Vulnerability of ecosystems to climate change moderated by habitat intactness

Introduction

Human-induced loss and fragmentation of natural land cover are major causes of the current global biodiversity crisis (Pimm & Raven, 2000; Millennium Ecosystem Assessment, 2005; Butchart et al., 2010), while additive or synergistic impacts of climate change are likely to exacerbate the pressures of habitat loss on species and ecosystems (Jetz et al., 2007; Brook et al., 2008; Pereira et al., 2010; Mantyka-Pringle et al., 2012). In particular, large-scale shifts in vegetation biomes due to climate change have altered boreal, temperate, and tropical ecosystems and future shifts may only leave limited refugia from land use and climate change (Scholze et al., 2006; Gonzalez et al., 2010).

Refugia are habitats to which components of biodiversity retreat, in which they persist, or from which they can potentially expand under changing environmental conditions (Keppel et al., 2012). The identification and protection of refugia from climate change and habitat loss offers an important adaptation strategy to conserve biodiversity. Agencies that manage national parks and other protected areas need spatial information on the locations of potential refugia and vulnerable areas to effectively prioritize adaptation actions for natural resource management (Baron et al., 2009; Oliver et al., 2012; Alagador et al., 2014).

While numerous efforts have projected the impacts of climate change on future distributions of biodiversity at scales ranging from regional (e.g. Carvalho et al., 2010) to global (e.g. Thomas et al., 2004; Foden et al., 2013), much less research has examined the additive effects of climate change and habitat loss. Most such efforts have analysed synergistic effects at either relatively local scales (e.g. Klausmeyer et al., 2011; Cabral et al., 2013; Ponce-Reyes et al., 2013; Riordan & Rundel, 2014) or globally, but focusing on individual taxa (e.g. Jetz et al., 2007; Visconti et al., 2011), or on the effectiveness of protected areas in the face of climate change (e.g. Hannah et al., 2007; Lee & Jetz, 2008).

In contrast, a recent study (Watson et al., 2013) analysed future vulnerability of ecoregions, rather than species, to climate change at the global scale based both on climate projections and the fraction of natural land cover in each ecoregion. Such an approach has considerable value for proactive conservation strategies as it offers a practical way of identifying refugia (Hodgson et al., 2009; Watson et al., 2013). Although global scenarios of land-use change (Millennium Ecosystem Assessment, 2005) suggest that current patterns of habitat loss may be poor predictors of future climate and land-use change-driven losses (Lee & Jetz, 2008), such possible future losses are not inevitable. Indeed, proactively preventing future habitat losses in intact regions rather than recreating habitat in vulnerable areas offers a potentially very effective climate change adaptation strategy (Hodgson et al., 2009).

Here, we analyse the vulnerability of ecosystems globally to climate change based on current levels of habitat intactness and vulnerability to biome shifts at two spatial scales based on three different measures of intactness. We then use this information to identify the locations of potential large-scale (macro) refugia. While conceptually similar to Watson et al. (2013), our analysis builds on existing efforts in five important ways.

First, we ran all analyses at two spatial scales – coarse (48 × 48 km) and medium (4.8 × 4.8 km) – thereby examining scale-dependency, a good practice for large-scale conservation planning (conservation biogeography) (Whittaker et al., 2005).

Second, we quantified the conservation value of remaining natural land cover in all analyses by calculating relative species richness (RSR) scores, which are based on the extent of natural habitat weighted by the species-area relationship (SAR) (Arrhenius, 1921). This is important because the relationship between the extent of habitat (measured by the extent of a given natural land cover) and the number of species it supports is not linear, but rather follows the power-law relationship described by the SAR (Rosenzweig, 1995).

Third, we quantified the effect of changing the definition of habitat intactness on our findings, as there is no single way of identifying the intactness of an ecosystem. We did this by comparing the extent of refugia and areas of high vulnerability at both coarse and medium spatial scales based on RSR scores that were calculated based on the coverage of (i) all natural land cover classes, (ii) areas that were at least 50% wilderness (sensu Sanderson et al., 2002), and (iii) areas with at least 50% designated as a protected area. We used these three measures of intactness to capture the generally diminished vulnerability to future habitat losses, and hence to climate change, that characterize both wilderness areas (Boakes et al., 2010) and protected areas (e.g. Hannah et al., 2007; Gillson et al., 2013).

