A climate-change adaptation framework to reduce continental-scale vulnerability across conservation reserves

. Rapid climate change, in conjunction with other anthropogenic drivers, has the potential to cause mass species extinction. To minimize this risk, conservation reserves need to be coordinated at multiple spatial scales because the climate envelopes of many species may shift rapidly across large geographic areas. In addition, novel species assemblages and ecological reorganization make future conditions uncertain. We used a GIS analysis to assess the vulnerability of 501 reserve units in the National Wildlife Refuge System as a basis for a nationally coordinated response to climate change adaptation. We used measures of climate change exposure (historic rate of temperature change), sensitivity (biome edge and critical habitat for threatened and endangered species), and adaptive capacity (elevation range, latitude range, watershed road density, and watershed protection) to evaluate refuge vulnerability. The vulnerability of individual refuges varied spatially within and among biomes. We suggest that the spatial variability in vulnerability be used to define suites of management approaches that capitalize on local conditions to facilitate adaptation and spread risk across the reserve network. We conceptually define four divergent management strategies to facilitate adaption: refugia, ecosystem maintenance, ‘‘ natural ’’ adaptation, and facilitated transitions. Furthermore, we recognize that adaptation approaches can use historic (i.e., retrospective) and future (prospective) condition as temporal reference points to define management goals.


INTRODUCTION
Rapid climate change heightens the need for coordinated reserve networks to accommodate dynamic ecological patterns (Halpin 1997, Hannah 2010. However, to be effective, conservation reserve networks must be coordinated at continental, regional and local scales (Soule and Terborgh 1999). This criterion of planning at multiple spatial scales for multiple resources within a reserve network is problematic because many climate change vulnerability assessments have been based on single species or resources, such as a habitat or ecosystem type (Dawson et al. 2011). A new approach is needed for assessing the vulnerability of reserve units, which are predefined parcels of land in dynamic landscapes, in order to promote a coordinated adaptation response to climate change and other environmental stressors within a conservation network.
Adaptation in a management context is defined as ''the adjustment in natural or human systems in response to actual or expected climatic stimuli or their effects, which moderates harm or exploits beneficial opportunities'' (IPCC 2007). Traditionally, natural resource managers have used historic condition as a management benchmark (Hunter 1996). Rapid climate change and other anthropogenic change have caused ecologists to reconsider whether historic condition is a viable goal (Millar andWolfenden 1999, Schroeder et al. 2004). Ecosystems are now seen as complex, adaptive systems with multiple possible trajectories (Chapin et al. 2009). Therefore, adaptation can be retrospective or prospective, which we define here as having different temporal reference points or benchmarks. Prospective adaptation is proactive and works with climate change trajectories; retrospective adaptation works against climate change, towards historic conditions. The former approach manages the system towards a new climate-changeinduced steady state, whereas the latter abates the impact by trying to maintain the current condition despite climate change.
Deciding when to apply retrospective or prospective strategies can be problematic for land managers (GAO 2007). Low-risk and nonintervention strategies have been advocated for facilitating adaptation on conservation reserves; e.g., increasing the redundancy and representation of habitat types and increasing landscape connectivity (Griffith et al. 2009, Heller and Zavaleta 2009, Mawdsley et al. 2009). However, the climate envelopes of many species have been forecast to move rapidly across large geographic areas (for a U.S. example, see Iverson and Prasad 2001). In addition, some large geographic areas have been forecast to experience novel species assemblages in the future due to high species turnover ). In response to directional change at continental scales, managers may need to engage in high-risk adaptation strategies, such as long-range translocations of species to places they have never occurred before (McLachlan et al. 2007). In these cases, to spread the risk of failure and/or unintended ecological consequences, managers of conservation reserves will need to strategically coordinate strategies for facilitating adaptation at scales larger than the landscape matrix surrounding any individual reserve.
In this paper, we use the National Wildlife Refuge System (NWRS) as a case study to demonstrate a new approach to managing reserve networks in a rapidly changing climate. The NWRS is a 600,000 km 2 reserve network managed by the U.S. Fish and Wildlife Service (USFWS). Although individual refuges have management priorities or purposes that originate from legislation outlined when they were established, the NWRS Improvement Act of 1997 (Public Law 105-57) unifies the 540 individual refuges into a coordinated system with an overarching mission to conserve fish, wildlife, and plants and their habitats, and a derivative policy that promotes the maintenance of biological integrity, diversity, and environmental health across the NWRS. This organic legislation makes the NWRS an ideal network to apply a vulnerability approach that focuses on minimizing species extinction at a continental scale.
Vulnerability is defined as ''the degree to which a system is susceptible to, and unable to cope with, the adverse effects of climate change'' (IPCC 2007). Vulnerability depends on exposure to climate change, the sensitivity of the system, and the adaptive capacity of the system (see Table 1 for definitions). Using these three components, we conduct a spatial analysis of vulnerability across the NWRS. Our final product is an adaptation framework, based on vulnerability that describes how prospective and retrospective approaches can be strategically applied across a continental-scale reserve network to facilitate adaptation while spreading risk.

