Improving the Accuracy of Tree Species Mapping by Sentinel-2 Images Using Auxiliary Data – A Case Study of Slyudyanskoye Forestry near Lake Baikal

Timely and accurate knowledge of forest composition is important for the conservation and management of ecosystems. Information on the land cover can be obtained by classifying satellite images. Still, satellite optical data are not always sufficient to get results of the required accuracy because of the similarity of spectral characteristics of tree species. One approach to improve the accuracy of tree species mapping is to use auxiliary data such as climatic, soil, topographic, and vegetation indices. The paper presents the study results for the Slyudyanskoye forestry of the Irkutsk region near Lake Baikal. A set of 101 features was collected, including both Sentinel-2 satellite images and data on soils, climate, forest canopy height, and topography. The spectral characteristics of five tree species at key sites were determined for the training dataset. Polygons corresponding to these species and common land cover types (land, grass, water, and clouds) were marked on the original image. Forest classification was performed using the Random Forests machine learning method. The paper presents classification results for eight sets of variables: spectral bands, their combinations with each type of auxiliary data, all 101 features, and a reduced set of 98 features. The results showed a strong influence of the auxiliary data on the performance of the tree species classification model – the overall accuracy increased from 49.59% for only Sentinel-2 bands to 80.69% for the set of 98 selected features. The addition of climate and soil features caused the greatest increase in accuracy, while the most important variables were the B11 band, forest canopy height, and growing season length. The result shows that auxiliary environmental data improves the accuracy of tree species mapping from satellite images.