Speaker
Description
The Souss-Massa basin, located south of the High Atlas Mountains, is a semi-arid region highly exposed to climate variability and extreme drought events. This study proposes an integrated, data-driven framework that combines multivariate statistical analysis and machine learning to analyze spatio-temporal precipitation patterns (1940–2023) and predict drought risk under future climate scenarios.
Principal Component Analysis (PCA) was employed to extract dominant rainfall patterns and reduce dimensionality, while the Standardized Precipitation Index (SPI) was used to quantify drought severity at multiple time scales. A Partial Least Squares (PLS) regression model demonstrated high predictive performance, showing strong agreement with historical drought events (R² = 0.99; RMSE = 0.039).
To enhance robustness and capture potential non-linear relationships, a Random Forest Regression model was also tested. All analyses were conducted using Python on Google Colab, ensuring reproducibility and scalability. The combined approach offers practical insights to support water resource management, drought risk mitigation, and climate change adaptation in vulnerable arid regions.
