Regression Models Enhance Fluorescence Spectra for Smart Surface Water Surveillance

Over the past three decades, fluorescence spectroscopy has been conventionally interpreted through peak picking, fluorescence regional integration, and parallel factor analysis to analyze aquatic dissolved organic matter. However, there is a growing need for advances in analytical toolkits to unlock the full potential of fluorescence spectra to tackle pressing challenges in smart water surveillance such as real-time surface water monitoring and wastewater source tracing. To this end, we established two types of easily implementable regression models. Through weighted linear regression (WLR), we constructed a novel correlation map for fluorescence excitation-emission matrices (EEMs) and dissolved organic carbon (DOC) based on 191 surface water samples from diverse aquatic environments. This map reveals that humic-like fluorescence intensity (FI) at excitation/emission wavelengths of 300-380/440-490 nm serves as a reliable indicator of aquatic DOC. In addition, a multivariable linear regression (MLR) model was developed to identify pharmaceutical wastewater blended into diverse surface water matrices, achieving fitting errors of 10-20%, despite fluorescence quenching at short excitation wavelengths. This work demonstrates that the regression-based toolkits developed herein can advance the application of fluorescence spectra for protecting aquatic environments.