Fire type and hemisphere determine the effects of fire on the alpha and beta diversity of vertebrates: a global meta-analysis

Authors


  • Editor: Thomas Gillespie

Abstract

Aim

We conducted a quantitative meta-analysis to investigate the responses of vertebrate diversity to fire, controlling for variables such as fire type, taxon and ecoregion to identify trends across studies and locations.

Location

World-wide.

Methods

We calculated indices of the difference in species richness (alpha diversity) and species composition (beta diversity) between burnt and unburnt habitats from studies reporting the species richness and assemblage composition of amphibians, reptiles, birds and mammals. We used a hierarchical approach to investigate the effects of fire on alpha and beta diversity. We tested first for the main effect of fire before investigating the potential influence of fire type (wildfire/prescribed burn), taxon, ecoregion and geographical location (hemisphere/continent).

Results

One hundred and four studies were evaluated: 56 studies on birds, 26 on mammals, 17 on reptiles and 5 on amphibians. The studies fell into 14 ecoregions, with the three most common being temperate broadleaf and mixed forests, temperate grasslands and savannas and shrublands, and temperate coniferous forest. The effect of fire on species richness and community assemblage composition was strongly influenced by fire type. Prescribed burns significantly increased alpha diversity, whereas wildfires had no overall effect. However, wildfire increased the alpha diversity of temperate coniferous birds in North America. The effects of fire on alpha diversity were stronger in the Northern than the Southern Hemisphere. Turnover in species assemblages (beta diversity) was influenced primarily by fire type. Species assemblages in burnt and unburnt habitats were more similar after prescribed burns and generated lower levels of beta diversity than did wildfires.

Main conclusions

The divergent effects of wildfires and prescribed fires on the alpha and beta diversity of vertebrates and the disparate responses of vertebrate diversity to fires in the Northern and Southern Hemisphere suggest that there is no general response of vertebrate diversity to fire. Our results provide little support for the patch mosaic burn theory or the intermediate disturbance hypothesis to predict post-fire responses of vertebrate diversity.

Introduction

Fire is a powerful force that has shaped the evolution and function of ecosystems in many parts of the world (Bond & van Wilgen, 1996; Bowman et al., 2009). Frequent and widespread fires can alter entire ecosystems, as seen in the transformation of expansive tracts of forest into fire-maintained savannas across South America, Africa, Australia and New Zealand (Keeley & Rundel, 2005; Bowman & Haberle, 2010). Fire-dependent ecosystems now cover a large proportion of the global land surface and contain primarily fire-tolerant and fire-adapted species, often at the expense of fire-sensitive species (Bowman et al., 2009).

Fire is often used intentionally by humans at relatively small scales (10–100 ha) with the aim of benefitting biodiversity (Parr & Andersen, 2006; Fuhlendorf et al., 2009). This approach is informed by the intermediate disturbance hypothesis (Connell, 1978) and the patch mosaic burn theory (Brockett et al., 2001), which each predict that a moderate level of disturbance, such as that produced by small-scale burns, will promote local biodiversity.

Despite the large body of literature on the responses of ecosystems to natural and prescribed fires, the effects of fire on faunal diversity remain uncertain. In many instances, wildfires eliminate habitat and food resources and increase the hunting efficiency of predators in post-fire landscapes (Letnic et al., 2005; Green & Sanecki, 2006; Kodandapani et al., 2008). In other cases, both low- and high-severity fires have been found to have positive effects on affected taxa (Fuhlendorf et al., 2006; Hutto, 2008; Fontaine & Kennedy, 2012). In some instances, however, these effects are short-lived and no lasting effects on faunal diversity are evident (e.g. Pons & Prodon, 1996; Cunningham et al., 2002; Letnic et al., 2013). Similarly, increasing numbers of studies are reporting that prescribed burning does not provide the expected biodiversity gains over a wide range of mammal, reptile and plant taxa (Parr & Andersen, 2006; Lindenmayer et al., 2008a). Another reason for uncertainty regarding the effect of fire is that the effects on individual species and subsequently species assemblages may vary depending on the intensity and extent of fires (Cleary et al., 2004; Pastro et al., 2011). For example, the relatively small patches of recently burnt habitats created by low-intensity prescribed fires may result in smaller differences in species assemblage composition between burnt and unburnt patches than the large patches of burnt habitats created by high-intensity wildfires (Cleary et al., 2004; Pastro et al., 2011).

