Unveiling sources of organophosphate esters in marine environments utilizing multi-factor multi-modal high-dimensional clustering algorithm

Abstract

In marine environments, the sources of organophosphate esters (OPEs), particularly emerging OPEs (eOPEs) remain primarily unclear and present significant challenges for accurate source tracing. Here, we developed an unsupervised machine learning framework termed a multi-factorial multimodal high-dimensional clustering (MFM-clustering) algorithm to efficiently attribute source tracing of these pollutants. Our approach integrates physicochemical properties auch as log K_ow and log BCF, along with geographical data, to comprehensively represent the environmental behavior of these compounds beyond traditional concentration data. The robustness of the MFM-clustering algorithm was validated, offering enhanced pollutant classification accuracy compared to conventional statistical methods by focusing on pollutant-specific features. We used a systematic framework comprising field investigations, target screening, risk assessment, and MFM-clustering-based source analysis. The methodology was applied to the Bohai Sea, China, as a case study, where 29 OPEs, including 15 eOPEs, were quantified in sediment samples. This application refined the clustering analysis and enabled detailed ecological risk assessments. Industries associated with OPEs production, sewage treatment plants, industrial discharges, surface runoff from automotive activities, atmospheric transport of volatile OPEs, and petroleum-related operations for most eOPEs have been identified as key contributors to OPE pollution in various regions of the Bohai Sea. Our results highlight the necessity of tracing upstream production processes and identifying environmentally safer alternatives as effective strategies for mitigating OPE emissions.