Remote Sensing Classification and Evaluation of Regional Scale Forest Types in Shangri-La, Yunnan

Based on the Google Earth Engine cloud platform and Sentinel–2 image, the random forest algorithm was used to identify and classify the forest types in Shangri-La region of Yunnan Province at 3 levels by different combinations of multi-temporal images and topographic features. The results showed that the classification accuracy of multi-temporal features combined with topographic information was the highest at 3 levels. The overall accuracy of forest and non-forest types was 98.15%, and the Kappa coefficient was 0.9624. In coniferous forest and broad-leaved forest, the overall accuracy was 89.74% and the Kappa coefficient was 0.7926. For 8 coniferous forest types, the overall accuracy was 92.87%, and the Kappa coefficient was 0.9180. The results concluded that the terrain information is beneficial to the extraction of forest type information, and the multi-temporal Sentinel–2 data has great potential for the accurate identification of forest type in a large range.