In addition to the effect of fire type (wildfire or prescribed burn), the effect of fire on diversity is influenced by factors such as study habitat type, fire regime and geographical location (Driscoll et al., 2010; Fontaine & Kennedy, 2012). For example, the fauna of many C4 arid and temperate grasslands has co-evolved with fire; the biota in these habitat types is often fire-adapted and thus less affected by fire events (Andersen et al., 2005). Fire regime can also strongly influence the effect of fire on biota, with factors such as fire interval, intensity and seasonality interacting to change community structure and either positively or negatively influence the survival of affected species (Charrette et al., 2006; Parr & Andersen, 2006; Fisher et al., 2009).

The choice of study taxon can also determine the findings of fire studies (Pastro et al., 2011). Birds, for example, are relatively mobile and are often able to escape fire. Consequently, mobile taxa may be less affected by fire than more sedentary taxa, such as amphibians, that can suffer high mortality (Bury, 2004; Cano & Leynaud, 2010). Similarly, endothermic and ectothermic vertebrates may respond differently to post-fire succession owing to their disparate metabolic requirements. For example, in arid Australia, the colonization by small mammals of habitat regenerating after fire is largely a response to the availability of food, whilst lizards appear to colonize habitats based on their suitability for thermoregulation (Letnic et al., 2004). These suites of interacting factors make it challenging to draw general conclusions about the effect of fire on biodiversity and to design appropriate fire management strategies for biodiversity conservation (Pastro et al., 2011).

In this study we use quantitative meta-analytic methods to investigate the effects of fire on community-scale biodiversity. Our approach allowed us to control for variables such as fire type, taxon and vegetation type to identify trends across studies and locations. Because previous studies show that fire can influence species diversity by increasing or decreasing diversity at the patch scale and also by driving differences in the relative abundances of species in burnt and unburnt habitats, we examined the effect of fire on two components of vertebrate diversity: alpha diversity (α, within sampling units) and beta diversity (β, between sampling units (Whittaker, 1972; Crist et al., 2003). Our specific aims were: (1) to quantify the effect of fire on alpha and beta diversity and (2) determine the influence, if any, of fire type, study taxon (amphibians, reptiles, birds and mammals), habitat and geographical location on the effects of fire.

Materials and Methods

Literature searches

We searched the electronic databases Web of Science, BIOSIS Previews, Zoological Record and Scopus using combinations of the search terms fire*, wildfire*, burn*, mammal*, rodent*, reptil*, lizard* herpeto*, avian*, bird*, vertebrate* and amphib*. The reference lists from earlier reviews (Kotliar et al., 2007; Fontaine & Kennedy, 2012) and of relevant papers were also searched. Only published, peer-reviewed journal articles were included in the analysis. For consistency, the large body of grey literature on fire effects in North America was not included. The end-date for literature inclusion was June 2011.

Papers were included if they addressed the effect of fire on the species richness of amphibians, birds, mammals or reptiles or at the community level. Studies that analysed the effects of fire on only one or a selected number of species were excluded. We did not set a threshold of allowable species numbers for inclusion in the study: as long as a study endeavoured to sample all available species in a study area it was deemed suitable for inclusion.

Where studies investigated the combined effects of fire and other disturbances, only data pertaining to the effects of fire were extracted. To avoid influencing our conclusions with selection criteria that were not specifically a function of the relevance of the data, no other quality-based inclusion criteria were applied (Englund et al., 1999).

Variables

The following data were extracted from the text, tables and/or figures of each study: experimental methodology (treatment–control or before–after); replication (replicated or unreplicated); fire type (wildfire or prescribed burn); study taxon (amphibian, reptile, bird or mammal); time since fire (including study time length); study spatial scale (summed area of the data collection patches) and geographical location (continent and hemisphere). To measure the effect of habitat, each study was also classified into one of 17 terrestrial ecoregions and one freshwater ecoregion following a modified version of the World Wildlife Fund (WWF) global classification system (Olson et al., 2001). It should be noted that these ecoregions are identified by biome type and not by geography. Ecoregions were as follows: arid and semi-arid grassland; boreal forests/taiga; flooded grasslands and savannas; mangroves; Mediterranean coniferous forest; Mediterranean oak forest; montane grassland and shrubland; small lake ecosystems; subalpine woodland; temperate broadleaf and mixed forest; temperate coniferous forest; temperate grasslands, savannas and shrublands; tropical and subtropical grasslands, savannas and shrublands; tropical and subtropical coniferous forests; tropical and subtropical dry broadleaf forests; tropical and subtropical moist broadleaf forests; tundra; and xeric shrubland.

