Cautions, Citations and Frequently Asked Questions

Read the following before interpreting the maps and tables While we believe that our tree atlas is an example of predictive vegetation mapping and contributes to the understanding of tree habitat distribution and colonization under changed climate, we want to emphasize what it is not. In order to avoid the mis-interpretation of our atlas, we want everyone to read the following section before making sense of the maps.

First of all, the results of DISTRIB-II give potential habitat distributions (or habitat suitability) for future General Circulation Model (GCM) scenarios (~ 2100) for 125 tree species. By potential habitat distribution, we mean that the habitat becomes suitable for a species to colonize, provided that the GCM-predicted climate of the future is accurate and our model captures all relevant climatic attributes pertaining to the current distribution of the species.

In addition to the geographic range-shift of habitats, our model predicts the potential change in abundance (importance value) of the species. Some species could potentially increase in abundance in some areas and decrease in other areas under future climates. These should be interpreted as relative changes compared to the current-modeled habitats and not as absolute changes in importance values.

We also use the migration model SHIFT to calculate the colonization likelihood of the tree species in ~ 100 years to approximately match the future habitat distributions predicted by DISTRIB-II. We combine the predicted future suitable habitats of DISTRIB-II (or habitat quality, HQ) with the colonization likelihood (CL) computed by SHIFT to estimate the colonization potential (HQCL) of future habitats.

GCM climate scenarios: We used the data from three GCMs (CCSM4, GFDL, and Hadley; see details in About the Models) for two future scenarios: the representative concentration pathways (RCP) 4.5 and 8.5 which result in projections of 4.5 W/m2 with ~650 CO₂ equivalent and 8.5 W/m2 with ~1370 CO₂ equivalent by 2100, respectively. The RCP 4.5 assumes that efforts to significantly reduce CO₂ and other greenhouse gas (GHG) emissions will occur to stabilize atmospheric concentrations by mid-century and decline to 2100, while RCP 8.5 assumes steady increases of GHG emissions through the century. These two RCP scenarios bracket most of the future emissions as outlined by the Intergovernmental Panel on Climate Change’s evaluation of emission scenarios and result in peak GHG emissions at roughly double (650 ppm-RCP 4.5) and quadruple (1370 ppm-RCP 8.5) the pre-industrial levels for CO₂ (~300 ppm, excluding non-CO₂ emissions). We also averaged the outputs of the three models for each RCP scenario to yield suitable habitat under an average high and average low set of conditions.

So, if species x has increased its range in our maps on one of the GCM scenarios, it would be accurate to assume that: if climate were to change as defined by that GCM model, then the suitable habitat for colonization for x could expand according to our model.

Please note here that we merely use the results of the GCM-climate scenarios - we have no control on its outputs. We would like to stress here that DISTRIB-II is not predicting migration of species x - but rather the change of suitable habitat for that species. In order to predict migration, we use the colonization likelihood estimated by the SHIFT model (based on current species distribution, historical migration rate, habitat quality of the colonizable cells and the generation times of individual tree species), and combine it with DISTRIB-II to estimate the colonization potential of newly suitable habitats. However, it should be noted that by colonization we just mean the likelihood of a propagule landing on a suitable site – we do NOT estimate its chances of establishment (which depends on species interactions and other local site factors) and maturity to reproductive stage, which are beyond the scope of our current effort.

How to cite the tree atlas

Peters, M.P., Prasad, A.M., Matthews, S.N., & Iverson, L.R. 2020. Climate change tree atlas, Version 4. U.S. Forest Service, Northern Research Station and Northern Institute of Applied Climate Science, Delaware, OH. https://www.nrs.fs.fed.us/atlas.

