Monitoring and predicting biodiversity is crucial for achieving biodiversity conservation and sustainable management. Traditional methods construct multivariate relationship models between the environment and biodiversity through field investigations. The development of spatial big data technologies, along with machine and deep learning algorithms, provides new perspectives and methods for exploring the environment-biodiversity relationship and predicting spatial patterns of biodiversity. In this study, we constructed prediction models based on the XGBoost algorithm, integrating plant diversity data from field surveys and environmental variable data from multiple sources databases. We developed models to predict the spatial pattern of plant diversity in the water-level fluctuation zone of Danjiangkou Reservoir by examining relationships between 34 environmental variables, including climate, topography, soil, hydrology, human activities, as well as species richness, species diversity, and phylogenetic diversity of plant communities. Additionally, we identified key environmental factors using the SHAP framework. Furthermore, we predicted the future spatial pattern of plant diversity in the water-level fluctuation zone of Danjiangkou Reservoir for 2050. The results showed that the XGBoost algorithm performed well in predicting plant diversity in the water-level fluctuation zones with better predictive ability for the phylogenetic diversity model compared to the species diversity model which had relatively lower predictive performance. Through SHAP analysis, we found that annual average flooding duration, human footprint, and mean temperature during the coldest season significantly influenced species richness, species diversity, and phylogenetic diversity of plant communities in the water-level fluctuation zone. The average annual duration of flooding had the most significant impact, with species richness, species diversity, and phylogenetic diversity decreasing as the average annual flooding duration increased. The interpretable prediction model constructed in this study effectively revealed spatial patterns of plant diversity in the water-level fluctuation zones, providing scientific evidence for conservation and sustainable management of biodiversity in these areas. It also offers a new method for biodiversity monitoring and management, which is of great significance for assessing the impacts of global changes on ecosystems and promoting biodiversity conservation.