Prediction of 3D spatial distribution, driving factors, and risk for heavy metal(loid)s in sediment of Dongdagou River based on ensemble learning.

Sediments contaminated with heavy metal(loid)s can cause harm to the environment. This study proposed the use of machine learning methods to predict, classify, and identify the three-dimensional distribution of heavy metal(loid)s in sediment. Based on the 1423 sampling data from Dongdagou, a total of 8 metal(loid)s and 18 covariates were used to train and test the model. The predictive performance of 2 traditional methods and 4 machine learning methods on the spatial distribution of heavy metal(loid)s was compared. The results demonstrated that the ensemble random forest model yielded a satisfactory performance (R² = 0.85). The 3D analysis revealed that heavy metal(loid) contamination in the sediment was concentrated at the upstream source and midstream areas, mainly within the top 2.2 m of the shallow subsurface. Finally, the pollution in the sediment was divided into four levels using the K-means clustering algorithm as a reference for remediation priority. Significant factors influencing heavy metal(loid) distribution included sediment texture (0.167) and pH (0.154). The Bayesian analysis showed that the pollution risk was below 20 % at depths below 2.8 m. In summary, this study provides guidance for the research and analysis of heavy metal(loid) distribution in sediment.