Forests cover about 30% of Earth’s land surface, affect the environment from local to global scales, and provide important services to humankind. Both ecological processes and ecosystem services depend greatly on which tree species are present. Unfortunately, mapping tree species across large areas is difficult. However, there are new opportunities for mapping tree species thanks to new satellites, more frequent satellite images, and new algorithms. The goal for my dissertation was to map main groups of tree species aggregated to forest types across large areas with satellite imagery. I analyzed data from three satellite programs: Landsat, Sentinel-2, and MODIS. I achieved three main outcomes. First, my results show a positive effect of the number of observations on classification accuracy. Increase in the number of image acquisitions improved the accuracy of my maps especially when the cloud-free scenes were not available. Second, I improved the methodology of generating dense time-series of vegetation indexes, which are useful for both monitoring forest phenology and forest type classifications. I did so by advancing the use of the STARFM algorithm, a tool for fusing Landsat and MODIS imagery, to conditions of low data availability, where the standard application of STARFM is suboptimal. Finally, I evaluated the usefulness of the imagery from constellation of three satellites: Landsat-8, Sentinel-2A and 2B. Forest type classifications based on imagery from all three satellites were as good as classifications based on Landsat imagery from three-year period. My results advance remote sensing technology, and are relevant for environmental sciences, forest management, and conservation. Managers can utilize forest type maps to assess the commercial value of the forests, develop climate change adaptation strategies, and manage wildlife habitat.