Fourth, we used data on the vulnerability of ecosystems to biome shifts due to climate change (Gonzalez et al., 2010) that use historical climate change trends and future vegetation projections of a dynamic global vegetation model (DGVM) to identify potential refugia. In contrast to the equilibrium climate niche methods used by Watson et al. (2013), DGVMs dynamically model the interactions of biogeography, biogeochemical cycling, and wildfire for every pixel in an analysis area at a time step of days or months.

Finally, the data on the vulnerability of ecosystems to biome shifts due to climate change (Gonzalez et al., 2010) use both historical and projected data, in contrast to Watson et al. (2013), who only used future projections. Because vegetation often responds slowly to changes in environmental conditions, a time lag between a change in climate and a shift in vegetation can commit an ecosystem to biome change long before any response manifests itself. Therefore, future vulnerability is partially a function of past climate change. Using historical data account for impacts of climate change that have already occurred. This can provide a more complete assessment of vulnerability than future projections alone.

Materials and methods

Results

Our results show that three-quarters of global terrestrial vegetated area is vulnerable to climate change resulting from high vulnerability to biome shifts and low levels of habitat intactness. However, the spatial extent of such areas and of refugia to future impacts of climate change is dependent on the definition of habitat intactness and on the spatial scale of the analyses of habitat intactness. Globally, 28% of terrestrial vegetated area can be considered refugia if all natural vegetated land cover (nlc) is considered (Table 1; Fig. 1). This, however, drops to 17% if only areas that are at least 50% wilderness (ltw) at a scale of 48 × 48 km are considered and to 10% at a scale of 4.8 × 4.8 km (Table 1; Fig. 2). Only 2% of vegetated area globally is classified as refugia when considering areas at least 50% protected by the global protected area network (wpa) at the 48 × 48 km scale; this drops to 1% at the 4.8 × 4.8 km scale (Table 1; Fig. 3).

While up to three-quarters of global terrestrial vegetated area is highly to very highly vulnerable to the combined effects of low habitat intactness and biome shifts (Table 1), the spatial extent of refugia varies considerably among continents. Europe has the lowest fraction of its area in refugia of all continents, followed by Asia (Table 1; Fig. 1). By contrast, over half of the area of Australia and approximately a third of the vegetated area of South America and Africa are classified as refugia, based on the ‘nlc’ definition of intactness (Table 1; Fig. 1). South and Northeast Asia have the most extensive areas of high vulnerability (Fig. 1), but Europe also has few areas with low or very low vulnerability (refugia), as do large sections of North America (Fig. 1). Again, refugia are consistently less widespread when only wilderness is considered, and much less widespread when only protected areas are considered.

The spatial scale of the analyses also affected our findings. For the wilderness and protected areas measures of intactness, refugia, and areas of high and low vulnerability are more widespread at the coarse than the medium spatial scale. By contrast, for the natural land cover measure of intactness, spatial scale did not affect the results when considering both vulnerability to biomes shifts and habitat intactness (Table 1).

Our results also reveal extensive variation in the extent of refugia and areas of high and very high vulnerability among different biomes. As a fraction of biome area, deserts, tropical evergreen broadleaf forest, and tundra and alpine have the greatest fraction of area in potential refugia at the medium scale for the three definitions of intactness (Fig. 4, Table S3). By total surface area, tropical evergreen broadleaf forest has the greatest area in potential refugia at both spatial scales and all definitions of intactness (Table S4). The deserts and tropical evergreen broadleaf forests of central Australia and the western Amazon, respectively, encompass the most extensive refugia globally. In contrast, a maximum of 6% of temperate mixed forests are in potential refugia, while other temperate forests and tropical woodlands are also poorly represented in potential refugia (Fig. 4, Table S3). Again, the definition of habitat intactness had a considerable impact on this finding. For example, wilderness-dominated refugia cover only a small fraction of temperate shrubland (1% of biome area at the medium scale), though nonwilderness refugia of this biome type remain relatively widespread (22% of biome area). By contrast, wilderness-dominated refugia of boreal forests remain fairly widespread, and cover over half of the roughly 25% of the area of this biome classified as refugia.