Data were collected at the end of each study. In the event that terminal data were not available, the mean values obtained over the course of the study were used. Where studies included once-burnt or multiple-burnt habitats, data from the once-burnt areas were used to minimize, but perhaps not remove, the confounding factor of repeated burning. Where a study included sampling units that varied in spatial scale, the means of these were used to indicate the spatial scale of the study.

Our literature searches yielded 368 publications that addressed the effects of fire on vertebrates. Eighty-seven of these addressed the effect of fire on the alpha or beta diversity of amphibians, reptiles, birds or mammals and thus were included in the meta-analysis (Appendix S1 in Supporting Information). Some publications reported the independent effects of fire on more than one taxon and thus the total number of studies included in the meta-analysis was 104 (Appendix S1).

Statistical analyses

Alpha diversity

Due to the limited data available in many published papers, we defined alpha diversity (α) as the species richness within a sampling unit. We used the log response ratio of species richness between burnt and unburnt habitats as our metric of alpha diversity. This effect size ln(Xe/Xc) was calculated for each study to indicate the effect of the fire on alpha diversity, where Xe and Xc represent the species richness in post-fire burnt and unburnt habitat (treatment–control studies) or in habitat before and after a fire (before–after studies; Rosenberg et al., 2000; Salo et al., 2010). This metric was chosen over more traditional effect sizes such as Hedges' d or ln(R) because it does not require within-study variance and a large proportion of our data set consisted of unreplicated or pseudo-replicated studies in which within-study variance was not reported (Salo et al., 2010).

We tested first for the main effect of fire before investigating the potential influence of fire type, taxon, habitat and geographical location (hemisphere and continent). An effect size ln(Xe/Xc) ± 95% confidence interval (CI) > 0 indicated that a fire event increased species richness, an effect size ln(Xe/Xc) ± 95% CI = 0 (i.e. the confidence intervals overlap 0) indicated that the fire event had no effect on species richness, and an effect size ln(Xe/Xc) ± 95% CI < 0 indicated that the event reduced species richness (Rosenberg et al., 2000).

Univariate tests were employed to compare alpha-diversity effect sizes within and between variable groupings. One outlier was removed from the prescribed burn data set. This outlier was identified using spss version 18 to make the data suitable for analysis using univariate tests. Relationships between the alpha-diversity effect size and (1) study length and (2) spatial scale were explored using Pearson correlations. The relationship between alpha-diversity effect size and study replication was tested using the Mann–Whitney U-test.

Beta diversity

The effect of fire on beta diversity was assessed using Sørensen's similarity index (Sørensen, 1957). The formula for the index is QS = 2C/(A + B),where A represents the species richness of the landscape prior to the burn or in control habitat and B represents the species richness of the landscape after the burn or in treatment habitat. C represents the number of species shared by both A and B. An increase in the Sørensen index signifies an increase in the similarity of species assemblages between burnt and unburnt habitats and thus a decrease in beta diversity. As with our alpha-diversity metric, metrics such as the Bray–Curtis dissimilarity index (Kessler et al., 2009) which take into account species abundances could not be calculated given the limited quantitative information that was presented in many studies.

We used a hierarchical approach to investigate the effects of fire on alpha and beta diversity (Rosenberg et al., 2000). We tested first for the effect of our primary variable of interest, fire, before investigating the potential contribution of fire type, taxon, habitat and geographical location (continent and hemisphere). The effect of fire on beta diversity within and between variable groupings was investigated using univariate tests. Outliers were identified and removed prior to analysis.

Our measures of alpha and beta diversity were each free to vary independently as we chose not to calculate beta diversity as a multiplicative or additive function of gamma and alpha diversity (Veech & Crist, 2010). All statistical analyses were undertaken in spss version 18.

Publication bias

Publication bias can arise in meta-analyses as authors may choose to publish only statistically significant results (the ‘file-drawer problem’; Rosenthal, 1979) or results that support currently popular theories (Simmons et al., 1999).

To check for publication bias, we plotted the sample size of each study against the alpha-diversity effect size ln(Xe/Xc) to produce a funnel plot (Appendix S2; Gates, 2002). This method was chosen over the more traditional normal quantile plot because our study contained many unreplicated studies for which the variance required by the normal quantile plot could not be calculated. The funnel plot produced no evidence of publication bias (Appendix S2).

