 This study evaluates the accuracy of a multi-factor combined linear regression model, MLR, and machine learning models for accurate lake depth mapping and estimation of changes in water level and storage on the Tibetan Plateau using satellite images and bathymetric data, and finds that the random forest, RF, model has the highest precision within a two value of 0.9311 and a mean absolute error, MAE, value of 1.13M in the test dataset. The water level increased by 9.36M at a rate of 0.47MY, and the water storage increased by 1.811km3 from 1998 to 2018 based on inversion depth, with results consistent with those obtained using the Shuttle Radar topography mission, SRTM, method. The study suggests that this method may be used for studying water depth distribution and changes in shallow lakes by combining bathymetric data and satellite imagery. This article was authored by Hongyang, Hanliangua, Wenhao Dai, and others. We are article.tv, links in the description below.