On the use of neural networks to evaluate groundwater levels in fractured media

This paper evaluates the feasibility of using artificial neural network (ANN) methodology for estimating the groundwater level in some piezometers implanted in unconfined chalky aquifer of Northern France. These aquifers are the most susceptible to depletion and contamination, with the recharge rate and dominant processes determining their level of vulnerability. Groundwater level simulation in a chalky media is regarded as a difficult subject in hydrogeology due to the complexity of the physical processes involved, the variability of piezometry in space and time and the aquifer response in fissured and matrix blocks. In the present project, after a detailed geologic and hydrogeologic study of the sector, we simulated the groundwater level by Neural Network. The first objective was to determine the most influential parameters which impact groundwater level in fissured chalky media. The second objective was to investigate the effect of temporal and spatial information by considering current and past data sets along with the use of a variety of piezometer readings. The third objective was to simulate the groundwater level in a selected piezometer. The reasonably good ANN-based simulations revealed the merit of using ANNs and specifically Multi Layer Perceptron (MLP) models. The proposed ANN methodology using minimal lag and number of hidden nodes, along with the optimal number of spatial and temporal variables consistently produced the best performing network-based simulation models.