A two-step approach to mapping particulate organic carbon (POC) in inland water using OLCI images

Abstract

Particulate organic carbon (POC) plays an important role in the carbon cycle in freshwater ecosystems, and estimating the POC concentration with remote sensing is a useful method of analyzing the spatiotemporal distribution of POC in inland waters. However, the complex origins of POC in inland waters require the application of remote sensing methods at large scales, and these methods are often challenging to implement. Therefore, a two-step method based on classification was developed to estimate the POC concentrations in inland water bodies. A total of 331 samplings were collected from 7 inland waters, Taihu Lake, Chaohu Lake, Dianchi Lake, Dongting Lake, Hongze Lake, Hengshui Lake, and the Jiajiang River, at different times. The POC concentration and other optical parameters were analyzed. A determination method using OLCI bands 6 and 11 was proposed to identify the POC originating from phytoplankton and from other origins. As a result, the water was classified into two water types. In water type I, the POC was mainly derived from phytoplankton, and in water type II, and the POC mainly originated from non-algal materials. Therefore, specific POC estimation algorithms were developed for different water types. OLCI bands 8, 12 and 16 were used for water type I, and bands 8, 11 and 12 were applied for water type II. An estimation accuracy comparison with two unclassified estimation methods, semi-analytical algorithm and empirical algorithm using an independent dataset was conducted, and the results showed that the MAPE decreased to 27.25% from 55.56% and 60.87%, and the RMSE decreased to 2.13 mg/L from 3.33 mg/L and 3.20 mg/L. The accuracy assessment demonstrated that the estimation performance of the two-step method was significantly improved. Finally, this method was validated using two OLCI images acquired on December 8th, 2016, and July 24th, 2017, for mapping the POC concentration in Hongze Lake and Taihu Lake.