Leveraging artificial intelligence and statistical models for predicting dam water storage: insights from a Moroccan case study

This study addresses sustainable water management challenges by proposing a forecasting approach applied to daily storage series from 45 Moroccan dams (2015-2023). Three modelling frameworks are compared: classical statistical models (autoregressive integrated moving average, ARIMA), machine learning models (support vector regression, SVR, and long short-term memory networks, LSTM), and hybrid models. Stationarity was tested using the Augmented Dickey-Fuller (ADF) test, revealing that 64% of the series are non-stationary. Trend analysis showed declining volumes in 46.7% of dams. Models were trained using five-fold cross-validation and optimised via grid search. Performance was assessed using mean absolute error (MAE) and root mean squared error (RMSE). ARIMA and SVR with linear kernel outperformed other models, suggesting mainly linear behaviour, possibly driven by anthropogenic influences. These results confirm the relevance of simple and interpretable models in semi-arid contexts, supporting strategic planning under natural and human-induced pressures.