Results

Fifty-six studies investigated the effect of fire on birds, 26 on mammals, 17 on reptiles and 5 on amphibians. Seventy studies were replicated (i.e. they included at least two treatment and two control plots or a before–after design where the treatment was reversed between plots) and 34 studies were unreplicated. Fifty-three studies examined the effects of prescribed burns and 51 examined the effects of wildfires (Appendix S1).

The studies took place in 14 ecoregions. The three most commonly studied ecoregions were temperate broadleaf and mixed forest, temperate grasslands, savannas and shrublands, and temperate coniferous forest (Table 1). Fifty-two studies were located in North America, 22 in Australia, 16 in South America, 7 in Europe, 5 in Africa and 2 in Asia. Overall, 61 studies were located in the Northern Hemisphere, 34 in the Southern Hemisphere and 9 in equatorial regions (Fig. 1).

Table 1. World Wildlife Fund ecoregion classifications (Olson et al., 2001; Coetzee et al., 2013) and the number of studies that fell into each. The ecoregion classification for each individual study is included in Appendix S1
EcoregionTaxonNumber of studies
Arid and semi-arid grasslandsMammals6
Reptiles3
Boreal forests/taigaBirds1
Mammals1
Flooded grasslands and savannasBirds5
Mediterranean coniferous forestBirds2
Mediterranean oak forestBirds2
Reptiles1
Mammals2
Montane grassland and shrublandMammals1
Subalpine woodlandBirds1
Temperate broadleaf and mixed forestAmphibians3
Birds9
Mammals3
Reptiles2
Temperate coniferous forestBirds12
Temperate grasslands savannas and shrublandsAmphibians2
Birds7
Mammals4
Reptiles5
Tropical and subtropical grasslands, savannas and shrublandsBirds3
Mammals3
Reptiles1
Tropical and subtropical dry broadleaf forestsBirds5
Mammals1
Reptiles2
Tropical and subtropical moist broadleaf forestsBirds6
Mammals1
Xeric shrublandBirds3
Mammals4
Reptiles3
Total studies 104
Figure 1.

The location of each of the studies included in the meta-analysis. Closed circles denote studies reporting wildfires, open circles denote studies reporting prescribed burns and closed triangles denote studies reporting both wildfires and prescribed fires.

Lists with scientific names for each species were provided for 94 studies (Appendix S1). Birds were the most diverse of the taxonomic groups evaluated, with a mean of 28.16 species per study (± 2.30 SE). Reptiles were the next most species-rich group, with an average of 11.76 species per study (± 1.89 SE). Amphibians returned an average of 6.40 species per study (± 1.96 SE) and mammals were the least species-rich group, with an average of 6.26 species per study (± 0.88 SE).

Alpha diversity

The effect of fire on alpha diversity was analysed for 96 studies (Appendix S1). The effect of fire on alpha diversity in each study, along with the number of species that were recorded in burnt and unburnt habitats, is included in Appendix S1. Fire had no overall effect on alpha diversity (see Appendix S3). Instead, the effect of fire was determined primarily by fire type, as prescribed burns significantly increased alpha diversity and wildfires had no effect (Fig. 2a). Fire effects were not influenced by study taxon or habitat, and North America was the only continent in which fire had significant effects (Figs 2b & 3a–c).

Figure 2.

Effect size ln(Xe/Xc) ± 95% confidence interval (CI) of prescribed burns (open symbols) and wildfires (closed symbols) on the alpha diversity of terrestrial vertebrates (a) across all studies and (b) in North America. (c) For birds in temperate coniferous forests (TCF) in North America and (d) for birds in temperate grassland and shrubland (TGS) in North America. Xe and Xc represent the mean species richness in post-fire burnt and unburnt habitat (treatment–control studies) or in habitat before and after a fire. The number of studies in each analysis is indicated above each bar. A significant effect is obtained if ln(Xe/Xc) ± 95% CI does not overlap zero. A significant difference between the prescribed burn and wildfire effect size in a given grouping is indicated by * (P ≤ 0.05).

Figure 3.

