Accurate identification of sludge contamination sources by classification-based PMF and machine learning with consideration of sewer network distribution differences

Abstract

The application of source identification such as PMF for large-scale pollution source analysis frequently produces ambiguous outcomes. In this study, we utilized a classification-based method to accurately track key pollution sources in the sludge. In the study, we categorized the wastewater treatment plants into two groups: T1 and T2, according to the pipeline network. T1 sewage treatment plants are the main sewage plants in urban areas, covering a large area and connected to industrial wastewater treatment plants for secondary treatment. T2 sewage treatment plants are typically smaller in size and usually responsible for treating sewage in rural or township areas. The PMF analysis indicates that industrial pollution sources contribute 3.4 times more to T1 sludge than to T2 sludge, making industrial pollution the primary factor causing the disparity. The application of Random Forest and Adaboost based on pollutant concentrations for classification and fitting of sludge resulted in the identification of the main pollutants: Zn, Cu, Ni, and Cyanide, which align with characteristic pollutants from the electroplating industry. The GIS analysis shows a significant correlation between the distance of wastewater treatment plants with abnormal environmental risk and electroplating industrial parks, all within a 20 km radius. Indeed, when conducting large-scale pollution source identification studies, utilizing classification-based analysis can effectively improve the accuracy of pollution source identification, leading to more valuable analysis results.