DISTRIB-II: habitat suitability of eastern United States trees

Modeled habitat suitability for 125 eastern United States trees species under 1981-2010 climate conditions and projected future conditions (2070-2099) were created using a statistical modeling approach that correlates mean importance values to environmental data. The associated files includes importance values, derived from tree basal area and number of stems, for the species according to USDA Forest Service Forest Inventory and Analysis data, modeled under current conditions, and for eight future scenarios. Potential suitable habitat was modeled using Random Forest (Breiman 2001), a decision-tree based ensemble approach that correlates mean importance values with environmental data on climate, elevation, and soil. Swapping 30-year mean climate conditions for a baseline period (1981-2010) with projections for the future (2070-2099) results in potential suitable habitat representing a species potential abundance. A hybrid lattice consisting of 10 × 10 and 20 × 20 kilometer (km) cells, defined by the density of forest inventory plots, was used to process environmental data and model potential suitable habitat as individual tree species importance values. Downscaled future climate projections were obtained from the NASA Earth Exchange Downscaled Climate Projections (NEX-DCP30) program (Thrasher et al. 2013). Output from three general circulation models (GCM) were used to explore possible changes in habitat suitability resulting from climate change, and included the NCAR Community Climate System Model (CCSM4, Gent et al. 2011), NOAA Geophysical Fluid Dynamics Laboratory Coupled Model 3 (GFDL CM3, Donner et al. 2011), and Met Office Hadley Global Environment Model 2 - Earth System (HadGEM2 - ES, Jones et al. 2011). Projections under the representative concentration pathways (RCP, Moss et al. 2008) 4.5 and 8.5 were used to encapsulate the range of plausible increases in greenhouse gases during this century. Reported values are based on a statistical process to include mean importance values supplemented by median importance values, when the median predicted values were zero and the mean predicted values were ≥ 2.75 times the coefficient of variation for each 10 × 10 or 20 × 20 km cell. An important caveat when interpreting these models is that they are predicting potential suitable habitat by year 2100 – not where the species will be found. See Iverson et al. (2019) and Peters et al. (2019) for more details.
Knowing where suitable habitat for a species exists and how changes may unfold as a result of changing climatic conditions allows resource managers to plan for future conditions.
This file/version supersedes those associated with Iverson et al. (2008). Additional information and products are available at https://www.fs.fed.us/nrs/atlas. Original metadata date was 07/10/2019. Metadata updated on 08/08/2019 to include reference to new publication. On 11/01/2019 the modeled importance values for each species were updated based on random forest models that used a corrected calculation of daylength coefficient of variation. For most species, this is a minor predictor variable and modeled importance values haven’t changed much, though a slight difference may be observed as a result of random sampling within the modeling framework. Model reliability for two species has changed due to the correction, Pinus clausa (sand pine, 107) is now high and Quercus texana (Texas red oak, 828) is now medium. Metadata were also updated to note this change.