We recommend the following publications be cited along with the atlas citation, depending on what you used:

For habitat suitability models on trees:

Iverson, L.R, Peters, M.P., Prasad, A.M., & Matthews, S.N. (2019). Analysis of Climate Change Impacts on Tree Species of the Eastern US: Results of DISTRIB-II Modeling. Forests, 10(4), 302. doi: 10.3390/f10040302 https://www.fs.usda.gov/treesearch/pubs/57857

Peters, M. P., Iverson, L. R., Prasad, A. M., & Matthews, S. N. (2019). Utilizing the density of inventory samples to define a hybrid lattice for species distribution models: DISTRIB‐II for 135 eastern U.S. trees. Ecology and Evolution. doi: 10.1002/ece3.5445 https://www.fs.usda.gov/treesearch/pubs/58353

Iverson, L. R., Prasad, A. M., Peters, M. P., & Matthews, S. N. (2019). Facilitating Adaptive Forest Management under Climate Change: A Spatially Specific Synthesis of 125 Species for Habitat Changes and Assisted Migration over the Eastern United States. Forests, 10(11), 989. doi: 10.3390/f10110989 https://www.fs.usda.gov/treesearch/pubs/59105

Prasad, A. M., Iverson, L. R., Matthews, S. N., & Peters, M. P. (2016). A multistage decision support framework to guide tree species management under climate change via habitat suitability and colonization models, and a knowledge-based scoring system. Landscape Ecology, 31(9), 2187–2204. doi: 10.1007/s10980-016-0369-7 https://www.fs.usda.gov/treesearch/pubs/50748

Prasad, A. M., Gardiner, J. D., Iverson, L. R., Matthews, S. N., & Peters, M. (2013). Exploring tree species colonization potentials using a spatially explicit simulation model: implications for four oaks under climate change. Global Change Biology, 19(7), 2196–2208. doi: 10.1111/gcb.12204 https://www.fs.usda.gov/treesearch/pubs/43705

Iverson, L. R., A. M. Prasad, S. N. Matthews, and M. Peters. 2008. Estimating potential habitat for 134 eastern US tree species under six climate scenarios. Forest Ecology and Management 254:390-406. http://www.treesearch.fs.fed.us/pubs/13412

For Adaptability of tree species:

Iverson, L. R., S. N. Matthews, A. M. Prasad, M. P. Peters, et al. (2012). Development of risk matrices for evaluating climatic change responses of forested habitats. Climatic Change 114(2): 231-243. doi: 10.1007/s10584-012-0412-x. https://www.fs.usda.gov/treesearch/pubs/41221

Matthews, S. N., L. R. Iverson, A. M. Prasad, M. P. Peters, and P. G. Rodewald. 2011. Modifying climate change habitat models using tree species-specific assessments of model uncertainty and life history factors. Forest Ecology and Management 262:1460-1472. http://www.fs.usda.gov/treesearch/pubs/38643

Frequent Asked Questions

Getting started; where to begin?

If this is your first time viewing the atlas or you would like more information about the contents offered, we suggest that you spend some time watching a tutorial video provided on the Climate Change Atlas main page, https://www.nrs.fs.fed.us/atlas/. The introductory tutorial is located on the right side and is about 5 minutes in length.

I don’t see the species I’m interested in.

Currently, we have modeled potential suitable habitat for 125 tree species east of the 100th meridian, using forest inventory data and 45 environmental variables. While we know other species are present within the eastern U.S., there is insufficient data to accurately model these species under current and future conditions. Additionally, we provide relative abundance for 24 species which we were unable to model but were reported by recent forest inventory data.

What is DISTRIB-II?

DISTRIB-II is an empirical statistical modeling framework that utilizes machine learning algorithms (random forest) to predict values based on updated training data and a hybrid lattice approach. We use DISTRIB-II to model the potential distribution of suitable habitat for 125 tree species under current and future climate conditions. More details about the modeling methods are in the papers in the Citations section above.

What is SHIFT?

SHIFT is a spatially explicit cell-based model, which considers current abundance of the species, landscape fragmentation, and historical migration rates to compute the likelihood of colonization in approximately 100 years depending on the generation time of the species (i.e., years to seed production maturity). When combined with DISTRIB-II, it estimates the colonization likelihood of future suitable habitats. This allows us to examine how much of newly suitable habitat beyond its current range boundary could be colonized by a species.

What is Adaptability?

