1. Tropical dry forests are among the most threatened ecosystems in the world. Rapid loss, degradation, and fragmentation of these native ecosystems in a changing climate have driven a time-sensitive need to improve our understanding and management of remaining dry forests. 2. We used advanced remote sensing technologies, combined with extensive field data and machine learning, to better understand how spatial drivers (e.g., climate, fire, human) of canopy species composition vary in importance and correlate with forest cover (total, native and non-native), within an endangered Hawaiian tropical dry forest. 3. Past introductions of non-native, drought-tolerant tree species into this Hawaiian dry forest have created a new forest canopy composition and a loss of native forest biodiversity and connectivity at the landscape scale. 4. Synthesis and application. Our findings help to spatially visualize the loss and transition of native Hawaiian forests and provide a new conservation planning tool. Conservation and restoration efforts can now be informed by spatial maps of canopy composition, connectivity, and determinants of forest cover for the region. For example, our models identified a climatic transition zone between 800-1000 m where native forests exist in high densities, and non-native forests are not yet dominant. This area may be optimal for cost-effective conservation and targeted management.

The raw data contains information from multiple sources. Most of the data was originally downloaded from the Hawaii GIS portal (https://planning.hawaii.gov/gis/) and is freely available. Hyperspectral data to create the forest cover map was obtained from the Global Airborne Observatory (https://gdcs.asu.edu/programs/global-airborne-observatory ). A more in-depth description of the data can be found in the associated paper.