Effect size ln(Xe/Xc) ± 95% confidence interval (CI) of fire on alpha diversity grouped by (a) taxon, (b) ecoregion and (c) continent. Xe and Xc represent the mean species richness in post-fire burnt and unburnt habitat (treatment–control studies) or in habitat before and after a fire. The number of studies in each analysis is indicated above each bar. A significant effect is obtained if ln(Xe/Xc) ± 95% CI does not overlap zero. Ecoregion codes are as follows: ASG, arid and semi-arid grasslands; FGS, flooded grasslands and savannas; TBF, temperate broadleaf and mixed forests; TCF, temperate coniferous forests; TGS, temperate grasslands, savannas and shrublands; TS, tropical and subtropical grasslands, savannas and shrublands; TSD, tropical and subtropical dry broadleaf forests; TSR, tropical and subtropical moist broadleaf forests; XS, xeric shrublands. Groupings without sufficient replication were excluded.

The effect of fire in North America was not a generalized trend as it varied according to fire type and ecoregion. Overall, prescribed burns increased alpha diversity but wildfires did not. However, prescribed burns increased the alpha diversity of birds in temperate grassland and shrubland but had no effect on bird alpha diversity in temperate coniferous forests. Wildfires resulted in increased alpha diversity of birds in temperate coniferous forests (Fig. 2c, d).

Hemisphere was also an important determinant, as fire significantly increased alpha diversity in the Northern Hemisphere and had a generally negative effect on alpha diversity in the Southern Hemisphere (Fig. 4). This held for both prescribed burns and wildfires. The effect of prescribed burns in the Northern Hemisphere was also significantly greater than in the Southern Hemisphere (t = −3.312, d.f. = 46, P < 0.001).There was no correlation between alpha-diversity effect size and study length, study spatial scale or replication.

Figure 4.

Effect size ln(Xe/Xc) ± 95% confidence interval (CI) of prescribed burns (open symbols) and wildfires (closed symbols) on alpha diversity in the Northern and Southern Hemisphere. Xe and Xc represent the mean species richness in post-fire burnt and unburnt habitat (treatment–control studies) or in habitat before and after a fire. The number of studies in each analysis is indicated above each bar. A significant effect is obtained if ln(Xe/Xc) ± 95% CI does not overlap zero. A significant difference between the effect of prescribed burns in the Northern and Southern Hemisphere is indicated by * (P ≤ 0.05).

Beta diversity

The effect of fire on the similarity of species assemblage composition between burnt and unburnt habitats (beta diversity) was calculated for 97 studies. As with alpha diversity, these effects were not influenced by study taxon, study habitat or continent (Fig. 5a–c). Instead, the effect of fire on beta diversity was influenced primarily by fire type, as species assemblages between burnt and unburnt habitats were less similar after wildfire (higher beta diversity) when compared with prescribed burns (Fig. 6). Wildfires increased beta diversity when compared with prescribed burns across all studies (t = 3.049, d.f. = 88, P = 0.0015) and within the groupings of birds (t = 1.798, d.f. = 44, P = 0.040), lizards (t = 1.753, d.f. = 14, P = 0.050), North America (t = 2.378, d.f. = 45, P = 0.011), North American birds (t = 2.165, d.f. = 23, P = 0.020) and birds in North American temperate coniferous forests (t = 4.981, d.f. = 10, P = 0.00028). Studies in the Northern Hemisphere (t = 2.693, d.f. = 42, P = 0.0005) and Southern Hemisphere (t = 1.707, d.f. = 37, P = 0.048) showed a similar trend. Analysis was limited to these particular habitat and continent groupings as there was not adequate replication in other groupings for analysis (Appendix S1).

Figure 5.

Sørensen's similarity index ± SE grouped by (a) taxon, (b) ecoregion, and (c) continent. The number of studies in each analysis is indicated above each bar. Ecoregion codes are as follows: ASG, arid and semi-arid grasslands; FGS, flooded grasslands and savannas; MOF, montane forest; TBF, temperate broadleaf and mixed forests; TCF, temperate coniferous forests; TGS, temperate grasslands, savannas and shrublands; TS, tropical and subtropical grasslands, savannas and shrublands; TSD, tropical and subtropical dry broadleaf forests; TSR, tropical and subtropical moist broadleaf forests; XS, xeric shrublands. Groupings without sufficient replication were excluded.

Figure 6.

