Hydro-meteorological drought assessment using machine learning algorithms: a case study of Kabul River Basin

This study comprehensively investigated the hydro-meteorological drought in the Kabul River Basin by utilizing machine learning models. Drought is a complex and poorly understood natural hazard. It causes water shortages that ultimately impact the communities and degrade the ecosystem. The standardized streamflow index and rainfall anomaly index are employed in this study at different timescales to track the hydro-meteorological drought in the basin. The two machine learning models—extreme gradient boosting (XGBoost) and artificial neural network (ANN)—were trained on 80% data and then tested on 20% remaining data for the four hydrometric and nine meteorological stations. Both the models have given exceptional high accuracy in training and testing, but the ANN model came out to be the most accurate model, in training with R2 of 1.0000 and root mean square error (RMSE) of 0.0032 and in testing with R2 of 1.0000 and RMSE of 0.0031. Then both the models were predicted for both scenarios, SSP245 and SSP585, in which XGBoost came out to be the most accurate model in prediction with R2 of 0.9855 and RMSE of 0.1704. To explore further, the geographical distribution of meteorological drought was achieved by using the inverse distance weighting technique in the region.