Prediction and control of nitrate concentrations in groundwater by implementing a model based on GIS and artificial neural networks (ANN)

Abstract

Groundwater modelling has become a major step for decision support in integrated water resource management, but groundwater models require accurate and spatially distributed data to provide reliable results. Hydrogeological modelling of these data can be implemented with physically based models (i.e. MODFLOW, MT3D…). Other approaches that are simpler to implement may be a good substitute for these numerical approaches. This is the case of probabilistic approaches and especially the statistical approach neural networks. The proposed method (coupling GIS/ANN) is especially suitable for the problem of large-scale and long-term simulation. It has been applied in the spatial prediction of nitrates in the chalk aquifer in Bethune (North of France). This confined chalk aquifer in its northern part provides natural denitrification and ensures a good drinking water quality, while in its southern part this aquifer is facing a high level of nitrate concentrations far above the European Nitrates Directive standard. A good groundwater management of this ecosystems service is therefore of great importance for regional water management. Thus, the spatial distribution of nitrate concentration obtained by GIS/ANN coupling model was compared with the results obtained from the numerical modelling (MT3D) and validated by the real measurements. ANN modelling seems to be more realistic than MT3D modelling both for 2003 and 2004. This is true for both of the nitrate concentrations and their difference. So, ANN modelling’s spatially distributed difference with observed data ranges from − 3.67 to + 1.24 mg/l in 2003 and − 10.8 to + 6.51 mg/l in 2004, whereas for the MT3D model, this difference ranges from − 11.5 to + 17.9 mg/l in 2003 and − 9.91 to + 16.9 mg/l in 2004. The satisfactory results of the ANN model allowed to launch prospective simulations for 2025 under two groundwater recharge scenarios: a deficit year (150 mm/year) and a rainy year (500 mm/year) show an expansion of the exploitable zone ([NO3–] < 50 mg/L) in the case of a rainy year. The results demonstrate the potential of ANN modelling of spatially distributed hydrogeological data for groundwater management of nitrate pollution. From a groundwater management point of view, the GIS/ANN modelling represents an alternative data analysis to obtain fast results using a less tedious method whose results are satisfactory.