Integration of machine learning and fluorescence-based Bayesian mixing model for real-time source apportionment in urban stormwater drainages

Yuansi Hu, Zhiwei Jiang, Shiyi Lei, Kailei Li, Wenwen Li, Dongdong Gao, Yu Xiang, Han Zhang,
Integration of machine learning and fluorescence-based Bayesian mixing model for real-time source apportionment in urban stormwater drainages,
Process Safety and Environmental Protection,
Volume 202, Part A,
2025,
107706,
ISSN 0957-5820,
https://doi.org/10.1016/j.psep.2025.107706.
(https://www.sciencedirect.com/science/article/pii/S0957582025009735)
Abstract: Urban expansion has intensified pollution from structural defects in stormwater drainage systems, such as illicit discharges and misconnections. Wastewater intrusion is diffuse, concealed, and sporadic, hindering timely source identification and real-time management by traditional monitoring. Based on the fluorescence fingerprint information of dissolved organic matter (DOM) in urban stormwater drainage systems, this study proposed a real-time quantitative source apportionment model by coupling machine learning with a fluorescence excitation-emission matrix (EEM)-based end-member mixing model, which is termed the ESML model. A virtual sensor was developed using machine learning algorithms to predict fluorescence signatures from eight online water quality indicators, addressing limitations of real-time EEM analysis. The model subsequently integrated EEM data with a Bayesian mixing framework to achieve accurate source apportionment within urban drainage systems. According to the results of machine learning performance evaluation and multivariate statistical analysis, the fluorescence peak ratio of humic-like to fulvic-like substances (Peak C/Peak A, C/A) and the humification index (HIX) were identified as the optimal parameter combination for pollution source apportionment. The ESML model quantifies contributions of four pollution sources across urban drainage systems. This study offers a practical approach for smart drainage management and precise pollution control.
Keywords: Stormwater drainage system; Dissolved organic matter; Fluorescence characteristics; Source apportionment; Machine learning