METHODS
We conducted a national-scale vulnerability assessment of NWRS lands in the United States with GIS data from high-quality, public sources. We assessed vulnerability using seven variables representing climate change exposure, sensitivity, and adaptive capacity of refuge lands (Table 1, Fig. 1). We calculated a Pearson correlation matrix for the vulnerability indicators to ensure v www.esajournals.org that the continuous variables provide relatively independent measures (r , 0.7).
We used the legislative boundaries to delineate NWRS lands (USFWS 2010). These boundaries include lands owned, lands with established management agreements or easements, and lands that have been authorized by Congress for future acquisition. Therefore, the legislative boundaries represent the planned future spatial distribution of refuge lands.

Climate change exposure
We used annual temperature change rate (8C/ yr) to summarize the historic  climate change exposure on refuge lands. We chose to use historic temperature change instead of future forecasts because we wanted the analysis to focus on areas already experiencing change. Temperature change has been linked with very high confidence to changes in natural systems (IPCC 2007). The gridded data, along with gridded statistical confidence estimates for the trend (pvalue), were distributed by Climate Wizard (www.climatewizard.org). We only used the rate estimates from grid cells with a p-value 0.10 associated with the trend. We conservatively assumed that pixels with a p-value . 0.10 had no trend. To generate an annual temperature change rate for each refuge, we averaged the pixels within the refuge.
For the contiguous U.S., Climate Wizard uses the 4-km resolution PRISM (Parameter-elevation Regressions on Independent Slopes Model) climate mapping system (www.prism.oregonstate. edu). The PRISM algorithm interpolates spatial climate data through a process in which individ-ual station data were weighted using expert knowledge to reduce bias caused by sparse or unrepresentative stations and factors that affect climate at finer spatial scales (Daly et al. 2002, Daly 2006 The IPCC (2007) estimates that 1.2-28C increase from pre-industrial temperature in the next 50 years would place 9-31% of species at high risk for extinction. Therefore, we used a 1.28C increase over 50 years to delineate an annual temperature change rate of 0.0248C/yr as a vulnerability threshold. Refuges experiencing temperature change .0.0248C/yr were categorized as having high exposure vulnerability (Fig.  1). We considered areas experiencing 0.0058C/yr (0.258C increase over 50 years) to have low exposure vulnerability because a 0.58C increase from pre-industrial temperature was not linked with ecosystem or biodiversity change (IPCC 2007).

Sensitivity
We defined sensitivity based on whether the Table 1. Variables used in vulnerability assessment. Vulnerability was defined based on Dawson et al. (2011) as the extent to which species or populations within a refuge are threatened due to climate change and has three components: exposure, sensitivity and adaptive capacity.