For the US National Park System, the wilderness (ltw) analyses indicate that up to one half of system area is in potential refugia while up to one quarter is in areas of high to very high vulnerability (Table S6). Refugia are mainly in Alaska and remote parts of the interior western United States (Figure S4). Boreal conifer forest has the highest fraction of biome area in potential refugia (48 ltw) while the tropical woodland and temperate mixed forest biomes are most vulnerable (Table S7). In terms of absolute area, the tundra and alpine biome has the most extensive refugia while temperate shrubland is most vulnerable (Table S8).

Discussion

Our results add to the growing evidence (Jetz et al., 2007; Lee & Jetz, 2008; Gonzalez et al., 2010; Watson et al., 2013) that the combination of low habitat intactness and vulnerability to climate change threatens biodiversity in many of the globe's terrestrial ecosystems. They are the first to show that the definition and scale of intactness has a major impact on the projected extent of vulnerability. Our results have considerable conservation implications, as habitat intactness gives a measure of the adaptive capacity of species and therefore can comprise an important component of the relative vulnerability to climate change in different regions (Watson et al., 2013).

While the value of quantifying habitat intactness for the identification of spatial priorities for climate change conservation is undisputed, identifying what constitutes sufficiently intact habitat is less straightforward. First, the spatial scale of the analysis and the natural land cover types that are considered as habitat are important, because what constitutes sufficient habitat is a species-specific concept (Wiens, 1989). For example, a region that contains forest blocks of up to 10 ha in size may offer sufficiently intact habitat for most forest beetles in the region. The extent of forest cover, however, is not a good measure of habitat intactness for grassland species that inhabit the same region, nor are ~10 ha forest fragments sufficient for large, forest-dependent top predators. Second, context matters in the sense that a region with extensive natural habitat but high levels of human activity (i.e. urbanization) in the nonhabitat (matrix) parts of the region is likely to have less biodiversity (Kupfer et al., 2006; Koh & Ghazoul, 2010) and lower adaptive capacity (Gillson et al., 2013) than a region with similar amounts of natural habitat but a more benign matrix (i.e. agroforestry). Finally, our results suggest that regions with high levels of habitat intactness and extensive protected areas are more resilient to biome shifts due to climate change than regions with similar levels of intactness but without protected area status (Gillson et al., 2013). Nevertheless, we recognize that protected area status has not seemed to mitigate some impacts of climate change occurring across extensive areas, such as increased wildfire or forest dieback. Furthermore, the high level of vulnerability under the analyses that used the protected area definition of intactness (wpa) suggests that the current configuration of the world protected areas system may be insufficient to guard against the combined effects of climate change and habitat fragmentation.

An additional issue is whether to only measure habitat extent or whether the configuration of the habitat (habitat fragmentation) also needs to be considered. For most species, habitat extent is much more important than configuration (Fahrig, 2003), but again the relative effects of extent and configuration are species-specific. The species-focused nature of most studies examining the combined effects of habitat intactness and climate vulnerability is the most likely reason for a lack of other work considering different metrics of intactness, as the measures of intactness chosen are simply the ones most appropriate for the species considered. As Watson et al. (2013) point out, however, ecosystem-based assessments of vulnerability to climate change provide an important complement to such species-based analyses. Our results show that, for ecosystem-focused analyses, how intactness is measured has a major impact on results.

Indeed, the degree to which our findings support previous studies depends on the definition of habitat intactness. Our results are broadly similar to those of Watson et al. (2013) when we use natural land cover alone to calculate our RSR scores (nlc). We both identify the most vulnerable areas globally to be the Indian subcontinent, southeast China, and parts of South America, Australia, and central Europe, though Watson et al. (2013) show larger areas of vulnerability in North America and Europe. The broad similarity in these results is not surprising since Watson et al. (2013) use extent of natural land cover per ecoregion as an indicator of habitat intactness. Our results, however, show larger refugia in parts of Siberia and northern North America, perhaps because our analyses used both historical and projected climate, while Watson et al. (2013) only used projected climate. If we use the wilderness (ltw) definition of habitat intactness, our results additionally highlight the near-complete lack of refugia in Europe, Central America and Southeast Asia (outside of Indonesia), while the protected areas (wpa) definition of intactness highlights the near-complete lack of protected refugia. Indeed, one of the more sobering results that emerges from our analyses is quantitative support for assertions (e.g. Hannah et al., 2007) that the current protected area system is wholly inadequate to protect ecosystems in the face of climate change, as refugia that fall at least 50% within protected areas cover a maximum of 2% of the vegetated areas of the Earth. As the strictly protected areas, we consider here (IUCN categories I – IV) cover 6% of the global terrestrial area (Jenkins & Joppa, 2009), this means that only 17% (medium scale) or 34% (coarse scale) of protected areas are also refugia.

