Biol. Rev. (2010), 85, pp. 35–53. 35 doi:10.1111/j.1469-185X.2009.00093.x Critical thresholds associated with loss: a review of the concepts, evidence, and applications

Trisha L. Swift* and Susan J. Hannon Department of Biological Sciences, University of Alberta, Edmonton, Alberta, T6G 2E9 Canada

(Received 6 July 2008; revised 30 June 2009; accepted 9 July 2009)

ABSTRACT

A major conservation concern is whether size and other ecological variables change linearly with habitat loss, or whether they suddenly decline more rapidly below a ‘‘critical threshold’’ level of habitat. The most commonly discussed explanation for critical threshold responses to habitat loss focus on habitat configuration. As habitat loss progresses, the remaining habitat is increasingly fragmented or the fragments are increasingly isolated, which may compound the effects of habitat loss. In this review we also explore other possible explanations for apparently nonlinear relationships between habitat loss and ecological responses, including Allee effects and time lags, and point out that some ecological variables will inherently respond nonlinearly to habitat loss even in the absence of compounding factors. In the literature, both linear and nonlinear ecological responses to habitat loss are evident among simulation and empirical studies, although the presence and value of critical thresholds is influenced by characteristics of the (e.g. dispersal, reproduction, area/edge sensitivity) and landscape (e.g. fragmentation, matrix quality, rate of change). With enough empirical support, such trends could be useful for making important predictions about species’ responses to habitat loss, to guide future research on the underlying causes of critical thresholds, and to make better informed management decisions. Some have seen critical thresholds as a means of identifying conservation targets for habitat retention. We argue that in many cases this may be misguided, and that the meaning (and utility) of a critical threshold must be interpreted carefully and in relation to the response variable and management goal. Despite recent interest in critical threshold responses to habitat loss, most studies have not used any formal statistical methods to identify their presence or value. Methods that have been used include model comparisons using Akaike information criterion (AIC) or t-tests, and significance testing for changes in slope or for polynomial effects. The judicious use of statistics to help determine the shape of ecological relationships would permit greater objectivity and more comparability among studies.

Key words: critical thresholds, habitat loss, fragmentation, configuration, time lag, Allee effects, landscape, simulation, conservation, statistics

CONTENTS I. Introduction ...... 36 (1) Purpose and structure of review ...... 36 II. Possible explanations for critical thresholds, and conservation implications ...... 37 (1) Configuration effects at low habitat cover ...... 37 (2) Allee effects at low habitat cover ...... 38 (3) Time lags ...... 39 (4) Habitat loss ...... 39 III. Evidence for critical thresholds ...... 39

∗ Address for correspondence: PO Box 668, Athabasca, Alberta, Canada T9S 2A6 E-mail: [email protected]

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 36 Trisha L. Swift and Susan J. Hannon

(1) Simulation studies of critical thresholds ...... 40 (a) Effects of simulated landscape characteristics on critical thresholds ...... 40 (b) Effects of species characteristics on critical thresholds ...... 44 (c) Evidence for causes of critical thresholds in simulations ...... 45 (2) Small-landscape studies of critical thresholds ...... 45 (a) Evidence for causes of small-landscape thresholds ...... 45 (3) Large-landscape studies of critical thresholds ...... 45 (a) A comparison of landscape-level study designs ...... 45 (i) Fragmentation effects at low habitat proportion ...... 46 (ii) Allee effects ...... 46 (iii) Time lags ...... 46 (iv) Habitat loss effects alone, for certain response variables ...... 46 (b) Evidence for presence and values of large-landscape critical thresholds ...... 46 (c) Evidence for causes of large-landscape thresholds ...... 47 IV. General comparisons of simulation, small-landscape, and large-landscape study results ...... 48 V. Utility of critical thresholds in conservation ...... 48 VI. Statistical considerations ...... 49 VII. Conclusions ...... 50 VIII. Acknowledgements ...... 50 IX. References ...... 50

I. INTRODUCTION also of conservation interest. For example, all else being equal an increase in the critical threshold level for a species’ abun- Numerous studies have documented the detrimental effects dance would be considered undesirable, since the steeper of habitat loss on various ecological responses (e.g. bird decline would begin ‘‘sooner’’ in relation to habitat loss. body condition: Burton et al., 2006; amphibian population declines: Cushman, 2006; plant reproduction: Aguilar et al., 2006). Recently, there has been growing interest in the (1) Purpose and structure of review shapes of these relationships. Do ecological responses change The purpose of this review is to present theoretical and linearly with habitat loss, or are there ‘‘critical threshold’’ empirical evidence for critical thresholds in species’ responses levels of habitat? A critical threshold is ‘‘an abrupt, nonlinear to habitat proportion in the landscape. We will discuss change that occurs in some parameter across a small range (a) possible explanations for critical thresholds, (b)evidence of habitat loss’’ (With & King, 1999b). The response variable for their occurrence and value in simulated and real land- undergoing this abrupt change may be individual behaviour, scapes, (c) the effect of species and landscape characteristics the of a species, or composition, on the existence and value of critical threshold levels, and among others. The key point is that the magnitude or slope (d) potential uses and misuses for critical threshold informa- of the relationship between the response and the amount tion in landscape management. The review will conclude of habitat in the landscape changes once that amount of with a summary of major trends and recommendations for habitat falls below a critical threshold level. For example, the future research. abundance of a species in a landscape may decrease more Included are studies for which the authors explicitly steeply with habitat loss once the amount of remaining habi- addressed the presence or absence of ‘‘critical thresholds’’, tat falls below some proportion of the total landscape area. ‘‘thresholds’’, ‘‘sudden changes’’, or ‘‘nonlinear’’ relation- For simplicity, this proportion will henceforth be referred to ships, as well as a few which only presented data suggesting as the ‘‘critical threshold level’’ or the ‘‘threshold level’’, and such relationships (e.g. data plots). Different authors seem to ‘‘habitat proportion’’ will mean the amount of habitat as a estimate exact critical threshold levels in different ways, or not fraction of total landscape area. at all. Therefore, when there was no formal statistical assess- The existence of critical thresholds in habitat proportion ment of critical threshold levels, we re-estimated them as is of conservation concern, because small additional losses either the point at which the slope changed (for sharp thresh- of habitat below the critical threshold level may lead to olds), the midpoint of the curve around which the slope abrupt population declines or other important ecological changed (for more gradual thresholds), or the range over changes. Unanticipated, such changes may preclude timely which the response value changed markedly (for categorical conservation measures. Such thresholds also indicate that data), rounded to the nearest 5% (Fig. 1). This was done to some other factor (such as fragmentation, see below) may make the results of different studies more comparable, not become substantive only below certain proportions of habi- necessarily because this is the best or only way to define the tat, compounding the effects of habitat loss at low habitat value of a critical threshold level, which in some cases may proportions. If critical thresholds do exist, then any factors be considered as a range rather than an exact point (e.g. see that may increase or decrease the critical threshold level are Huggett, 2005 for a broader view of ‘‘ecological thresholds’’).

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 37

A B C

Ecological response

0 20 40 60 0 20 40 60 10-20 20-30 30-40 40-50 Habitat cover (%) Habitat cover (%) Habitat cover (%)

Fig. 1. Various types of critical threshold relationships, and estimation of critical threshold value: A) sharp threshold; B) smooth thresholds; C) categorical threshold. Dashed line or bracket indicates estimated threshold value or range, respectively.

We were fairly liberal in what we considered a ‘‘threshold’’, that at low levels of habitat, negative effects of habitat in terms of accepting even rather gradual or ‘‘curved’’ rather fragmentation compound those of habitat loss, such that than sharp thresholds. However, we tried to include only the rate of change in the ecological response is greater than curves/relationships in which there were identifiable ‘‘above expected from habitat loss alone (e.g. Andren,´ 1994). Habitat threshold’’ and ‘‘below threshold’’ portions with different fragmentation is distinct from habitat loss, reflecting aspects overall slopes (generally steeper below than above), rather of habitat configuration (e.g. number of habitat fragments, than curves that were so gradual that there was no reasonably edge density, patch shape), rather than the total amount of clear point of division between the two segments. There habitat in a landscape. Negative fragmentation effects may were two reasons for this. First, we felt it was unlikely that include increased and brood (Donovan critical thresholds in habitat proportion would be absolutely et al., 1997), harsher microclimate (Dolby & Grubb, 1999), abrupt. For example, if negative fragmentation effects start decreased food (Zanette, Doyle & Tremont, 2000), and to compound habitat loss when habitat proportion becomes decreased ability of animals to move across the landscape sufficiently low (see below), then those negative effects may among habitat patches (Belisle,´ Desrochers & Fortin, 2001). become increasingly strong as habitat proportion decreases. Why might fragmentation effects compound those of A perfectly sharp threshold would require an all-or-none habitat loss only at low habitat levels? Simulation models scenario: no fragmentation effects above 30% habitat, for based on percolation theory have shown that some aspects of example, and equally negative effects between 1 and 29% structural fragmentation itself may increase abruptly below habitat. The second reason for our liberal interpretation critical proportions of habitat. In the physical sciences, was related to limitations in ecological data collection. The percolation models have been used to study the movement scatter that is invariably present in ecological data will blur of liquids through material aggregates. This is reviewed the underlying, ‘‘true’’ relationship. Thus, even if very sharp briefly by Orbach (1986) and is summarized as follows. thresholds did exist, they may still be difficult to detect. Toms Percolation models assume a square lattice of cells that & Lesperance (2003) found that even with sufficient data are either randomly occupied or unoccupied by material. to estimate statistically the threshold level in an ecological Two occupied cells are considered connected only if directly data set, the nature of the transition around the threshold adjacent along a horizontal or vertical edge (not diagonally), (sharp or smooth) cannot accurately be determined without and a cluster is made up of connected occupied cells. sufficient data points near the threshold. Liquid can flow only between connected occupied cells, Only studies conducted at the ‘‘landscape’’ level were and thus cannot flow between clusters. When the proportion reviewed (including studies for which species’ responses of occupied cells in the lattice falls below the ‘‘percolation were measured within individual habitat patches but habitat threshold’’ (near 59%), the largest cluster no longer spans proportion was measured from the surrounding area). the lattice from one side to another (Orbach, 1986), ‘‘Landscapes’’ were considered to be mosaics of habitat and thus the liquid can no longer ‘‘percolate’’ through and non-habitat patches (or patches of varying quality), large the lattice. Percolation models that have been adapted enough to be relevant to the response variable and focal to model landscape patterns have found similar patterns: organism (McGarigal & McComb, 1995; Chust et al., 2004). when habitat covers less than 59% of the landscape, the largest habitat patch size decreases abruptly and no longer spans the landscape (Gustafson & Parker, 1992; Andren,´ II. POSSIBLE EXPLANATIONS FOR CRITICAL THRESHOLDS, AND CONSERVATION 1994; Bascompte & Sole,´ 1996). In addition, mean inter- IMPLICATIONS patch neighbour distance increases rapidly below about 40% habitat (Gustafson & Parker, 1992; Andren,´ 1994), and patch number peaks near 30% (Gustafson & Parker, 1992). If a (1) Configuration effects at low habitat cover species is sensitive to these aspects of structural fragmentation, There are several explanations for critical threshold their apparent responses to habitat proportion might also be relationships with habitat cover. The most common is expected to be nonlinear. This is particularly true if an