Component Definition Variables
Exposure extent of climate change experienced by a species or locale.
(1) annual temperature change rate (8C/yr) Sensitivity degree to which species survival, persistence, fitness or regeneration may be affected by climate change.
(2) critical habitat for threatened or endangered species in refuge (yes/no); (3) refuge boundary contains biome boundary (yes/no) Adaptive capacity capacity of a species to cope with climate change, including adaptation responses such as shifting to more suitable local microhabitat or migrating to more suitable regions (4) latitude range within refuge boundary (decimal degrees); (5) elevation range within refuge boundary (m); (6) road density of watershed(s) in which refuge is embedded (m/ ha); (7) percentage of watershed(s) with permanent conservation protection (%) refuge contained critical habitat for threatened and endangered species, and whether a refuge was located at a biome edge. The USFWS maintains a geodatabase of critical habitats (USFWS 2011). Critical habitats are lands designated under the Endangered Species Act to be occupied by an endangered species or to contain essential physical or biological features for a listed species. Threatened and endangered species are more likely to be sensitive to climate change because of their restricted ranges, weak dispersal abilities, or small population sizes  (Wilcove et al. 1998) and therefore are sensitive to environmental change. Olson et al. (2001) delineated 14 global biomes based on flora and fauna. Biomes have been forecast to undergo large-scale shifts under future climate change scenarios (Gonzalez et al. 2010, Murphy et al. 2010. Species responses to climate change are influenced by population changes at range margins which are often associated with biome boundaries (Hampe and Petit 2005). Therefore, biome edges are expected to be more sensitive to climate change. We used the presence of critical habitat and biome edge within the refuge boundary to define high, moderate, and low sensitivity (Fig. 1).

Adaptive capacity
Adaptive capacity, or the capacity of species in a refuge to cope with climate change, increases when species are able to shift to more suitable local microhabitats or to migrate to more suitable regions (Dawson et al. 2011). Both latitudinal and elevational range within a refuge increases the potential for species migration along climate gradients (McNeely 1990). Species in many taxa have already responded to recent climate change by shifting northward in latitude and upward in elevation (Parmesan and Yohe 2003). Therefore, a refuge with a large latitude range or elevation range has greater adaptive capacity. We used the northern and southern extent of each refuge to calculate latitude range, and the minimum and maximum elevation of a refuge (USGS 1999) to calculate the elevation range as indicators of adaptive capacity (Table 1). We sorted refuges into large, moderate, and small climate-gradient categories, using a threshold of 0.28 decimal degrees of latitude range and a 31 m elevation range ( Fig. 1). In the northern hemisphere, species ranges have expanded an average of 6.1 km/decade northward and 6.1 m/decade upward in response to recent climate change (Parmesan and Yohe 2003). Our thresholds are equivalent to 6.1 km/decade and 6.1 m/decade over 50 years.
Other anthropogenic drivers, such as road development and land-use conversion, also influence the capacity of species to move and migrate and therefore affect adaptive capacity (McNeely 1990). Roads increase mortality and avoidance behaviors, creating a partial barrier to population movements and affecting population persistence (Forman et al. 2003). Lands outside of the conservation network are subject to land-use conversion and corresponding habitat fragmentation, habitat degradation, and reduced landscape connectivity (Forman 1995). We used road density (m/ha; U.S. Census Bureau 2001) and the percentage of protected lands (The Conservation Biology Institute 2010) in the watershed(s) (USGS 2006) in which a refuge was located as indicators of anthropogenic factors that influence adaptive capacity (Table 1). Refuges embedded in watersheds with high road density and low percentage of lands in protection were considered to have less adaptive capacity than refuges in watersheds with low road density and a high percentage of protected lands. We used thresholds of 12 m/ha of roads and 25% watershed protection to delineate large, moderate, and small anthropogenic footprint (Fig. 1). To define the road density threshold, we doubled the 0.6 km/km 2 threshold above which populations of large mammals, such as wolves and cougars, decline (Forman et al. 1997). In the conservation literature, thresholds for the percentage of protected lands vary from 8 to 80%, depending on the conservation target (Svancara et al. 2005). We chose a threshold of 25% protected because it corresponded with conservative recommendations for biodiversity protection (Noss 1996). Finally, we categorized refuges with high, moderate, and low adaptive capacity based upon the climate gradient and anthropogenic footprint thresholds ( Fig. 1). The thresholds that we selected were reasonable, based on current literature, but could be modified based on improved information or to address particular issues (e.g., birds vs. trees vs. fire risk).

