The landscape-moderated dissimilarity of local communities determines landscape-wide biodiversity and overrides negative local effects of habitat fragmentation on biodiversity
The theory of island biogeography, which predicts decreasing species richness with decreasing area and increasing isolation of a habitat (MacArthur & Wilson, 1967), has been applied to predict diversity on oceanic islands, for which it was developed, as well as habitat islands on the terrestrial mainland. The species-area relationship, i.e. increasing species richness with area, has been claimed to be of universal importance as one of the few laws in ecology (Rosenzweig, 1995; Lawton, 1999; Holt, 2010), and along with island biogeography theory shaped early thinking about fragmented landscapes. Habitat loss and habitat fragmentation per se (i.e. altered spatial arrangement of remaining habitat) are widely regarded as central drivers of biodiversity loss (Wiens, 1996; Collinge, 2009). However, the concept of habitat fragmentation has also been heavily criticised as having ambiguous value (Haila, 2002; Fahrig, 2003; Lindenmayer & Fischer, 2006, 2007; Yaacobi, Ziv & Rosenzweig, 2007; Laurance, 2008; Prugh et al., 2008; Collinge, 2009; Bennett & Saunders, 2010, but see Ewers & Didham, 2007). There are at least three distinct arguments why fragmentation per se has been often overestimated as a driver of landscape-wide biodiversity losses and why beta diversity has been underestimated as a driver of landscape-wide biodiversity.
Mechanisms driving biodiversity patterns in fragmented landscapes include not only the separate effects of habitat loss, fragmentation per se
and landscape context (i.e. matrix effects), but also the interaction of these factors with the underlying spatial heterogeneity of the landscape prior to habitat modification and the landscape extent covered (Fig. 2
). For a given total area of habitat, all other things being equal, the greater the degree of underlying spatial heterogeneity in the landscape, the greater the degree of community dissimilarity among habitat patches that are separated by increasing distances. This is because of the well-known fact that sites close to each other are usually more similar in environmental conditions than are distant sites (e.g. Fortin & Dale, 2005
). Hence, when a given amount of habitat area is spread out in a landscape (via
fragmentation) instead of remaining as a single large patch, beta diversity is increased (Fig. 2
; although local population persistence decreases, see below, this section). The open question is whether this spatial pattern arises due to similar patches sampling different species from a regional pool which is related to the random sampling hypothesis of Coleman et al. (1982)
or whether it reflects that there is almost always greater environmental variation among separate patches than found within a single patch. This observation relates to the question of whether conservation of “single large or several small” (SLOSS) habitats better maximizes biodiversity (Simberloff & Abele, 1976
; Simberloff, 1986
; Quinn & Harrison, 1988
; Peintinger, Bergamini & Schmid, 2003
). For instance, Cook et al. (2005)
found in an experimentally fragmented landscape of patches undergoing succession that beta diversity was greater among small patches, than in similarly spaced arrays of samples within large quadrats. Tscharntke et al. (2002b
) showed that 10 ha of protected area made up of 29 small grassland remnants harboured many more species than the same area made up by just 1-2 large habitats (Fig. 3A
). This effect was not simply caused by common and generalist species, as endangered species showed the same pattern (Fig. 3B
). Hence, assuming that small patches do not have any extinction debts (Kuussaari et al., 2009
) or that extinction debts have at least been overestimated (He & Hubbell, 2011
), this case provides clear evidence of the benefit of many small island habitats, collectively spanning a large geographic distance and so covering greater environmental heterogeneity, thereby maximizing landscape-wide biodiversity. Such an increase in beta diversity is sometimes attributed entirely to fragmentation, when in fact it may stem at least partially from underlying spatial heterogeneity and variation in the geographical coverage of patch networks in differing landscapes. The importance of separating the purely spatial components of habitat fragmentation from underlying environmental causes of beta diversity in the landscape has not yet been grasped conceptually in the literature, let alone empirically tested, and thus merits more research attention.
