A classification-based approach to mapping particulate organic matter (POM) in inland water using OLCI images

Abstract

Particulate organic matter (POM) plays a major role in freshwater ecosystems by serving as a bridge for the conversion of various nutrients. The composition and sources of POM in inland lakes are complex, making it difficult to estimate its concentration accurately via remote sensing. Therefore, a classification-based method based on the sources and composition of POM is proposed for estimating POM concentrations in inland lakes. In this study, 379 samples were collected from ten lakes in the Yangtze River Delta (YRD) at different times. A water-type classification method based on OLCI [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] was developed for POM estimation based on biological and optical characteristics. Water type 1 is relatively clear, and POM may originate from aquatic vegetation or sediment. Water type 2 was dominated by inorganic suspended matter, and POM mainly originated from the attachment and entrainment of inorganic minerals. Water type 3 is an algae-dominated water body, and POM is mainly derived from fresh algal particles and the microbial degradation of phytoplankton. Therefore, specific POM estimation algorithms were developed for each water type. OLCI [Formula: see text], [Formula: see text], [Formula: see text], and [Formula: see text] were used for water type 1; [Formula: see text], [Formula: see text], and [Formula: see text] were adopted for water type 2; and [Formula: see text], [Formula: see text], and [Formula: see text] were selected for water type 3. Using an independent dataset to evaluate the estimation accuracy of the developed algorithm, the results show that the estimation performance of this algorithm is significantly improved compared to the two other algorithms used; the mean absolute percentage errors (MAPE) decreased from 72.56% and 52.21% to 32.61%, and the root mean square errors (RMSE) decreased from 3.05 mg/L and 2.24 mg/L to 1.75 mg/L. A random error analysis of the atmospheric correction demonstrated that this algorithm is robust and can still perform well within a random error of 30%. Finally, this method was successfully applied to map the POM concentrations in the YRD using OLCI images acquired on November 12, 2020.

Keywords: Classification estimation; Particulate organic matter; Remote sensing; Sources and composition.