Combining random forest and XGBoost models for source apportionment and health risk assessments of heavy metals in suburban farmland soils

With the rapid development of industry, heavy metal pollution in farmland soil environments has become a common concern globally. To explore the environmental risks in farmlands, three typical heavy metals (Hg, Pb and Cd) in 232 surface soil samples (0-20 cm) from representative farmland areas in Midong District, Urumqi city, Xinjiang, were comprehensively investigated. The results revealed the highest soil heavy metal content for Pb (32.7 mg/kg), followed by Cd (0.74 mg/kg) and Hg (0.41 mg/kg). The average Hg and Cd concentrations in the rice and sweet potato cultivation areas were 0.071 and 0.21 mg/kg and 0.065 and 0.21 mg/kg, respectively, which exceeded the background values in Xinjiang (0.017 and 0.12 mg/kg, respectively). The average concentrations of the three heavy metals in the cotton and corn cultivation areas all exceeded the Xinjiang background values. The soil pollution level was evaluated via the pollution load index and Nemerow composite index. The results indicated varying soil Pb (slight: 65.52 %), Cd (moderate: 31.03 %) and Hg (severe: 1.72 %) pollution levels. The risk of heavy metals in soil was quantified via the human health risk assessment and Monte Carlo simulation methods, thereby calculating the risk of soil heavy metals to human health. In human health risk assessment, the overall noncarcinogenic risks of adults and children were low, whereas the overall carcinogenic risk of children was greater than that of adults. The Monte Carlo simulation results revealed that Pb yielded the lowest overall cancer risk in children and adults, whereas Cd mots notably influenced the carcinogenic risk in children. Among the four crops, the noncarcinogenic and carcinogenic risks at the corn planting sites were greater than those at the other planting sites. The average total carcinogenic risk (TCR) values at the sweet potato, cotton and rice planting sites all exceeded 1E-06. Combining the random forest and XGBoost models, the following three heavy metal sources were determined through principal component analysis: F1 (anthropogenic activities), F2 (industrial activities), and F3 (long-term phosphorus fertilizer use). Overall, this study emphasized the importance of monitoring heavy metal pollution in farmland soil, providing a sound basis for effectively managing heavy metal pollution in farmland soil and ensuring farmland soil quality maintenance.