New Remote Sensing Technology Solves Challenge of Acquiring 3D Underwater Topography of Lakes

A research team led by Professor LU Shanlong from the Aerospace Information Research Institute (AIR), Chinese Academy of Sciences, has developed a machine learning approach to simulate three-dimensional (3D) underwater topography of lakes. The study, published in Big Earth Data, introduces a method based on the XGBoost algorithm that requires only a digital elevation model (DEM) as input. This study provides a promising solution for generating lakebed topography in regions where data are limited.

Traditional underwater terrain mapping relies heavily on sonar-based techniques, which, although accurate, demand substantial manpower, equipment, and financial resources. As a result, only a small fraction of lakes worldwide has comprehensive bathymetric datasets, limiting progress in precise water resource management and aquatic ecosystem protection.

To address this challenge, the team proposed an innovative machine learning–based approach. By analyzing topographic features embedded in DEMs—such as elevation, slope, and aspect—they established a mathematical relationship between above-water and underwater terrain. This enables rapid simulation of lakebed topography without the need for specialized equipment or in-situ measurements, significantly lowering the data acquisition threshold.

The method was tested on representative lakes of the Tibetan Plateau through a workflow of DEM preprocessing, topographic feature extraction, and model construction. Validation shows an average depth error of -11.83%, a maximum depth error -26.94%, and a water volume estimation error 19.36%. These results confirm the method’s capacity to produce reliable bathymetric data in areas lacking field measurements.

This technology has already been applied to hydrological monitoring in small reservoirs and integrated into the team's self-developed virtual hydrological station system. It offers a new approach for constructing underwater terrain models of small reservoirs and river cross-sections. While the method is not yet as precise as direct measurements, its advantages—low cost, high efficiency, and ease of deployment—make it a valuable complement to traditional techniques.

This study offers a promising solution for the global challenge of inaccessible lakebed data. "By combining DEM-based features with XGBoost model, we can generate reliable bathymetric data without costly field surveys" said Prof. LU Shanlong. "With further optimization, this method will greatly support hydrological modeling and water resource management."