Land Use and Water Storage Dynamics in the Krishna River Basin: Insights from Satellite Observations and Machine Learning

Abstract. Water scarcity and recurrent droughts threaten agricultural productivity and water security in the semi-arid Krishna River Basin of southern India. This study integrates satellite observations and machine learning techniques to assess long-term terrestrial water storage (TWS) and drought dynamics from 1992 to 2022. The Extreme Gradient Boosting (XGBoost) algorithm was employed to reconstruct GRACE-based Terrestrial Water Storage Anomalies (TWSA) using precipitation, temperature, evapotranspiration, and soil moisture as predictors. The reconstructed TWSA showed strong consistency with GRACE/GRACE-FO data (R2 = 0.92, RMSE = 43.18 mm), extending the GRACE record to earlier decades. The GRACE Drought Severity Index (GRACE-DSI), applied at a 3-month scale, identified 15 major drought events during 1992–2022, with the most severe occurring in 2015–2017 (minimum DSI = −2.0) and 2018–2019 (minimum DSI = −2.63). Droughts typically recurred every 5–7 years, showing increased intensity after 2010. Land use and land cover (LULC) analysis from ESA-CCI data revealed declining agricultural areas and shrublands, alongside expansion of forests and urban land. Correlation between LULC changes and TWSA was weak, indicating stronger climatic control on basin water storage. The study underscores the value of fusing remote sensing, statistical tools, and hydrological indices to support better monitoring and governance of land and water systems in drought-prone basins.