Abstract: Grasslands of central Inner Mongolia are a crucial component of ecological security barrier in northern China. By integrating field-measured quadrat data, remote sensing imagery, and environmental variables of grasslands in central Inner Mongolia, we developed aboveground biomass estimation models using machine learning algorithms, and generated high-resolution spatial distribution datasets for the period 2000-2020. We further analyzed the spatiotemporal variations and driving factors of aboveground biomass by the trend analysis and GeoDetector methods. The results showed that among multiple machine learning models, the gradient boosting machine (GBM) algorithm demonstrated optimal performance, with the coefficient of determination (R²), mean absolute error (MAE) and root mean square error (RMSE) being 0.58, 42.40 g·m^-2, and 56.99 g·m^-2, respectively. From 2000 to 2020, aboveground biomass showed a fluctuating upward trend, with a multi-year average value of 148.72 g·m^-2. Spatially, aboveground biomass displayed a pattern of low values in the northwest and high values in the southeast. Overall, 77.9% of the region experienced increases in aboveground biomass, while only 0.3% showed significant degradation. Factor detection revealed that annual precipitation, growing season precipitation, soil nitrogen, and soil organic carbon content were the primary drivers of spatial heterogeneity in aboveground biomass, and all interactions exhibiting enhanced effects. Our results could provide scientific basis for the management and sustainable development of grassland resources in central Inner Mongolia.

Key words: central Inner Mongolia, grassland aboveground biomass, machine learning, spatiotemporal change, driving factor