Water Transparency Inversion Method of Dongting Lake Based on Feedforward Neural Network Learning

Shirui Hu,Xiuguo Liu,Qingwu Hu
DOI: https://doi.org/10.1109/geoinformatics57846.2022.9963859
2022-01-01
GEOINFORMATICS
Abstract:Water transparency reflects the nutritive degree of inland lakes. Remote sensing inversion of lake transparency can be used for rapid monitoring of water quality in a large range, which is of great significance for timely management of lakes and protection of aquatic ecological environment. Taking Dongting Lake as an example, based on Landsat 5 remote sensing images, this paper proposes a time-series remote sensing monitoring method of lake transparency based on feedforward neural network learning. This method combines the measured water quality observation points with Landsat 5 remote sensing images, and adopts feedforward neural network to establish a remote sensing inversion model of water transparency. The temporal and spatial distribution of water transparency in Dongting Lake was inversion based on remote sensing images. The results show that it is possible to apply the feedforward neural network learning to the transparency inversion of Dongting Lake. The $R^{2}$ between the predicted value and the measured value is 0.638, the root mean square error (RMSE) is 0.048m, and the average relative error is 0.042m. The inversion results can be effectively applied to the water quality analysis and monitoring of the transparency of Dongting Lake.
What problem does this paper attempt to address?