Climate change vulnerability assessment of forests in the Southwest USA

Climate change effects are already apparent in some Southwestern US forests and are expected to intensify in the coming decades, via direct (temperature, precipitation) and indirect (fire, pests, pathogens) stressors. We grouped Southwestern forests into ten major types to assess their climate exposure by 2070 using two global climate models (GCMs) and two emission scenarios representing wetter or drier conditions and current or lowered emission levels. We estimate future climate exposure over forests covering 370,144 km2 as the location and proportion of each type projected to experience climate conditions that fall outside 99% of those they currently occupy. By late century, 27–77% is climatically exposed under wetter or drier current emission levels, while lowered emission levels produce 10–50% exposure, respectively. This difference points to the benefits of reducing emissions from the RCP8.5 to the RCP4.5 track, with regard to forest retention. Exposed areas common to all four climate futures include central Arizona and the western slope of the Sierra Nevada. Vulnerability assessments also comprise sensitivity and adaptive capacity, which we scored subjectively by forest type according to the number of key stressors they are sensitive to and the resilience conferred by life history traits of their dominant tree species. Under the 2070 RCP8.5 emissions, four forest types are critically and six are highly vulnerable under the hotter GCM; and eight are highly and two moderately vulnerable under the wetter GCM. We discuss forest management adaptation strategies and the barriers to and co-benefits of such plans.

pathogens-to define sensitivity. These represent current stressors that may be amplified under future climate. For adaptive capacity, we rank forest types according to presence of fireadaptive traits, the number of modes of seed dispersal, and seed longevity. The sensitivity and adaptive capacity components of this vulnerability assessment are subjective and more speculative.

Forest types analysed
The continental Southwestern USA spans a broad gradient of environmental conditions and encompasses a diverse array of forest types, including cool and moist coastal and high-elevation forests and arid woodlands and savannahs. We defined ten general forest types derived from combinations of 53 mapped forest and woodland types found in the national LANDFIRE existing vegetation map for our region (Landscape Fire and Resource Management Planning Tools; www.landfire.gov; accessed August 2014). The 53 types were grouped according to similar species composition and geographical distribution (crosswalk for each forest type in Online Resource 1). LANDFIRE map data was based on 30-m grid cells. We resampled the LANDFIRE existing vegetation model to 300 × 300-m grids, a resolution at which the patterns of dominant vegetation could still be observed, and to which we also downscaled the climate data. We used a majority rule process for the vegetation aggregation. The result somewhat generalizes the vegetation patterns, which may translate to a lower level of spatial precision of the spatial results for areas with fine-scale vegetation patterns or with very steep topographic relief such as the sky islands in the region. The area of each of the ten types, portrayed by in the resulting grid cells, occupies between 8126 and 135,775 km 2 in the region.

Climate data
We used current and future climate grids provided on the WorldClim website (WorldClim Data V1.4; Hijmans et al. 2005;worldclim.org; further details also in Fick and Hijmans 2017). WorldClim data are provided at 30′ resolution and have been used or cited in over 5670 instances (Web of Science, accessed Oct 1, 2016). Among many approaches to downscaling climatic data, WorldClim uses current climate grids to bias-correct and spatially downscale projected future conditions. We further downscaled the WorldClim climate data using bilinear interpolation to the 300 × 300-m grid resolution of the vegetation data. We examined change from current time and relative change among the four future projections. We used 50-year means (1950-2000) of 19 BioClimatic variables (BioClim; Hijmans et al. 2005;Online Resource 2) to represent current climate. We used the same variables for future conditions under two climate models and two emission scenarios (general circulation models; IPCC 2013) that represent a bracketing of predicted future climate conditions (Online Resource 3; Hijmans et al. 2005;worldclim.org). Under current emission levels represented by the RCP8.5 emission scenario by 2061-2080, these models produce warming of 2.9-5.6°C and +43 to −39-mm mean annual precipitation for the MRI-CGCM3 and MIROC-ESM-CHEM models, respectively (Online Resource 3). Under the RCP4.5 emission scenario, the changes are 1.6-4.0°C of increasing temperature and +38 to −8 mm in annual precipitation, for the MRI-CGCM3 and MIROC-ESM-CHEM models, respectively. We suggest that users of our results consider the relative change from current to future time under the four scenarios as the primary application of the climate projections used, and not expect actual climate values at any given location to be more accurate than ground-based measures taken at those locations, where available.

