Spatially explicit representation of ecological states is of value to range managers because they already use state-and-transition models associated with ecological sites for range assessment and monitoring. However, in arid rangelands, state classes rarely exhibit simple, direct relationships with remotely sensed reflectance variables and are challenging to map using traditional classification algorithms. We tested machine- and deep learning algorithms with multi-temporal, multi-spatial datasets over a Chihuahuan Desert location previously mapped using expert interpretation of aerial imagery. Algorithm success varies by ecological site, but initial results indicate maps comparable to expert interpretations and potential for broader scale implementation.
