Permanganate Index Variations and Factors in Hongze Lake from Landsat-8 Images Based on Machine Learning

Abstract

The permanganate index (CODMn), defined as a comprehensive index to measure the degree of surface water pollution by organic matter and reducing inorganic matter, plays an important role in indicating water pollution and evaluating aquatic ecological health. However, remote sensing monitoring of water quality is presently focused mainly on phytoplankton, suspended particulate matter, and yellow substance, while there is still great uncertainty in the retrieval of CODMn. In this study, the Landsat-8 surface reflectance data set from Google Earth Engine and in situ CODMn measurements were matched. The support vector regression (SVR) machine learning model was calibrated using the matchups. With the SVR model, this study estimates the CODMn in Hongze Lake, presents the historical spatiotemporal CODMn distributions, and discusses the affecting factors of the change trend of the CODMn in Hongze Lake. The results showed that the SVR model adequately estimated CODMn, with a sum squared error of 1.49 mg2/L2, a coefficient of determination (R2) of 0.95, and a root mean square error of 0.15 mg/L. CODMn in Hongze Lake was high in general and showed a decreasing trend in the past decade. Huai River, Xinsu River, and Huaihongxin River were still the main sources of oxygen-consuming pollutants in Hongze Lake. The wetland natural reserve near Yugou had a significant effect on reducing CODMn. This study provides not only a scientific reference for the management of CODMn in Hongze Lake, but also a feasible scheme for remote sensing monitoring of CODMn in inland water.