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

Prolonged droughts and water scarcity have become more frequent in recent years, exacerbating the problem of artificial reservoir management in the Mediterranean area. This study proposes a methodology that combines a Nonlinear AutoRegressive network with eXogenous input (NARX) data-driven model with seasonal forecast (SF) 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 few or no measurements 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 modelling system strictly depends on: (i) the quality of climate forecasts and (ii) the strength of the autocorrelation for the water volumes.