Gestion de la contamination des eaux souterraines par les fertilisants agricoles: application du modèle AgriFlux

Abstract

La contamination croissante des eaux souterraines par les fertilisations agricoles nécessite une gestion efficace de ces ressources impliquant l'utilisation combinée d'outils informatiques et de suivis de terrain. Le modèle AgriFlux a été développé afin de combler une lacune existant entre les modèles de recherche très complexes et les modèles de gestion peu flexibles. AgriFlux est un modèle de type mécaniste-stochastique, c'est-à-dire utilisant une représentation conceptuelle des mécanismes combinée à une prise en compte de la variabilité des paramètres et processus. Il permet l'évaluation de la contamination potentielle des eaux souter- raines par les fertilisants agricoles. Les modules HydriFlux (bilan hydrique) et NitriFlux (cycle de l'azote) sont actuellement disponibles. Une application du modèle est présentée dans l'optique de la gestion environnementale d'un système agricole. Le site expérimental étudié est localisé près de la ville de Québec (Canada). Il s'agit d'un limon sableux sous culture de maïs sucré (Zea Mays, L.) et recevant des fertilisations inorganiques selon les doses recommandées. Un échantillonnage de l'eau interstitielle a été réalisé sur un réseau de lysimètres avec tension durant deux saisons végétatives ainsi qu'un échantillonnage du sol durant un été. Le contenu en nitrates est déterminé dans les deux cas. Les concentrations en nitrates dans les eaux interstitielles simulées à l'aide d'AgriFlux représentent relativement bien les concentrations mesurées. Les différences observées peuvent être expliquées en partie par les conditions de grande sécheresse ayant prévalu durant la période d'étude. Les contenus en nitrates mesurés dans le sol sont moins bien représentés par le modèle. En début de saison, les variations rapides des contenus en nitrates observées au champ ne sont pas reproduites par le modèle alors que les valeurs de fin de saison sont mieux obtenues par le modèle. Malgré ces différences, la concordance au niveau des ordres de grandeur des concentrations dans l'eau obtenues du modèle et des mesures de terrain confirme l'intérêt d'un tel outil pour la gestion environnementale des contaminations agricoles des eaux souterraines.Groundwater contamination by agricultural practices is a problem of growing concern with water resources managers. Considering the environmental importanve of the situation and the complexity of agricultural systems, models are used more than ever in parallel with field investigaffons to assist in the decision-making process. Most available models are either too complicated (many non-measurable parameters) or too simple (semi-empirical or site-specific) to be used as management tools. Such tools should conform to known theory and be structured to enable efficient analysis of field situations with minimal requirements for parameters. The AgriFlux model was developed according to these criteria. It is a mechanistic-stochastic model simulating groundwater contamination by agricultural fertilizers. It combines reliability and conceptual representations with a limited number of parametersMost mechanistic models are also deterministic, using a single value for every parameter. Because of the important field variability of most parameters, it is pertinent to use a stochastic model, incorporating the variability resulting from field heterogeneity, measurement errors and intrinsic uncertainty related to parameter definition. AgriFlux represents soil water and nitrogen dynamics at the scale corresponding to a homogeneous area (soil type, crops, fertilizers, ..), usually the agricultural field. The soil profile is divided in homogeneous horizons or compartments. A daily tlme step is used in the calculations. The model is based on a modular structure to facilitate the inclusion of future modules describing the fate of phosphorus and pesticides. At present, AgriFlux contains two modules. The first one named HydriFlux, simulates water-related processes (precipitation snowmelt, infiltration, runoff, water uptake by plants, evaporation, percolation and drainage using soil water characteristic functions and the unsaturated hydraulic conductivlty function. The second module, NitriFlux, represents the nitrogen cycle processes (fertilizer inputs, mineralization, nitrification, denitrification, nitrogen uptake by plants and nitrate leaching). Biochemical processes are influenced by soil temperature and humidity.To illustrate the use of AgriFlux in a resource management perspective, an application of the model was performed on an agricultural field located near Québec City (Québec, Canada). The cultivated soil consists of a loamy sand underlain by a coarse till. Sweet corn (Zea Mays, L) was grown on the plot for ten years using standard inorganic fertilizations. The environmental studies performed on the site consisted of sampling interstitial water through a series of tension Iysimeters (June 1990-November 1991) to determine the nitrate concentration of water reaching the groundwater. Soil samples were collected (June 1991-November 1991) to evaluate the potential for nitrate leaching. The parameters required to represent the field were identified from site characterization (soil data), from available agricultural data (crop and fertilization data), from literature (nitrogen cycle parameters) and from other available values (climate data). Special care was put into the identification of nitrogen cycle parameters because of their importance on the simulated results. The necessary parameters require no adjustement or optimization because they represent physically measurable values.The simulated nitrate concentrations represent the measured values relatively well with the exception of some periods during the year especially during the summer for which the measured values are slightly overestimated by the model. This discrepancy can be explained partially by the drought of the 1991 summer months which limited the number of water samples thus restricting their representativeness. The spatial variability of measured concentrations was underestimated by AgriFlux. This can be related to the small number of field measurements compared to simulated values (the standard deviation decreases when the number of values increases), to a possible underestimation of parameter variability in the model (mainly nitrogen cycle parameters which are difficult to estima te) and to the influence of macroporosity (matric flow and macropore flow having different nitrate contents). This observation confirms the importance of using a stochastic model and the necessity of sampling at many locations in the field. The soil nitrate contents are represented with less precision by the model. Rapid variations in measured values are not found in the simulated soil nitrate contents. The overestimation of simulated summer nitrate concentrations and underestimation of simulated soil nitrate contents can be related to the drought conditions prevailing during the summer of 1991. When the soil is very dry, water extracted from tightly bonded water can contain more nitrates than freely flowing water, an effect which AgriFlux does not take into account.The case study shows an example of an application of AgriFlux in a resource management perspective. Overall, simulated results represent adequately measured values, thus confirming the parameter values selected to represent the field under study. A water resources manager could use this set of parameters to represent a large number of scenarios consisting of alternative agricultural practices. These scenarios can be confirmed afterwards by a limited number of field investigations. The use of a model such as AgriFlux in the decision-making process facilitates and accelerates the implementation of governmental intervention .