Data driven analysis of the discharge variations at Mercure spring South Italy

Mercure Spring collects groundwater from a large catchment on the north-western side of Mount Pollino massif. This spring area is characterized by several small springs in a zone where fluvial and lacustrine deposits overlay the limestone of Pollino Unit. Fluvial-lacustrine deposits of Mercure river are generally impermeable due to the silty clayey fraction and hinder the flow through the limestone with a permeability threshold that force groundwater to come out as spring. Groundwater hydrology of mount Pollino Massif is strictly conditioned by the relative permeability of the geological units and by tectonic structure and karst phenomena and groundwater circulation are quite complex, so there are flow paths of quite different length supplying the spring. The average discharge of the spring is about 1.8 m3/s with a minimum value of 1.10 m3/s and a maximum one of more than 2.50 m3/s. A detailed analysis of hydrograph allows to observe interesting short term variations up to 15-20% of the discharge values, occurring during the year. These variations are related to specific rainfall events or periods, in fact these local peaks are normally during January and February when severe rainfall periods are more common. The available discharge data are not long enough to use a data-driven approach to relate discharge to rainfall for all the annual hydrograph. Anyway available data permit an analysis of short-term variations of discharge in relation to the rainfall. For this purpose it was used a symbolic regression technique namely EPR: Evolutionary Polynomial Regression based on a Genetic Algorithm implemented in the tool EPRMOGA-XL. As input data rainfall measured at Rotonda rain gauge station were used and, as output, data of the spring discharge time-series. Daily data on a period of about 4 years (2007-2010) were used. Data-driven analysis allows to recognize a quite quick flow path of 5-7 days and a longer one of about 15-20 days that determine the short-term discharge variability. In addition, it is possible to try to relate this variation to the geological structure. Anyway, there are long term variations like those occurring during the entire year that cannot be caught by the model, due to the limited set of available data.