Abstract
Heavy metal contamination in agricultural soils is a critical global concern, threatening ecosystem safety and food security. The wheat–corn rotation system, vital for food production in regions like Northern China, is particularly vulnerable. However, comprehensive studies investigating vertical migration, future dynamics under climate change, and predictive modeling of heavy metals within this system are still limited. This study combined field sampling of soil profiles (0–200 cm) with geochemical modeling (the PROFILE and SSCL models) and machine learning techniques (Multiple Regression, Neural Networks, and Random Forest). Key findings revealed that atmospheric deposition was the primary input source for most heavy metals, contributing 49.50–93.27%. The release rates (Rm) of heavy metals were significantly higher during the corn season than the wheat season and are projected to increase by 1.2–1.5 times under the RCP4.5 climate scenario. Vertical distribution analysis showed a significant accumulation of heavy metals in the middle soil layer (20–120 cm), with Arsenic (As) and Cadmium (Cd) exhibiting the strongest migration potential, posing a threat to groundwater. The Random Forest model demonstrated superior performance (R2 > 0.95) in predicting heavy metal behavior, identifying Fed and soil TOC as the dominant controlling factors. This study provides a unique and significant contribution by integrating geochemical fate modeling with climate projections and advanced machine learning to offer a predictive, multi-faceted risk assessment framework, thereby supplying a scientific basis for targeted pollution control and sustainable soil management in wheat–corn rotation systems under a changing climate.