Evaluating vulnerability
We combined the sensitivity and adaptive capacity information into an index of resilience. The properties of resilience include both the ability to absorb disturbance without fundamental change (sensitivity) and the ability of the system to reorganize, learn and adapt (adaptive capacity; Carpenter et al. 2001). We used the categories of high, moderate and low resilience and high, moderate, and low exposure to assign a vulnerability category to each refuge ( Fig. 1). We then used both resilience (sensitivity and adaptive capacity) and exposure to define four v www.esajournals.org management strategies for refuges to coordinate adaptation efforts across reserve networks: refugia, ecosystem maintenance, ''natural'' adaptation, and facilitated transitions (Fig. 2).

RESULTS
After limiting the analysis to the U.S. and excluding refuges with no climate data, 501 refuges were assessed for vulnerability. The average size for these refuges was 1,356 km 2 with a median size of 43 km 2 . Alaska accounted for most of the large refuges, with their 16 refuges contributing 57% of the total land area currently in the NWRS or slated for acquisition. Alaskan refuges have a median size of 15,618 km 2 . At the opposite extreme, nearly 20% of refuges are ,5 km 2 in size.
Climate change exposure NWRS refuges have warmed an average of 0.0108C/yr (SD ¼ 0.011) over the past 50 years. Warming trends ranged from À0.008 to þ0.0438C/ yr. We classified 229 of 501 refuges as having low exposure based on annual temperature change rate: 11 refuges had slight cooling trends ( 0.0088C/yr), 180 refuges had no statistically significant temperature trend (p . 0.10), and 38 refuges had warming trends ,0.005. The remaining 272 vulnerable refuges included 206 with warming !0.0058C/yr but ,0.0248C/yr (moderate exposure) and 66 refuges exceed the vulnerability threshold of 1.28C with warming !0.0248C/yr (high exposure). In Alaska, the Yukon Flats and Bercharof NWRs have already experienced warming trends .0.048C/yr (Appendix).

Sensitivity
Critical habitat for threatened and endangered v www.esajournals.org species occurs on 111 refuges. Sixty-three refuges are located on a biome edge. We assigned 21 refuges as having high sensitivity because they included critical habitat and biome edge, 132 with moderate sensitivity because they had either critical habitat or biome edge, and 348 with low sensitivity.

Adaptive capacity
Refuges encompass an average of 0.339 (SD ¼ 1.181) decimal degrees in latitude, which is equivalent to approximately 37.3 km. The median latitude range was 0.111 decimal degrees. The refuges with the smallest latitude range (0.001 degrees) were the 0.56 ha Susquehanna NWR and the 3.8 ha Caloosahatchee NWR. Maritime NWR, which includes many islands distributed across the state of Alaska, had the largest latitude range of 19.1 degrees. Refuges contain an average of 135.7 m (SD ¼ 344.4) of elevation. Elevation ranges of refuges vary from 0 to 2621 m. The distribution of elevation range within the NWRS is skewed to small values with a median value of 28 m. Thus most refuges are small and have modest latitudinal and elevational ranges.
Refuge watersheds had road densities that averaged 15.4 m/ha (SD ¼ 12.33) with a median 12.8 m/ha of roads in the watershed(s) where they are embedded. Road density ranged from 0.07 m/ha at Koyukuk NWR in Alaska to 104.6 m/ha at Seal Beach NWR near Los Angeles, California. On average, 20.4% (SD ¼ 23.2) of the watershed(s) in which refuges are embedded are in permanent conservation protection. However, the median watershed protection was 9.2% and ranged from ,1% to 97.2% for the Elk NWR in Wyoming.
Given their climate gradient and anthropogenic footprint, 300 refuges were categorized as having low adaptive capacity. Of the remaining 201, 112 were categorized as having moderate adaptive capacity and 89 as having high adaptive capacity.

Evaluating vulnerability
When our categories for sensitivity and adaptive capacity were combined into an index of resilience ( Fig. 1), we categorized 144 refuges as having high resilience, 284 with moderate resilience, and 73 with low resilience. Vulnerability was widely distributed across the U.S. and not delineated by biome or region, although many of the reserves that were highly or moderately vulnerable were in northern or in populous coastal zones (Table 2, Fig. 3). Seventy-six refuges were classified as having high vulnerability and 264 as having low vulnerability. Of the 161 refuges classified as having moderate vulnerability, 104 had intermediate exposure and resilience, 27 had high exposure and high resilience, and 30 had low exposure and low resilience. Refuges were separated into management strategies based on resilience and exposure and a range of management strategies existed across biomes and regions (Table 3, Fig. 4).

