Synthetic Tree Species: labeled species data with more variety and fewer resources
Labeled tree species data is among the most expensive data to produce in Earth observation. It requires field surveys and trained botanists, exists only for a limited set of well-studied forests, and is sparse elsewhere. That scarcity is what constrains species-level detection and segmentation.
Tree AI generates this data synthetically. It produces drone and satellite imagery of forest scenes with tree species included and labeled, each scene supplied with its segmentation masks. It varies resolutions ranging from drone or aerial data to VHR satellite imagery and can incorporate as many species as a task requires. Tree AI can generate the same forest under many different conditions, across sensors, growth stages or even after events like drought or disease, which when working with real data, would mean waiting years to observe and collect. Alongside the imagery, it can provide matched elevation maps and above-ground biomass, so a model can learn species, structure and carbon from one single consistent labeled source.
