Bathymetry mapping of shallow lakes on the Tibetan Plateau Using Sentinel-2 imagery and Machine learning

Due to the harsh environment and high operational costs on the Tibetan Plateau (TP), traditional bathymetric surveying is difficult to conduct efficiently over large areas. The clear water and low suspended sediment concentrations of TP lakes provide ideal conditions for optical remote sensing-based bathymetric inversion. The blue, green and red bands of Sentinel-2 imagery of Sentinel-2 imagery show strong correlations with water depth, and the three-band combination delivers the best performance. Taking Sentinel-2 multispectral imagery as the data source, this study selected 10 typical shallow lakes on the TP as research objects. Composite images of Sentinel-2 bands b2, b3 and b4 were adopted, combined with multi-scale spatial statistical features extracted by the kernel median method to construct the input feature vector. A random forest model was used to establish the depth-feature mapping relationship. Bathymetric inversion and mapping were achieved with Frost filtering, median filtering for noise reduction, manually optimized resampling points and spatial interpolation. The results indicate that the model hyperparameters optimized via a heuristic two-stage random search yield the highest inversion accuracy. Filtering performance varies significantly across lakes and water depth zones, requiring targeted selection of filtering methods and parameters. Lake basin morphology and water depth dominate inversion accuracy. Flat shallow areas achieve high precision, with only slight error increases in local deep waters due to optical path attenuation. Lakes with complex terrain, steep slopesand deep troughs have significantly larger biases under the combined effects of depth gradients, topography, and substrate, leading to reduced accuracy. This efficient, low-cost and scalable approach is applicable to large-scale lake depth estimation on the TP, and can provide reference for bathymetric inversion of other clear high-altitude lakes.