Prediction of root zone soil moisture and flash drought at short lead times over the Narmada Basin using machine learning

Root zone soil moisture (RZSM) plays a crucial role in land–atmosphere interactions, agricultural water availability, and the antecedent moisture conditions during floods and droughts. Accurate short-term (7-15 days) prediction of RZSM is particularly important for the early detection of flash droughts, which develop rapidly during the monsoon season and pose significant risks to both rainfed and irrigated agriculture. However, most existing soil moisture prediction studies focus on surface soil layers, seasonal averages, and show limited skill in capturing rapid, sub-seasonal RZSM variability during the monsoon period, particularly at basin level. In this study, we investigate the spatio-temporal variability of RZSM over the Narmada River Basin, India, and develop deep learning-based models to predict RZSM anomalies at 7-day and 15-day lead times during the monsoon season (June-September). Multi-layer soil moisture observations are combined to estimate RZSM, and gridded daily precipitation and near-surface air temperature are used as predictors in a long short-term memory (LSTM) network trained in a grid-wise framework to capture both temporal persistence and spatial heterogeneity of soil moisture dynamics. Model performance is evaluated using spatial patterns of the coefficient of determination (R²), root mean square error (RMSE), and observed-predicted relationships across the basin. The predicted RZSM anomalies are further used to identify flash drought events based on rapid soil moisture depletion during the monsoon season. Results indicate robust predictive skill at 7 and 15 day lead times, with consistent spatial performance across the basin and improved detection of rapidly evolving drought conditions. The proposed framework highlights the utility of RZSM anomaly prediction for early flash drought monitoring and provides insights for adaptive irrigation planning and drought risk management in semi-arid river basins.