Satellite algorithms for retrieving dissolved organic carbon concentrations in Chinese lakes.

Abstract

Water quality and the carbon cycle in lakes are strongly related to the concentration of dissolved organic carbon (DOC). Several regional algorithms have been proposed to remotely retrieve lake DOC concentration at a regional scale, but further efforts are needed to reliably retrieve DOC concentration over a large area. Based on bio-optical measurements from 55 lakes across China, this study investigates feasible satellite algorithms for retrieving DOC concentrations from OLCI/Sentinel-3 imagery. The results revealed that the bio-optical characteristics of DOC were different in freshwater and saline lakes. Compared to saline lakes, freshwater lakes had lower DOC concentrations (9.89 ± 3.97 mg/L vs. 32.97 ± 42.07 mg/L) but similar levels of colored dissolved organic matter as indicated by its absorption coefficient at 280 nm (a_CDOM(280), 12.8 ± 6.94 1/m vs. 17.15 ± 22.97 1/m). Moreover, DOC concentrations in freshwater lakes were exponentially related to a_CDOM(280) (r = 0.74) and linearly correlated with red-to-green reflectance ratios. However, DOC concentration in saline lakes was linearly related to a_CDOM(280) (r = 0.93) and exponentially correlated with red-to-blue reflectance ratios. Then, although we discriminated freshwater and saline lakes with a conductivity threshold of 2000 μs/cm, the three commonly used linear regression methods for estimating DOC concentrations still obtained mean absolute percent difference (MAPD) of 55.68-66.44 %. Alternatively, we developed a hybrid machine learning algorithm (MAPD = 18.16 %), that used water reflectance and lake/basin properties to model DOC concentrations in freshwater and saline lakes, respectively. Satellite monitoring of 370 large lakes (> 20 km²) showed that DOC concentration was high in the northwest and low in the southeast of China. This study has implications for dynamic monitoring of DOC concentrations in lakes using satellite imagery.