Calculation of exposure to climate change
We used a spatial modelling approach that draws on vegetation biogeography to assess climate exposure. The method leverages advances in land cover maps and uses the spatial distribution of each vegetation type to measure and provide a regional perspective of relative climate exposure. We used current and future climate projections to map the change in climate of each forest type's grid cells and mapped which areas of the range of each type projected to enter climatically marginal (or non-analogue) conditions in the future. We combined the 19 BioClim variables representing current time and two future time steps for each future climate model and emission scenario. We used principal component analysis (PCA) to transform these variables into two-dimensional climate space for each of the four climate scenarios tested. We consider the locations where a forest type becomes climatically Bexposed^as those where the future climate condition is in, or beyond the most marginal 1% of the type's current climate space. The methodological details of the exposure modelling are described in Online Resource 4 and Thorne et al. (2016).
We extracted the values of the first two PCA axes from every mapped location of each forest type and used them to quantify the frequency with which current climatic conditions are occupied by each vegetation type (Thorne et al. 2016). We used the current climate distribution of each type to classify conditions that are common, which we assumed to be equivalent to unstressed conditions, and to identify areas with marginal (infrequently occupied) climates. We assumed that climates occupied by less than 1% of a forest type's grid cells represent marginal climatic conditions for that type and that the type at those locations was likely already stressed in some way (Fig. 1).
We scored exposure of each forest type according to what percentage of its mapped area extent becomes highly exposed (i.e. at or beyond the outermost 1% of its current climate envelope) under each of the four futures by 2061-2080. Scores range from 1 to 6 as follows: 1 (1-5%), 2 (5-15%), 3 (15-30%), 4 (30-60%), 5 (60-80%), and 6 (80-100%). These climate exposures can be considered as Bnot,^Blow,^Bmoderate,^Bhigh,^Bvery high,^and Bcritical,r espectively. High climatic exposure does not necessarily mean that the current forest will disappear at highly exposed locations or that all dominant tree species comprising the type will be extirpated. Rather, highly exposed areas are where the existing vegetation is likely to be highly climatically stressed and potentially more at risk. Because climatic change is predicted to be in the direction of increasing temperatures, increasing climatic water deficit , and potentially decreasing water availability, many locations will have a decreasing probability of retaining their current forest type. We summarize climate exposure scores for the whole region by forest type and by US Level III ecoregion (US EPA 2012).

