Because of the difficulty in obtaining large-scale distribution data on tree species in Rhinopithecus roxellana habitat in Shennongjia, we attempted to use multi-source and multi-temporal remote sensing data combined with expert knowledge to identify species at different levels. Firstly, after analyzing the discrimination of sample trees, we used winter Landsat8/OLI image data to extract evergreen and deciduous forest, respectively. Secondly, we used summer Worldview-2 high resolution image data to for the recognition of tree species, which included the evergreen species (Abies fargesii, Pinus armandii, Picea wilsonii, Quercus spinosa) and deciduous tree species (Betula albo- sinensis, Larix kaempferi, Fagus engleriana, Toxicodendron vernicifluum, Quercus aliena, Populus wilsonii), respectively. Thirdly, combining the vegetation quadrats and expert knowledge on elevation, we corrected the classification results based on the second step. Finally, making use of GIS spatial analysis, we analyzed the terrain and geographical distribution on the dominant species. The experiment revealed that accuracy was higher in evergreen forests, such as Abies fargesii, Pinus armandii, Quercus spinosa, and Pinus armandii affected by pests, whereas relatively higher in deciduous forest, such as Betula albo-sinensis and Toxicodendron vernicifluum. Some species, such as Populus wilsonii and Quercus aliena, showed poor accuracy. In general, evergreen species had higher accuracy than deciduous trees. By combining plant geography, remote sensing, and GIS, we integrated the multi-source, multi-temporal remote sensing data, phenological characteristics of the tree species, and expert knowledge to propose a method for identifying tree species. This method (1) provides an effective way to identify dominant tree species in complex mountainous environments, and it has the versatility for a variety of geographical environments; (2) makes full use of the integration of species phenological features and characteristics of remote sensing data to reduce data costs; (3) uses ground sampling and expert knowledge, ensuring the classification results are correct, which can avoid excessive reliance on spectral characteristics, and reduce the possibility of misclassification. This method will provide more accurate data for the protection and restoration of the habitat of Rhinopithecus roxellana in Shennongjia.