Our analyses also generally support work based on climatic stability alone (e.g. Iwamura et al., 2010). They highlight that, despite ongoing deforestation, tropical rainforests show relatively low vulnerability to shifts at the biome level and contain extensive areas of potential refugia (Malhi et al., 2008), although vulnerability may be higher at the species level (Reu et al., 2011). The intactness of upper Amazonian forests leads our analyses to place higher value on these areas than climate-only analyses. Our results also highlight that, despite the relatively low climatic stability of boreal forests (Gonzalez et al., 2010; Watson et al., 2013), the intactness and wildness of this biome allow it to contain fairly extensive climate refugia compared to temperate forests, which are highly vulnerable due to low habitat intactness and relatively low climatic stability.

The relatively large differences in results between the medium and coarse-scale analyses indicate that medium-scale pixels with high (and low) natural land cover frequently cluster within coarse-scale pixels. For example, most coarse-scale pixels (48 × 48 km) with less than 50% wilderness will tend to have some medium-scale (4.8 × 4.8 km) pixels within them with over 50% wilderness. The converse is also true; coarse-scale pixels with over 50% wilderness tend also to include medium-scale pixels below this threshold. As a result, any of our analyses that involved thresholds (50% wilderness or protected area) tended to have more extreme values (high or low vulnerability) at coarse scales; medium-scale analyses gave more intermediate vulnerability scores due to the averaging of the relative richness scores. These results are unsurprising in that it is well-established that larger pixel sizes (coarse grain analyses) can lead to reductions in perceived deforestation (e.g. Nepstad et al., 1999), but they nonetheless have important conservation implications. For example, the medium-scale analyses are important as some areas identified as having high vulnerability to climate change at coarse scales in the wilderness analysis have only moderate vulnerability at medium scales. This indicates that some wilderness-dominated 4.8 × 4.8 km pixels exist and therefore have considerable conservation value for species with smaller habitat requirements. As such mismatches between scales are frequently located in relatively human-dominated regions (e.g. parts of the conterminous United States and Scandinavia in the wilderness analyses) with no remaining large wilderness areas, these medium-scale areas of moderate vulnerability may be of considerable importance as they may represent the areas of the highest conservation value in such regions. Simply running the analyses at medium scales, however, can also be misleading because many areas have low vulnerability at coarse scales (i.e. they are climate refugia) but moderate vulnerability at medium scales. Such areas have considerable conservation value for most species but are characterized by pockets of high anthropogenic activity. It is important to note that changing pixel sizes would quantitatively change our findings, but not the main conclusion – that the perceived habitat intactness of any given region can depend on the spatial scale of the analysis. Changing the thresholds in our analyses (i.e. to 75% wilderness) would also have a minor quantitative effect on our findings (Table S5), but again would not affect our main conclusion.

Our study has a number of limitations with management implications. The first set of these relate to the scale at which our results are useful for managers. Our results are limited by the quality of the land cover data [the associated uncertainty in classification accuracy for land cover varies 68–74% (Defourny et al., 2010)], class aggregation to the biome level, and the quality and resolution of the biome shift data [discussed further in Gonzalez et al. (2010)]. Moreover, the GlobCover land cover class definitions do not provide any direct information on land use. For example, a broadleaf class might contain natural and/or managed forest, creating spatial variation in habitat quality within a single class. In addition, by focusing on the dominant natural land cover type in each pixel, due to the overwhelming importance of habitat extent on biodiversity (Fahrig, 2003), we may undervalue regions with high natural habitat heterogeneity. Areas of high habitat heterogeneity have been shown to promote population stability, albeit at a much finer spatial scale (Oliver et al., 2010). As such, our results are intended to inform broad-scale landscape conservation planning, rather than fine-scale local efforts.