Evaluation of sensitivity and adaptive capacity to climate change
Sensitivity and adaptive capacity can be evaluated by a number of means (Glick et al. 2011;Friggens et al. 2013). For example, sensitivity could include anthropogenic issues such as habitat loss and fragmentation, while adaptive capacity could imply land management practices to increase resilience. We define sensitivity and adaptive capacity by the inherent traits of the constituent dominant tree species of the forest types and address forest management in the discussion as possible response to climate vulnerability.
Sensitivity We used a literature review to identify the potential sensitivity of species in each forest type to three types of disturbance that are important in the region: (1) wildfire (Dwire and Kauffman 2003;Westerling 2016), (2) drought (Breshears et al. 2005;Williams et al. 2010;Huang and Anderegg 2012;Anderegg et al. 2015), and (3) pests, principally in the form of beetle outbreaks (Bentz and Schen-Langenheim 2007;Bentz et al. 2010) and pathogens (Rizzo et al. 2005). Climate change may amplify the intensity and/or frequency of these disturbances, and so the natural degree to which different forest types are sensitive to each can be used as a component of vulnerability (Comer et al. 2012). We ranked how impacting each disturbance is to the dominant species of each forest type under normal ecosystem functioning. We scored the forest types according to their relative response to drought, pests and pathogens, and fire. Forest types sensitive to none were scored as B0,^to one of the three disturbances were scored as low sensitivity (=1), to two of the disturbances as moderate (=2), and to all three disturbances as highly sensitive (3). Scores for drought and fire were derived from Thorne et al. (2016), modified in some cases by further literature review. Scores for pathogens were derived from the literature.
Adaptive capacity We considered adaptive capacity to be mechanisms in species of each forest type that permit a response to a disturbance. We used attributes compiled by Thorne et al. (2016) to score forest types according to (a) the presence of fire-adaptive traits (e.g. stump sprouting or serotinous cones), (b) how many modes of dispersal are available (e.g. gravity, wind, animal), and (c) the estimated seed longevity of the dominant species or groups of species composing the forest type (e.g. 1 year, 10 years). We used Thorne et al. (2016) for forest types that were previously scored and for a mean score of individual species comprising other forest types. We confirmed and modified the scores with further literature review. Species were scored as low adaptive capacity if they had none or one adaptive trait (=3), moderate if they had two adaptive traits (=2), and high if they had three adaptive traits (=1).
We added the sensitivity and adaptive capacity (SAC) rank scores to the climate exposure score for an overall possible range from 2 to 12, with higher numbers being progressively more vulnerable. This approach is likely an incomplete measure of forest types' sensitivity and adaptive capacity, but it permits relative ranking among our ten types, and because the ranks are explicit, they can be further modified as more information becomes available.

Evaluation of vulnerability
We summed exposure (by model), sensitivity, and adaptive capacity scores to create an overall vulnerability score for each forest type by each future climate projection. No forest type received 4 or fewer points to be ranked as low vulnerability. Forest type vulnerability was classified as moderate (5-7 points), high (8, 9 points), and critical (>9 points).

Exposure
Future extents of Southwest forests in the 1% or most marginal of current climate conditions cover about three times more area under the RCP8.5 than under the RCP4.5 emission scenarios; and the differences in forest climate exposure between the wetter and drier GCMs are fivefold under RCP4.5 emissions and threefold under the RCP8.5 emission scenarios. This section presents detailed results for the RCP8.5 since that is the emission level currently closest to actual emissions; however, the RCP4.5 results are included in the tabular and map presentations (Fig. 2, Table 1; and details in Online Resources 1, 5).
Climate exposure projections for the ten forest types under the hotter and drier future (MIROC-ESM-CHEM RCP8.5) indicate an increase to 77.6% of forested areas that are in the last 1% or beyond of their current climatic envelope by 2061-2080 (Table 1). Under this climate future, oak woodlands and red fir, subalpine, and redwood forests are exposed on over 90% of their area. Under this projection, even the least-exposed types (aspen and mixedmontane forests) are expected to have 63 and 70.5% of their area, respectively, in climatically stressed conditions by 2061-2080.
The wetter GCM (MRI-CGCM3 with RCP8.5) predicts considerably less exposure to climate change. Under this forecast, 27.3% of forested areas are in the final 1% or beyond of current climate conditions 2061-2080 (Table 1). Some individual forest types still have high proportions of exposure under this climate projection, including 97% of redwood forest and 77% of oak woodlands. The most widespread type we analysed is pinyon-juniper woodland, which is projected to be one of the least exposed forested ecosystem types by the end of the century; 13.7% of current area falls in the 1% climate exposure class under the wetter, RCP8.5 scenario and 72% under the hotter and drier RCP8.5 scenario by 2061-2080 (Table 1).
The ecoregions (US EPA 2012) with the most remaining suitable forest climates under the hotter and drier scenario include the Southern Rocky Mountains, the Wasatch and Uinta Mountains, and portions of the Cascades and Klamath Mountains. Remaining suitable areas under the wetter scenario are more wide spread and also include the Arizona/New Mexico Mountains and higher elevations in the Central Basin and Range and Sierra Nevada Mountains (Online Resource 5; Online Resources 1 and 6 provide more detail by forest type and ecoregion).

