Meteorological-to-hydrological drought propagation and its drivers in the Upper Indus Basin, Pakistan: A multiscale analysis using explainable machine learning

Study region The Upper Indus Basin of Pakistan is a critical headwater region for the country’s water supply. Study focus This study investigated meteorological and hydrological drought characteristics and the propagation of meteorological drought into hydrological drought in the Upper Indus Basin of Pakistan during 1982–2022. Meteorological drought was quantified using the Standardized Precipitation Index at 1-, 3-, 6-, 9-, and 12-month scales, and hydrological drought using the one-month Standardized Runoff Index. Drought propagation time was estimated using the maximum Pearson correlation coefficient method, while rate, counts, and intensity were derived from event-based analysis, and an explainable machine-learning framework (XGBoost-SHAP) was used to identify the dominant drivers. New hydrological insights for the region Meteorological drought was most frequent at short time scales, whereas hydrological drought was less frequent but more persistent. Mean propagation time was 5.84 months, with 31.61% of the basin showing a one-month response. Mean propagation counts and rate were 18.25 and 0.32, respectively. The model performed well, with a mean R² of 0.75 and a maximum of 0.84 for propagation intensity. Vegetation condition, temperature, and runoff were the dominant drivers of propagation time, whereas temperature, runoff, and potential evapotranspiration mainly controlled propagation counts and intensity. By integrating multiscale drought propagation metrics with explainable machine learning, this study provides new regional evidence to support drought monitoring and climate adaptation.