Abstract: Taking Shangri-La City, Yunnan Province as the research area, based on ICESat‒2 / ATLAS data, the remote sensing forest canopy density estimation models were established by random forest regression, gradient boosting tree regression and nearest neighbor regression, respectively. The optimal model was selected to invert the forest canopy closure within the study area spots. The results showed that random forest modeling was the best method to estimate forest canopy closure, the coefficient of determination(R²) was 0.9446, mean square error (RMSE) was 0.0560 and the prediction accuracy(P) was 90.60%. The predicted values of canopy closure corresponding to 74873 effective forest spots in Shangri-La City were obtained, and the spatial distribution map of canopy closure of all forest spots in the city was obtained by combining the spot center coordinates. The results can provide a reference for remote sensing estimation of forest canopy closure at low-high altitude areas.