Sørensen's similarity index between burnt and unburnt habitats ± SE after prescribed burns (open symbols) and wildfires (closed symbols) across all studies and grouped according to fire type, taxon, continent, ecoregion and hemisphere. The number of studies in each analysis is indicated above each bar for prescribed burn studies and below each bar for wildfire studies. A significant difference between the prescribed burn and wildfire Sørensen similarity index is indicated by * (P ≤ 0.05), ** (P ≤ 0.01) and *** (P ≤ 0.001).

Discussion

Alpha diversity

Small-scale prescribed burns (10–100 ha) are often thought to promote alpha diversity by creating local habitat heterogeneity, as the resulting habitat matrix provides for species with varying habitat requirements (Fuhlendorf et al., 2009; Pastro et al., 2011). However, recent studies have found that this effect is not widespread and is influenced by factors such as habitat type and the taxon studied (e.g. Cleary et al., 2004; Lindenmayer et al., 2008a; Pastro et al., 2011). In our study, the positive effect of prescribed burning on alpha diversity was relatively small and associated with a large confidence interval. We suggest that this result was driven by a high proportion of habitat- and species-specific responses within the data set.

In support of this interpretation, the meta-analysis further revealed that taxon was a poor predictor of fire effects on alpha diversity (Fig. 3a). In addition, North America was the only continent where prescribed burns significantly increased alpha diversity (Fig. 3c); however, the effect differed between ecoregions (Fig. 2c, d). The absence of a broad-scale ‘wildlife response’ to prescribed burning has been identified in other recent studies (e.g. Kotliar et al., 2007; Pastro et al., 2011; Fontaine & Kennedy, 2012), which have similarly highlighted the importance of variables such as habitat, fire severity and species in determining fire responses. These findings have important ramifications for the application of the patch mosaic burn hypothesis and the intermediate disturbance hypothesis because they suggest that a mosaic of habitat in different seral stages, as produced by patchy prescribed burns, does not as a general rule enhance alpha diversity within specific vertebrate taxa. Recent research suggests that fire-sensitive species could be further adversely affected by such practices (Kelly et al., 2012).

In our meta-analysis, wildfires did not on average significantly affect alpha diversity. Neutral responses may represent true ecological insensitivity in variables of interest, or reflect inconsistency among studies. We suggest that the latter is more likely the cause of this result. Wildfires included in our analysis occurred in a range of locations and ecoregions (Fig. 1), and probably produced contrary variations that contributed to the overall neutral result. Such ‘neutralization’ in meta-analyses is not surprising. For example, ecoregions where wildfires occur frequently and are well studied, such as parts of the Mediterranean, South Africa, Australia and the temperate coniferous forests of North America, may be more adapted to severe fires and thus resilient to the effects of fire (Andersen et al., 2005). Species-specific responses in regions where severe burn specialists occur may also have contributed to the overall neutral effect and positive effect evident in North American coniferous forests (Hutto, 2008).

Differing severities among wildland fires, and the varying animal responses that they engender, also may have masked species responses to wildfire. Fire severity can influence the effects of wildfire on terrestrial biota, as vertebrates respond differently to low-intensity and stand-replacing, high-intensity fires (Smucker et al., 2005; Fontaine & Kennedy, 2012). The effect of low-, mid- or high-severity fires on vertebrate diversity at the patch scale may vary further with habitat type. For example, in Australia, wildfires in stony desert habitats dominated by sparse grass cover have less effect on habitat structure and consequently species richness of small mammals than wildfires in sandy desert habitats dominated by perennial grasses of the genus Triodia (Letnic et al., 2005, 2013). While wildfires in both sandy desert and stony desert habitats produce vast expanses of post-fire environment that are devoid of vegetation, the fauna of normally sparsely vegetated stony desert habitats appear better able to tolerate the removal of vegetation than species in sandy desert habitats which make extensive use of the habitat matrix provided by Triodia (Letnic et al., 2013).

In addition, while high-severity, stand-replacing fires often reduce species diversity (Green & Sanecki, 2006; Kodandapani et al., 2008), high-severity fires may also be important for a few species by providing habitat conditions not found in unburned or lightly-burned habitats (Letnic et al., 2004; Fontaine & Kennedy, 2012). These effects may also occur over very long time-scales. For example, high-severity fire adversely affects northern spotted owls (Strix occidentalis occidentalis) in north-western California, but creates suitable habitat after 20 years and high-quality habitat after about 60 years (Franklin et al., 2000). Analysis of wildfires as a single category may therefore preclude the detection of finer-scale responses to differing fire severities. In addition to this, many high-severity fires contain mixtures of burn severities which may further confound the detection of species responses (Kotliar et al., 2007; Halofsky et al., 2011; Fontaine & Kennedy, 2012). Analysis at this finer scale may allow the identification of positive and negative species and community responses to fire that are masked by a single ‘high-severity’ analysis. Of our wildfire studies, 23 reported a high-severity fire, 10 a low-severity fire and 14 studies did not report fire severity (Appendix S1). As each fire type affected a range of habitats, we did not have adequate observations to test for the independent effects of fire severity within the wildfire grouping.