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 38 Trisha L. Swift and Susan J. Hannon organism behaves like the liquid of the percolation model, (c) disruption of social facilitation, such as reduced encounter able to move within habitat patches but not between them. rates between potential mates, or reduced effectiveness of This is clearly not the case for all species. The effects of species anti-predator strategies that depend on a sufficient number characteristics on critical thresholds in habitat proportion will of participants. Per capita population growth rate is often be discussed in more detail later. used as the measure of fitness (Courchamp et al., 1999). Another possible explanation for Below a threshold , growth rates may even responses to habitat loss is that there may be a nonlinear become negative (Courchamp et al., 1999), though this is not relationship between habitat loss and biological responses always the case (Stephens et al., 1999). Because habitat loss to structural fragmentation (i.e. functional fragmentation). reduces the of a landscape (e.g. Siffczyk For example, if the number of habitat patches is held et al., 2003), there may be a critical threshold level of habitat constant, then inter-patch distances will tend to increase below which Allee effects come into play. Assuming growth as habitat proportion decreases (Andren,´ 1994; Trani & rates become negative below a habitat population threshold, Giles, 1999), but this could have little effect until the ability the population will no longer have the ability to reach or or willingness of the organism to traverse that distance is maintain itself at the carrying capacity of the remaining surpassed. Some forest bird species in forest-agricultural habitat (Greene, 2003). landscapes readily flew across small gaps in the forest, but the A Levins model incorporating both likelihood of doing so dropped steeply for gaps wider than habitat loss and Allee effects (Amarasekare, 1998) supports 50 m if forested detours were available (St. Clair et al., 1998). the idea that Allee effects could lead to critical threshold Similarly, small gliding marsupials were rarely found in tree relationships between population size and habitat amount. patches that were more than 75 m from a network of wooded Before discussing this, some metapopulation terms must be strips, where they were common; 75 m corresponds to the defined. A metapopulation is comprised of a set of local maximum distance of a single glide (van der Ree, Bennett (individuals that are highly likely to interact) & Gilmore, 2004). Patch size also tends to decrease as that are connected through dispersal; metapopulation size habitat proportion decreases (Trani & Giles, 1999). Habitat refers to the number of local populations comprising it patches that fall below a minimum size can presumably no (Hanski, 1991). In metapopulation terminology, a ‘‘patch’’ longer individually support a population (Bevers & Flather, is a unit area of suitable habitat that may be occupied 1999) or home range. As habitat proportion increases, a or unoccupied by a local population (Hanski, 1991). It similar level of subdivision would result in a system of is akin to a percolation model’s habitat cell (not habitat patches that become on average large enough to avoid local patch), except that a metapopulation ‘‘patch’’ has no and/or close enough together to allow frequent particular location relative to other patches because simple recolonization (Fahrig, 1998) or multi-patch home range metapopulation models are spatially implicit. A simple Levins movements. Simulation models that vary habitat proportion metapopulation model without Allee effects predicts that and fragmentation independently have suggested that some the total number of occupied patches will decrease linearly organisms (particularly habitat specialists) are more sensitive with habitat loss. Deterministic extinction occurs at some to the same degree of fragmentation when total habitat proportion of habitat, x (the eradication threshold), equal proportion is low than when it is high (Henein, Wegner & to the proportion of patches that are unoccupied when Merriam, 1998; Fahrig, 1998). habitat = 100% For example, if 80% of patches are occupied The existence of a fragmentation-related threshold would at 100% habitat, then the metapopulation becomes extinct at mean that when habitat cover falls below the threshold, 20% habitat (x = 100–80; the value of x depends on various reducing fragmentation of the remaining habitat may be an model assumptions that are not central to this discussion). effective management strategy. More specifically, reducing However, if Allee effects occur once the proportion of fragmentation should help to maintain the ecological occupied patches drops below a certain threshold, t,then parameter (e.g. population size) closer to what would be there is a range of habitat proportion (x to x + t)in expected from habitat loss alone. which the predicted equilibrium proportion of occupied patches is unstable (Amarasekare, 1998). Within this range, (2) Allee effects at low habitat cover the metapopulation may go extinct, even if the amount Allee effects could also potentially lead to critical threshold of available habitat is above the eradication threshold, x relationships with habitat amount, even in the absence of (Amarasekare, 1998). Thus, the decline in occupied patches fragmentation effects. The term ‘‘Allee effects’’ refers to may become nonlinear (decline more steeply) below an ‘‘Allee situations in which individual fitness is positively related threshold’’ level of habitat (x + t). An Allee-effects-related to population density or size (Stephens, Sutherland & critical threshold could therefore occur in the absence of Freckleton, 1999). Courchamp, Clutton-Brock & Grenfell fragmentation. However, fragmentation could conceivably (1999) describe three main mechanisms to explain Allee interact with habitat loss to divide a population into smaller, effects in small populations: (a) inbreeding depression isolated subpopulations that are individually subject to Allee and a corresponding decrease in fitness; (b) demographic effects, thus increasing the critical threshold level of habitat stochasticity, such as random fluctuations in sex ratios, that would otherwise be observed. On the other hand, which could have a large effect on small populations; and when subdivided populations are connected through random

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 39 dispersal, the population as a whole may be more ‘‘resistant’’ time and the threshold would be more abrupt (see Fig. 4 of to Allee effects than a single, intact population (reviewed by Schrott et al., 2005b), thus reducing the potential for managers Greene, 2003), particularly when dispersal involves habitat to restore population viability before it is too late. selection rather than being random (Greene, 2003). Allee-effects-related habitat thresholds may be of partic- (4) Habitat loss ular conservation concern, because they may severely limit our ability to reverse the effects of habitat loss and frag- Habitat loss alone is likely to lead to nonlinear changes in mentation. In particular, habitat loss and fragmentation some ecological responses. Conceivably, binomial population may cause populations to become demographically limited, responses such as persistence or occurrence could remain at rather than habitat limited (Schrott, With & King, 2005a). 100% over the initial range of habitat loss, particularly if Thus, below some ‘‘restoration threshold’’ for population the initial population size is large. However, as the amount size/habitat amount, habitat restoration would not pre- of habitat moves closer to zero, persistence or occurrence vent further declines. Instead, effective management would probability must at some point also decline towards zero. require efforts to increase reproduction and decrease mor- Thus, a threshold relationship will be evident, with a steeper tality (Schrott et al., 2005a). slope below than above the threshold level. Community-level responses such as (number of species present) (3) Time lags may also exhibit threshold declines in response to habitat loss alone. Consider a group of species, all of which average Time lags (Tilman et al., 1994; Hanski & Ovaskainen, 2002) 5–20 individuals per landscape, depending on the species. occur when an ecological response to a perturbation becomes Each species declines in exact proportion to (linearly with) apparent some time after the perturbation itself. For example, habitat loss, and reaches an average abundance of zero for as habitat loss progresses organisms may crowd into remain- landscapes with no habitat remaining. Such linear responses ing habitat. The affected population may thus maintain high imply that additional factors do not compound habitat loss occupancy levels while habitat loss is occurring, only to crash at low habitat levels. Yet for this community, species richness after habitat loss has stopped (Lamberson et al., 1992). Sim- should decline little until the amount of habitat drops below ulation models have suggested that ecological time lags are a threshold level, and then it will decline more steeply. pronounced when the rate of habitat loss is rapid relative to This is because the ‘‘cloud’’ of scatter around each species’ the demographic response time (e.g. generation time) of the average abundance is unlikely to encompass zero until the organism (Schrott, With & King, 2005b). Indeed, several pop- average abundance (determined by habitat amount) becomes ulations are better predicted by past than by current habitat sufficiently low. Thus, as habitat amount decreases below levels (Hanski & Ovaskainen, 2002; Cowlishaw, 1999; Gu, some threshold level, the average number of species that Heikkila & Hanski, 2002). Time lags could lead to apparently are present in a given landscape should decrease sharply. nonlinear responses to habitat loss. Consider a population We confirmed this expectation by calculating ‘‘species that declined linearly with habitat loss when habitat loss was richness’’ for a community of 15 artificially generated species’ slow, disappearing below 20% habitat. If habitat loss was abundance data sets with Poisson distributions (data not rapid enough to produce a time lag, then the population shown). Although we made the average abundance of each would initially decline less steeply with habitat loss than ‘‘species’’ decline linearly with habitat loss as described above, expected. The corollary is that if habitat loss is ongoing, then richness fell off sharply below a threshold level of habitat. The at some point (e.g. after habitat levels reach or fall below exact threshold level is not important here; it will depend 20%), the decline must become steeper (plotted against % on the initial sizes of the populations in the community. habitat) as the population moves towards extinction. This Smaller population sizes will lead to higher threshold levels, nonlinearity would be accentuated if the rate of habitat loss and a greater range of sizes among populations will lead to a itself was nonlinear over time–initially rapid and then slower, shallower decline below the threshold level. for example. The slower rate would allow the population to ‘‘catch up’’ with current habitat levels, and abundance would appear to decline in a threshold manner over a relatively small range of habitat loss. In addition, for relationships that III. EVIDENCE FOR CRITICAL THRESHOLDS are already nonlinear for other reasons (e.g. fragmentation or Allee effects at low habitat cover), time lags may cause a Evidence for the presence and value of critical threshold lev- decrease in the apparent threshold level of habitat. els comes from three types of studies: simulation models, and When habitat loss is rapid and ongoing or recent, time- two empirical approaches that we call ‘‘small-landscape’’ and lagged responses make it more difficult to assess habitat loss ‘‘large-landscape’’. Simulations allow the experimenter to effects empirically, since they initially mask the full ecological manipulate landscape and organism properties in the absence consequences (Schrott et al., 2005b; With, Schrott & King, of environmental noise. They explore species or landscape 2006). A population experiencing rapid habitat loss may characteristics associated with the occurrence of critical appear to be relatively unaffected over a large range of thresholds, and compare the relative effect of such character- habitat loss, compared to a population experiencing slower istics on threshold levels (Fahrig, 2001). However, simulations rates of loss. Yet the former would go extinct sooner in do not capture the full range of environmental complexity

