Water Quality Variations in the Lower Yangtze River Based on GA-RF Model From GF-1, Landsat-8, and Sentinel-2 Images

Total phosphorus (TP) and total nitrogen (TN) are critical water quality indicators in the Yangtze River and remote sensing techniques can inverse these parameters. However, current models suffer from shortcomings such as lower accuracy due to the fewer spectral bands available from a single satellite. In this article, GF-1, Landsat-8, and Sentinel-2 data are jointly used to develop a genetic algorithm-random forest (GA-RF) water quality inversion model weighted by the entropy method. These models are validated and applied to derive long-term time series of TP and TN in the lower Yangtze River from 2018 to 2023. The results indicate that the three-satellite GA-RF joint model shows the best estimation performance from the in-situ measurements: TP with MAE 0.0108 and RMSE 0.0132, and TN with MAE 0.32 and RMSE 0.40. From 2018 to 2023, the water quality shows an improved trend with TP decreasing by 8.91% and TN decreasing by 11.34% . The annual average TP shows a decreasing trend with 0.0017 mg/L per year, while TN shows a decreasing trend with 0.0557 mg/L per year. In terms of seasonal distribution, the highest values of TP and TN are mostly distributed in summer, and the lowest values are mostly distributed in winter. Spatially, both TP and TN increase from west to east. Furthermore, the effects of hydrometeorological factors on water quality are discussed as well as water environmental factors such as pH and NH3-N.