Modeling mammal response to fire based on species’ traits

Fire has shaped ecological communities worldwide for millennia, but impacts of fire on individual species are often poorly understood. We performed a meta‐analysis to predict which traits, habitat, or study variables and fire characteristics affect how mammal species respond to fire. We modeled effect sizes of measures of population abundance or occupancy as a function of various combinations of these traits and variables with phylogenetic least squares regression. Nine of 115 modeled species (7.83%) returned statistically significant effect sizes, suggesting most mammals are resilient to fire. The top‐ranked model predicted a negative impact of fire on species with lower reproductive rates, regardless of fire type (estimate = –0.68), a positive impact of burrowing in prescribed fires (estimate = 1.46) but not wildfires, and a positive impact of average fire return interval for wildfires (estimate = 0.93) but not prescribed fires. If a species’ International Union for Conservation of Nature Red List assessment includes fire as a known or possible threat, the species was predicted to respond negatively to wildfire relative to prescribed fire (estimate = –2.84). These findings provide evidence of experts’ abilities to predict whether fire is a threat to a mammal species and the ability of managers to meet the needs of fire‐threatened species through prescribed fire. Where empirical data are lacking, our methods provide a basis for predicting mammal responses to fire and thus can guide conservation actions or interventions in species or communities.


INTRODUCTION
Fire shapes ecological communities worldwide (Bowman et al., 2009;He et al., 2019). The impact of fire on animal populations can depend on characteristics of the fire and traits of the species. Two different fires can affect the same population in different ways, just as the same fire may affect populations of 2 species with differing requirements in different ways (Mowat et al., 2015;Radford et al., 2020;Swan et al., 2016). Understanding how fire affects communities is becoming more important as human-driven changes, such as climate change, habitat modification, and fire suppression, change how fire behaves in many ecosystems (Cochrane et al., 1999). In some ecosystems, large, intense fires occur more frequently today than centuries or decades ago as a direct consequence of human land-use changes (Bowman, 1998;Cochrane et al., 1999). For species evolved to cope with a certain fire regime, these relatively rapid changes can be disastrous, and many land managers are using prescribed fire to try to reestablish desired fire regimes. Land managers use prescribed burning to protect human structures and lives and to conserve habitat or species of interest (Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, Macgregor, Possingham, et al., 2010). When fire is implemented for conservation, the regime is usually based on the optimal fire interval to restore or maintain habitat in a desired state (Dellasala et al., 2004;Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, Mac-Gregor, Russell-Smith, et al., 2010). In turn, this is thought to benefit animal communities that have evolved in these habitats and have been negatively affected by changed fire regimes (Pausas & Parr, 2018). To ascertain whether fire is having the predicted effect, researchers may assess the impacts of fire on species, with measures of abundance, occupancy, behavior, or fitness (Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, MacGregor, Russell-Smith, et al., 2010). However, other simultaneous, interacting human-induced changes such as introduced species, habitat fragmentation, and climate change can lead to unexpected outcomes (Legge et al., 2019).
Fire affects species through mortality and indirect effects, such as changes to food or shelter resources or increased exposure to predators, which may increase mortality or invoke behavioral changes (Engstrom, 2010;Jolly et al., 2022;Santos et al., 2022). The ability of a species to avoid direct or immediate impacts of a disturbance, such as fire, can be referred to as its resistance, and the ability of a species to withstand the indirect effects of disturbance can be referred to as its resilience (Díaz-Delgado et al., 2002;Grime et al., 2000). Some organisms have evolved to depend on fire, for example, directly for seed release for reproduction in some plants and indirectly because the animal depends on a particular habitat structure to avoid exposure to predators (Ames et al., 2017;Bowman et al., 2009Bowman et al., , 2014Lawson et al., 2010;Leahy et al., 2015). Fire at the wrong time or with the wrong frequency, severity, or extent can have deleterious impacts on fauna (Ames et al., 2017;Lawson et al., 2010;Santos et al., 2022).
Certain traits predispose species to declines and extinction, and traits that drive a population response to an extrinsic source of potential mortality vary between threats (Fisher & Owens, 2004;Hernández-Yáñez et al., 2022). For example, specialized species are particularly vulnerable to habitat loss, and large-bodied species are more prone to overharvesting and overpredation by invasive predators due to their slow reproductive rates (Fisher & Owens, 2004). Just as a certain trait or a set of traits makes animals more prone to being threatened by poaching or introduced predators, there are likely to be traits that contribute to a species' sensitivity to fire (Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, MacGregor, Russell-Smith, et al., 2010;Nimmo et al., 2021).
Shifting fire regimes may exacerbate the impacts of other threatening processes. For example, in species that are prone to predation and require ground cover vegetation for shelter, more frequent or more severe fires result in more exposure to predators (Doherty et al., 2015;Hradsky, 2020). Predators may benefit from fire in the short term due to an increase in hunting success in the simplified landscape , although groundcover can recover to prefire conditions within 2 years of fire (McKenna et al., 2018). Conversely, ambush predators may be disadvantaged by fire; therefore, their prey species may respond positively to fire in the landscape (Geary et al., 2020). Geary et al. (2020) used a meta-analysis to test whether they could predict the response of predators (mammals, reptiles, and birds) to fire, based on a set of species and fire traits, and found that none of the traits they tested were predictive. Conservation practitioners and government agencies must base decisions about postfire funding and management on their assessment of species vulnerability to fire based on limited species-specific evidence . Following severe wildfires, intervention may be required to assist wildlife populations, for example, a huge amount of effort was invested following the 2019 and 2020 fires in Australia to locate, triage, and treat injured koalas (Phascolarctos cinereus) (Dunstan et al., 2021;Parrott et al., 2021). Resources are limited, so understanding which species are likely to be most affected by fire can help managers prioritize where resources are allocated to have the greatest return on investment. We assessed mammal species vulnerability to fire according to their ecological and biological traits.
We investigated the influence of species traits, study-, fire-, or habitat-specific variables, and the threats a species faces on the abundance or site occupancy of mammals in response to fire. We carried out a global meta-analysis on mammal population response to fire to identify which traits and habitat and study variables are useful for predicting how mammal populations respond to fire. Though data may not discriminate between changes driven by death or birth and those driven by immigration or emigration, they provide an insight into site-level population changes. We sought to help land managers predict how mammal populations will respond to prescribed fire and provide a tool to help prioritize management in the event of wildfires.