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 40 Trisha L. Swift and Susan J. Hannon present in real , and thus cannot predict exact about the simulated organism and landscape, the effects of threshold levels for real species (Lamberson et al., 1992). which are discussed below. ‘‘Small-landscape’’ studies focus on landscapes and organisms that are small from a human perspective. They (a) Effects of simulated landscape characteristics on critical thresholds tend to be manipulative experiments (sensu McGarigal & Cushman, 2002), which allow strong inference because non- In simulation models, habitat pattern influences the critical treatment factors are held constant, and treatments/controls threshold level. For a given proportion of habitat, land- are applied randomly (McGarigal & Cushman, 2002). scape maps with fractal (clustered) patterns of habitat have However, such experiments may not reflect long-term fewer, bigger habitat patches with less edge than do random responses to changes in real landscapes, because they tend to maps (With, Gardner & Turner, 1997), and are thus less examine short-term responses in artificial landscapes. structurally fragmented. For random maps the percolation ‘‘Large-landscape’’ studies of critical thresholds are threshold level (see above) occurs at 59% habitat, but this so-called mensurative experiments (sensu McGarigal & threshold becomes smaller and more variable for fractal Cushman, 2002), measuring ecological parameters in relation patterns [45–54% for a binary habitat/nonhabitat system, to pre-existing variation in real landscapes, rather than (With & King, 1999a); 29–50% for a three-habitat system experimentally manipulating the independent landscape (With et al., 1997)]. Similarly, the threshold levels for sim- variables of interest (McGarigal & Cushman, 2002). They ulated ecological responses to habitat loss tend to decrease are subject to a greater amount of uncontrolled variation as the distribution of habitat becomes more clustered (patch than small-landscape/manipulative studies (McGarigal & occupancy: Hill & Caswell, 1999; With & King, 1999a; Cushman, 2002). In addition, while many of the simulations population survival probability: Fahrig, 2001; proportion and small-landscape studies (Tables 1 and 2) focus on risk of of species pool persisting: McLellan et al., 1986; population extinction or details of movement patterns, these variables growth rates: Schrott et al., 2005b; movement path complex- are more difficult to measure in large landscapes. Thus, ity: With, Cadaret & Davis, 1999). For plant migration rates, most empirical large-landscape studies have used surrogate the threshold level increased as fragmentation decreased measures such as species occurrence, abundance, or diversity (Collingham & Huntley, 2000). This is because faster migra- (Table 3), under the assumption that these reflect habitat tion requires that plant propagules spread quickly across the quality and/or individual fitness. This assumption may landscape to new, unoccupied habitat patches, not just to any suitable habitat such as within the parent patch. Smaller not hold under some conditions (Van Horne, 1983; Bock levels of fragmentation mean that although individual habitat & Jones, 2004). However, a few studies suggest that this patches are larger, they are also further apart, and thus more assumption is often (though not always) a reasonable one. difficult to reach (Collingham & Huntley, 2000). For example, per capita and per land area was The quality of the matrix (‘‘non-habitat’’ or less preferred positively related to adult bird density in 72% and 85% habitat portions of the landscape) may have an even larger of northern hemisphere studies (Bock & Jones, 2004). In effect on critical threshold levels than habitat configuration. addition, simulated population size (Fahrig, 1998; Flather & In naturally patchy landscapes or in landscapes that have Bevers, 2002) and patch occupancy (Vos et al., 2001) were experienced habitat loss and fragmentation, organisms may positively correlated with persistence, thus lending general be obliged to move through the matrix, either during support to the use of presence or abundance measures as home range movements (Siffczyk et al., 2003; Turcotte indicators of habitat/landscape quality. & Desrochers, 2003) or during dispersal (Matthysen, Among the studies compiled for this review, there is a large Adriaensen & Dhondt, 1995) or migration (Higgins, Lavorel amount of variation in terms of whether thresholds were & Revilla, 2003). For some species and matrices, mortality found for the ecological response, and at what proportion of may be higher in the matrix than in habitat patches habitat (Tables 1, 2, and 3). Below, we outline the results from (Baguette et al., 2003), due to such factors as predation simulation, small-landscape, and large-landscape studies, and (Schneider, 2001), mortality from vehicles (Fahrig et al., examine some possible reasons for this variation. 1995; Trombulak & Frissel, 2000), or unfavourable physical conditions (Hertzberg et al., 2000; Higgins et al., 2003). (1) Simulation studies of critical thresholds A simulation model by Fahrig (2001) found that reducing the Many simulation models have found critical threshold rate of mortality in the matrix could reduce the threshold level relationships between habitat loss and various ecological of habitat for population persistence by up to 58% (depending responses (Table 1), including: (a) plant migration rates; on assumed species and landscape characteristics). In other (b) population size; (c) patch occupancy; (d) population per- words, the population could persist with up to 58% less sistence probability; (e) population demographic parameters; habitat in landscapes with the highest matrix quality, (f ) species extinction rates in communities; (g)dispersalsuc- compared to landscapes with the lowest matrix quality. cess; and (h) individual movement parameters. The threshold By contrast, reducing fragmentation from very high to very levels range across nearly the entire continuum of habitat low reduced the threshold by up to only 17% (Fahrig, 2001). proportion, from about 1% to 99% (though most fall between The rate of landscape change over time may also have an 10 and 50%), and linear relationships also occur (Table 1). important influence on critical threshold levels. As discussed Some of this variation reflects different assumptions made above, as the rate of habitat loss increases the critical

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 41 4 2 (*) on threshold presence/value Species (S) or landscape (L) effects looked for; effect ments, reproductive rates) or haracteristics). dency to be aggregated or cluster hes). Different authors use different Linear? s) to have an effect on threshold presence 40 yes S*, L* 3 50 yes S*, L* | − 1–7 yes L* S*, 5 (%)? ∼ < ome proportion of the cells are randomly designated Threshold (%) 5–100 10–50 no S* Range of clustered: habitat cells have a ten habitat cover = 1 Landscape characteristics Relationship(s) found Habitat pattern implicit) ape characteristics appeared (e.g. in data plot o a more precise threshold estimate was not possible y characteristics of the species S (e.g. dispersal distance, area require random: in landscapes composed of a lattice of cells, s ing, the less fragmented the landscape tends to be (i.e. fewer, larger patc = ships were found (depending, for example, on assumed species or landscape c ells do not have any tendency to cluster together. C size) bird n/a (spatially tion size) bird C 10–90 30–50 no L* /clustering). * indicates where species or landsc abitat cells across the landscape. R indicate that habitat proportion was categorical, s | ) movement (dispersal success) generic R or C 1–99 10–40 yes S, L* ) prevalence (patch occupancy) generic C 20–80 5–85 yes S*, L* ) demographic (growth rate) birds C 0–100 5–90 yes S*, L* b a b ´ e (1996) prevalence (patch occupancy) generic R 0–100 25–95 yes S*, L* ., (1997) movement (dispersal success) generic C 2–24 8 no n/a . (1992) prevalence (patch occupancy) bird R 0–100 20–30 no S*, L* . (1997) prevalence (population . (1986) community (proportion of species pool persisting) generic C 0–80 10–40 no S*, L* et al . (2000) prevalence (patch occupancy) generic R 10–90 25–50 no S*, L* et al . (2003) movement (migration rate) plant C 1–90 . (2005 et al et al . (2006). (2006) prevalence (population size) persistence (probability) mammal C mammal C 50–100 n/a 50–100 yes n/a yes L n/a . (2006). (2006) demographic (proportion females mated) demographic (births, death rate) mammal C mammal C 50–100 n/a 50–100 yes 80–90 yes L L* . (1999) movement (individual patterns) insect C 0–80 20 et al . (2004) community (species richness) generic R 10–80 10–40 no n/a et al et al et al et al et al et al et al et al ´ e Habitat pattern–the distribution of h Relationship found–whether threshold, linear,Threshold or ranges both separated types by of relation The authors often examined whether or not model predictions were affected b Collingham & Huntley (2000)With movement (migration rate) plant C 1–90 10–25 no L* Flather & Bevers (2002) prevalence (popula Jager Flather & Bevers (2002)Fahrig (1997)Jager persistence (probability) persistence (probability) bird C generic C 10–90 0–100 30–50 no 20 no L* n/a Keymer Lamberson Higgins Carlson (2000)Bascompte & Sol Wimberly (2006)Hill & Caswell (1999)With & King (1999 dit Durell prevalence (patch occupancy) prevalence prevalence (patch (patch occupancy) occupancy) bird generic generic C R R or C 10–90 10–100 0–100 20–70 25–40 10 no no yes S*, L* S* L* together, to a degree determined by the researcher. The greater the cluster Sol Jager King & With (2002) movement (dispersal success) generic R or C 1–90 Schrott Jager With & King (1999 Ruckelshaus as habitat cells (e.g. King & With, 2002). Thus, habitat c algorithms for generating these patterns. or level. Table 1. Simulation studies of the shape of the relationship between habitat proportion and ecological responses SourceMcLellan Response variable Taxon 1 2 3 4 landscape L (e.g. degree of

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 42 Trisha L. Swift and Susan J. Hannon 4 2 (*) on threshold presence/value Species (S) or Landscape (L) effects looked for; effect Linear? icates an effect of species or landscape on 3 40 yes S*, L* | 80 yes n/a 20 yes n/a | | 60 no n/a 20, 20 | uniform: habitat is clustered and patches are uniform in size = 2 16m 10 15m no yes n/a 15m no not sig. n/a 5m 50 15m5m no 0 yes n/a size Threshold (%)? × × × × × × Landscape g. degree of clustering) was examined. * ind (%) Range of clustered: as for Table 1. U habitat cover = Landscape characteristics Relationship(s) found 1 random, C Habitat pattern = hip between habitat proportion and ecological responses the effect of landscape characteristics (e. = (# patch visits) insects R 20–100 15 abitat cells across the landscape. R . (1997) movement (patterns) insect R 0–80 5 . (2004) movement (percolation) fungus R 40–100 154 cm . (2002) distribution (lacunarity) insect C 10–80 16 . (1999) movement (patterns) insect C 0–80 5 et al et al responses of different species were compared. L et al et al = As for Table 1. Habitat pattern–the distribution of h S threshold presence or level. and distribution across the landscape. Table 2. Small-landscape studies of the shape of the relations SourceSummerville & Crist (2001) community (species richness) insects Response R variable1 2,3 4 20–100 Taxon 15 Parker & Mac Nally (2002) community (species richness) insects U 10–100 15 With Summerville & Crist (2001) movement With Wiens Otten

