Meteorological drought forecasting via soft computing models in Gharb perimeter (Northwest Morocco)

Abstract

Drought forecasting has implications for managing water and irrigation. Currently, with improved technology like artificial intelligence, forecasting can be more accurate. In this research, we chose standardized potential evapotranspiration index (SPEI) to characterize drought pattern. To achieve this, the data used was acquired from five meteorological stations in an irrigated Moroccan perimeter from 1976 to 2015. Besides, we predict SPEI at two scales (SPEI-3 and SPEI-6) with two inputs combination by exploring the capabilities of M5 pruned (M5P) and Light Gradient Boosting Machine (LightGBM), along with their hybrid model (LightGBM-M5P). To assess their effectiveness, we employed three statistical metrics (R2, MAE and RMSE). The findings indicated that using a larger time scale for analysis enhances the ability to forecast SPEI values more accurately. Moreover, the forecasting analysis revealed that M5P model demonstrated superior performance compared to the other studied models.