An integrated framework coupling process driven model and machine learning to simulate soil heavy metal(loid)s...

Existing approaches for assessing soil heavy metal(oid)s (HMs) accumulation at the regional scale remain limited in their ability to consistently link spatiotemporal input and output processes and to accurately quantify fluxes without relying on extensive environmental monitoring data. To address these limitations, we developed an integrated modeling framework that couples a bottom-up multi-source emission inventory, the AERMOD atmospheric dispersion model, and a HYDRUS-based water-solute transport module, further extended with established empirical approaches to quantify output processes, enabling long-term, spatially explicit simulation of soil HMs dynamics, with a machine learning module incorporated to calibrate process-derived net HMs fluxes. The framework was applied to an industrial agglomeration located in eastern coastal Zhejiang province, China. Results indicated that, with rapid industrialization and urbanization, atmospheric deposition inputs have steadily increased, whereas agricultural inputs have declined, with high-deposition zones concentrated within 10 km away from the point source. Simulation of output processes revealed that leaching and soil erosion were the dominant pathways of HMs loss, while runoff fluxes exhibited a gradual increase in response to growing impervious surface. A pronounced decline in soil environmental carrying capacity was predicted after 50 years, with overload zones expanding substantially, particularly for As (30.82 %), Cu (31.04 %), Cd (19.96 %), Pb (28.48 %) and Hg (32.66 %). Risks were dominated by As and Cu in forestland due to their high ecotoxicity, while the carrying capacity declined sharply in cultivated and residential land owing to receptor sensitivity and intensive human activities. This framework provides a robust tool for tracing emission sources, elucidating transport and accumulation dynamics, and supporting long-term risk zoning and early warning for regional environmental management.