Combining a data-driven approach with seasonal forecasts data to predict reservoir water volume in the Mediterranean area

Abstract

Prolonged droughts and water scarcity have become more frequent in recent years, exacerbating the problem of the artificial reservoirs management in the Mediterranean area. This study proposes a methodology which combines a Nonlinear AutoRegressive network with eXogenous input (NARX) data-driven model with Seasonal Forecasts (SFs) data, with the aim to predict the water volume stored in reservoirs at a mid-term scale, as requested by the local authority. The methodology is applied to four Sicilian reservoirs that experienced water scarcity in the recent past. SFs produced at the European Centre for Medium-Range Weather Forecasting are used to force the NARX models. Also, the reservoirs are in a typical data-scarce environment, where very scarce or no measures at all are available. The results show that the NARXs have the capability to reproduce the volumes stored in the considered reservoirs for the investigated period up to four months in advance. The performance of the modeling system strictly depends on: (i) the goodness of climate forecasts and (ii) the strength of the autocorrelation for the water volumes.