Louis Iverson , Northern Research Station; William Hargrove , Southern Research Station
Species distribution models (SDMs) project future species habitats based on statistical associations between species occurrence or abundance and environmental (predictor) variables thought to influence habitat suitability (1). Such models often examine how species habitats are related to a combination of soil characteristics, elevation, climate, and land fragmentation. For example, environmental variables were statistically related to sugar maple abundance to map its abundance across the eastern United States (2). Using current or projected future climate information in these models allows the prediction and mapping of suitable habitat, though not whether or not a species will reach those habitats. SDMs thus predict potential ranges based on the realized niche rather than the fundamental niche; that is, the niche into which the species is “squeezed” by predators, parasitoids and competitors. Numerous statistical methods have been used to build SDMs. Some of the more recent methods, such as Random Forests (5), stochastic gradient boosting (6), or maximum entropy (7) have been shown to perform better in prediction than most other methods.
SDMs have some advantages compared with other modeling techniques; they are purely empirical, computationally less demanding, and allow occurrence patterns to emerge freely from the available survey or census data. SDMs can approximate potential species habitat for scores of species using multiple future scenarios of climate and human-based decisions, and thus can be useful for conservation and management, especially since ignoring the inevitability of future changes in a rapidly changing climate is not a realistic option (8).
SDMs have limitations, however, including assumptions that the selected variables reflect the niche requirements of a species, that species are in equilibrium with their suitable habitat, that predictions can be made into novel climates and land covers, that the effects of adaptation and evolution are minimal over the modeled timeframe, and that the effects of biotic interactions (including human interactions) are minimal (9, 10, 4, 11 ). It is also important to realize that not all SDMs are equal – within a series of species run with a particular methodology, some models will perform much better than others, as a function of the quantity and quality of input data. And SDM methods and subsequent outcomes can also vary widely with various algorithms and input variables so one must carefully scrutinize the methods and reliability of the models (12).
Several groups have been using SDMs to project potential changes in habitat for the trees of North America. The “ForeCASTS” project of Hargrove and Potter (13) uses 17 climatic, soil-related (edaphic), and topographic variables to model environmental niches and geographic ranges for more than 200 tree species under current and future climates, nationally and globally. ForeCASTS uses the PCM and HadleyCM3 climate models and A1 and B1 emissions scenarios to project future climates and uses multivariate spatial clustering as the statistical prediction method. Each tree species range is predicted with and without using elevation as an environmental factor. For example, in the A1 emissions scenario, the prediction for red spruce (Picea rubens) including elevation describes habitat gains through 2050, but then losses from 2050-2100. ForeCASTS partitions climatic change risk into three separate metrics: percent change in suitable area under future climates, percent overlap of present and future suitable ranges, and average nonzero Minimum Required Migration (MRM) distance. Potter et al. (13) estimated the straight-line MRM distance from each 4 km2 grid cell in a species' current suitable habitat to the nearest favorable future habitat. The greater this distance, the less likely that the species will be able to reach the nearest refuge, and the more likely that the species will become locally extirpated without intervention. Less than a dozen of the ~225 tree species evaluated in ForeCASTS were predicted to have an increase in suitable habitat under the Hadley B1 combination by 2050. Eastern species were predicted to experience a greater decline in suitable area and had less range stability than western species, although predicted MRM distance did not differ between the regions. Eastern species were more likely, on average, to be habitat generalists. ForeCASTS lists all 17 environmental factors in order of their relative importance for each tree species.
McKenney and colleagues (14) (15) used heat and moisture climatic variables to model tree species, and projected an average northward movement of the climate habitat for 130 North American tree species of roughly 700 km, across three scenarios, by end of century.
Crookston and colleagues (16) modified the Forest Vegetation Simulator to assess presence/absence for 74 tree species of the western United States and to modify site index to account for expected climate effects under various GCM/emission scenarios. This was done by modifying Forest Inventory and Analysis plot data with outputs from the Random Forests classification tree (3).
