Trophic state assessment of optically diverse lakes using Sentinel-3-derived trophic level index

ABSTRACT

An accurate estimation of trophic state of lakes with satellite remote sensing is a challenge due to the optical complexity and variability associated with inland waters. Match-up data from 393 sampling stations that has concurrent Sentinel-3 OLCI images were acquired across Wuhan lakes. Trophic Level Index (TLI) algorithms were developed within a global Optical Water Type (OWT) classification system. The performance of algorithms with limited training data gathered by using spectral similarity of highest Sowt was not improved compared with that on basis of no classification. In contrast, using spectral similarity of Sowt > 0.9 rather than the highest Sowt to group more training data with similar traits for each OWT can help build more robust algorithms, which performance is better than that on basis of no classification. Algorithm performance statistics of the test dataset for the stepwise multiple linear regression (SMLR) method were the following Mean Absolute Error (MAE) = 5.56;Mean Absolute Percentage Error (MAPE) = 11.02 %;Root Mean Square Error (RMSE) = 7.24 and for the back propagation neural network on the basis of the Levenberg-Marquardt-Bayesian regularization algorithm (LMBR-BPNN) method MAE = 4.56;MAPE = 8.33 %;RMSE = 5.98. We detected 8 different OWTs (2,3,4,5,9,10,11,12) in Wuhan lakes and clear spatio-temporal patterns of the trophic state between 2018 and 2020.Our results revealed that the trophic state of Wuhan lakes did not decrease as expected during the COVID-19 lockdown period.