Machine learning uncovers dominant fractions of heavy metal(loid)s in global soils

Global heavy metal(loid) (HM) pollution in soils threatens food security and human health, yet HM mobility—governed by chemical fractions—remains poorly understood at large scales. Here we develop an eXtreme Gradient Boosting model for classifying soil HM dominant fractions using a globally compiled dataset of 9489 field observations. We show that organic carbon and pH positively affect high-mobility of HMs in soil. Using mercury as a case study, we identify both known and previously unreported high-mobility hotspots. The global map indicates that mercury has high mobility across 17.85% of global regions. Combining population and cropland distribution maps with model predictions shows that an estimated 15.1 million people and 100.9 million hectares of farmland are situated in high-mobility regions, with Asia being disproportionately impacted. This study facilitates the efficient identification of dominant HM fractions on a large scale, providing a critical foundation for developing targeted HM stabilization/solidification strategies. Dominant fractions of soil-heavy metal(loid)s are mapped globally using a machine learning model trained on 9489 field measurements. Using mercury as a case study, the model shows that soil mercury exhibits high mobility across 17.8% of global regions.