Microplastics (MPs, <5 mm) are a pervasive global pollutant, posing particularly acute challenges for soils in China. However, nationwide and systematically classified assessments of soil MPs across diverse land-use types remain limited. Here, we applied machine learning to 1497 spatially diverse soil samples across mainland China, revealing that over 72 % of sites exhibited moderate contamination (100 - 10,000 items/kg). Agricultural lands exhibited significantly higher MPs abundance (2966 items/kg) compared to forests (1979 items/kg), grasslands (1134 items/kg), community public domains (940 items/kg), and pristine ecological lands (191 items/kg). Notably, croplands with long-term plastic mulching exhibited MPs abundances approximately 6 times higher than those of non-mulched fields. Among agricultural land types, orchards emerged as a previously overlooked high-risk scenario, with a median MPs concentration of 2396 items/ kg, exceeding that in dryland and vegetable fields. We established an effective random forest model (Train = 0.96, Test = 0.84), which identified land-use patterns as the most influential driver of MPs variability, followed by soil sampling depth, while environmental variables such as solar radiation and wind speed contributed moderately. MPs pollution risks (PLIzone) were spatially heterogeneous, with Central China, South China, and Northeast China showing the highest ecological risk, while East and North China exhibited the lowest levels. This large-scale, data-driven analysis provides a robust scientific foundation for developing targeted, regionspecific strategies to mitigate soil MPs pollution across China's diverse landscapes.