Hemisphere had a strong influence on the effect of fire on alpha diversity. On average, fire increased alpha diversity in the Northern Hemisphere and somewhat decreased it in the Southern Hemisphere across all fire types, ecoregions and taxa (Fig. 4). It is interesting to note here that the effect of fire type disappeared when studies were grouped by hemisphere. The Northern Hemisphere effect was most likely influenced by the high proportion of North American studies in that grouping (51 of 58 Northern Hemisphere studies) and the generally positive effect of North American fires on alpha diversity. Many North American studies involve the introduction of prescribed fire into fire-suppressed or excluded habitats (Allen et al., 2006) or the effects of wildfire in fire-suppressed forests (Mendelsohn et al., 2008). Our meta-analyses suggest that these fires may not have the negative effects on biodiversity that often accompany wildfires in the Southern Hemisphere. For example, high-severity wildland fire in south-western montane forests in America did not have long-term negative ecological effects on bird communities (Kotliar et al., 2007). Further to this, Fontaine & Kennedy (2012), in their meta-analysis of bird responses to fire in fire-prone habitats across the United States, found that while many bird species responded negatively to high-severity fires (predominantly canopy-nesting and foliage-foraging species), many others did not.

The Southern Hemisphere effect is probably due to regional factors including climate, human population density (Archibald et al., 2013) and interactions with other processes such as introduced predators and the cultural and economic activities of humans (Driscoll et al., 2010). Like the Northern Hemisphere, the incidence of wildfire across the Southern Hemisphere is largely climate-driven, but has been less disrupted by human activities owing in part to lower population densities (Archibald et al., 2013). The extensive and intense wildfires that occur in the Southern Hemisphere are increasingly being found to have strongly negative effects on alpha diversity in some areas (Cleary et al., 2004; Green & Sanecki, 2006; Pastro et al., 2011). In other areas, by contrast, wildfires have been found so far to have had few long-term effects on the diversity of flora and fauna (Williams et al., 2008; Lindenmayer et al., 2008b).

The prescribed burns grouping was dominated by Australian studies (22 of the 32 Southern Hemisphere studies) and prescribed burns are known to interact in complex ways with other processes on that continent (Driscoll et al., 2010). The significant pressure that Australian fauna experience from invasive predators and human modification of fire regimes means that prescribed burns may not produce the expected benefits for alpha diversity (Driscoll et al., 2010). Debate is currently high over what constitutes a ‘correct’ fire regime in various parts of the continent (Bradstock et al., 2005; Driscoll et al., 2010; Andersen et al., 2012), and in some regions too-frequent prescribed burns are thought to be negatively affecting diversity (Driscoll & Henderson, 2008). In addition to this, introduced predators such as the feral house cat (Felis catus) and European red fox (Vulpes vulpes) may preferentially hunt on newly burnt habitats and have catastrophic effects on recolonizing biota in the bare habitat (Legge et al., 2011; Andersen et al., 2012).

Beta diversity

The effect of fire on community similarity between burnt and unburnt habitats was influenced primarily by fire type, as wildfires consistently increased beta diversity when compared with prescribed burns (Fig. 6). This effect occurred equally in both hemispheres and held for most variable groups including birds, lizards, North American biota, North American birds and birds in temperate coniferous forests. Our results concur with those of Cleary et al. (2004), who found that different taxa can show largely independent patterns of alpha diversity in response to fire but that beta diversity remains convergent.

Prescribed burns and wildfires each create different scales of landscape heterogeneity which are likely to be the drivers behind the divergent patterns of beta diversity (Pastro et al., 2011). Animals are often able to move between the typically smaller burnt and unburnt patches after a prescribed burn and thus species turnover between the two habitat types is low. In contrast, wildfires often burn intensely and affect larger areas of habitat (Archibald et al., 2013). By profoundly altering habitat structure, often across large tracts of land, wildfires typically promote the beta diversity of vertebrate fauna by producing distinctive burnt and unburnt habitats that support unique species assemblages (Holdsworth & Uhl, 1997; Pastro et al., 2011). The typically large area of burnt habitat created by wildfire (Archibald et al., 2013) is likely to enhance the high species turnover associated with these fires, as species recolonizing burnt areas must travel further to reach burnt patches (Pastro et al., 2011).

