National mapping of compound soil erosion across China: Integrating 137Cs tracer technique and random-forest machine learning

Soil erosion remains a pervasive global environmental challenge, with spatial mapping serving as a cornerstone for conserving regional soil and mitigating land degradation. Current mapping efforts predominantly focus on single types of erosion (e.g., erosion by wind or water), leaving a large void in regional-scale assessments of compound soil erosion. Through a dataset of 137 Cs-derived erosion rates extracted from nearly 1000 research articles, the first spatial distribution map of China's compound soil erosion was produced via a random-forest machine-learning approach, elucidating its primary driving factors. The modelled national average rate of soil erosion was 17.8 t ha −1 y −1 , with a total annual soil loss of approximately 21.4 Pg y −1 , a finding that challenges traditional estimates derived from single erosion type. Spatially, higher values were found in the northern and northwestern regions, whereas lower values dominated the eastern and southern areas. Soil, climate, and vegetation were the dominant controllers, contributing 30.2, 28.6, and 25.8%, respectively, while topography and anthropogenic activity each accounted for 7.7%. Specifically, soil organic matter, runoff, silt content, standard deviation of the normalized difference vegetation index, wind speed, and population density were identified as the six most influential predictors. The impacts of these factors exhibit threshold effects and spatial heterogeneity, indicating that distinct mechanisms govern soil erosion across different regions. These findings provide critical support for understanding the spatial distribution of compound erosion and offer a scientific basis for formulating macro-scale erosion mitigation policies in China.