Whenever the matrix surrounding habitat fragments is not entirely hostile to species, but rather contains usable resources, island biogeographical and early metapopulation theory have limited applicability (Daily, Ehrlich & Sanchez-Azofeifa, 2001
; Haila, 2002
; Sekercioglu et al., 2007
; Lindenmayer et al., 2008
). Models of population dynamics in fragments need to be modified to consider colonisation and extinction dynamics due to organism spillover from the matrix (e.g. Janzen, 1983, 1986
; Krauss, Steffan-Dewenter & Tscharntke, 2003a
; Cook, Anderson & Schweiger, 2004
; Pereira & Daily, 2006
; Collinge, 2009
; Holt, 2010
). Matrix habitats can be surprisingly rich, even in species normally found in the fragments. In the Amazonian rainforest, for example, 40-80% of frogs, small mammals, birds and ants typical of primary forest were detected outside forest fragments, in a matrix composed of pastures and regenerating forest (Gascon et al., 1999
). However, edge effects can change with seasons (Lehtinen, Ramanamanjato & Raveloarison, 2003
) and occurrence does not automatically imply long-term individual survival or population viability in modified habitats (e.g. Kuussaari et al., 2009
). The conditions under which these effects are most important might vary systematically between ecosystems or biogeographic regions. For instance, matrix effects appear to be more important in forested than open, and in tropical than temperate landscapes (Sodhi, Brook & Bradshaw, 2007
), but this issue has received scant attention to date.
Most studies do not effectively discriminate habitat fragmentation from habitat loss (see Fig. 2
and review by Fahrig, 2003
), and the few published studies explicitly discriminating these two effects do not support the hypothesis that fragmentation per se
(i.e. fragmentation in addition to habitat loss) reduces landscape-wide diversity (e.g. Yaacobi et al., 2007
). Historically, most research has focused on fragmentation at the patch, not landscape scale, and the landscape-wide amount of habitat was not simultaneously considered as a causal covariate (Fahrig, 2003
; Lindenmayer & Fischer, 2006
; Collinge, 2009
; Mortelli et al., 2010
). According to Fahrig (2003)
, the famous Amazonian forest fragmentation study (Laurance et al., 2007, 2011
) is in this sense a study of forest loss, not fragmentation.
Figure 2. A conceptual figure showing four scenarios of fragmented landscapes, separating habitat loss from habitat fragmentation per se. In addition to habitat loss and fragmentation as two distinct factors, the figure illustrates that splitting a single habitat patch into many habitat fragments leads often (but not necessarily) to a broader geographical area covered by habitat. Because landscape extent covered by fragments, overall habitat heterogeneity and dissimilarity of communities are positively correlated, beta diversity should increase in such a scenario of habitat fragmentation.
Download figure to PowerPoint
Figure 3. Cumulative number of butterfly species in relation to cumulative grassland area (ha) of 33 calcareous grassland fragments (Tscharntke et al., 2002b). These results suggest that on a landscape scale many small habitats capture more heterogeneity and thus higher species richness than single large habitats. For example, 10 ha from 29 small fragments support many more species than 10 ha from 1-2 fragments (3A). (A) % of all (N = 61) species; (B) % of N = 38 species listed in the Red Data Book of the German state Lower Saxony.
Download figure to PowerPoint
Of the three arguments for the possible overestimation of fragmentation effects outlined above, the beta diversity effects have been the least considered. However, a growing body of evidence indicates the dominance of beta diversity in driving overall biodiversity across landscapes (at least in the majority of cases), mitigating negative fragmentation effects. For example, even though individual smaller habitat islands harbour impoverished communities, experience reduced frequency and strength of biotic interactions (Holt, 2010), show higher extinction probability (e.g. Krauss, Steffan-Dewenter & Tscharntke, 2003b; Kuussaari et al., 2009), and are often lacking rare, fragment-area-sensitive species (see below, this section), these negative local patch effects on biodiversity are numerically overcompensated in terms of total species richness by the higher beta diversity among patches (Tscharntke et al., 2002b). Habitat islands rarely have an identical suite of species and are therefore not completely nested subsets of mainland habitat (Tscharntke et al., 2002b; Peintinger et al., 2003; Fischer & Lindenmayer, 2005; Kier et al., 2009). Thus, the mainland-island concept underestimates the importance of heterogeneity in driving community dissimilarity, which increases with distance in a poorly understood way (Gering & Crist, 2002; Gering, Crist & Veech, 2003; Cook et al., 2005). A meta-analysis of over 150 data sets from habitat fragments and oceanic islands showed that 74% of the total landscape-level richness was due to species turnover among habitat fragments and 84% of the combined-island richness was due to turnover among islands. By contrast, species turnover among habitats or islands of different area explained only 27 and 41% of combined richness for habitat fragments and true islands, respectively. Beta diversity (turnover patterns or species gain/loss patterns) therefore has a dominant role in determining biodiversity patterns among fragments or islands, less than half of which can be attributed to variation in habitat or island area (T. O. Crist, in preparation).
