Machine learning-based source apportionment and nonlinear associations of influencing factors on potentially toxic elements in floodplain profiles of Pearl River Delta, China.

Floodplains function as sinks for sediments and potentially toxic elements (PTEs), where PTEs contamination is particularly pronounced. Assessing environmental risks and sources of PTEs in floodplain sediments is essential for regional sustainability. However, precisely identifying PTEs sources and quantifying nonlinear associations of influencing factors remain challenging. In this study, sediment profiles were collected from floodplains of Pearl River Delta (PRD), and probabilities of biological effects and overload risks of PTEs were evaluated using sediment quality guidelines (SQGs) and environmental capacity risk (ECR) method. Furthermore, RF-SHAP and t-SNE-RF-SHAP models were constructed to elucidate dominant factors governing PTE distribution and their nonlinear associations, and to apportion pollution sources. Results indicated that more than 50% of floodplain sediment samples from West River (WR) and North River (NR) exhibited severe ecological risk from PTEs, notably Cd, As, Cu and Ni in WR, and Cd, As, Pb and Ni in NR. Mn and Corg are primary controlling factors for most PTEs in WR and NR, respectively, and most factors exhibited nonlinear associations. Natural sources constituted the predominant contribution in both WR (48.91%) and NR (47.29%), followed by a mixed source of mining and transportation activities (31.15%, 37.10%). Taking Cd as an example, contribution of natural sources decreases from upper to lower reaches, by approximately 77 and 48 percentage points for WR and NR, respectively. Whereas mixed mining-transportation source contributed slightly more to surface sediments in middle and lower reaches of study area compared to upper reaches. These differences in source contributions across sediment profiles reflect spatial heterogeneity of material inputs within watershed. This study provides a theoretical foundation for establishing pollution control priorities and optimizing monitoring strategies in floodplain ecosystems.