Interpretable soil moisture prediction based on the SABO-CONV1D-BiLSTM model on the Tibetan Plateau

The limited regional adaptability of soil moisture prediction models constrains their application under complex climatic conditions. Enhancing modeling accuracy and predictive capability is crucial for improving the precision of climate simulations and the effectiveness of extreme weather early warnings. This study proposes an interpretable and generalizable soil moisture prediction approach that employs the SABO algorithm to optimize CONV1D-BiLSTM model. The model's performance was evaluated and validated at Linzhi, Dangxiong, Mozhugongka, and Xietongmen in the southwestern Tibetan Plateau. Results indicate that, across four distinct environmental settings, the proposed model achieved an average root mean square error (RMSE) of 1.859 and an average coefficient of determination ( R 2 ) of 0.929. Furthermore, using the SHapley Additive exPlanations (SHAP) method, soil temperature and relative humidity were identified as key features across multiple stations. This method enhances the understanding of soil moisture dynamics in the context of climate change and provides a powerful tool for climate risk assessment and early warning on the Tibetan Plateau region.