Three-and-six-month-before forecast of water resources in a karst aquifer in the Terminio massif (Southern Italy)

• Ability of artificial neural networks to model the rainfall–discharge relationships of karstic aquifers.
• Three month before forecast of water resources.
• Six month before forecast of water resources.
• Error in forecasting discharge values of just 5% (three months before).
• Error in forecasting discharge values of just 10% (six months before).

The ability of artificial neural networks (ANN) to model the rainfall–discharge relationships of karstic aquifers has been studied in the Terminio massif (Southern Italy), which supplies the Naples area with a yearly mean discharge of approximately 1–3.5 m3/s. The Mediterranean climate causes a rapid increase in evapotranspiration and a decrease in rainfall towards spring–summer. Especially during drought, and in combination with highly sensitive climatic parameters, there are dramatic changes in the discharge amount especially during the July and August months. A neural network model was developed based on MLP (multi-layer perceptron) network to forecast of water resources three and six month before the main stress months of July and August. Example data were extracted on an ultra-centenarian hydrological serial. The training and validation phases, confirmed by a ten fold cross validation methodology, led to a very satisfactory calibration of the ANN model, with errors in forecasting discharge values of just 5% (three months before) and 10% (six months before).

Figure optionsDownload as PowerPoint slide