Landscape-wide beta diversity has been shown to be poorly, and in some cases not at all, related to local alpha diversity, to contribute much more to overall diversity than does the latter, and generally, to be a better indicator of overall biodiversity patterns (Clough et al., 2007; Hendrickx et al., 2007; Kessler et al., 2009; Flohre et al., 2011). Beta diversity patterns may even provide a picture that contrasts with alpha diversity. For example, comparing bee and wasp communities along a land-use gradient, Tylianakis, Klein & Tscharntke (2005) found that while plot (= alpha) diversity was highest in intensively used agroecosystems, beta diversity was significantly greater in less-intensively-used systems (due to higher habitat heterogeneity and associated higher community dissimilarity), resulting in overall higher biodiversity in the less-intensive systems, and contributing strongly to landscape-scale (gamma) diversity. We should caution that these patterns of beta-diversity dominance should hold for structurally simple and complex landscapes (as defined in section D, hypothesis 7), whereas cleared (extremely simplified) landscapes such as large-scale agricultural monocultures may sustain only a few surviving populations and nested, spatially homogenized communities.
One little-recognised component of the dominance of beta diversity hypothesis is that the same arguments made regarding the partitioning of species diversity across landscape scales can also be made in relation to intraspecific genetic variation. Parallel processes may govern spatial patterns in genetic diversity, as in community diversity. For instance, as with community dissimilarity, genetic dissimilarity increases with geographic distance between habitat fragments and with the landscape extent covered, providing evidence for a ‘parallel’ landscape-moderated intraspecific genetic diversity hypothesis, predicting that the underlying patterns of landscape-wide genetic dissimilarity of local populations determine intraspecific diversity following landscape modification and potentially override negative local effects of habitat fragmentation. The importance of intraspecific genetic differentiation across landscapes has been shown, for example, by studies in which pollen diversity increases the chance of selecting a particularly “good” donor for fertilization - a sampling effect of diversity (Paschke, Abs & Schmid, 2002). Such effects suggest that an increase in dispersal can have positive effects on adaptation. By contrast, reducing dispersal, and thereby maintaining population dissimilarity, can be important because intraspecific diversity differences between small and large habitats are small compared to the effect of intraspecific beta diversity reflecting local adaptation (Eckert, Samis & Lougheed, 2008). Gene flow can impede local adaptation in many plants and animals, which may be critical for effective habitat use (Kawecki, 2008; Leimu & Fischer, 2008). This has been shown for local adaptation of blue tits (Parus caeruleus) to peaks of food availability in deciduous forests, which can be prevented by genetic homogenization with populations that are adapted to use resources several weeks later in evergreen forests (Blondel et al., 2006). Acting through high turnover in local genetic structure, landscape-scale genetic variability maintained by local adaptation may provide insurance against changing environmental conditions (Hughes & Stachowicz, 2004). In the same way, reducing gene flow is considered to be necessary in agriculture to keep a maximum diversity of different breeds (e.g. in farm animals, Simianer, 2005) with a low but non-zero level of cross-type breeding to prevent homogenization while at the same time maintaining a pool of genetic variation that may be needed to deal with environmental novelties (e.g. emerging infectious diseases).
At the same time, beta genetic diversity should become increasingly strongly controlled by the way humans manage landscapes, both in terms of landscape structure and land-use intensity (Harrison, Ross & Lawton, 1992; Dormann et al., 2007). Increasing connectivity between spatially separated populations can be a double-edged sword and depends on whether a local (intraspecific “alpha” diversity) versus landscape (intraspecific “beta” diversity) perspective is adopted. From a local perspective, connectivity decreases random allele losses by genetic drift, inbreeding depression and extinction probability (the so-called rescue effect), for example in bumble bees (Herrmann et al., 2007). This means that from a landscape perspective, landscapes containing originally connected populations with (mostly non-adaptive) genetic differentiation increasing with distance between populations, should be managed to maintain that connectivity according to common conservation advice (e.g. Pullin, 2002). By contrast, connectivity can also enhance homogenization, resulting in the erosion of potential local adaptations. Maintaining genetic variation across landscapes may also have intrinsic conservation value, which is in tension with the goal of enhancement of gene flow for local survival, which occurs at the expense of permitting a strong response to local selection. The relative value of these opposing effects under climate and land-use change is unknown: connectivity will certainly be essential to prevent extinctions, and to maintain a pool of variation to cope with environmental change, but can sub-populations at times be better off with a lack of genetic exchange that facilitates precise evolutionary adaptations to local conditions? Theoretical studies (e.g. Holt, Barfield & Gomulkiewicz, 2005; Kawecki, 2008) suggest that low to moderate levels of gene flow are often optimal for generating and maintaining local adaptation. However, the optimal level of connectivity is likely to differ among species, and may be difficult or even impossible to ascertain without sophisticated models that specify the nature of the environment, the suites of traits under selection, and the nature of the genetic architecture underlying those traits. There is no single level of connectively that is likely evolutionarily optimal for all species at once.