Outputs from DISTRIB-II do not consider many biological or disturbance factors which favor or limit tree establishment, growth, or mortality. We have scored each species by 12 disturbance factors and nine biological factors, collectively called Modification Factors, based on information obtained from the literature, to help interpret DISTRIB-II models. For example, the life history characteristics of red maple allow it to thrive under most conditions, and likely will do better under climate change than the outputs of the DISTRIB-II models suggest. The Adaptability variable is a single score derived from the Modification Factors (or ModFacs), and are one way to locally integrate our models into management decisions.

What is meant by current?

The term current is relative and can differ among variables. With the climate data, we use current to refer to the 30-year period of climate (1981–2010) used as a baseline. With FIA, current would mean the most recently completed inventories for each state during the period 2000–2016. For more information about the variables used in the models, Peters et al. 2019, Iverson et al. 2019, and About the Models.

What is meant by future?

The term future here refers to a future time into which our models attempt to forecast potential suitable habitat for trees, based on estimated climates for 2070–2099. In these models, the non-climate environmental variables (elevation and soils) are assumed to remain constant into the future.

What is the difference between weather and climate?

The major difference is the measure of time. Weather refers to events that occur over a short period of time (i.e., hours, days, weeks), while climate refers to trends (average weather) over a longer period (i.e., months, years, decades). Our models use average climate values for a 30 year period.

What is a General Circulation Model?

A General Circulation Model (GCM) is a complex mathematical model parameterized with information about the atmosphere, Earth, and oceans to simulate hourly or daily climatic conditions. GCMs are run for a length of time prior to the period of future simulations in order to assess the accuracy of known events. The Intergovernmental Panel on Climate Change (IPCC) has included results from many GCMs into their assessments. These GCMs differ in their sensitivity of climate to respond to various levels of CO₂ in the atmosphere. We present the results for three GCMs (CCSM4, GFDL CM3, and HadGEM2-ES) obtained from NASA Earth Exchange Downscaled Climate Projections.

What is the difference between a General Circulation Model (GCM) and a Representative Concentration Pathways (RCP)?

GCMs simulate climatic conditions based on information related to the atmosphere, Earth, and ocean, whereas RCP scenarios provide information about global populations, greenhouse gas emissions, socioeconomical interactions, technological advances, land-use, conservation efforts and other aspects that could influence conditions that affect climate within the GCM. The RCPs were developed by various groups and the Intergovernmental Panel on Climate Change chose four that provide unique trajectories through the century. The combination of GCM and RCP scenario dictates the climate conditions predicted into the future.

Why were the 4.5 and 8.5 RCP scenarios chosen?

Of the various representative concentration pathways included in the Intergovernmental Panel on Climate Change (IPCC) Assessment Report 5 (2013), scenarios 4.5 and 8.5 represent the lower and upper bounds of greenhouse gas (GHG) emission levels, respectively. RCP 4.5 projects GHG emissions peak at about 650 ppm of CO₂ equivalents mid-century and are substantially reduced due to technological advances and conservation efforts resulting in 4.5 W/m2 of radiative forcing or increases of 1.1 to 2.6°C (2.0 to 4.7°F) globally by 2081–2100 relative to 1986–2005. RCP 8.5 assumes the current GHG emission trends continue, resulting in a quadrupling of CO₂ equivalent (1370 ppm) by the end of the century and increases of 2.6 to 4.8°C (4.7 to 8.6°F) globally. Thus, these RCP are intended to capture a wide range of possibilities over this century.

Why were the CCSM4, GFDL CM3, and HadGEM2-ES models chosen?

These three GCMs represent a wide range in sensitivity to CO₂ in the atmosphere. Each has been run with RCP 4.5 and 8.5 greenhouse gas emissions and downscaled with a consistent approach. Thus, the combination of GCMs and RCPs are intended to provide a realistic range of possibilities for changes in suitable habitat. In particular, averaged across the eastern US, the HadGEM2-ES and RCP 8.5 combination projects the greatest increase in mean annual temperatures (6.3°C or 11.4°F), while the CCSM4 and RCP 4.5 combination projects the least amount of change in mean annual temperatures (2.3°C or 4.2°F). Similarly, the GFDL CM3 and RCP 8.5 combination projects the greatest increase in annual precipitation (18.6 mm or 0.7 inches), while the HadGEM2-ES and RCP 8.5 projects the smallest increase of 6.2 mm or 0.2 inches. Locally and among regions within the eastern US, the end of century projected changes will vary in the amount of change (positively or negatively).

