Multi-model source apportionment and community-level ecological risk assessment of organophosphate esters in surface...

The ubiquitous distribution of organophosphate esters (OPEs) poses poorly quantified risks in coastal environments, where highly mixed pollution signals obscure accurate source identification and may underestimate community-level impacts. To address uncertainties in standalone receptor models, we employed a multi-model cross-validation approach (PCA-MLR, PMF, and Unmix) to quantify OPE sources in surface sediments from northern Liaodong Bay, China. Additionally, a robust framework utilizing 10 species sensitivity distribution (SSD) models and equilibrium partitioning evaluated individual and joint ecological risks. OPEs were detected across all sites (10.0-34.8 ng/g dw), with nine congeners showing detection frequencies ≥ 90%. While all models captured spatial variability (r² = 0.849-0.994, p < 0.001) and explained > 97% of total concentrations, their physical interpretability varied. Three consistent sources emerged: polyurethane foams (51.3%-59.4%), lubricants/wetting agents (20.9%-30.8%), and industrial polymers (17.4%-20.6%). PMF demonstrated superior robustness via its uncertainty-weighted algorithm, effectively overcoming PCA-MLR's negative-value artifacts and Unmix's geometric instability. Cross-validating these models extracted consistent signals, significantly reducing single-model prediction uncertainties. Multi-model SSD analysis revealed that arbitrary model selection introduces significant bias; optimal selection requires evaluating both statistical performance and ecological relevance. Risk characterization indicated negligible ecological impacts, with all risk quotients and hazard indices below 0.1. Utilizing multi-model comparative analysis alongside optimized SSDs significantly reduces apportionment bias, providing a reliable baseline for tracking OPE emissions and prioritizing mitigation in industrialized coastal systems.