Predicting broad-scale occurrences of vertebrates in patchy landscapes

Spatially explicit landscape-scale models that predict species distributions, where patches of habitat are shown as having potential to be occupied or unoccupied, are increasingly common. To successfully use such data, one should understand how these predicted distributions are created and how their relative accuracies are assessed. Geographic ranges, defined upon observations (e.g., atlases), literature review, and expert review, are a primary data layer. A map of land cover is created, often from interpretation of satellite imagery or other remotely-sensed data. Species/habitat associations are defined based upon a literature review and expert review, describing associations for habitats derived from the cover map. Included as ancillary associations are how species relate to physical features, where appropriate, such as elevation and hydrography. The three layers of information (range, land cover, and associations) are merged, often using raster-based algebraic statements that exclude unused habitats or patches outside the range of a species. The accuracy of predictions for a suite of species is typically assessed with surveys by comparing the species predicted to occur in an area to the species observed. Omission (i.e., present in species lists but not predicted) and commission (i.e., predicted but not present in lists) errors are reported. Errors may be due to many sources. For example, ranges of species change, cover types may be misidentified, species/habitat associations may be incorrect or change, or species may be rare and unlikely to be seen in surveys and judged in-error even though the species may be present. An example is given of an appropriate use of broad-scale species predicted distributions, in which patterns and threats to Maine terrestrial vertebrate diversity are summarized.