【Objective】To improve the accuracy of forest stock volume inversion and provide a reference for remote sensing estimation of forest stock volume in areas with complex terrain, this study aims to construct a multi-source remote sensing dataset and examine the impact of terrain correction at different slope classifications on the estimation results.【Method】Using Sentinel-2 and GF-6 remote sensing images, combined with field measurement data from Wangyedian Forest Farm in Chifeng, Inner Mongolia, this study constructed multiple traditional non-parametric models and ensemble learning models to invert the forest stock volume of Wangyedian Forest Farm. To reduce the influence of terrain fluctuations on inversion results, terrain corrections were performed on the images using the Teillet, VECA, and SCS+C methods at different slope classifications to improve the accuracy of forest stock volume inversion.【Result】The estimation performance of ensemble learning algorithms was generally superior to that of traditional non-parametric models, with the random forest model demonstrating the best performance among all models. Compared with the random forest model constructed using a single data source, combining Sentinel-2 and GF-6 data improved the inversion results of forest stock volume inversion performance, reducing the root mean square error (RMSE) of the models using Boruta by 7.41% and 9.61%, respectively. After terrain correction based on slope classification, the RMSE of the model decreased by 18.48%, and the spatial distribution of forest stock volume showed a high degree of consistency with the actual situation in Wangyedian Forest Farm.【Conclusion】Using Sentinel-2 and GF-6 as data sources, the constructed ensemble learning algorithms can more effectively estimate forest stock volume. Terrain correction based on slope classification significantly improves the estimation accuracy of forest stock volume.