Risk identification and tracing of heavy metals in rice fields: An integrated framework of machine learning, Google...

Heavy metals in rice have raised public concern due to health hazards. To achieve effective risk identification and tracing of heavy metals in rice fields on a regional scale, this study innovatively integrated machine learning, Google Earth Engine (GEE), geochemical modeling, and pollution fingerprint analysis, aiming to improve target field recognition, speciation prediction, and source apportionment in heavy metal risk assessment. GEE coupling random forest (RF) model achieved to accurately identify rice fields (overall accuracy, 0.98), facilitating the zoning for risk assessment. A multi-surface speciation model (MSM) coupled with RF model further identified speciation distribution, bioavailability, and leaching risks of heavy metals in rice fields, and simulated the adsorption-desorption behavior of heavy metals with soil pH changes. Fulvic acid (FA) can improve rice uptake of heavy metals, while dissolved organic matter (DOM) in soils contribute to the leaching potentials of heavy metals, especially Pb and Cu. Positive matrix factorization (PMF)-RF integrating approach identified the sources of heavy metals, which were mainly from soil parent material, fertilizer application, traffic pollution, and industrial exhaust with relative contributions of 27.8 %, 24.5 %, 26.2 %, and 21.5 %, respectively. The integrated framework for risk identification and tracing in this study may inform regional-scale agricultural pollution management.