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 43 4 2 (*) on threshold presence/value Species (S) or Landscape (L) effects looked for; effect % early-seral habitat (Cushman no n/a yes S*, L* Linear? 3 , 6 7 5 30 no n/a | 20 | 34–55 e) was examined. *indicates an effect of species or landscape on size (ha) Threshold (%)? Landscape loss of several early-seral species below 0–20 (%) Range of habitat cover F 2–60 10,000 10 no n/a 1 Landscape characteristics Relationship(s) found F 0–100 314, 1256 no yes n/a > V 5–70 various 10 type various. = lizards birds, mammals urban; V = 8 hip between habitat proportion and ecological responses the effect of landscape characteristics (e.g. size of landscap e, threshold value with largest test statistics only). suburban; U ies evenness) birds F 0–100 250–300 no yes n/a = = rasitism rate) insects A 3–65 177 20 yes S 0–100% late-seral forest corresponded to an abrupt esence) amphibians A 10–98 314 no yes S, L (presence) bird A 0–40 10,000 10–20 no L* ce (presence) bird F 8–100 28 no yes n/a Forestry; S random sample hypothesis = Agriculture; F = WL) prevalence (abundance) bird V 8–18 Finland 10 no n/a . (2005) (PL) prevalence (presence) birds F 0–100 314, 1256 no yes n/a . (2005) (PL) community (species richness) birds, ∼ . (2004) (WL) prevalence (presence) mammal F 0–90 100 40 no n/a et al et al . (2004) (PL) prevalence (presence) insects A 0–30 7850 5–15 no S* . (2000) (WL) prevalence (presence) mammal A, U 3–70 90,000 no yes n/a . (2005) (WL) community (species richness) birds A . (2004) (PL) prevalence (presence) amphibians S, A 0–100 0.3–314 10–50 et al et al et al et al et al responses of different species were compared. L ´ en (1994) (PL) consistency of results with = patch-landscape; WL: whole-landscape) Response variable Taxon As for Table 1. Disturbance type. A S An observed threshold decrease in richness above 8 According to piecewise regression (forAccording each to species/scal change-point analysis. Meta-analysis. Schmidt & Roland (2006) (PL)Schmidt & Roland (2006) (PL)Carlson (2000) ( Thies & Tscharntke community (1999) (species (PL)Imbeau richness) & Desrochers community (2002) (total (PL)Lindenmayer abundance) insectsBergman prevalence insects (pa prevalen Radford & Bennett (2004) (PL)Guerry & Hunter (2002) (PL)Kerkhoff A prevalence A prevalence (pr 5–95 20–95 50–113 12.6 40–50 20 no no S* n/a Homan Lindenmayer Radford Cushman & McGarigal (2003) (WL) community (species richness) birds F 0–100 250–300 0 Table 3. Large-landscape studies of the shape of the relations Source and study design (PL: Cushman & McGarigal (2003) (WL) community (spec 1 2,3 4 5 6 7 8 threshold presence or level. & McGarigal, 2003). Gibbs (1998) (WL)Reunanen Andr prevalence (presence) amphibians U 8–98 180? 50–60 yes S*

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 44 Trisha L. Swift and Susan J. Hannon threshold for simulated population growth rate appears to below 20–50% habitat, while movement parameters for a occur at a lower level of habitat. This reduced threshold cybercricket following the 12-neighbour rule exhibited no level is misleading because it is the result of time-lagged threshold (With et al., 1999). responses of the organism to habitat loss (Schrott et al., 2005b). Critical threshold levels are also influenced by dispersal Other measures of landscape change over time may also be characteristics, although this effect depends on whether important, such as the rate of habitat patch turnover (e.g. dispersal ability is modeled as a purely species characteristic, concurrent habitat patch destruction/senescence and habitat or as an interactive effect of both species and landscape patch creation/regrowth, without necessarily a change in the characteristics. Studies that assume colonization rates are overall amount of habitat in the landscape). As simulated a consequence of a species’ dispersal ability find that as a habitat patch turnover rates increase, the threshold level of species’ dispersal distance or search effort increases, critical habitat for patch occupancy increases (Keymer et al., 2000; threshold levels either decrease (proportion of species pool Wimberly, 2006). This suggests that in landscapes with more persisting: McLellan et al., 1986; patch occupancy: With & rapid turnover, less habitat loss can be tolerated before King, 1999b,Lambersonet al., 1992; Carlson, 2000) or the metapopulation size undergoes a threshold decrease. This is thresholds are less precipitous (dispersal success: With & because an increase in habitat patch turnover rates means King, 1999a). This suggests that more vagile species may that the individual persistence times for local populations be less sensitive to habitat loss/fragmentation. By contrast, if are reduced. Past a critical turnover rate local populations mortality is assumed to be higher in the matrix than in habitat become extinct faster than empty patches are colonized. then populations with greater emigration rates require more Thus the entire metapopulation cannot persist, even if habitat to ensure persistence, especially if matrix quality the amount of habitat would otherwise be sufficient in is low (Fahrig, 2001). Higgins et al. (2003) also found that a static landscape (Keymer et al., 2000). Finally, temporal the shape of the relationship between plant migration stochasticity in environmental conditions may also influence rates and habitat loss depended on a complex interaction the threshold level. For example, Lamberson et al. (1992) between species dispersal and landscape characteristics. Such found that if environmental stochasticity is introduced into species-landscape interactions may be highly relevant in the model (simulating annual fluctuations in food supply), anthropogenically modified systems, where organisms may the threshold for population persistence occurs at a higher encounter new sources of mortality or barriers to movement level of habitat proportion, although the threshold is less (e.g. roads: Trombulak & Frissel, 2000; Belisle´ & St. Clair, abrupt. 2001; urban development: Hitchings & Beebee, 1997; habitat gaps: Desrochers & Hannon, 1997; St. Clair et al., 1998), to (b) Effects of species characteristics on critical thresholds which they may be poorly adapted. As the ability of an organism to move through and survive Simulations have also demonstrated a strong negative in nonhabitat increases, the critical threshold decreases relationship between reproductive rate and threshold level or disappears. As discussed earlier, in percolation-based (patch occupancy: With & King, 1999b;Keymeret al., models the percolation threshold occurs when 59% of the 2000; Carlson, 2000; persistence: Fahrig, 2001; population lattice/landscape is occupied by material/habitat (Gustafson size: Dit Durell, Goss-Custard & Clarke, 1997), suggesting & Parker, 1992; Andren,´ 1994; Bascompte & Sole,´ 1996). In that species with low reproductive rates are particularly the ecologically geared models, this means that habitat patch sensitive to habitat loss/fragmentation. The importance of no longer spans the landscape when habitat proportion falls reproductive rate in determining critical threshold levels may below this level. However, this threshold level depends on a exceed that of dispersal, fragmentation, and matrix quality definition of ‘‘habitat patch’’ that is based on the assumption (Fahrig, 2001). With & King (1999b) suggest that species that an organism can move only into habitat cells and only with low reproductive rates have a limited ability to respond between cells that are directly adjacent along a horizontal to the effects of environmental disturbances. By contrast, or vertical edge (i.e. belonging to the same patch). Since species with high reproductive rates are able to ‘‘flood’’ the each square cell has four such neighbours, this is the ‘‘four- landscape and ensure population persistence (Fahrig, 1998). neighbour rule’’ (Cardille & Turner, 2001). If patches are A high abundance of organisms may act as a buffer; even if a instead defined by the ‘‘eight-neighbour rule’’ this implies large proportion of offspring experience dispersal mortality, that movement is also possible to the diagonal neighbours. for example, a relatively large absolute number of individuals In this case, the threshold decreases to 40% (Plotnick & will be successful. Gardner, 1993), and to 29% when movement even between Finally, threshold levels for population growth rates habitat cells separated by the width of one nonhabitat cell increased with increasing sensitivity to patch area and habitat can occur (twelve-neighbour rule; With & King, 1999a). edge (Schrott et al., 2005b). In this study, these characteris- Thus, as the ability of the organism to cross nonhabitat tics were intended to reflect empirically observed patterns for increases, greater levels of habitat loss can be tolerated some bird species: area- and edge-sensitive simulants avoided without compromising the potential for the organism to settling in smaller patches, and had lower reproductive suc- traverse the landscape from one side to another. Similarly, cess in patches with a high edge: area ratio (Schrott et al., simulated ‘‘sand cybercrickets’’ following the four-neighbour 2005b). Thus, fragmentation-sensitive species may require rule exhibited sharp declines in movement-path complexity more habitat to persist.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 45

(c) Evidence for causes of critical thresholds in simulations The ability of the fungus to ‘‘percolate’’ from the centre to the edge of the grid decreased strongly below 60% habitat Few studies were designed to investigate the causes of (Otten et al., 2004), remarkably near the critical percolation critical threshold relationships. However, several studies have threshold. suggested that fragmentation effects become more important Other small-landscape studies (Table 2) demonstrated at low levels of habitat, which could potentially lead to either lower threshold levels (20% for beetle movement in threshold relationships with habitat loss. For example, various random habitat: Wiens, Schooley & Weeks, 1997; 20–40% population parameters were consistently found to decline for insect spatial aggregation in fractal arrangements: With more steeply in more fragmented landscapes or for spatially et al., 2002), or linear responses (butterfly richness and explicit (versus implicit) simulations, but only at lower levels frequency of patch visits: Summerville & Crist, 2001; but of habitat (McLellan et al., 1986; Bascompte & Sole,´ 1996; note the high range of habitat: 20–100%). One small- Fahrig, 1997; Hill & Caswell, 1999; Collingham & Huntley, landscape study showed no effect at all of habitat loss or 2000; Keymer et al., 2000; Flather & Bevers, 2002; King & fragmentation on species richness or mean abundance of With, 2002; Jager, Carr & Efroymson, 2006; see Table 1 for terrestrial invertebrates (Parker & Mac Nally, 2002). This response types). Further, two simulation studies that found may be explained by the fact that both ecological responses thresholds also tested for interactions between habitat loss reflected the combined (and possibly contrasting) responses and fragmentation effects and found them to be significant of a variety of different species (Parker & Mac Nally, 2002). predictors of the ecological response (plant migration rates: Alternatively, since habitat loss and fragmentation were rapid Collingham & Huntley, 2000; population size: Flather & and the responses measured shortly thereafter (periodically Bevers, 2002). An interaction between habitat loss and from about two weeks to three months post-disturbance), the fragmentation is consistent with the idea that fragmentation lack of a response may reflect a time lag (e.g. Schrott et al., effects will be strongest at low habitat proportions (Trzcinski, 2005b). Fahrig & Merriam, 1999). More direct evidence of a causal relationship is provided by Flather & Bevers (2002). Predicted population declines were nearly linear for the least (a) Evidence for causes of small-landscape thresholds fragmented landscapes, but became threshold-like when As with most simulation studies, most small-landscape studies fragmentation levels were high (see Fig 4b of Flather & were not designed to test for time lag or Allee effects on the Bevers, 2002). occurrence or value of the threshold level. Many focused The idea that Allee effects could compound habitat loss on thresholds in individual movement parameters (Table 2). below a threshold level of habitat has been supported Fragmentation effects are thus a more likely explanation for by at least one study. Below a certain proportion of thresholds in these studies. One study compared a mobile habitat, simulated patch occupancy declined more steeply and less mobile species (With et al., 2002). The more mobile (and extinction occurred at a higher proportion of habitat species closely tracked the distribution of their habitat, red remaining) when dispersing females were required to search clover (Trifolium pratense). Both the aggregation of the habitat for mates (Lamberson et al., 1992). and species increased more rapidly below about 20% habitat Most studies measured long-term, equilibrium responses cover (i.e. there were larger and more variable distances to habitat proportion in static landscapes. Thus, time lags between habitat patches or occupied patches, respectively; were not usually an issue, though as mentioned above time With et al., 2002). Thus, the species’ threshold appeared lags had the ability to decrease the threshold level in one to be related to a similar threshold for structural habitat study that incorporated time (Schrott et al., 2005b). Many configuration. By contrast, the less mobile species had a studies measured ecological response as a proportion or higher threshold level of 40% (With et al., 2002), suggesting probability (e.g. proportion of species persisting or of females that below 40% habitat, its ability to locate and occupy mated, birth/death rates, dispersal success, patch occupancy, empty habitat patches was compromised. persistence probability; Table 1). In these cases, the presence of any thresholds may have been a response to habitat loss alone (see rationale in Section II.4), with fragmentation or (3) Large-landscape studies of critical thresholds Allee effects influencing the threshold level in some cases. (a) A comparison of landscape-level study designs There have been two basic designs applied to large-landscape (2) Small-landscape studies of critical thresholds studies of critical thresholds, which we will refer to as Do the critical thresholds predicted by simulation models ‘‘whole-landscape’’ and ‘‘patch-landscape’’ studies. Whole- appear in experimental small-landscapes with real organ- landscape studies measure the dependent variable (e.g. isms? Recall that the percolation threshold for randomly species abundance or richness) across several representative arranged habitat is 59% under the four-neighbour rule. points in the landscape. The dependent variable, often This scenario was replicated in an experiment involving a expressed as a single value per landscape, is then compared grid of randomly arranged agar ‘‘habitat’’ dots. The dis- across landscapes with different amounts of habitat. Sampling tances among these dots allowed the spread of a fungus only points may be arranged in a systematic/grid-like pattern between neighbouring dots (Otten, Bailey & Gilligan, 2004). across the landscape (e.g. Cushman & McGarigal, 2003).

