EVALUATION OF THE ABILITY OF AN ARTIFICIAL NEURAL NETWORK MODEL TO SIMULATE THE INPUT-OUTPUT RESPONSES OF A LARGE KARSTIC AQUIFER: THE LA ROCHEFOUCAULD AQUIFER (CHARENTE, FRANCE)

Abstract

The ability of artificial neural networks (ANN) to model the rainfall-discharge relationships of karstic aquifers has been studied in the La Rochefoucauld karst system, south-west France, which supplies water to the city of Angouleme. A neural networks model was developed based on MLP (multi-layer perceptron) networks and the Levenberg-Marquardt optimization algorithm. Raw rainfall data were used without transformation into effective rainfall. This allowed for the elimination of certain non-verifiable simplifying assumptions and their subsequent introduction into the modeling procedure. The raw rainfall and discharge data were divided into three groups for the training, the validation and the prediction test of the ANN model. The training and validation phases led to a very satisfactory calibration of the ANN model. The attempt to predict discharges showed that the ANN model is able to simulate the karstic aquifer discharges. The shape of the simulated hydrographs was found to be similar to that of the actual hydrographs. These encouraging results make it possible to consider interesting and new prospects for the modeling of karstic aquifers, which are highly non-linear systems.