DISCUSSION
In our analysis, the vulnerability of individual refuges, including the main components of resilience (sensitivity and adaptive capacity) and exposure to climate change, varies spatially within and among biomes. Therefore, we suggest that spatial variability in resilience and exposure be used to define suites of management actions that captilize on local conditions to facilitate adaptation and help spread ecological risk across the reserve network.
Various management approaches are available to facilitate adaptation in reserves, and the rationale underlying the choice of adaptation goals for any individual refuge will be influenced by local goals, planning timescales, uncertainty, and risk (Heller and Zavaleta 2009). Retrospective strategies, which maintain historic conditions, are generally risk-averse because they focus on tested conservation practices such as the mitigation of non-climatic stressors, habitat restoration, and land acquisition based on current ecological patterns. Retrospective strategies are most likely to meet conservation goals in refuges with slow rates of environmental change. Even in areas with rapid change, retrospective strategies may be valuable over the short term as a bet-hedging strategy when the uncertainty about future conditions is high and to give extant species time to transition to future conditions.
Prospective actions, which seek to facilitate ecological transitions that are congruent with future climatic conditions, are riskier in the short term because they mold future conditions based on expectations or model outcomes (Heller and Zavaleta 2009). Prospective strategies can increase the likelihood of systems adapting without intervention (natural transition) by ensuring landscape connectivity along climate gradients or with forecasts of ecosystem change that inform more active interventions (Chapin et al. 2007, Murphy et al. 2010). More radical prospective strategies may include management actions to foster ecological transition to a desirable future condition. Desirable future conditions imply active choices by managers about future habitat and species composition, and might involve translocating plant and animal species or genotypes to places they have never occurred (McLachlan et al. 2007), developing genetically modified organisms (e.g., acid tolerant corals), or hydrologic management in anticipation of sealevel rise.