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 46 Trisha L. Swift and Susan J. Hannon

Alternatively a variety of landscape elements may be selected distance (Radford & Bennett, 2004). Other configuration in proportion to their occurrence in the landscape (e.g. small metrics such as edge density or patch size in the surrounding remnants, large remnants, riparian vegetation, etc.; Radford, landscape would likely not have a direct effect on the focal Bennett & Cheers, 2005). In other studies, multiple habitat patch, unless it would perhaps be to decrease the suitability patches were surveyed across each landscape (e.g. Gibbs, of surrounding patches and thereby increase the distance to 1998; Reunanen et al., 2004). By contrast, patch-landscape source patches. studies measure the dependent variable within a single habitat (ii) Allee effects. At the whole-landscape level, if Allee patch, and determine how it changes with the proportion effects were to occur below a threshold habitat level/ of habitat in the surrounding landscape. Sampling points population size, this may reduce population density within may be arranged systematically across the habitat patch (e.g. habitat (Stephens et al., 1999) as discussed earlier. If so, this Radford & Bennett, 2004), or each habitat patch may be would translate into a decline in local abundance (and at sampled using a single transect (e.g. Lindenmayer, Fischer & some point, occurrence) within individual patches. This sort Cunningham, 2005) or point (e.g. Schmidt & Roland, 2006). of threshold would look similar to that described for the patch- The patch-landscape studies are thus not directly landscape fragmentation threshold above: a zero slope above comparable to the whole-landscape studies, or to the the threshold, and a decrease in abundance with habitat simulation and small-landscape studies which are also cover below a habitat threshold level. In contrast to the designed in a whole-landscape way. However, we include fragmentation effects threshold described above, however, them here because (a) a large proportion of large-landscape this would not reflect an effect of the surrounding landscape threshold studies follow a patch-landscape design, and (b)we on the focal patch. Rather, the situation within the focal feel that some of the same principles to explain or predict patch (declining local abundance) would represent a sample critical threshold responses to habitat loss in whole-landscape of the situation in the whole landscape (declining density scenarios can also apply to patch-landscape studies, if the within habitat). differences between the two approaches are evaluated. (iii) Time lags. In a whole-landscape scenario with a If a species’ abundance is only affected by the amount time lag, abundance may initially decrease less steeply of habitat in the landscape, then that species should decline than expected from habitat loss alone (as individuals crowd in exact proportion to habitat loss. This means that the into remaining habitat), and then decrease more steeply density of the species within the remaining habitat should than expected (as the population ‘‘catches up’’ to previous not change. Thus in a patch-landscape scenario, abundance habitat loss). Translated into a patch-landscape scenario, this in the patch should remain constant as habitat is lost from the relationship would instead look rather quadratic, because surrounding landscape, as long as patch area or the sampled density within habitat would initially increase and then area within each patch remains constant (or at least is not decrease as habitat is lost. On the other hand, if the response correlated with the amount of surrounding habitat cover). variable is occurrence, then a quadratic pattern would only Similarly there should be no change in species’ occurrence be possible for less common species whose occurrence is within patches, assuming the sampled area is small relative to < 100% even at high levels of habitat. the total area of habitat in the landscape (Rhodes et al., 2008). (iv) Habitat loss effects alone, for certain response variables. In a We now consider whether the four mechanisms proposed to whole-landscape sampling design, binary response variables explain critical thresholds in whole-landscape designs hold such as occurrence or community-level responses such as for patch-landscape designs. species richness can change in a threshold manner in response (i) Fragmentation effects at low habitat proportion. If fragmen- to habitat loss alone. As discussed earlier however, in a patch- landscape sampling design there should be no effect of pure tation effects in the surrounding landscape have a negative habitat loss from the surrounding landscape on occurrence effect on the patch population when habitat proportion is within a patch (Rhodes et al., 2008). This is because habitat low, then abundance or presence of that species should start loss alone should affect only the total abundance of a species to decline (recall that there should be no slope above the in the landscape, but not population density within habitat threshold). Configuration effects from the surrounding land- patches. Since individual species’ occurrence should not scape on the focal patch would probably be mainly restricted change, species richness should also remain constant. to isolation effects. As habitat cover in the surrounding land- scape becomes low, habitat patches will tend to become spaced further apart (Gustafson & Parker, 1992; Andren,´ (b) Evidence for presence and values of large-landscape critical 1994). This would increase dispersal distances through a thresholds potentially hostile matrix. The patch-occupancy thresholds Andren’s´ (1994) meta-analysis of bird and mammal studies for a woodland bird found by Radford & Bennett (2004) suggested a critical threshold level of 10–30%, the proportion near 20% habitat probably reflect the point at which inter- of habitat remaining in the landscape below which the effects patch distances exceed the species’ dispersal ability. In their of habitat loss/fragmentation were greater than expected study, distance to the nearest occupied patch (‘‘demographic from habitat loss alone (but see criticism by Monkk¨ onen¨ & isolation’’) made a large independent contribution to pre- Reunanen, 1999, and counter-response by Andren,´ 1999). dicting patch occupancy, and the authors also found strong Subsequent studies have varied in their support of Andren’s´ occupancy thresholds in response to demographic isolation (1994) proposed threshold level. Linear and critical threshold

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 47 relationships have been suggested for a variety of taxa, Interestingly, Gibbs (1998) also noted that among responses, habitat types, landscape types and spatial scales three species with threshold relationships, the threshold (Table 3). These large-landscape studies are too variable and level increased with increasing dispersal tendency. This limited in number to suggest clearly whether the presence observation is consistent with Fahrig’s (2001) simulations of critical thresholds or their level depends on any of these showing that species with greater emigration rates required factors. However, when thresholds were apparent, most more habitat for persistence, a result which depended on occurred within Andren’s´ (1994) proposed 10–30% range mortality being greater in the matrix. Indeed, Carr & (Table 3). Fahrig (2001) found that of two frog species, the more vagile one was more prone to traffic mortality. Gibbs (1998) (c) Evidence for causes of large-landscape thresholds further suggested that habitat specificity might mediate the influence of dispersal tendency on sensitivity to habitat loss Andren´ (1994) suggested that his observed 10–30% threshold and fragmentation: highly dispersive species may be more level of habitat was probably due to stronger negative sensitive if habitat specificity is high, but a combination of fragmentation effects at low habitat cover. None of the large- high dispersal tendency and habitat flexibility may confer landscape studies reviewed here directly compared the effects tolerance. of habitat fragmentation at low versus high habitat cover. Finally, Homan et al. (2004) examined the occurrence of Two studies that found thresholds tested for interactions two amphibian species that migrated seasonally between between habitat amount and fragmentation, which would breeding ponds and wintering forests. Threshold relation- be consistent with a stronger effect of fragmentation at ships were apparent when the proportion of wintering habitat low habitat cover (Fahrig, 2003). Neither found significant was measured in landscapes of 100 and 300 m radii (1 ha and interactions (Cushman & McGarigal, 2003; Radford et al., 2005). Other landscape studies not specifically looking 28 ha) around the breeding ponds. However when examined at critical thresholds have found significant interactions at smaller or larger spatial extents (30 m, 500 m, or 1000 m consistent with a larger effect of habitat configuration at radii), the relationship appeared linear (based on our sub- low habitat cover, though only for a small subset of the jective visual assessment of the data plots rather than the species examined (Trzcinski et al., 1999: the occurrence of authors’ statistical analyses; see Section VI). By themselves, only one of the 31 species examined responded significantly these results do not suggest the cause for the thresholds. to an interaction between habitat amount and configuration; Fragmentation could conceivably cause a threshold rela- Cumming & Schmiegelow, 2001: the occurrence of up to tionship if, when forest proportion was low, its effects were eight of 34 species showed an interaction, depending on how to disrupt migration or increase winter mortality at forest habitat was defined; Betts et al., 2006: the occurrence of one edges or in small fragments, for example. If so, it would be of two species showed an interaction). By contrast, others the amount and fragmentation of forest within the typical have not found an effect (on species prevalence: Guerry & migration distance of the breeding pond that would arguably Hunter, 2002; Holland, Fahrig & Cappuccino, 2005; on be the most important in terms of producing a threshold. avian community diversity: Cushman & McGarigal, 2003). Interestingly, at least one of the focal species from the Homan Comparisons within and among three amphibian stud- et al. (2004) study has the highest wintering densities within ies suggest indirectly that greater fragmentation effects at 300 m of breeding ponds (Regosin et al., 2005), suggesting low habitat cover may have played a role in the observed that most individuals do not migrate further than this. Thus, thresholds. First, two studies conducted in areas with urban the occurrence of thresholds when forest cover was measured or suburban matrices found threshold-like declines in occur- within, but not beyond, the migration distance (300 m) from rence for two species that were included in both studies breeding ponds is consistent with (though not direct evidence (Gibbs, 1998; Homan, Windmiller & Reed, 2004). By con- of) greater fragmentation effects at low forest proportion as trast, a third study in an agricultural landscape found linear an underlying cause. responses to habitat loss for these same two species (Guerry None of the large-landscape studies presented information & Hunter, 2002). Amphibians are vulnerable to traffic on changes in growth rates with habitat loss, so it is mortality (Fahrig et al., 1995; Carr & Fahrig, 2001) and uncertain whether Allee effects influenced the presence or are more genetically isolated among ponds in urban than level of thresholds in these cases. However, one patch- rural environments, suggesting that migration is inhibited level study (Groom, 1998) suggests an interaction between (Hitchings & Beebee, 1997). In other words, urban and sub- habitat loss and fragmentation in producing an Allee-related urban matrices may form barriers to amphibian movement threshold. Specifically, patches of a plant species experienced that enhance the degree of functional fragmentation of their reproductive failure past a critical threshold distance from breeding habitat, leading to threshold declines that are oth- pollen donors. Because isolation distances reflect the amount erwise absent in rural environments. This idea is generally of habitat loss around the focal patch (Fahrig, 2003), this consistent with the results of Fahrig’s (1998) simulation. She suggests that Allee effects may occur below a threshold level found that several conditions were required for fragmenta- of habitat in the landscape. Because the threshold occurred tion to have an effect on population persistence, including only for small patches (Groom, 1998), this suggests that a low overall proportion of habitat in the landscape and a fragmentation interacted with habitat loss to produce the higher rate of mortality in the matrix than in the habitat. threshold.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 48 Trisha L. Swift and Susan J. Hannon

