Ecological niche models predict plant responses to climate change by circumscribing species distributions within a multivariate environmental framework. Most projections based on modern bioclimatic correlations imply that high-elevation species are likely to be extirpated from their current ranges as a result of rising growing-season temperatures in the coming decades. Paleoecological data spanning the last 15,000 years from the Greater Yellowstone region describe the response of vegetation to past climate variability and suggest that white pines, a taxon of special concern in the region, have been surprisingly resilient to high summer temperature and fire activity in the past. Moreover, the fossil record suggests that winter conditions and biotic interactions have been critical limiting variables for high-elevation conifers in the past and will likely be so in the future. This long-term perspective offers insights on species responses to a broader range of climate and associated ecosystem changes than can be observed at present and should be part of resource management and conservation planning for the future.

These data were used to estimate models relating climate and land cover to wetland densities and develop projections under climate and land use change. Data for model estimation were derived from historical climate data, estimates of hydrological processes based on the Variable Infiltration Capacity model, National Wetlands Inventory, and the National Land Cover Database. Wetland densities were based on observations from the Waterfowl Breeding Population and Habitat Survey. Projected climate conditions were derived from ten Global Climate Models, and projected changes in land use were based on an economic model of the effects of climate on land use transitions. These data support the following publication: Sofaer, H. R., Skagen, S. K., Barsugli, J. J., Rashford, B. S., Reese, G. C., Hoeting, J. A., Wood, A. W. and Noon, B. R. (2016), Projected wetland densities under climate change: habitat loss but little geographic shift in conservation strategy. Ecol Appl. Accepted Author Manuscript. doi:10.1890/15-0750.1.

This landcover raster was generated through a Random Forest predictive model developed in R using a combination of image-derived and ancillary variables, and field-derived training points grouped into 18 classes. Overall accuracy, generated internally through bootstrapping, was 75.5%. A series of post-modeling steps brought the final number of land cover classes to 28.

This landcover raster was generated through a Random Forest predictive model developed in R using a combination of image-derived and ancillary variables, and field-derived training points grouped into 18 classes. Overall accuracy, generated internally through bootstrapping, was 72.7%. A series of post-modeling steps brought the final number of land cover classes to 28.

This study had two objectives: first, to generate a landcover map for the Charles M. Russell Wildlife Refuge (CMR) emphasizing the distribution of land cover types in relation to greater sage grouse ( Centrocercus urophasianus) habitat needs, and second, to provide data that would allow a determination of whether results were better with SPOT imagery or Landsat 8 imagery. SPOT imagery is provided at a 10m pixel resolution, while Landsat 8 is at 30m. Results from this classification will allow managers to determine which resolution provides the accuracy needed for habitat planning and management.

Training points collected in the field between 2012 and 2013 were grouped into 18 classes: Forested Burn (66), Foothill Woodland Steppe Transition (73), Greasewood Flat (73), Greasewood Steppe (239), Greasewood Sage Steppe (277), Great Plains Badlands (166), Great Plains Riparian (255), Low Density Sage Steppe (776), Medium Density Sage Steppe (783), Mixed Grass Prairie (555), Mixed Grass Prairie Burned (278), Ponderosa Pine Woodland and Shrubland (512), Riparian Floodplain (223), Semi-Desert Grassland (103), Sparsely Vegetated Mixed Shrub (252), Silver Sage Flat (70) , Silver Sage Steppe (64), and Water (246). When insufficient field data were available for a class, we augmented it through photointerpretation of 15 cm aerial imagery, using expert knowledge and field experience to guide us. The final dataset had 5,011 training points.