Quantifying contaminant concentration in mixed water systems using spectro-polarimetric measurement

Water pollution is a physicochemical change in water bodies that threatens water quality and availability. Traditional in-situ measurements face several challenges and often fail to capture the variability across ecosystems. Remote sensing with multispectral imagery, leveraging surface water reflectance and physicochemical properties, offers a promising alternative. However, accurate spectral decomposition remains crucial because of the optical complexity of natural waters and factors such as specular reflection, wave effects, and depth variations. Here, we demonstrate the integration of spectral measurement with polarization to develop an inverse model that improves the estimation of contaminant concentrations under varying water surface conditions. The proposed model achieves r² values of 0.86 for estimating chlorophyll (Chl) and suspended sediment (SS) contaminants, with RMSE <10%, outperforming commonly used indices. These findings indicate that the currently available indices are insufficient for optimizing contaminant estimation. Additionally, two-band indices demonstrate better performance than three-band index, highlighting the need for a balanced approach in selecting the number of spectral bands. However, the model exhibits limitations when applied to mixed waters with equal contaminant concentrations, yielding an error of 35% for Chl estimation and over 100% for SS estimation. Furthermore, the study demonstrates that viewing angles and wave conditions significantly affect the accuracy of the water quality predictions. These results highlight the potential of integrating polarization measurements with regression modeling to achieve more accurate water quality assessments, particularly under dynamic conditions such as wave environments. The findings also suggest a need for further refinement in addressing challenges associated with mixed contaminants, specular reflection, and particle distribution.