Estimating CDOM Concentration in Highly Turbid Estuarine Coastal Waters

Abstract

AbstractWe proposed a support vector machine (SVM) algorithm to retrieve colored dissolved organic matter (CDOM) concentration (using ag(443) as a proxy) in the highly turbid Pearl River estuary. Two band ratios, namely, Rrs(443)/Rrs(547) and Rrs(488)/Rrs(547), were used as inputs. Comparisons between the estimated and measured ag(443) illustrated high accuracy of the SVM algorithm, yielding R2s of 0.98 and 0.89, mean absolute percentage errors of 5.18% and 13.1%, and root‐mean‐square deviations of 0.012 and 0.034 m−1 for the training and validation data sets, respectively. The SVM algorithm was also evaluated against existing ones for the study area and gave the best accuracy with a R2 of 0.9, a mean absolute percentage error of 10.23%, and a root‐mean‐square deviation of 0.025 m−1. The diurnal dynamics of CDOM in the Pearl River estuary was revealed showing complicated variations and influenced by the combined effects of wind, tide, circulation, and river discharge. As for remote sensing applications, the SVM‐based CDOM product exhibited great potential to trace the Pearl River plume and the satellite‐derived plume area agreed well with the FVCOM model simulation result. SVM is an accurate and fast tool for retrieving CDOM concentration, especially in highly turbid estuarine coastal waters, and thus, river plume dynamics can be traced.