Abstract
Excitation-emission matrix (EEM) spectroscopy offers rapid and informative water monitoring, but its reliability is limited by chemical composition variability, which disrupts the relationship between fluorescence signals and contaminant concentrations. Recognizing this limitation, the lack of a robust and physically interpretable tool for assessing prediction reliability has become a critical bottleneck. In this work, the composition and photophysical inconsistencies among fluorescent compounds underlying the same fluorophore signal were identified as key sources of predictive inaccuracy. To detect these inconsistencies, fluorescence quenching was incorporated into EEM analysis with parallel factor analysis (PARAFAC). Apparent F(0)/F─the ratio of PARAFAC component intensity before and after extrinsic quencher addition─was proposed as an indicator for model failure and treatment anomaly detection. Validations with both model compound mixtures and real-world greywater samples showed that shifts in apparent F(0)/F reflect changes in the relationship between F(max) and target concentrations of total cell count (TCC) and dissolved organic carbon (DOC). Two practical tools were developed based on apparent F(0)/F: a clustering method for post hoc chemical composition analysis, and a thresholding method for outlier detection in real-time monitoring. This work highlights the added value of fluorescence quenching for improving the reliability and interpretability of EEM-based water monitoring at the subfluorophore level.