The TreeAtlas uses 38 environmental variables including seven climate variables to assess potential future habitat changes in eastern U.S. trees, through Random Forest modeling (2). For 134 tree species, the ‘mean centers’ of habitat were predicted to move northeastward by 2100; distances ranged from up to 400 km for the less CO2-sensitive model (PCM) with high energy-resource efficiency (B1) to 800 km in the more sensitive (HadleyCM3) model with a ‘business as usual’ scenario (A1F1). Under HadleyCM3 A1F1, 66 species would gain and 54 species would lose at least 10% of their suitable habitat under climate change. A lower emission pathway would reduce both losers and gainers. The iconic sugar maple would lose most of its habitat across the east central U.S . under the most severe scenario (17, 2), but it would still maintain a reduced presence throughout. These species models have been enhanced by 21 species-specific Modification Factors, which include considerations such as species capacities to compete for light or endure fire, drought, flood, or browsing 18). A large vulnerability assessment for northern Wisconsin using these models 19) found that the most vulnerable species were those in the northern part of their range. However, though these species’ habitats decline drastically, refugia are expected to remain in places like coves or low-lying north-facing slopes.
Many studies have used SDMs (see for example, some additional descriptions of studies worldwide in (12). Although the above four studies differed in data and methods, there was generally a consistent pattern of projections of poleward movement of habitat throughout this century. For example, both Crookston et al. (16) and ForeCASTS predict areas in Colorado that are suitable for whitebark pine, Pinus albicaulis, even though these areas are not included in Little’s present range polygons. The outputs in the TreeAtlas are somewhat different than the rest in that they are based on abundance, not a binary presence/absence, and the Modification Factors add a level of interpretability by including information on disturbances and biological characteristics not readily included in SDMs (20). The combined TreeAtlas-Modification Factors have also been used to create risk matrices for various species for the National Climate Assessment (21, 22). In an evaluation of several SDMs, (23) asserted that for conservation purposes, there is, at least sometimes, an inherent bias for models to over-estimate climate-driven vulnerability to extirpation, and are therefore generally more useful in estimating future habitat (poleward or upslope) than in estimating where current habitat will no longer exist under future states.
A key element for managers to remember is that outputs from SDMs are best used as guidance in a general way; they are not refined sufficiently for stand-level prescriptions. However, they do provide an idea on which species may have habitat gaining in the region, and which species may lose habitat in the future due to added pressures. If multiple models, especially models using different methodologies like SDMs and process models, have a tendency to agree on the general outcomes, one can have increased confidence in the guidance. Additionally, the fingerprint of potential climate change impacts for individual tree species may suggest potential adaptation strategies; a species having little overlap between present and future ranges, and with a large average minimum required migration distance may be a candidate for human-assisted migration (discussed elsewhere on this site), for example. This general information, along with local knowledge, should provide managers a base from which to start in developing strategies to respond to the impacts of climate change.
Some important examples can be found within the Climate Change Response Framework as developed by the Northern Institute of Applied Climate Science. The Framework is an integrated set of tools, partnerships, and actions to support climate smart conservation, and it provides a collaborative approach to incorporate climate change into forest management. The Framework covers three large regions across the eastern United States: the Northwoods, the Central Hardwoods, and the Central Appalachians. Each regional project interweaves four components - science and management partnerships, vulnerability assessments, adaptation resources, and demonstration projects. Example outputs from these projects include vulnerability assessments (e.g. 19, 24, 25, 26) and adaptation guidelines (27).
As mentioned above, it is important that managers interpret any outputs from SDMs as general guidelines and not specific instructions for the forest stand of interest. Fine-scale topography, soil, other threats, and management history play a huge role in final outcomes of most management actions. It is also important to remember that trees growing in the central parts of their ranges are very often also affected (positively or negatively) when the northern and southern extents of their ranges are shifting northward.