Groundwater level prediction in Apulia region (Southern Italy) using NARX neural network

Abstract

In the Mediterranean area, the high water demand frequently leads to an excessive exploitation of the water resource, which involves a qualitative degradation of the freshwaters stored in coastal karst aquifers, as a result of phenomena such as sea saltwater intrusion. In this study, the NARX network was used to predict the daily groundwater level fluctuation for 76 monitored wells located on the Apulian territory. A preliminary analysis on reference wells was performed in order to assess the impact on the groundwater level prediction of two input parameters, rainfall and evapotranspiration, and the sensitivity to changes of training algorithm and input time delay. Based on the findings of the preliminary analysis, a comprehensive regional analysis and extensive sub-regional analyses were performed, proving the reliability of the NARX-BR network for the groundwater level prediction in wells located on different hydrogeological structures. The accurate results obtained for the Apulia region suggest the NARX network application for groundwater level prediction in other areas affected by groundwater resource management issues.