Analysis of the dynamic of aquifer by evolutionary data-driven modelling

Data mining techniques are powerful approaches to model any kind of environmental system and their popularity improved during the last two decades by the large availability of monitoring/measurement information and the development of calculation instruments. These techniques provide reliable mathematical descriptions of relationships between the physical variables of the systems. Among these techniques, evolutionary modeling constitutes and interesting approach able to combine the regressive feature of data driven modeling with the power of evolutionary optimization. This return results, which consists in optimized models, achieving both a good fitness to data and a good structural parsimony. Multi-Objective Evolutionary Polynomial Regression, EPRMOGA is an evolutionary modeling technique implemented at the Technical Univ. Bari, which proved particularly effective at modeling the dynamics of groundwater in terms of response to rainfall thus giving an important contribution in groundwater management. Here EPRMOGA is used to model South Italy aquifers response to rainfall. In particular the response to rainfall of some aquifers is analyzed, based on data from the old “phreatic network” managed by the former National Hydrographic Bureau. It was a diffused network of phreatic wells, covering Murgia and Salento involving both the deep karst and the shallow quaternary aquifers. Date were collected on the period 1950-1996; therefore long time-series of data are available, which are here used to identify models, where the groundwater level is predicted as function of past measured water levels and past rainfall heights. EPRMOGA identifies a model in terms of closed-form equations for each chosen well. These equations show relatively parsimonious structures, which allow both for accurately predicting groundwater level oscillations and for giving evidence of the main rainfall components influencing groundwater levels and the lag between rainfall and water table oscillations. The identified models were consistent with the hydrogeological characters of the studied sites. It is interesting the comparison of the equations from different wells, representative of the same large aquifer. In fact, the comparison among the different equations, allows for emphasizing different modalities of infiltrations of the monitoring well, related to different local geological layouts.