Predicting meteorological drought in the Kabul River Basin using machine learning techniques and climate change projections

ABSTRACT Droughts on water resources and agricultural sectors can cause intense economic losses and loss of human life. In this analysis, the Standardized Precipitation Evapotranspiration Index (SPEI) from 1981 to 2022 is used for the prediction of drought conditions in the Kabul River Basin. The study is done with respect to short-term and long-term conditions, which prevail for SPEI-3 and SPEI-6, respectively. The dataset is split into training and testing subsets to provide approximately 80% for training and 20% for testing various models of machine learning, with k-nearest neighbors (kNN), extreme gradient boosting, AdaBoost, decision tree, random forest (RF), CatBoost, gradient boosting, and neural networks (NNs). The best model, extreme gradient boosting, presented an R2 of 0.98 with an RMSE of 0.12 from the testing dataset, while the worst model, RF, achieved an R2 of 0.69 with an RMSE of 0.51. The study further proceeded to give future drought predictions under SSP245 and SSP585 scenarios. Spatial distribution mapping by means of the inverse distance weighting (IDW) method in ArcMap showed continuous moderate drought occurrence in the basins. Results of drought forecast compare the potential of SPEI in line with being a useful tool for drought forecasting.