Methodological considerations

It should be noted that, as is the nature of meta-analyses, our results are the average effects of all studies tested. Although comprehensive species lists were provided for 94 of the studies included in the meta-analysis, the examination of fire responses of individual species or groups of species, or the investigation of habitat use associations and how they may influence species responses to fire, was outside the scope of this study. Many prescribed burns are undertaken to protect IUCN Red List species, particularly in south-western North America, and a post-fire analysis of just IUCN Red List species and changes in their density or abundance may yield substantially different findings.

Replicated studies of animal responses to fire regimes have been identified as a major research gap and remain a research priority in fire-prone communities (Bury, 2004; Driscoll et al., 2010). Our results highlight these shortcomings, as we found that many studies were either unreplicated or poorly replicated as they were replicated only once at the landscape scale. Constraints on experimental design are a problem inherent in unplanned natural ecological experiments, with other researchers acknowledging the difficulties that are attendant when single fires occur over large spatial areas with varying degrees of intensity (Kotliar et al., 2007). In these situations, multiple studies are needed to fully characterize species response patterns, and data collection along burn gradients will help to identify spatiotemporal variation. In addition, sampling methodology such as sample unit size, the count/detection metric that is used and granularity of habitat classification may also contribute to variation among studies (Kotliar et al., 2007).

It is also worth considering the effect that differing species detection probabilities may have on data quality. Easier detection of species in newly simplified burnt environments or increased post-fire movements of some species may artificially inflate estimates of species numbers in these habitats (Driscoll et al., 2012). While differing detection probability is certainly an issue in many fire studies, we believe that the landscape scale of our study, and the focus on assemblage composition and species richness rather than population-level abundance and indices, provide some degree of protection from the effects of differing detection probabilities. However, this potentially confounding factor should still be accounted for in the interpretation of fire studies (Driscoll et al., 2012).

Another important factor that could not be addressed by our study is the effect of fire regime, including variables such as fire severity, frequency and size. Despite an abundance of research, the ecological significance of fire regimes remains contentious or poorly understood, particularly in the Southern Hemisphere (Parr & Andersen, 2006). The effect of repeated fires is also likely to have quite different results between prescribed burns, low-severity wildfires and high-severity wildfires, and the effect of repeated high-severity wildfires in particular is not well understood (Thompson et al., 2007; Fontaine et al., 2009). The propensity of ecologists to study the most severe wildfires adds an additional potential source of bias to our study.

Conclusions

Our study has returned the intriguing result that the effect of fire on alpha diversity and landscape-scale species assemblages cannot be grouped by taxon, ecoregion or geographic location. Instead, fire type is the primary determinant of fire effects on both aspects of diversity, although the two diversity metrics studied here were affected in opposite ways. These results highlight the important fact that there is no ‘one size fits all’ approach that can be incorporated into land management practices, and that the effects of management practices based on the intermediate disturbance hypothesis (Connell, 1978) and the patch mosaic burn theory (Brockett et al., 2001) are context dependent, especially in parts of the Southern Hemisphere. Small-scale prescribed burns will almost definitely not produce the desired benefits to alpha or beta diversity across all habitats, taxa and locations. Our results also suggest that fire management programmes need to consider the effect of management actions at both the patch and landscape scale if they are to be truly effective. Our study identifies the need for more replicated, landscape-scale studies of the effects of fire on terrestrial animal communities, especially lizards and amphibians. Further research on the effects of fire on community assemblage composition would also be of great interest in montane, alpine and boreal habitats, in tropical forests and in wetland environments, which are currently under-represented in ecological studies of fire.

Additional references to the data sources used in this study may be found at the end of Appendix S1 at [http://onlinelibrary.wiley.com/doi/10.1111/geb.12195/suppinfo].

Louise A. Pastro is undertaking her PhD in fire ecology at the University of Sydney. Her dissertation focuses on the effect of broad-scale wildfire on vertebrates in arid Australia. Her research interests include landscape ecology and fire ecology.

Ancillary