The decline of the white-backed woodpecker (Dendrocopos size of one bird species over time suggested a possible time- leucotos) in Finland in response to several decades of habitat lag-related threshold (Carlson, 2000). Allee effects were not loss suggests a possible time-lag-related threshold. This usually addressed in the studies reviewed here, and support species initially declined less than expected from the amount for Allee-effects-related thresholds is mainly limited to of habitat loss, and then much more rapidly below an simulations (e.g. Lamberson et al., 1992, Amarasekare, 1998). apparent habitat threshold level of about 10%, suggesting Overall, threshold responses were more commonly found a time-lagged response for this long-lived species (Carlson, in simulation than in small- or large-landscape studies. There 2000). The rate of habitat loss during the first 15 years may be a number of reasons for this. First, in simulation immediately preceding the threshold was much more rapid studies there tended to be a wider range in the proportions (0.52% year−1) than that in the subsequent 20 years (0.09% of habitat cover represented, which may decrease the chance year−1; calculated from Table 1 in Carlson, 2000). As of ‘‘missing’’ a threshold. Simulation studies more often outlined previously, we speculate that this nonlinear change did not sample the lower (∼0–20%) range of habitat in the rate of habitat loss could have produced or accentuated cover, while large-landscape studies more often did not the apparent threshold response to habitat loss. sample the upper (∼50–100%) range, for various theoretical and logistic reasons. In addition, various response variables were not equally represented among all three study types, IV. GENERAL COMPARISONS OF SIMULATION, which may create a bias if some response types are more SMALL-LANDSCAPE, AND LARGE-LANDSCAPE or less likely to exhibit a threshold. For example, there STUDY RESULTS was a bias in small-landscape studies towards movement parameters as the response variable, while large-landscape studies focused primarily on numeric responses such as It would be unwise to get overly specific when comparing individual species’ prevalence or number of species present. the results of the simulation, small-landscape, and large- Some simulation studies also examined these responses, but landscape studies. As outlined earlier, each approach several others examined responses not considered in small- has different strengths and weaknesses, and the various or large-landscape studies, such as dispersal success, various approaches are generally geared towards addressing different demographic variables, and species’ persistence. Another questions or issues. Most simulation studies, for example, are possible reason for the greater prevalence of thresholds in not intended to predict actual threshold values for real simulations is that simulations make assumptions about the organisms, but rather to understand the conditions under characteristics and behaviour of the simulated organism that which they may occur and factors that may influence the are not representative of the particular real organisms used threshold level (Fahrig, 2001). By contrast, large-landscape in the small- and large-landscape studies. In particular, the studies can provide estimates of threshold presence and values lower prevalence of thresholds observed in real organisms for particular organisms or communities in real landscapes, could reflect a less negative effect of fragmentation than is but most typically are not designed to enable strong inference assumed in simulations. This could arise from a tendency about cause and effect (McGarigal & Cushman, 2002). Small- for real-landscape studies to focus on common species, landscape studies are useful for studying response variables which may be more resilient to negative habitat loss and that are difficult to measure in large landscapes, but are fragmentation effects (Davies, Margules & Lawrence, 2000; generally carried out over a short time period on artificially Henle et al., 2004). constructed or manipulated landscapes that may not reflect realistic conditions. The value in considering all three types of studies is in the different perspectives they provide, and also in determining whether there are any general similarities V. UTILITY OF CRITICAL THRESHOLDS IN in their results despite the different approaches utilized. CONSERVATION In comparing the results of simulation, small-landscape, and large-landscape studies, a few general statements can be There are several potential uses for critical threshold made. Both linear and critical threshold relationships were information, though some of these are subject to important found in all three types of study. While this observation does practical and conceptual limitations. The idea of using critical not provide any simple answers, it suggests strongly that the thresholds in habitat proportion to make broad management occurrence of critical thresholds may be highly dependent decisions has been criticized because threshold levels are on species or landscape characteristics or other conditions. expected to vary by species, landscape type, and spatial scale, There was some degree of support in all three study types and thus the results of one study do not necessarily apply to that an increasing effect of habitat configuration at low another situation (Huggett, 2005; Lindenmayer et al., 2005; habitat cover may be associated with a critical threshold Lindenmayer & Luck, 2005). This review supports this idea. response, although this was generally most rigorously tested However, such variation doesn’t preclude the possibility that in the simulation studies. This explanation was also the most useful generalizations can be found. For example, simulations commonly investigated. Simulation evidence suggested that suggest that critical thresholds are closely related to life- time lags could lead to threshold responses (Schrott et al., history and landscape characteristics (see above). If sufficient 2005b), and one large-landscape study following population empirical support existed, such trends could be used to

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 49 predict: (a) the presence and/or level of critical thresholds for empirically. Whether it is acceptable to use threshold values different species, assemblages, and landscapes, (b)therelative from substitute measures of persistence is debatable, though sensitivity of different species (see caveat below), (c) the range measures of fitness such as reproductive success (Gunn et al., of habitat proportions over which it may be most fruitful 2000) may be better than prevalence. Regardless, even a to look for (and potentially manage or account for) habitat fitness threshold should be interpreted cautiously, since per- fragmentation effects, Allee effects, or other factors, or (d)the sistence probability is determined by many different factors. form of landscape management with the largest potential to decrease the threshold level (e.g. by reducing fragmentation, increasing matrix quality, or increasing the demographic VI. STATISTICAL CONSIDERATIONS potential), and thus ameliorate the effects of landscape change. Such information could help to target conservation Assuming that the threshold level of habitat for some efforts more efficiently to the appropriate landscapes, species, ecological response is an acceptable criterion for setting and methods. a management target (given the particular management The presence of a critical threshold relationship does goal), there are still the practical problems of distinguishing not necessarily indicate greater ‘‘sensitivity’’ of a population threshold from linear relationships, and estimating threshold compared to one exhibiting linear responses to habitat loss. levels. The majority of studies reviewed here did not use Consider two species, one which declines gradually as habitat any formal statistical method to do so. Visual estimates is lost from a landscape, then declines more rapidly below a of threshold presence and level from data plots are likely critical threshold level of 20% habitat cover, and disappears to be inconsistent among observers (T. L. Swift, personal below 10% habitat; and another which declines linearly (but observation) and possibly biased. steeply) with habitat loss, disappearing below 40% habitat. Methods to distinguish linear from threshold/non-linear The latter species exhibits no critical threshold in habitat shapes included: (a) comparing linear versus piecewise cover (as defined above), but is arguably the more sensitive regression and other non-linear regression models [based on to habitat loss because it becomes extinct below 40% habitat the Akaike information criterion (AIC): Radford et al., 2005; (versus 10% for the former). or an unspecified measure of ‘‘fit’’: Lindenmayer et al., 2005], Critical thresholds are often viewed as a potential tool to (b)at-test comparing observed values versus predicted values set conservation targets for habitat retention or restoration under the assumption of linearity (Imbeau & Desrochers, (Huggett, 2005). However, there are some important points 2002), (c) significance of the difference between the upper to consider before designating an observed threshold level and lower slopes of piecewise regression models (Homan as the ‘‘required’’ amount of habitat to retain. What is the et al., 2004), and (d) significance testing of quadratic and response variable, what does the particular critical threshold cubic effects (Summerville & Crist, 2001). Although Homan under consideration mean, and what is the management et al. (2004) suggested that a significant ‘‘change-point test’’ goal? For example, a critical threshold for species richness indicates a non-linear relationship, this is incorrect. It merely represents a precipice, below which an increasingly larger indicates a significant change (i.e. a non-null relationship). portion of the community will be absent as habitat cover In fact, many of the relationships for which Homan et al. decreases. Thus the target level of habitat should be well (2004) found a significant change-point appeared (visually) above this threshold level (Radford et al., 2005; Lindenmayer to be quite linear, and did not show a significant difference & Luck, 2005). Indeed, Schmidt & Roland (2006) found that between the upper and lower piecewise slopes. The other while a threshold for moth occurred near methods are, to our understanding, tests of ‘‘nonlinearity’’. 20% forest, total moth abundance declined below a threshold Most are based on traditional significance testing, while the level of 40–50% forest. Further, community-level or average AIC method is a form of multi-model selection (see Johnson, threshold levels may underestimate the habitat requirements 1999, and Anderson, Burnham & Thompson, 2000, for an of the more sensitive species (Monkk¨ onen¨ & Reunanen, in-depth criticism of the former and promotion of the latter). 1999). Even population-level threshold levels may not One major advantage of multi-model selection is that it allows necessarily be appropriate targets for individual species. For simultaneous comparison of a set of models that need not be example, just because habitat fragmentation may compound nested (Johnson & Omland, 2004). For example, one could the effects of habitat loss on population size below 30% compare the weight of evidence for a linear, change-point, habitat (thus producing a threshold), it does not follow that piecewise, and polynomial relationship. habitat loss alone has not already had a profound effect on Methods that have been used to estimate the threshold population viability above this threshold level (or conversely level include piecewise regression (Radford et al., 2005; that the population is not viable below the threshold level). Lindenmayer et al., 2005; Homan et al., 2004), and change- Thismaybeparticularlytrueiftimelagsareoperating. point analysis (Homan et al., 2004). One study tried both If the management goal is population persistence, the most methods (Homan et al., 2004). For a given species and reliable target level of habitat is the threshold level for popula- spatial scale, the threshold value depended on the statistical tion persistence itself, and then only if persistence probability technique used (Homan et al., 2004). Because the ‘‘true’’ is acceptably high above this threshold level of habitat. threshold levels were unknown (if they existed; see above), it However, persistence probability is difficult to measure is uncertain which technique was more accurate.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 50 Trisha L. Swift and Susan J. Hannon