Similarly, our results are affected by the limitations of the Last of the Wild and WPDA datasets. A major limitation of the Last of the Wild – already highlighted by Sanderson et al. (2002) – is the age of the underlying datasets. Even the updated version (published 2005) that we use here is based on data collected between 1999 and 2004 (Wildlife Conservation Society, Center for International Earth Science Information Network, 2005). This means that, in some rapidly urbanizing areas (e.g. Southeast Asia), the extent of refugia may be considerably less than our results suggest. The main issues with the WDPA are those of omissions – e.g. parks for which vector boundary data is missing and considerable variability in the de facto protection provided by the protected areas. The former is likely a relatively minor issue for the global and continental-scale analyses we conduct here (Jenkins & Joppa, 2009) but the latter is potentially important in many areas (e.g. Leroux et al., 2010). That said, evidence strongly indicates that, on the whole, the global protected area network – as mapped by the WDPA – reduces natural land cover conversion (Joppa & Pfaff, 2011).

A second limitation results from our focus on biomes, rather than species. Use of the biome as a unit of analysis may understate vulnerability because it allows for some changes in species composition without conversion into a different biome. We do not examine how the distributions of the refugia that we identify overlap with global patterns of biodiversity or with global, largely species-driven conservation prioritization schemes [e.g. hotspots (Myers et al., 2000)]. Our results are therefore of themselves insufficient for prioritization of specific taxa and not intended as such. Instead, our species-neutral but theoretically grounded measure of intactness (relative species richness) offers an important complement to species-focused analyses and conservation approaches. This is because an ecosystems approach avoids a number of issues arising from species-related conservation. For one, we avoid the issue of taxonomic bias, which can, for example, affect conservation network design (Grand et al., 2007) and spatial conservation prioritization efforts (Grantham et al., 2010; Saetersdal & Gjerde, 2011). We also avoid the problem of assessing the conservation value of under-surveyed areas; a recent study shows that highly intact tropical forests are likely to contain both larger numbers and proportionately more undiscovered species than less intact areas (Giam et al., 2012). Finally, taxon-specific analyses of climate vulnerability can miss potentially very important impacts of interactions between species (Van der Putten et al., 2010).

A third limitation is that our analyses of habitat intactness are based on current land cover and do not use future projections, which may reveal even more substantial vulnerability of ecosystems to habitat loss (e.g. Lee & Jetz, 2008). Again, our analyses are intended to complement such studies and encourage proactive conservation measures, as reduction in nonclimate change related threats is an important climate change adaption strategy (Morecroft et al., 2012) and maintaining high levels of existing natural habitat is a particular priority (Hodgson et al., 2009). Finally, more study is required to integrate our results into global and broad-scale conservation planning efforts since our analyses contain no information on returns on investment (e.g. Iwamura et al., 2010) or considerations of other social goals such as maintaining ecosystem services (e.g. Venter et al., 2009).

While the spatial resolution of the present analysis is too coarse for use in small areas, the results can be useful for broad landscapes. The US National Park Service, which manages an extensive system of protected areas across North America, considers the vulnerability of ecosystems to biome shifts as one potential factor for the prioritization, across the system, of natural resource management adaptation measures, such as strict conservation of refugia and prescribed burning of vulnerable areas (Gonzalez, 2011). Similarly, the US Fish and Wildlife Service considers vulnerability to biome shifts in managing its separate system of wildlife refuges (Griffith et al., 2009). Our results indicate substantial vulnerability of the US National Park System to biome shifts and habitat loss.

In conclusion, our research adds to the few efforts that have analysed the vulnerability of ecosystems to both climate change and habitat intactness globally and illustrates that how habitat intactness is measured has a major impact on such analyses. We identify the locations of potential refugia and vulnerable areas, spatial information potentially useful for the adaptation of natural resource management planning at a broad scale. Our spatial results (data files available from the corresponding author) can help resource managers more effectively develop adaptation measures, including the identification of refugia for strict conservation and potential corridors to facilitate dispersal as vegetation and wildlife ranges shift. Despite the magnitude of the challenges of habitat loss and climate change, refugia may continue to provide hope for the conservation of wild species.