Valorisation de l’information pédologique numérisée pour la modélisation de la percolation des pesticides vers les eaux souterraines | Valorization of the Digitalized Soil information for the modelling of pesticide lixiviation into groundwater

In Wallonia (southern Belgium), 81% of human drinking water is captured from groundwater tables. In spite of actions implemented under the European directives and thematic strategies, the required standard (0.1 µg L-1 by pesticide and 0.5 µg L-1 for all pesticides) for human drinking water is most often exceeded at some catchments. Tools such as mechanistic models are appropriate since they allow an a priori estimation of pesticide concentrations leached at the bottom of soil profile, which could reach the groundwater tables. However, simulations often lead to a degree of uncertainty due to the approximations noted in the models, or due to the reliability of available data. The main objective of this thesis was to improve the usefulness of the Digital Soil Map of Wallonia (DSMW) and the associated soil analysis database named “Aardewerk” in the framework of the evaluation of the groundwater table contamination by pesticides, in order to reduce approximations associated with the GeoPEARL model and its meta-model MetaPEARL in the area of Condroz. In order to cover the broad spectrum of pesticides used in agriculture in Wallonia, nine theoretical pesticides have been defined. The first research question is to measure the sensitivity of GeoPEARL to two soil data parameterization schemes at the regional scale, depending on whether we first aggregate soil data at a regional scale before application of the model (AC approach); or we first apply the model to soil profiles and aggregate simulated leached concentrations afterwards (CA approach). The results show a significant difference between the two approaches, compared on the basis of a spatial threshold (80th percentile) calculated by soil type. Moreover, this difference varies depending on pesticide type, and is more important when the simulated concentrations are very low. It is also clear from this analysis that only pesticides located in certain ranges of DT50 (30 and 67 days) and Kom (58 and 252 dm3 kg-1) give concentrations more often above water potability standard of 0.1 µg L-1. The second research question is to calibrate MetaPEARL metamodel (simplified model) based on GeoPEARL simulated concentrations, from which it was derived by considering only the most sensitive parameters of GeoPEARL. The objective by developing a metamodel is to provide a simple tool, usable for a first evaluation of pesticides leaching in order to identify the most risk-prone areas, but also readily transposable for a global decision support system. In contrast to the approach of MetaPEARL calibration used until now, taking into account soil morphology through soil horizons (layers) types explains about 80% of the spatial variability of pesticide concentrations obtained from the complete original GeoPEARL model. The third research question is to take into account the spatial variability of input variables within the same soil typological unit using stochastic simulations to assess the uncertainty associated with deterministic simulations (taking into account average values by soil type for the input variables at the regional scale) obtained from MetaPEARL. The random spread of soil characteristics variability leads to a very significant spreading of simulated pesticide concentrations. Moreover, we obtain values of 80th spatial probabilistic percentile higher than those obtained from the deterministic approach. Consequently, for a risk mapping designed to aid decision-making tools, the probabilistic approach should be used since it allows, by taking into account the spatial variability of the model input parameters, to obtain more realistic concentrations of leached pesticides.