Early-Season Streamflow Prediction in the N’fis Basin (Morocco) Using Teleconnection indices and Machine Learning

Anticipating early-season streamflow is essential for water management in semi-arid basins where reservoir decisions remain largely reactive. In the N’fis Basin (Morocco), we investigate whether large-scale climate signals, combined with machine-learning methods, can improve short-lead streamflow outlooks. Using monthly observations from 1982–2021, we evaluate three approaches—Random Forest (RF), Partial Least Squares Regression (PLSR), and Multiple Linear Regression (MLR)—for lead times of one to three months (t+1 to t+3). Predictor selection is based on correlation analysis and multicollinearity diagnostics, and model skill is assessed through RMSE and R². Streamflow anomalies are expressed using the Standardized Streamflow Index (SSI), which provides a normalized measure of hydrological drought directly linked to water availability. Results show that incorporating climate indices improves early identification of low-flow conditions relative to persistence-based benchmarks. Predicted SSI anomalies capture major drought periods, demonstrating the value of climate-informed models for anticipatory reservoir management. These findings could support the potential development of forecast-informed reservoir operations (FIRO) in the region, contributing to more proactive drought forecasting.