Unraveling spatial distribution and associated factors of bisphenol analogues in the Yangtze River Delta: A machine learning approach.

Bisphenol analogues (BPs) are widely detected in aquatic environments; however, comprehensive spatial characterization across large river systems remains limited, and their associated factors have rarely been quantitatively assessed. This study investigated the occurrence and spatial patterns of nine BPs across the Yangtze River Delta (YRD) using 190 surface water samples collected in September 2024 and applied advanced machine learning models to reveal the complex relationships between environmental factors and BP concentrations. At least one BP was detected in every sample, with the total concentration of ΣBPs ranging from 0.14 to 1081 ng/L (median: 28.3 ng/L). While BPA remained the primary pollutant, emerging substitutes like BPAF and BPS reached their highest concentrations in areas where BPA levels were relatively low, such as near Dianshan Lake. This spatial contrast suggests a possible transition in chemical usage patterns across the region. Pollution was generally lighter in northern Qingpu, Kunshan, and Songjiang, but remained high in traditional manufacturing hubs. Among the nine models, the XGBoost model achieved the best predictive performance (R2train = 0.96, R2test = 0.85). Model interpretation further revealed that socioeconomic indicators and land use were primary factors associated with these distribution patterns. The findings underscore the value of machine learning in revealing complex pollution processes and indicate the need for more targeted control of cumulative pollutant loads in regions with dense manufacturing activities.