Machine learning-based source apportionment and source-oriented probabilistic ecological risk assessment of heavy metals in urban green spaces.

Global urbanization has significantly contributed to soil contamination by heavy metals (HMs), posing serious ecological risks, particularly within urban green spaces (UGS). This study focused on UGS soils in Lanzhou, a major river-valley city in China. Multiple pollution indices, including geo-accumulation index (I), enrichment factor (EF), and Nemerow integrated enrichment factor (NIEF), were combined with Monte Carlo simulations (MCS) to assess probabilistic contamination levels. Machine learning methods, including SOM super-clustering and random forest (RF), were integrated with positive matrix factorization (PMF) to quantify the sources of soil HMs. Ecological risk index (RI) was combined with MCS analysis and PMF model to apportion the source-oriented probabilistic ecological risks. Results showed that the average concentrations of Cd (0.38 mg kg), Cu (35.51 mg kg), Hg (0.07 mg kg), Pb (34.59 mg kg), and Zn (130.58 mg kg) exceeded local soil background values, except for As (8.56 mg kg), Cr (62.77 mg kg), and Ni (27.68 mg kg). Notably, exceedance rates for Cd, Hg, Pb, and Zn were 90.91 %, 94.95 %, 80.81 %, and 87.88 %, respectively. Elevated concentrations, particularly of Zn, Cd, Pb, and Hg, displayed distinct spatial patterns linked to industrial activities and urban development. Overall contamination reached moderate levels, primarily driven by Cd and Hg. Source apportionment identified traffic emissions, industrial activities, and coal combustion as the principal HM sources. Probabilistic ecological risk assessment confirmed that Cd and Hg pose the greatest ecological risks, primarily stemming from industrial activities and coal combustion. These findings provide important insights for developing source-specific remediation to mitigate and manage HM pollution in urban green spaces.