Abstract: Lake topography, as a core element of the geographical environment, holds multifaceted significance in geographical research. It exerts fundamental influence on surface processes, particularly in hydrological and hydrodynamic modeling. Given that traditional methods for acquiring bathymetric data in large lakes are cost-intensive, time-consuming, and yield infrequent updates, it is imperative to develop rapid satellite remote sensing-based approaches for lake topography mapping. This study utilizes the Random Forest (RF) algorithm combined with Landsat remote sensing imagery and measured elevation data to inverse the local topography of Poyang Lake during the dry season. To address the spatial non-stationarity of topographic features and the spatial autocorrelation of prediction residuals, this study integrates Geographically Weighted Regression (GWR) with Ordinary Kriging (OK) methods to optimize the inversion results and analyzes its errors. The results show that: (1) compared with the RF model, the accuracy of the geographically weighted regression random forest kriging hybrid model (GWR-RF-OK) is significantly improved, and the coefficient of determination (R2) of the measured and inverted elevations in the two study areas are increased, and the mean absolute error (MAE) and mean relative error (MRE) are decreased. (2) The hybrid model has better inversion effect in both the bare beach area with single surface cover type and the Nanji Wetland National Nature Reserve of Poyang Lake (hereinafter referred to as Nanji Wetland Area), which has relatively complex surface cover types, with the R2 of 0.71 and 0.56, the MAE of 0.34m and 0.35m, and the MRE of 5.26% and 3.06%, respectively. After segmentation analysis, the model has better inversion effect in areas with topographic elevation greater than 10m. (3) The degree of topographic relief and the type of surface cover affects the accuracy of the inversion, with the error being smaller in areas with gentle topography than in areas with steep topography, and the accuracy of the topographic inversion of the same type of surface cover in areas with a single surface cover is significantly better than that in areas with a mixture of multiple surface cover types.