a comparison of componentsidentified by a physically based runoff model and thosedetermined by chemical mixing techniques. Hydrol. Process.6, 199–214.

Skop, E., Sorensen, P.B., 1998. GIS-based modelling of solutefluxes at the catchment scale: a case study of the agriculturalcontribution to the riverine nitrogen loading in the VejleFjord catchment, Denmark. Ecol. Model. 106, 291–310.

Vachaud, G., Vauclin, M., Addiscott, T.M., 1988. Solutetransport in the vadose zone: a review of models. In:DeCoursey, D.G. (Ed.), International Symposium on WaterQuality Modeling of Agricultural Non-Point Sources.United States Department of Agriculture, Logan, Utah.

Van Der Linden, A.M.A., Van Veen, J.A., Frissel, M.J., 1987.Modelling soil organic matter levels after long-term applica-tions of crop residues, and farmyard and green manures.Plant Soil 101, 21–28.

Van Grinsven, H.J.M., Driscoll, C.T., Tiktak, A., 1995. Work-shop on comparison of Forest–Soil–Atmosphere models:preface. Ecol. Model. 83, 1–6.

Walter, C., Curmi, P., 1998. Les sols du bassin versant duCoet-Dan: organisation, variabilite spatiale et cartographie.In: Cheverry, C. (Ed.), Agriculture intensive et qualite deseaux, Science Update, pp. 85–105.

Wolock, D.M., McCabe, G.J.J., 1995. Comparison of single andmultiple direction algorithms for computing topographicparameters in TOPMODEL. Water Resour. Res. 31 (5),1315–1324.

.

modelling the effect of the spatial distribution of agricultural...

Documents