Even when statistical estimates are used, these are subject (3) The study of critical thresholds in landscape to uncertainty, as are all such estimates, and the variability is still new, and many questions remain for future inherent in most ecological data sets will increase this research. How common are critical thresholds, and uncertainty. This may not be of great concern when the what are their causes? Many simulations have object is to find general trends between threshold presence suggested that increasing fragmentation effects at low or levels and life-history traits, or to approximate the range levels of habitat can produce threshold relationships of habitat proportions in which to look for fragmentation with habitat proportion. Empirical studies rarely test or Allee effects. However, if the object is to set a minimum for this, or consider the possibility of time lags or Allee target for conservation, then accuracy is more important, and effects. Which species are likely to exhibit critical underestimation of the threshold level could have profound thresholds, and in what types of landscapes? At consequences. what spatial scales should we expect to find critical thresholds, and can these be predicted from species characteristics (e.g. dispersal distances)? There is also VII. CONCLUSIONS a need for the use of statistical methods to detect and estimate threshold levels, both to increase objectivity (1) Threshold responses to habitat loss were common within individual studies, and to facilitate comparison among the studies reviewed here, although their among studies. This is essential if the questions above exact value depended on many factors. For example, are to be addressed. Answers to these questions would simulation studies suggested that threshold levels tend help land managers to direct resources more efficiently to increase with various landscape characteristics: (a) to where they would have the greatest impact. increasing fragmentation of habitat; (b)decreasing matrix quality; (c) increasing environmental variance or patch turnover rates; and (d) decreasing rates of VIII. ACKNOWLEDGEMENTS habitat loss. Threshold levels may also depend on species characteristics, increasing with: (a)decreasing Research was funded by the Natural Sciences and Engi- ability to enter the matrix; (b) decreasing dispersal neering Research Council of Canada, Alberta Conservation distance (assuming no matrix mortality); (c)increasing Association, North American Waterfowl Management Plan emigration rate (assuming matrix mortality); (d) Initiative, Alberta Sport, Recreation, Parks and decreasing reproductive rate; and (e)increasing Wildlife Foundation, University of Alberta and the Canadian sensitivity to fragmentation. These trends have yet Circumpolar Institute. We are grateful for the helpful com- to be rigorously tested empirically. ments and suggestions on a previous draft of this manuscript (2) While evidence was limited, some of these same by Andrew Bennett and another anonymous reviewer. patterns were supported in empirical studies. When present, most empirical thresholds fell near Andren’s´ (1994) proposed range of 10–30% habitat cover. IX. REFERENCES Higher critical habitat threshold levels have occa- sionally been reported, notably for fungal spread in a Aguilar,R., Ashworth,L., Galetto,L. & Aizen,M.A. percolation-like laboratory model with random habi- (2006). Plant reproductive susceptibility to habitat fragmentation: tat (60%, Otten et al., 2004), for total moth abundance review and synthesis through a meta-analysis. Ecology Letters 9, (40–50%, Schmidt & Roland, 2006), and for amphib- 968–980. ian occurrence with urban/suburban disturbances (up Amarasekare, P. (1998). Allee effects in metapopulation to 55 or 60%, Gibbs, 1998; Homan et al., 2004). dynamics. American Naturalist 152, 298–302. For the fungus, the high threshold level probably Anderson,D.R.,Burnham,K.P.&Thompson, W. L. (2000). reflects the random habitat arrangement and the ‘‘four Null hypothesis testing: problems, prevalence, and an alternative. neighbour rule’’ manner of fungal spread (reflecting Journal of Wildlife Management 64, 912–923. a percolation model). For moths and amphibians, the Andren´ , H. (1994). Effects of habitat fragmentation on birds and mammals in landscapes with different proportions of suitable high threshold levels may reflect a sensitivity to frag- habitat: a review. Oikos 71, 355–366. mentation after only moderate habitat loss. Linear Andren´ , H. (1999). Habitat fragmentation, the random sample relationships occurred somewhat more frequently in hypothesis and critical thresholds. Oikos 84, 306–309. empirical than in modeling studies, perhaps reflecting Baguette,M.,Mennechez,G,Petit,S&Schtickzelle,N. greater behavioural flexibility of real organisms (e.g. (2003). Effect of habitat fragmentation on dispersal in the butterfly ability to move through non-habitat) than is sometimes Proclossiana eunomia. Comptes Rendus Biologies 326, S200–S209. parameterized in simulation models. Alternatively, this Bascompte,J.&Sole´, R. V. (1996). Habitat fragmentation and may merely reflect a sometimes smaller range of habi- extinction thresholds in spatially explicit models. Journal of Animal tat proportions represented in empirical studies (such Ecology 65, 465–473. that the threshold may be ‘‘missed’’), or to some Belisle´ ,M.,Desrochers,A.&Fortin, M.-J. (2001). Influence degree differences in the response variables examined of forest cover on the movements of forest birds: a homing in simulation versus empirical studies. experiment. Ecology 82, 1893–1904.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 51

Belisle´ ,M.&St. Clair, C. C. (2001). Cumulative effects of to winter habitat loss. Journal of 134, barriers on the movements of forest birds. Conservation Ecology 5,9. 155–1164. Bergman,K.-O.,Askling,J.,Ekberg,O.,Ignell,H., Dolby,A.S.&Grubb, T. C. Jr. (1999). Effects of winter weather Wahlman,H.&Milberg, P. (2004). Landscape effects on on horizontal and vertical use of isolated forest fragments by butterfly assemblages in an agricultural region. Ecography 27, bark- birds. Condor 101, 408–412. 619–628. Donovan,T.M.,Jones,P.W., Annand,E.M.&Thompson, Betts,M.G.,Forbes,G.J.,Diamond,A.W.&Taylor,P.D. F. R. I. (1997). Variation in local-scale : mechanisms (2006). Independent effects of fragmentation on forest song- and landscape context. Ecology 78, 2064–2075. birds: an organism-based approach. Ecological Applications 16, Fahrig, L. (1997). Relative effects of habitat loss and fragmenta- 1076–1089. tion on population extinction. Journal of Wildlife Management 61, Bevers,M. & Flather, C. H. (1999). Numerically exploring 603–610. habitat fragmentation effects on populations using cell-based Fahrig, L. (1998). When does fragmentation of breeding habitat coupled map lattices. Theoretical population Biology 55, 61–76. affect population survival? Ecological Modeling 105, 273–292. Bock,C.E. & Jones, Z. F. (2004). Avian habitat evaluation: Fahrig, L. (2001). How much habitat is enough? Biological should counting birds count? Frontiers in Ecology and the Environment Conservation 100, 65–74. 2, 403–410. Fahrig, L. (2003). Effects of habitat fragmentation on biodiversity. Burton,N.H.K., Rehfisch,M.M., Clark,N.A.& Dodd, Annual Review of Ecology, Evolution, and Systematics 34, 487–515. S. G. (2006). Impacts of sudden winter habitat loss on the body Fahrig,L., Pedlar,J.H., Pope,S.E., Taylor,P.D.& Weg- condition and survival of redshank Tringa totanus. Journal of Applied ner, J. F. (1995). Effect of road traffic on amphibian density. Ecology 43, 464–473. Biological Conservation 73, 177–182. Cardille,J.A.& Turner, M. G. (2001). Understanding land- Flather,C.H.& Bevers, M. (2002). Patchy reaction-diffusion scape metrics I. In Learning : A Practical Guide and population abundance: the relative importance of habi- to Concepts and Techniques (eds. Gergel, S. E. & Turner, M. G.), tatamountandarrangement.The American Naturalist 159, pp. 85–100. Springer, New York. 40–56. , A. (2000). The effect of habitat loss on a deciduous forest Carlson Gibbs, J. P. (1998). Distribution of woodland amphibians along a specialist species: the white-backed woodpecker (Dendrocopos forest fragmentation gradient. Landscape Ecology 13, 263–268. leucotos). Forest Ecology and Management 131, 215–221. Greene, C. M. (2003). Habitat selection reduces extinction of Carr,L.W.& Fahrig, L. (2001). Effect of road traffic on two populations subject to Allee effects. Theoretical Population Biology amphibian species of differing vagility. 15, 64, 1–10. 1071–1078. Groom, M. (1998). Allee effects limit population viability of an Chust,G.,Pretus,J.LI.,Ducrot,D.&Ventura, D. (2004). annual plant. The American Naturalist 151, 487–496. Scale dependency of insect assemblages in response to landscape Gu,W., Heikkila,R. & Hanski, I. (2002). Estimating the pattern. Landscape Ecology 19, 41–57. consequences of habitat fragmentation on extinction risk in Collingham,Y.C.& Huntley, B. (2000). Impacts of habitat dynamic landscapes. Landscape Ecology 17, 699–710. fragmentation and patch size upon migration rates. Ecological ,A.D.& , M. L. (2002). Amphibian distributions Applications 10, 131–144. Guerry Hunter in a landscape of forests and agriculture: an examination of Courchamp,F., Clutton-Brock,T.& Grenfell, B. (1999). Inverse and the . Trends in Ecology landscape composition and configuration. Conservation Biology 16, & Evolution 14, 405–410. 745–754. Cowlishaw, G. (1999). Predicting the pattern of decline of Gunn,J.S., Desrochers,A., Villard,M.-A., Bourque,J. African primate diversity: an from historical & Ibarzabal, J. (2000). Playbacks of mobbing calls of black- deforestation. Conservation Biology 13, 1183–1193. capped chickadees as a method to estimate reproductive activity Cumming,S.G. & Schmiegelow, F. K. A. (2001). Effects of of forest birds. Journal of Field Ornithology 71, 472–483. habitat abundance and configuration, and the forest matrix, Gustafson,E.J.&Parker, G. R. (1992). Relationship between on distributional patterns of boreal birds. Sustainable Forest landcover proportion and indices of landscape spatial pattern. Management Network working paper. Landscape Ecology 7, 101–110. Cushman, S. A. (2006). Effects of habitat loss and fragmentation Hanski, I. (1991). Metapopulation dynamics: brief history and on amphibians: A review and prospectus. Biological Conservation conceptual domain. Biological Journal of the Linnean Society 42, 3–16. 128, 231–240. Hanski,I. & Ovaskainen, O. (2002). Extinction debt at Cushman,S.A. & McGarigal, K. (2003). Landscape level . Conservation Biology 16, 666–673. patterns of avian diversity in the Oregon Coast Range. Ecological Henein,K., Wegner,J. & Merriam, G. (1998). Population Monographs 73, 259–281. effects of landscape model manipulation on two behaviourally Davies,K.F., Margules,C.R. & Lawrence, J. F. (2000). different woodland small mammals. Oikos 81, 168–186. Which traits of species predict population declines in Henle,K., Davies,K.F., Kleyer,M., Margules,C.& Set- experimental forest fragments? Ecology 81, 1450–1461. tele, J. (2004). Predictors of species sensitivity to fragmentation. Desrochers,A.&Hannon, S. J. (1997). Gap crossing decisions Biodiversity and Conservation 13, 207–251. by forest songbirds during the post-fledging period. Conservation Hertzberg,K., Yoccoz,N.G., Ims,R.A. & Leinaas,H.P. Biology 11, 1204–1210. (2000). The effects of spatial habitat configuration on recruit- dit Durell,S.E.A.leV., Goss-Custard,J.D. & Clarke, ment, growth and population structure in arctic Collembola. R. T. (1997). Differential response of migratory subpopulations Oecologia 124, 381–390.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society 52 Trisha L. Swift and Susan J. Hannon

