Abstract: The classification and distribution mapping of tree species are indispensable to precision forestry management. However, the classification of tree species faces challenges such as the high dimensionality of remotely sensed time-series data, difficulty in feature extraction, and similar spectral features of tree species. To address these challenges, this study proposed a remote sensing method for classifying dominant tree species based on the modified Transformer network. By combining the capability of the Transformer model in capturing global features, the proposed method improved the sensitivity to the spectral-temporal features of different tree species and the identification accuracy through the optimization of time series modeling. With Ningyuan County as the study area, the dominant tree species were classified and mapped using the Sentinel-2 time series data. The results show that the TransformerToST algorithm could adaptively extract typical spectral-temporal features of key phenological stages from the satellite image time series (SITS), improving the overall accuracy and Kappa coefficient by about 5% (to 89.39%) and 0.066 0 (to 0.867 2), respectively, compared to the traditional Transformer algorithm. Additionally, the cross-regional model validation in the Huangfushan forest farm confirmed the significant accuracy improvement of the modified model. The tree species map generated in this study provides data support for the dynamic monitoring, ecological conservation, and management of forest resources in the study area, as well as a technical reference for forest resource survey and assessment.