Adaptation framework based on vulnerability
To coordinate individual reserves under the continental-scale adaptation goal of minimizing species extinction, we conceptually define 4 divergent management strategies that facilitate adaptation based on resilience (sensitivity and adaptive capacity) and climate change exposure (Fig. 2). The categories of refugia, ecosystem Fig. 4. Refuges sorted into management strategies for facilitating adaptation based on climate change exposure and resilience (sensitivity and adaptive capacity). Major biomes (Olson et al. 2001) are also shown.
v www.esajournals.org maintenance, ''natural'' adaptation, and facilitate transitions help to delineate whether retrospective or prospective approaches are more appropriate. In all cases, an adaptive management framework will be vital to measure progress toward adaptation goals and to maintain flexibility to react to emerging conditions (Griffith et al. 2009. Refugia.-We suggest that reserves with high resilience and low exposure to climate change could serve as refugia. These reserves will function as strongholds where historic ecological conditions and the associated species assemblages may be maintained over foreseeable climatechange scenarios. Appropriate management activities in these reserves, at least in the short term, would be retrospective and focused on maintaining historic conditions (e.g., managing invasive, exotic species). To maintain species assemblages in refugia, managers may use standard conservation principles to ensure that the reserve size and connectivity are adequate to maintain viability. If not already in place, inventories to document which species are represented and protected in these areas should receive high priority (Dawson et al. 2011). Refugia are also potential sources of biodiversity for other transitioning reserves, so these lands should be assessed for their potential to serve as population sources within the larger region or biome. To this end, partnerships and other collaborative land ownership regimes can help to maintain or enhance connectivity and other landscape qualities that confer resilience. Even refugia may eventually experience climate change and reduced resilience, so maintaining historic conditions in perpetuity may not be a viable long-term management goal.
In our spatial analysis of ecosystem vulnerability, we identified refuges with high resilience. However, in agreement with Scott et al. (2004), our analysis indicates that most refuges within the NWRS are small islands within an anthropogenic and fragmented matrix. In addition, most refuges are undergoing some directional climate change. Therefore, modeling and monitoring of exposure (e.g., climate, sea level) and resilience (e.g., watershed protection and connectivity) provide forewarning of the need to reassess management strategies.
Ecosystem maintenance.-Reserves with low resilience and low exposure to climate change may function as areas where tested conservation principals can work toward ecosystem maintenance. We suggest that adaptation options in these areas, at least in the near future, should be retrospective with the goal of maintaining or restoring historic conditions. Reserves working toward ecosystem maintenance will benefit from standard conservation approaches that manage anthropogenic stressors such as fragmentation, land-use change, invasive species, contamination, and over-exploitation. Within the NWRS, where many refuges are small islands in a fragmented landscape, managers should ensure that the plants, animals, and habitats represented are redundant within the network (Griffith et al. 2009). The low resilience of these refuges may increase the likelihood of ecological transition into non-desirable states. In this case, these refuges may be important for testing the viability and cost of retrospective restoration efforts. In addition, ecosystem maintenance reserves have potential to serve as stepping stones for species shifts across the landscape. ''Natural'' adaptation.-We suggest that reserves with high resilience (low sensitivity and high adaptive capacity) and high exposure to climate change are compatible with ''natural'' adaptation. Within a reserve network, these areas present an opportunity to study how species and ecosystems adapt to directional change without deliberate human intervention. Scientific uncertainty about how ecosystems will respond to climate change is high, so there is a need for some reserves to function as research areas to learn about the costs and benefits of novel assemblages, phenological shifts, dispersal constraints, and functional reorganization. Monitoring in ''natural'' adaptation reserves will provide the background and context to understand how rapid climate change affects extant ecosystems and landscapes. Context monitoring, which tracks a suite of variables that are not related to specific management actions, may prove valuable in these reserves (Holthausen et al. 2005). Context monitoring has been criticized for being inefficient and unfocused (Nichols and Williams 2006), but climate change will likely interact with other local social-ecological changes to create unexpected ecological changes that may not be captured by narrowly focused monitoring prov www.esajournals.org grams.
Although we suggest there is a need to learn about how species and ecosystems will adapt without intervention, uncertainty about ecosystem change is high, so management interventions may be necessary when unanticipated threats emerge (e.g., novel, injurious, invasive species) or species extinction is likely. In addition, the landscape matrix where these reserves occur may foster a diverse, spatial mosaic of adaptation strategies. Geographic diversity in adaptation approaches would allow for learning about adaption without intervention, testing prospective approaches, and engaging in retrospective strategies that maintain historic conditions. The use of retrospective approaches may be costly or impossible to achieve in the long term (Hobbs andHarris 2001, Choi 2007). However, in the short-term, retrospective adaptation may be an important precautionary strategy where future conditions are highly uncertain or when rare and/or endemic species would benefit from additional time to cope with changes.
Understanding directional climate change should be a priority for research and adaptive management on these reserves because they will function to reduce the uncertainty about future conditions for the entire reserve network. Therefore, an understanding of likely future conditions will be necessary. Spatial forecasts based on climate models and vulnerability assessments for species of concern provide tools to understand future conditions. Forecasts should be treated as hypotheses and linked to monitoring efforts in order to reduce uncertainty (Lawler et al. 2008).
Finally, these reserves also present an opportunity to form conservation partnerships that protect resilience elements in regions where development has not yet irreparably impaired landscape integrity and connectivity adjacent to reserves. For example, many rural refuges in the contiguous U.S. would benefit from these partnerships because, although natural cover is available in the surrounding landscape, protection tends to be low and human populations are increasing (Svancara et al. 2009). Even large Alaskan refuges would benefit from efforts that maintain large-scale connectivity as biomes shift (Murphy et al. 2010).
Facilitated transitions.-Reserves with low resil-ience (high sensitivity and low adaptive capacity) and high climate-change exposure can function as areas to test active management to facilitate transitions. These areas have a high probability of ecological reorganization, so managers need to be aware of probable future climate conditions and the species assemblages that could be supported under emerging conditions. When transformations are likely, managers must consider whether the likely future conditions are desirable. In reserves managed to facilitate transitions, managers will engage in prospective actions that include risk and uncertainty. Lowerrisk prospective actions include assessing the potential of the reserve to serve as a ''stepping stone'' for dispersal to other areas and increasing landscape connectivity based on probable future development patterns. However, in some cases, these areas may benefit from higher-risk management due to their isolation and small size.
Higher-risk prospective actions include habitat manipulation and introduction of new species assemblages. These activities provide an opportunity to document and disseminate information about whether prospective management can successfully facilitate the non-linear and complex responses of ecosystems.
Reserves with high levels of anthropogenic stressors (low resilience) may have difficulty identifying the impacts of climate change because these effects may be masked or operate synergistically with other drivers. Managers mistakenly focusing on the wrong drivers of change may apply ineffective (and costly) conservation strategies.

