Abstract: The 81 permanent sample plots of Pinus kesiya var. langbianensis from 3 Yunnan forest surveys were combined with remote sensing, and the remote sensing factors with significant correlation were selected as the alternative variables for inverse modeling of carbon stock in Pinus kesiya var. langbianensis forests after independent variable screening. Three models of K-NN, RF, and PLSR were established to optimize the best model for remote sensing of carbon stock estimation of natural forests of P. kesiya var. langbianensis, and a variety of models were developed to predict the carbon stock of P. kesiya var. langbianensis natural forest in 2030. The results showed that the coefficient of determination(R²) of the random forest model was the maximum at 0.83, the root mean square error was the minimum at 5.55, and the prediction accuracy was the maximum at 83.2%, and the AICc was the minimum at 162, which has the best fitting effect and can be used to estimate the carbon storage of P. kesiya var. langbianensis forest. In this study, it was found that the carbon stock of P. kesiya var. langbianensis natural forest showed a "single peak" change with the increase of elevation, and the carbon stock was mainly concentrated in the distribution of elevation 1 500–2 000 m. According to the prediction results of the power function model and the polynomial regression model, it was shown that the accuracy of the polynomial regression model was better than that of the power function model, and the R² and RMSE of the model were 0.960 and 0.054, respectively. The prediction result of P. kesiya var. langbianensis forest carbon stock shows that the carbon stock will reach 2 961.204 t in 2030, which is an increase of 42.095 t/hm² compared with that of 2 000, and this study provides a new solution for better understanding the change law of forest carbon sink.