When mapping land surface characteristics with remotely sensed data, it is widely accepted that the field training data needs to be representative in space and time for the area of interest. Such principal is implied from the first law of geography. However, it is not uncommon to lack in situ training data in areas where we wish to map with remotely sensed data. We present the concept of ‘ecological distance’ in mapping the land surface, which allows identification of regions with relevant in situ data that can be used as substitutes for training in other regions that lack in situ data. Using an extensive dataset of in situ data from the western US, we are able to show that ecoregions that are geographically distant can be similar enough ecologically (i.e., low ecological distance) to allow surface prediction models trained in one ecoregion to map vegetation indicators in another with similar performance in prediction mean absolute error. We further developed a method using satellite data time series analysis derived harmonic coefficients to measure the "Ecological Distance" between ecoregions with no in situ training to those that do. Such ecological distance measures allowed us to provide a mappability score for all the ecoregions with on in situ training data. Further, we provide a map to predict vegetation indicator mapping accuracy using the aforementioned remote sensing time series data analysis approach.