Uncertainty
Several sources of uncertainty are inherent in any vulnerability assessment. These range from model-based uncertainty (the model structure and variables that were included) to uncertainty in parameter values used in the application of the model (IPCC 2007). Our model structure was derived from the definition of vulnerability (exposure, sensitivity and adaptive capacity). We chose the historic trend in annual temperature to represent exposure to climate change because there is more certainty that ecosystem change and biodiversity loss can be linked to temperature than to other climate variables (IPCC 2007). Other potentially important meav www.esajournals.org sures of climate change include precipitation and sea level rise. Any of these variables could be represented as historical or projected means, extremes, or variance. For sensitivity and adaptive capacity, we selected variables that were likely to influence many types of ecosystems and species. Variables tailored to a specific ecosystem or species could also be used and would likely change the outcome of the analysis. The outcome of our vulnerability assessment was also influenced by the thresholds (parameters) we chose to represent breaks between high, moderate and low categories. We used scientific literature to define these thresholds, but in most cases, a range of values may be meaningful. Finally the accuracy of the GIS data layers provides a source of uncertainty.
Adaptive management is a process for planning and managing in the face of inevitable uncertainty (Nichols et al. 2011). Adaptation to changes in climate, land use, and societal goals requires adaptive adjustments that incorporate new information as it becomes available and to respond to emerging conditions, many of which will be unanticipated. We suggest that transparent vulnerability assessments can be useful for strategic planning because the spatial variation in vulnerability helps spread ecological risk across a conservation network. However, the information used and variables engaged should be constantly reassessed and refined to make sure that the refuges categorized as refugia, ecosystem maintenance, ''natural'' adaptation, and facilitated transitions are able to function in that capacity. In addition, the general approach of using spatial variability in vulnerability for strategic planning may be useful at different spatial scales (e.g., regional) or when tailored to a species of concern. CONCLUSION In a world with accelerating climate change, we suggest that conservation reserve networks should be focused on minimizing species extinction by facilitating the adaptation of fish, wildlife, and habitats to emerging conditions (Scott et al. 2008). However, adaptation approaches used by individual managers within a reserve network need to be strategically coordinated to meet the continental-scale goal of minimizing species extinction while being responsive to local condi-tions and stressors. We use the concept of vulnerability to develop a strategic adaptation framework for coordinating management approaches across conservation reserves. Based on spatial variability in resilience and exposure to climate change, managers could tailor local-scale adaptation approaches toward refugia, ecosystem maintenance, ''natural'' adaptation, or facilitated transitions. This adaption framework helps define the role of individual reserves in responding to climate change (and other stressors) within a larger network. We suggest that this adaptation framework be used to identify opportunities for individual reserves to contribute substantively to continental-scale species conservation. The actual strategy or strategies selected on a particular reserve would integrate this network-scale goal with local needs and priorities. In our case study, we applied the framework to the NWRS, a network of over 500 refuges across the U.S. However, the framework could also be applied at regional or smaller scales or across networks with diverse management objectives (e.g., wilderness areas, National Park Service network, and private or public lands managed as working landscapes). In addition, assessments of ecosystem vulnerability could include other variables not considered in this study (e.g., future projections of climate and population). We suggest that our approach to the strategic landscape-level conservation of biodiversity in the face of rapid climate change has broad application to reserve networks elsewhere in the world.