Despite the foregoing arguments, it is important to keep in mind that from a conservation point of view, community composition may change dramatically, even if the level of habitat fragmentation per se has little negative effect on overall biodiversity. There may be individual species that are strongly patch area and edge sensitive and in need of large contiguous areas of natural habitat, and these species may experience high extinction probabilities in small reserves. Conservation managers need to consider the fate of these often endemic and frequently charismatic flagship species, not just species richness, to craft effective conservation policies (Sodhi et al., 2007; Collinge, 2009). The debate on habitat loss versus fragmentation has at times focused almost exclusively on species richness, rather than on community composition, which usually changes greatly and often at the cost of endangered and “valuable” species. For example, rainforest fragmentation reduces biomass and carbon sequestration, because the biggest trees die first (Laurance et al., 2000). Additional ecosystem processes that can suffer from habitat fragmentation include the resistance of fragments to species invasions (Janzen, 1983; Collinge, 2009) and to external disturbances such as hunting and fire (Cochrane & Laurance, 2002). Protection of a complex of both large and small reserves (Noss, 1983; Tscharntke et al., 2002b) over as large an area as possible and maintaining flexible strategies (as to prioritizing few larger or many smaller reserves depending on the conservation focus) will be critical. In any case, expanding ecological research to incorporate a broader, landscape perspective will clearly be preferable to the traditional management focus on single reserves optimizing local biodiversity. Finding sustainable solutions that integrate different conservation goals such as enhancing local biodiversity across different functional or taxonomic groups, promoting landscape-wide biodiversity and associated ecosystem services, as well as maintaining viable populations of patch-area-sensitive (e.g. large or vagile) organisms, is difficult and underlying conservation priorities need to be more clearly articulated.
Future research should focus on separating effects of fragmentation per se from habitat loss in real landscapes, which has been described as “virtually logistically impossible” (Damschen et al., 2006; Collinge, 2009) but is an innovative and, we think, still manageable enterprise (see Krauss, Steffan-Dewenter & Tscharntke, 2003a; Ewers & Didham, 2008; Smith et al., 2009; Brückmann, Krauss & Steffan-Dewenter, 2010). In comparative studies, landscapes should be selected that have a similar total area of remaining habitat but differing levels of fragmentation (number of fragments), or similar levels of fragmentation but different total habitat area. There needs to be an explicit consideration of pre-existing gradients and other forms of environmental variation, which should be reflected in community composition both before and after fragmentation. Furthermore, we need a better mechanistic understanding of, and empirical evidence evaluating, the relative importance of small versus large or isolated versus non-isolated habitat fragments (and dissimilarity in their species composition) for landscape-wide biodiversity conservation. Landscape effects on biodiversity patterns may change along landscape-complexity gradients, latitudinal gradients and between different landscape types or biomes. For example, corridors in forested landscapes and in the tropics may be more effective than in open habitats and temperate regions (Sodhi et al., 2007). Is this prediction based upon the traits of the species involved, or the magnitude of the contrast in the environment between corridors and surrounding matrix habitats? Edge effects may vary depending on fragment size and should be much better quantified (Ewers & Didham, 2006, 2008). The role of beta diversity in contributing to overall diversity patterns needs to be better explored, in particular assessing changes along landscape gradients or successional stages (e.g. Cook et al., 2005). Patterns of beta diversity in relation to the increasing environmental variation with distance need to be explored, while the problem of limited sample sizes potentially causing overestimation of beta diversity needs to be taken into account. The conservation value of small as well as large habitats for reducing extinction probability and maintaining food-web interactions needs to be analyzed in different landscape contexts and across different groups of organisms (from microbes to mammals). However, habitat loss results in a concomitant change in the spatial structure and patterning of habitat, so it is worthwhile looking for integrated assessments of impact that explicitly combine the effects of habitat loss and fragmentation (Ewers et al., 2009). Future research should also address the question of the minimum cumulative habitat area necessary to provide the array of resources needed in a heterogeneous landscape to sustain landscape-scale diversity. Changes in species composition or evenness (Crowder et al., 2010) can have larger effects on ecosystem function than do changes in species richness (Symstad et al., 1998; Wilsey & Potvin, 2000), and this needs to be explored within a landscape framework. The genetic and population-level consequences of short-term rescue versus long-term homogenization effects also need to be empirically evaluated, to define trade-offs between maintaining genetic variation across landscapes versus viable local populations. Such studies should test whether translocation of individuals between long-established isolated populations can swamp local adaptations, while re-connecting populations that have been isolated only recently contributes to reduced extinction probability.