Satellite estimation of dissolved organic carbon in eutrophic Lake Taihu, China

Dissolved organic carbon (DOC) in lakes serves as a substrate for heterotrophic bacterial growth, a regulator of the global carbon cycle, and a light absorption agent. DOC in eutrophic lakes is greatly influenced by phytoplankton phenology and terrigenous input by rivers. Therefore, it is necessary and significant to dynamically monitor the concentration, storage, and riverine exchange flux of DOC. By using in-situ DOC measurements from 2004 until 2018 (N = 2019), a machine learning algorithm, namely, a multilayer back-propagation neural network (MBPNN) model, was developed in this work to improve the remote sensing estimation of DOC concentrations in eutrophic Lake Taihu. The model yielded a mean estimation error of 15.14% for the testing dataset. The monthly mean DOC concentration significantly increased from 2003 to 2018 (N = 192, p < 0.01). High DOCs were observed in lake bays with high chlorophyll a (Chl-a) levels, and phytoplankton growth explained more than 50% of the monthly DOC variations. Then, given the evenly mixed DOC in the water column on the monthly and annual scales, we further estimated the monthly mean DOC storage in the lake from 2003 until 2018 and DOC fluxes (input and output) due to rivers during 2008-2018. Although the mean net riverine DOC input (50.56 +/- 32.22 x 10(3) t C) was approximately 5.2 times the average DOC storage (9.73 +/- 1.23 x 10(3) t C), phytoplankton growth controlled DOC variations, which indicated that much terrigenous DOC was transformed into other carbon forms after entering Lake Taihu. This study verified the feasibility of remote sensing of DOC (surface concentration, storage, and riverine exchange flux) in eutrophic lakes.