Abstract: Tropical forests play a vital role in biodiversity conservation and research on global climate change. However, the complexity and diversity of vegetation types pose challenges to the fine remote sensing-based classification of tropical forests. The classification of tropical forests in the Jianfengling area, Hainan Province was analyzed using the multi-temporal Landsat8 data of the Google Earth Engine (GEE) platform. Based on the analysis of the impacts of the size and combination of multi-temporal data on the classification accuracy, this study proposed a classification method based on multi-temporal Landsat8 images for the vegetation type groups of tropical natural forests, such as typical tropical rain forest, tropical monsoon forest, and evergreen broad-leaved forests. The results are as follows. ① The classification accuracy of tropical natural forests was significantly improved as the size of multi-temporal data increased. The classification accuracy of the vegetation type groups of natural forests in Hainan Island reached 91%. ② When the multi-temporal data reached a certain size, the classification accuracy tended to be stable. Different combinations of multi-temporal data can improve the classification accuracy of tropical forests, especially when the classification accuracy of individual data involved was low. This finding reflects the broadness of the selection of temporal data. The proposed method, taking advantage of the temporal changes in remote sensing data, provides an effective reference for the remote sensing-based classification of tropical natural forests in Hainan Island.