Author(s): Tang Zecong; Ma Yicheng; Ma Chenxi; Qin Chao; Xue Yuan; Xu Ximeng; Fu Xudong

Linked Author(s): Xudong Fu

Keywords: River cross section morphology; Deep learning; Terrain prediction; Correlation analysis

Abstract: The cross section morphology of rivers is fundamental for studies on river hydrological processes and material fluxes. The acquisition of cross section morphology is primarily based on field measurements, which limits the ability to obtain cross sections in inaccessible areas and across entire river basins. Multi-source remote sensing observations from integrated air-space systems have enabled the large-scale extraction of river cross section morphology above the lowest water level. However, non-contact measurement methods for morphology below the lowest water level have been rarely reported. This study focuses on a typical data-scarce mountainous river, specifically the six major external river systems of the Qinghai-Tibet Plateau. Utilizing 88 measured cross sections, we constructed an underwater cross section morphology prediction model based on terrain above the lowest water level, employing the Encoder-Decoder architecture of the Long Short-Term Memory (LSTM) deep learning model. The study also identifies the optimal function for fitting underwater cross section morphology and employs correlation analysis to determine the key factors influencing it. The main findings are: (1) The LSTM deep learning model demonstrates potential in predicting the underwater cross section morphology of single-threaded rivers, with an average Root Mean Square Error (RMSE) of 0.296 m on the test set; (2) The underwater cross section morphology of single-threaded rivers can be fitted with a hook function or an exponential function, with R² values of 0.542 and 0.781, respectively; (3) The main factors influencing river cross section morphology include climate type, mean annual temperature, potential evaporation, mean annual runoff, vegetation cover, elevation, and latitude. The research results provide accurate boundary conditions for hydro-sediment dynamics simulation in data-scarce areas. They also offer research insights and technical support for the automated, systematic, and detailed extraction of river cross section morphology and other river information in data-scarce regions or large basins, contributing to the development of digital twin basins.

DOI: https://doi.org/10.64697/IAHR-APD2024_P533

Year: 2024