Understanding carbon dynamics in China's forests is essential for achieving national carbon neutrality goals, yet large-scale data on key processes such as litter decomposition remain scarce. To address this gap, we integrated 1003 field observations with 39 environmental variables into an interpretable machine learning framework (XGBoost, R2 = 0.68) to generate China's first 1-km resolution map of forest litter decomposition rates. Our model identifies a hierarchical network of controlling factors, reaffirming the dominant role of temperature and revealing a distinct nonlinear threshold for Mean Annual Precipitation (MAP). Decomposition rates peak at 1000-1800 mm of MAP before declining in hyper-humid regions. This climatic threshold, influenced by vegetation type, creates a biogeographic gradient of decomposition from the warm, humid southeast to the cold, arid northwest. Projections under future climate change scenarios reveal divergent vulnerabilities: temperate broadleaf forests in North China emerge as hotspots for accelerated decomposition, increasing the risk of soil carbon loss, while arid regions in Northwest China show a deceleration of decomposition, suggesting shifts in nutrient cycling dynamics. By quantifying these key nonlinear controls and spatial vulnerabilities, this study provides a data-driven foundation for climate-adaptive forest management across China.