Abstract: The trophic state of lakes is a direct reflection of the water quality, but its comprehensive, accurate assessment is often a challenge if a large lake is influenced by complex environmental factors. This study synthesizes multi-source information such as remote sensing data and on-site measured data, and cleans the data outlines using the methods of spectral curves and the interquartile range principle (IQR), so as to invert lake trophic state characterization metrics: chlorophyll-a concentration (Chl-a) and the trophic state index (TSI). Key environmental factors－including pH, temperature, average wind speed, and sediment content－are selected through statistical analysis; A trophic state inversion model SSA-BP-NN is constructed based on the back-propagation neural network (BP-NN) and sparrow search algorithm (SSA). The results show that the inversion accuracies of Chl-a and TSI based on the BP-NN model are 0.843 and 0.834 respectively, while they are improved to 0.918 and 0.936 through optimization using the SSA algorithm. In the case study of Hongze Lake, this model has been applied to analysis of the spatiotemporal distribution characteristics of the trophic state: TSI is higher in the western and northern lake than the eastern lake. An analysis of raster point data shows that Chl-a and TSI are closely correlated, and the range of TSI change is more stable. In this lake, water blooms are complex due to their highly transient nature and the influence of various factors; TSI, as a comprehensive indicator that can reflect the basic situation of eutrophication in water bodies, provides as background information for early warning of water bloom risks. This study verifies that the assessment accuracy of a remote sensing inversion model can be improved significantly by introducing environmental factors correlated closely with the lake trophic state, so as to help useful assessment of the health state of lake ecosystems and risk early warning.

Key words: large lakes, remote sensing, environmental factors, machine learning algorithm, chlorophyll-a, trophic state index