Sensitivity and adaptive capacity
We considered the sensitivity of forest types to three stressors at their average historical levels of presence, which might be amplified with changing climate. None of the forest types analysed is considered to have high sensitivity. Five types were considered to have moderate sensitivity (two stressors), while six types had low sensitivity (zero to one stressor) ( Table 2).  Table 1 A Projected climate exposure and B total area (km 2 ) exposed for ten major forest types in the Southwestern USA listed in descending order of total forested area. Projected exposure to climate change is based on two global circulation models and two emission scenarios. Forest type nomenclature follows the USDA Plants Database (plants.usda.gov See the text for further description of component scores. Scores were then summed and overall vulnerability was classified as moderate (5-7 points), high (8, 9 points), and critical (10 or more points). No forest type score less than 5 points, to receive a ranking of  We defined adaptive capacity as the inherent ability of species comprising forest types to respond to three types of disturbance. Five types have low (scores of 3 in Table 2), three types have moderate, and two types have high adaptive capacity (Table 2).

Overall vulnerability analysis
Under the wetter RCP8.5 scenario, three forest types are moderately vulnerable and seven are highly vulnerable. Under the hotter and drier scenario, six types are highly vulnerable and four are ranked as critical, the highest level of vulnerability in our ranking system (Table 2).

Discussion and conclusions
Our results suggest that forests in the Southwestern USA are likely to experience significant stress due to their climate exposure, sensitivity to climate-related stressors, and adaptive capacity. For many forest types at many locations, this stress may be sufficient to catalyse a change from forest to non-forest, while the climatic conditions suitable for current species composition of the forest types are likely to shift to other locations (Hansen et al. 2001;Walther et al. 2002). The factors catalysing mortality in these forests are the subject of intensive current research and include physiological stress from increasingly hot droughts and secondary effects such as increased beetle outbreaks and fire frequency or intensity (e.g. Allen et al. 2015;Anderegg et al. 2015;Moritz et al. 2014;Mann et al. 2016;Asner et al. 2016). This paper adds spatial predictions of the forested areas most likely to experience climatic stress to the context, which represents in situ risk.
The vast majority of Southwest forests appear highly climatically exposed under the RCP8.5 scenarios by end century, particularly under the hotter and drier GCM tested, the MIROC-ESM-CHEM. These forests are considerably less vulnerable under the less extreme RCP4.5 scenarios (Table 1), which would require significant modifications of anthropogenic greenhouse gas emissions. This suggests that natural resource managers face a high likelihood of difficult choices, of whether to manage for resilience, resistance, or realignment or simply to exercise restraint and allow forests to change as these combined stressors drive them (Millar and Stephenson 2015). In addition, our results suggest that policies and management actions that lower the trajectory of global climate forcing this century may noticeably lessen the climatic impacts to the Southwest's forests.
Our assessment is based on multi-year aggregates that we use to bracket relatively drier and warmer future conditions. This approach portrays a rate of climate change that could directly impact dominant vegetation but does not account for annual variability or large-scale disturbances that may serve as tipping points for vegetation transition from one type to another, such as wildfire or severe drought (e.g. Allen and Breshears 1998;Breshears et al. 2005;Asner et al. 2016). While we do not predict the likelihood of one climate future over another, we assert that even if there is greater precipitation in the future, plant stress is likely to become higher due to increasing temperatures .
Some forested regions appear climatically at risk under all scenarios tested, including the western slope of the Sierra Nevada Mountains, northern NV, western AZ, and parts of northern UT; while some less impacted areas include mountains around Mt. Shasta, the Klamath Mountains, and parts of northwestern CO (Fig. 2). The factors of sensitivity, adaptive capacity, and land management techniques may be most relevant for sustainable management in regions that fall between these two extremes of climate exposure. For example, eastern UT and parts of CA's Klamath Mountains show greater variability with regard to the degree of climate exposure among the four futures tested. In such areas, management and inherent biotic characteristics may have a better chance to influence forest trajectories, perhaps in places where changing climate conditions may change forest composition but the locations remain