How current is the FIA data and how is it used to calculate importance values?

Forest Inventory and Analysis records from >84,000 plots surveyed during 2000–2016 were used to calculate Importance Values (IVs) for 148 tree species. Importance Values(X) = (50 * basal area(X) / basal area(all species)) + (50 * number of stems(X) / number of stems(all species)), where X is a single species. Thus within a cell, the IV for a particular tree species represents the relative abundance for its potential suitable habitat.

Where are the forest type maps corresponding to this version of the Tree Atlas?

Forest types can be defined in several ways. As such, efforts are underway to evaluate the methods previously used to define and classify suitable habitat for individual species.

What is Little’s range?

Elbert L. Little developed range boundaries for many tree species across North America and published these ranges in a series of atlases during the 1970’s. The ranges defined by Little used data from field surveys, herbarium records, and expert knowledge to delineate boundaries to encompass the natural distribution of a tree species. We use “Little’s Boundaries” in many of the maps represented here as one estimate of the range boundary for the species. Even though these boundary estimates are now quite old, these data still remain the primary estimate of range boundaries for the trees of North America.

How realistic are the DISTRIB-II models?

DISTRIB-II uses the statistical learning algorithm of Random Forest to correlate 45 environmental predictors across the eastern U.S. with tree species Important Values (IVs) derived from forest inventory data. Because we use many predictor variables other than the seven climate variables, DISTRIB-II is not a ‘climate envelope model’, but could be referred to as a ‘realized-niche model’. All statistical models lack the capability to include various disturbance and biological features of the species being modeled, and assume that the species is in equilibrium with its environment and has integrated those other features over a long time (e.g., capacity to withstand drought, ice, competition) to allow the species to survive in particular places.

How reliable is a model with low model reliability?

We assign models reliability as high, medium and low. Species with low model reliability have poor prediction performance due to low model fit (equivalent of R-square). Often poor models are the result of a) low sampling abundance within the forest inventory data, or b) the species is relatively rare or occupies a small geographic area. For these species, the current distribution does not allow DISTRIB-II to accurately predict suitable habitat. The results from low-model reliability should be used with extreme caution.

Are the GIS data used to create the maps in the atlas available for my own use?

For each of the 125 species which we were able to model suitable habitat, shapefiles containing the Importance Values derived from the current forest inventory and modeled under current and future climates, as well as the current forest inventory for 24 species we were unable to model, are available at https://www.fs.usda.gov/rds/archive/catalog/RDS-2019-0029.

Why are the models confined to the eastern United States?

Ecosystems do not follow political boundaries, however, datasets collected and managed by various agencies and governments often do not use the same methods for sampling or definitions for reporting. For reasons related to data incompatibility between the USDA Forest Service Forest Inventory and Analysis data and the Canadian equivalent, our model outputs reported here do not extend north of the U.S./Canadian border. Nor have we ventured west of the 100th meridian so far, due to limitations in funding, data, and knowledge on western ecosystems. However, efforts are underway to expand the scope of our models to the western United States as well as Canada. Please stay tuned!

Can I view results for multiple species?

While it would be difficult to visualize mapped modeled results for multiple species, we have summarized for various regions (state, National Forests, National Parks, HUC6 watersheds, 1×1° of latitude and longitude) information pertaining to the current and future changes to suitable habitats, adaptability, capability, and colonization likelihoods for species currently present or modeled to have suitable habitat now and into the future. See the combined species outputs section.

What is the smallest area that should be used for analysis or interpretation?

We recommend that at least 8000 sq. km (3088 sq. mi) be considered when interpreting the model outputs. This area, when averaged, will allow reasonable confidence in the outputs, so that spurious model outliers are not driving the results. Summaries of combined species outputs for areas less than 8000 sq. km have been buffered to the minimum area, but often slightly larger through an iterative process.