Abstract

Shallow groundwater in the Dongting Lake area is an important resource for domestic, agricultural, and industrial use, and its quality is essential for regional sustainable development and public health. Therefore, effective protection of this resource is urgently needed. In this paper, we integrate Positive Matrix Factorization (PMF) and Self-Organizing Map (SOM) machine-learning algorithms to conduct an in-depth analysis of the distribution, sources, and risks of toxic elements in shallow groundwater along the southern shore of Dongting Lake. The results indicate that Fe and Mn in the groundwater of the study area are at a severe pollution level, while As is at a light pollution level. The model analysis identified four pollution sources: natural sources (Fe, Mn) accounting for 31.33%, agricultural production (Zn) for 18.96%, traffic-mining mixed source (Pb, Cu, Cd) for 32.67%, and mineral dissolution-redox driven (As) for 17.04%. The average concentrations of Fe and Mn exceeded the standard limits. Although the carcinogenic metal Cd did not pose a health risk, the health risk value of As exceeded the maximum acceptable level, which requires serious attention. The PMF model quantified four potential sources of toxic elements, while SOM was used as a complementary nonlinear clustering tool to examine the consistency of the PMF-derived source contribution patterns. The integrated PMF–SOM framework, together with spatial distribution and geochemical evidence, improved the interpretability and robustness of source identification.

IPC Classification

Keywords