Climate variability and man-made impacts have severely damaged forests around the world in recent years, which calls for an urgent need of restoration aiming toward long-term sustainability for the forest environment. This paper proposes a new three-level decision tree (TLDT) approach to map forest, shadowy, bare and low-vegetated lands sequentially by integrating three spectral indices. TLDT requires neither image normalization nor atmospheric correction, and improves on the other methods by introducing more levels of decision tree classification with inputs from the same multispectral imagery. This approach is validated by comparing the results obtained from aerial orthophotos (25 cm) that were acquired at approximately the same time in which the Formosa-2 images (8 m) were being taken. The overall accuracy is as high as 96.8% after excluding the deviations near the boundary of each class caused by the different resolutions. With TLDT, the effectiveness of forest restoration at 30 sites are assessed using all available multispectral Formosat-2 images acquired between 2005 and 2016. The distinction between natural regeneration and regrowth enhanced by restoration efforts were also made by using the existing dataset and TLDT developed in this research. This work supports the use of multitemporal remote sensing imagery as a reliable source of data for assessing the effectiveness of forest restoration on a regular basis. This work also serves as the basis for studying the global trend of forest restoration in the future.