forested. We note that CA forest types in some areas are shifting from conifer dominance toward dominance of hardwoods (McIntyre et al. 2015;Thorne et al. 2008). While these changes may impact economic, recreational, aesthetic, and wildlife values (Shaw et al. 2011), they at least may retain dominant tree cover. Similarly, at higher elevation sites, adaptive management might target actions with lower impact changes-conifers in these areas might be replaced by other conifer species (Lenihan et al. 2003); while some studies suggest that for higher elevations that are conifer dominated, cover could become more conifer-dense (Notaro et al. 2012) and management objectives to retain tree cover could prevail. These results point to the need for development of strategic adaptive management plans for such areas.
Given the declining budgets dedicated to public forest management (United States Department of Agriculture 2015), restraint may be the de facto dominant management response. In this case, managers of public lands should use the data from studies such as these to predict and expect realignment of systems with changing climates.
Previous work has presented strategies for adaptation along a spectrum from resisting ecosystem change to realigning ecosystems in the face of change (Millar et al. 2006;Bierbaum et al. 2014;Garfin et al. 2014;Millar and Stephenson 2015). Although there is no Bone size fits all^adaptation, there are similarities in approaches across regions and sectors. Proactive forest adaptation efforts generally aim for one or more of the following goals: building resistance to change (e.g. pathogen treatment), increasing ecosystem resilience (e.g. stand thinning), helping forests respond in novel ways to perturbations (e.g. altering seed zones, planting densities, and thinning cycles), or realigning ecosystems to future conditions (e.g. changing managed forest cover type, timber rotation cycle, or assisted migration of species).
We expect the full range of active management to occur across the span of Southwestern forests. For example, special circumstances (e.g. iconic giant sequoia groves) may necessitate management of forests to slow the influence of climate change (Parker et al. 2000). Among adaptation options, striving for forest resilience is the most hopeful (Hansen et al. 2003) as resilient forests may accommodate gradual changes and return toward a prior condition after disturbance, either naturally or with management. Restoring resilience has become a touchstone for forest management across much the study region (North et al. 2009;Stine et al. 2014), perhaps because it is a conservative adaptation strategy, fundamentally following the definition of restoration in terms of maintaining historical representation in composition, structure, pattern, and processes (US Forest Service 2014). Management tools to increase resilience include combinations of mechanical treatment, prescribed fire, managed wildfire, and untreated control sites.
Managing for gradual realignment may be the best strategy to achieve species persistence with sufficient abundance to maintain viable populations at broad ecoregion scales (Bierbaum et al. 2014). One strategy for this realignment would be to create porous landscapes through which target tree species can move and shift distributions (Stephens et al. 2010). Alternatively, planting broad mixes of species and genotypes particularly in areas of high uncertainty, or perhaps in areas where high climate exposure is predicted by many models, may enhance local persistence (Stephens et al. 2010).
Overlain on climate as a stressor is variability in stakeholder interest in different ecosystem services and preferences for forest management strategies to maximize their values (Rauscher 1999;Fernandez-Gimenez et al. 2008). For example, ecologists studying the role of highintensity fire in forested ecosystems of CA agree that fire suppression has degraded ecosystem integrity (Minnich et al. 1995;Stephens and Ruth 2005;Spies et al. 2006), but disagree on the appropriate response to the potential risk of high-intensity wildfire to reduce that stand density (Lee and Irwin 2005;North et al. 2009;Hanson and Odion 2014). Disagreement on management objectives and on methods for achieving them could constrain the adaptive management of given forest types. We recommend varying forest management practices and assumptions about outcomes be tested for climate adaptation suitability in large-area experiments, in which continuation of current management is considered one of the treatments, and landscape facets (Brost and Beier 2012), watersheds, or other large-area units comprise treatment units. Several assumptions and model constraints of our study bear noting. We only considered the final 1% of current climates to be marginal, likely making our assessment of climate exposure very conservative. A far greater fraction of forest could be exposed to climate stress than this 1% extreme margin, particularly as climatic water deficit accelerates under warming conditions, even if precipitation is stable or modestly increases Thorne et al. 2015). However, we map climate exposure with many more categories to permit more nuanced interpretations (Online Resource 4).
