Abstract: To address the threat of seasonal drought to rubber plantation production in tropical monsoon climate regions, this study investigated the soil moisture dynamics and meteorological driving mechanisms of typical rubber plantations in Xishuangbanna. Based on hourly monitoring data of soil moisture content at 0-80 cm depth and meteorological factors, prediction models were constructed using Extreme Gradient Boosting (XGBoost) andRandom Forest (RF) algorithms. The results showed that: Soil moisture content exhibited significant seasonal variations, with moisture content during the dry season being 3.75%-14.72% lower than that during the rainy season, and the fluctuation amplitude was greatest in the 0-20 cm layer. SHAP analysis revealed a seasonal transition in driving mechanisms: the rainy season was dominated by 7-day rainfall frequency (with a maximum contribution rate of 39.1%), while the dry season was dominated by daily average temperature (with a contribution rate of approximately 33.5%), reflecting the transition of soil moisture from rainfall supply to evaporation consumption. Both models demonstrated excellent predictive performance in shallow soil layers (0-20 cm) (R² > 0.94), with prediction accuracy exhibiting a V-shaped pattern with depth. XGBoost showed significant advantages in deep soil layers (50-80 cm), demonstrating stronger applicability for short-term soil moisture prediction in rubber plantations. The purely meteorology-driven machine learning models can achieve hourly-scale accurate prediction of soil moisture in rubber plantations, providing a scientific basis for precision irrigation and drought early warning in rubber plantations.