Higgins,S.I., Lavorel,S.& Revilla, E. (2003). Estimating bird communities: empirical observations and theoretical impli- plant migration rates under habitat loss and fragmentation. Oikos cations. In Wildlife (2000): Modeling Habitat Relationships of Terrestrial 101, 354–366. Vertebrates (eds. Vertner, J. & Morrison, M. L.), pp. 305–313. Hill,M.F.& Caswell, H. (1999). Habitat fragmentation and University of Wisconsin Press, Madison, Wisconsin. extinction thresholds on fractal landscapes. Ecology Letters 2, Monkk¨ onen¨ ,M.&Reunanen, P. (1999). On critical thresholds 121–127. in landscape connectivity: a management perspective. Oikos 84, Hitchings,S.P.&Beebee, T. J. C. (1997). Genetic substructur- 302–306. ing as a result of barriers to gene flow in urban Rana temporaria Orbach, R. (1986). Dynamics of fractal networks. Science 231, (common frog) populations: implications for biodiversity 814–819. conservation. Heredity 79, 117–127. Otten,W., Bailey,D.J.& Gilligan, C. A. (2004). Empirical Holland,J.D.,Fahrig,L.&Cappuccino, N. (2005). Fecundity evidence of spatial thresholds to control invasion of fungal determines the extinction threshold in a Canadian assemblage of parasites and saprotrophs. New Phytologist 163, 125–132. longhorned beetles (Coleoptera: Cerambycidae). Journal of Insect Parker,M.&Mac Nally, R. (2002). Habitat loss and the 9, 109–119. fragmentation threshold: an experimental evaluation of impacts Homan,R.N., Windmiller,B.S. & Reed, J. M. (2004). on richness and total abundances using grassland invertebrates. Critical thresholds associated with habitat loss for two vernal Biological Conservation 105, 217–229. pool-breeding amphibians. Ecological Applications 14, 1547–1553. Plotnick,R.E.& Gardner, R. H. (1993). Lattices and land- Huggett, A. J. (2005). The concept and utility of ‘ecological scapes. In Lectures on Mathematics in the Life Sciences: Predicting thresholds’ in biodiversity conservation. Biological Conservation spatial effects in ecological systems, Vol. 23 (ed Gardner, R. H.), 124, 301–310. pp. 129–157. American Mathematical Society, Providence, RI. Imbeau,L.&Desrochers, A. (2002). Area sensitivity and edge Radford,J.Q.& Bennett, A. F. (2004). Thresholds in land- avoidance: the case of the three-toed woodpecker (Picoides scape parameters: occurrence of the white-browed treecreeper tridactylus) in a managed forest. Forest Ecology and Management 164, Climacteris affinis in Victoria, Australia. Biological Conservation 117, 249–256. 375–391. Jager,H.I.,Carr,E.A.&Efroymson, R. A. (2006). Simulated Radford,J.Q., Bennett,A.F. & Cheers, G. J. (2005). effects of habitat loss and fragmentation on a solitary mustelid Landscape-level thresholds of habitat cover for woodland- predator. Ecological Modelling 191, 416–430. dependent birds. Biological Conservation 124, 317–337. Johnson, D. H. (1999). The insignificance of statistical significance Regosin,J.V.,Windmiller,B.S.,Homan,R.N.&Reed,J.M. testing. Journal of Wildlife Management 63, 763–772. (2005). Variation in terrestrial habitat use by four pool-breeding Johnson,J.B. & Omland, K. S. (2004). Model selection in amphibian species. Journal of Wildlife Management 65, 1481–1493. ecology and evolution. Trends in Ecology and Evolution 19, 101–108. Reunanen,P., Monkk¨ onen¨ ,M.,Nikula,A.&Nivala,V. Kerkhoff,A.J.,Milne,B.T.&Maehr, D. S. (2000). Toward a (2004). Assessing landscape thresholds for the Siberian flying panther-centered view of the forests of South Florida. Conservation squirrel. Ecological Bulletins 51, 277–286. Ecology 4,1. Rhodes,J,R., Callaghan,J.G., McAlpine,C.A., de Keymer,J.E., Marquet,P.A., Velasco-Hernandez,J.X.& Jong,C., Bowen,M.E., Mitchell,D.L.& Possingham, Levin, S. A. (2000). Extinction thresholds and metapopulation H. P. (2008). Regional variation in habitat-occupancy thresholds: persistence in dynamic landscapes. American Naturalist 5, 478–494. a warning for conservation planning. Journal of Applied Ecology 45, King,A.W.&With, K. A. (2002). Dispersal success on spatially 549–557. structured landscapes: when do spatial pattern and dispersal Ruckelshaus,M.,Hartway,C.&Kareiva, P. (1997). Assess- behavior really matter? Ecological Modelling 147, 23–39. ing the data requirements of spatially explicit dispersal models. Lamberson,R.H., McKelvey,R., Noon,B.R. & Voss,C. Conservation Biology 11, 1298–1306. (1992). A dynamic analysis of northern spotted owl viability in Schmidt,B.D. & Roland, J. (2006). Moth diversity in a a fragmented forest landscape. Conservation Biology 6, 505–512. fragmented habitat: importance of functional groups and Lindenmayer,D.B.,Fischer,J.&Cunningham, R. B. (2005). landscape scale in the boreal forest. Ecology and Population Biology Native vegetation cover thresholds associated with species 99, 1110–1120. responses. Biological Conservation 124, 311–316. Schneider, M. F. (2001). Habitat loss, fragmentation and Lindenmayer,D.B.& Luck, G. (2005). Synthesis: Thresholds predator impact: spatial implications for prey conservation. in conservation and management. Biological Conservation 124, Journal of Applied Ecology 38, 720–735. 351–35. Schrott,G.R., With,K.A.& King, A. W. (2005a). Demo- Matthysen,E., Adriaensen,F. & Dhondt, A. A. (1995). graphic limitations of the ability of habitat restoration to rescue Dispersal distances of nuthatches, Sitta europaea, in a highly declining populations. Conservation Biology 19, 1181–1193. fragmented forest habitat. Oikos 72, 375–381. Schrott,G.R., With,K.A.& King, A. W. (2005b). On the McGarigal,K.& Cushman, S. A. (2002). Comparative eval- importance of landscape history for assessing extinction risk. uation of experimental approaches to the study of habitat Ecological Applications 15, 493–506. fragmentation effects. Ecological Applications 12, 335–345. Siffczyk,C., Brotons,L., Kangas,K & Orell, M. (2003). McGarigal,K. & McComb, W. C. (1995). Relationships Home range size of willow tits: a response to winter habitat loss. between landscape structure and breeding birds in the Oregon Oecologia 136, 635–642. Coast Range. Ecological Monographs 65, 236–260. Sole´,R.,Alonso,D.&Saldana˜ , J. (2004). Habitat fragmenta- McLellan,C.H., Dobson,A.P., Wilcove,D.S.& Lynch, tion and biodiversity collapse in neutral communities. Ecological J. F. (1986). Effects of forest fragmentation on new and old world Complexity 1, 65–75.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society Habitat loss and critical thresholds review 53

St. Clair,C.C.,Belisle´ ,M.,Desrochers,A.&Hannon,S. Van Horne, B. (1983). Density as a misleading indicator of habitat (1998). Winter responses of forest birds to habitat corridors and quality. Journal of Wildlife Management 47, 893–901. gaps. Conservation Ecology 2, 13. Vos,C.C.,Verboom,J.,Opdam,P.F.M.&Ter Braak,C.J.F. Stephens,P.A., Sutherland,W.J. & Freckleton,R.P. (2001). Toward Ecologically Scaled Landscape Indices. American (1999). What is the Allee effect? Oikos 87, 185–190. Naturalist 157, 24–41. Summerville,K.S.&Crist, T. O. (2001). Effects of experimen- Wiens,J.A., Schooley,R.L.& Weeks, R. D. (1997). Patchy tal habitat fragmentation on patch use by butterflies and skippers landscapes and animal movements: do beetles percolate? Oikos (Lepidoptera). Ecology 82, 1360–1370. 78, 257–264. Thies,C.& Tscharntke, T. (1999). Landscape structure and Wimberly, M. C. (2006). Species dynamics in disturbed land- biological control in agroecosystems. Science 285, 893–895. scapes: when does a shifting habitat mosaic enhance connectivity? Tilman,D.,May,R.,Lehman,C.&Nowak, M. (1994). Habitat Landscape Ecology 21, 35–46. destruction and the extinction debt. Nature 371, 65–66. With,K.A., Cadaret,S.J. & Davis, C. (1999). Movement Toms,J.D.& Lesperance, M. L. (2003). Piecewise regression: responses to patch structure in experimental fractal landscapes. a tool for identifying ecological thresholds. Ecology 84, Ecology 80, 1340–1353. 2034–2041. With,K.A., Gardner,R.H.& Turner, M. G. (1997). Land- Trani,M.K.&Giles, R. H. Jr. (1999). An analysis of deforesta- scape connectivity and population distributions in heterogeneous tion: Metrics used to describe pattern change. Forest Ecology and environments. Oikos 78, 151–169. Management 114, 459–470. With,K.A.&King, A. W. (1999a). Dispersal success on fractal Trombulak,S.C.&Frissel, C. A. (2000). Review of ecological landscapes: a consequence of lacunarity thresholds. Landscape effects of roads on terrestrial and aquatic communities. Ecology 14, 73–82. Conservation Biology 14, 18–30. With,K.A.& King, A. W. (1999b). Extinction thresholds for Trzcinski,M.K., Fahrig,L. & Merriam, G. (1999). Inde- species in fractal landscapes. Conservation Biology 13, 314–326. pendent effects of forest cover and fragmentation on the With,K.A.,Pavuk,D.M.,Worchuck,J.L.,Oates,R.K.& distribution of forest breeding birds. Ecological Applications 9, Fisher, J. L. (2002). Threshold effects of landscape structure 586–593. on biological control in agroecosystems. Ecological Applications 12, Turcotte,Y.&Desrochers, A. (2003). Landscape-dependent 52–65. response to predation risk by forest birds. Oikos 100, With,K.A., Schrott,G.R. & King, A. W. (2006). The 614–618. implications of metalandscape connectivity for population Van der Ree,R., Bennett,A.F.& Gilmore, D. C. (2004). viability in migratory songbirds. Landscape Ecology 21, 157–167. Gap-crossing by gliding marsupials: thresholds for use of Zanette,L.,Doyle,P.&Tremont, S. M. (2000). Food shortage isolated woodland patches in an agricultural landscape. Biological in small fragments: evidence from an area-sensitive passerine. Conservation 115, 241–249. Ecology 81, 1654–1666.

Biological Reviews 85 (2010) 35–53 © 2009 The Authors. Journal compilation © 2009 Cambridge Philosophical Society