We also modelled current climatic envelopes based on current forest type distributions. Some forests we analysed likely established under climate conditions different from those they now occupy. In this respect, our analysis is only forward-looking and does not recognize a potentially higher level of stress inherent because establishment conditions may have been different from current conditions. The climate exposure analysis has the advantage of high numbers of replicates of climate space derived from the spatial pattern of each forest type. If a forest type extends greatly beyond the boundaries of the study area, it may occupy climate conditions we did not assess. However, given that the least area used to define forest type's climate conditions is 8131 km 2 , we feel that the climate envelopes of each type are reasonably defined.
The approach we used does not attempt to project future ranges of species or forest types. This is in contrast to many approaches including global dynamic vegetation models (GDVMs) that estimate shifts in vegetation (e.g. Lenihan et al. 2003;Notaro et al. 2012) and species distribution models (SDMs) that predict where individual species might find suitable future climates (Friggens et al. 2013;Comer et al. 2012). However, such approaches can potentially be used collectively, to more fully understand the spatial congruence (Seo et al. 2009) of projections. Combining our spatially explicit predictions, of where climatic stress would be higher and the GDVMs and SDMs providing estimates of what types of species or vegetation could be likely to move into the areas of high stress should they be vacated (Notaro et al. 2012;Thorne et al. 2016), could help focus future research and guide management actions.
We focused on landscape-scale patterns of forest types with moderate spatial resolution to analyse relative climate exposure. The distribution of our forest types may not be the same as the distribution of their constituent species, and the species may respond independently to climate change. Further, even when individual 300 × 300-m pixels are recorded as highly exposed, there may be microclimates within them that allow survivorship (Dobrowski 2011). Studies of shifting species distribution patterns have recognized the importance of microclimates for population persistence (Hannah et al. 2014). However, we chose the 300-m grid resolution, a general level of vegetation classification and in situ impact assessments because they are management-relevant scales, where most public resource agencies set management goals based on land units described by dominant land cover types.
Finally, we assessed vulnerabilities to a limited set of stressors (fire, pests, pathogens) currently influencing the landscape. In addition to changing climate, land use change ) and many associated anthropogenic disturbances can affect forests, such as amplification of drought risk (Diffenbaugh et al. 2015). We did not conduct a detailed assessment of each potential stressor to forest condition under changing climates. The forests of the Southwest are likely to encounter additional stressors that interact with changing climates in the future. In their favour, Southwest forests may have a high capacity to resist these stressors, until or unless the processes of mortality and regeneration are directly affected.

Conclusions
Given the distributed nature of Southwest forest ownership and management, enhancing collaboration among landowners, government, and stakeholders will likely be necessary to build regional forest ecosystem resilience. The fact that a large fraction of Southwestern forests are on public land makes reconciling management actions both simpler (fewer landowners with the potential for more alignment of values and practices) and more difficult (due to the diverse goals of non-owner stakeholders). Providing an environment in which resource managers can experiment with climate adaptation strategies could have major advantages over an environment where legal actions constrain land managers to business-as-usual approaches.
This paper specifies explicit spatial hypotheses about the location and degree of climate exposure and potential forest vulnerabilities. Despite the uncertainty in these and other forecasts, they point to the growing need for forest adaptation management that anticipates climate change effects. Such adaptation actions can also often help achieve other societal goals, such as sustainable development, disaster risk reduction, and improvements in quality of life and can therefore be incorporated into existing decision-making processes (Bierbaum et al. 2014). Continuing to improve our understanding of the climate exposure, sensitivity, and adaptive capacity of Southwestern forests will help inform adaptation plans that help preserve their biodiversity, ecosystem services, and societal value.