Literature search and data extraction
We used Web of Science to search the literature for studies published from 2018 to 2021 relating to mammal responses to fire (Appendix S1). Year of publication was limited to 2018-2021 due to the large number of studies returned by our search, resulting in significant time commitment screening papers and extracting data. Studies from outside this date range were added at later stages through unstructured, targeted searching or taken from references of studies returned by the search.
This search returned 2437 results, and we screened these in 3 steps. First, the title and abstracts of all results were checked for potential suitability. Second, studies in which mammal responses were observed at burned versus unburned sites (BU), at sites with different time since fire, and at sites with differing fire severity and studies with before-after (BA) or before-after, control-impact (BACI) designs were selected. Response variables could include changes in abundance indices, occupancy, or site use. Thirds, we added papers returned from similar previous Web of Science searches (carried out by Diana Olwyn Fisher (DOF) in 2018-2019) and carried out a targeted search, across all years, for studies pertaining to taxa underrepresented across both the systematic searches. We also extracted data from Geary et al.'s (2019) data set. Results from remaining studies were checked to extract effect sizes where possible. To include responses in the meta-analysis, we needed to extract an effect size and standard deviation related to population abundance or occupancy for BACI, BA, or BU study designs (Appendices S1 & S2).
We extracted sufficient data to calculate 161 effect sizes from 35 studies, representing 115 species. Studies came from 11 countries, dominated by Australia (13 studies, 51 effect sizes), South Africa (1, 32), and the United States (9, 31), and covered 40 families (Cricetidae 15 species, 17 effect sizes; Muridae 12,20;Canidae 8,26;and Felidae 8,14). Study-specific fire variables are summarized in Table 1. We calculated effect sizes as Hedges' g (Hedges, 1981), which compares control and impact groups based on mean and standard deviation reported in studies or measured from plots. Several studies were not included due to not reporting results in a way that means or standard deviations could be calculated. We subtracted means and standard deviations of before measurements from after measurements for BACI design studies to obtain control-impact values (Eales et al., 2018). This typically resulted in lower standard deviations for effect sizes from BACI design studies. We used a linear mixed-effects model to predict effect sizes with 95% confidence intervals (CIs) for each species to obtain a single value for species with multiple reported effect sizes. We obtained a phylogenetic tree for the species included in this meta-analysis with the phylogeny subsets tool accessed through www.vertlife.org as described by Upham et al. (2019).

Trait data
We compiled species trait data to model whether any traits predicted mammal species response to fire. We sourced data from the PanTHERIA database or directly from literature where possible. Information was inferred from closely related species for those with little or no data available or from reputable internet sources (e.g., International Union for Conservation of Nature [IUCN] Red List). We also recorded study-specific variables representing fire and study site characteristics. These were mostly reported in the relevant papers; if not, we used an internet search for this information. Traits, variables, and predicted associations with effect size are summarized in Table 2.

Modeling
We used phylogenetic generalized least squares regression and nlme (Pinheiro et al., 2022) in R 4.1.1 (R Core Team, 2021) to model effect size as a function of species traits and study variables. We imported the phylogeny with ape (Paradis & Schliep, 2019) and incorporated the phylogeny through a phylogenetic correlation function. Allometric data were log-transformed prior to modeling. We first defined the null model with only the phylogeny incorporated with Grafen's correlation structure (Grafen, 1989). With the Grafen method, a height is designated for each node of the phylogeny based on the subtree's number of leaves. Heights are scaled relative to root height, and branch lengths are calculated as the difference between heights of lower and upper nodes (Grafen, 1989). We then defined all single-variable models and ranked these by Akaike information criterion adjusted for small sample size (AICc) to determine which variables may be important. We used this insight, along with informed predictions about which variables may interact with one another (e.g., an interaction term with fire type was fitted for many variables) to build a set of candidate models.

Publication bias
We constructed a funnel plot of effect size and variance with metafor (Viechtbauer, 2010) in R to visually assess the data for asymmetry, which could suggest publication bias toward statistically significant effect sizes. We also carried out a Kendall's rank correlation test on the full data set with metafor to assess correlation between effect size and variance; a significant correlation indicated publication bias.

Meta-analysis and modeling
Nine out of 115 species (7.83%) returned a statistically significant effect size in response to fire ( Figure 1). The species that returned significant negative effect sizes were the beech marten (Martes foina), cerrado climbing mouse, (Rhipidomys macrurus), Australian swamp rat (Rattus lutreolus), western chestnut mouse (Pseudomys nanus), and leopard (Panthera pardus). The red-legged pademelon (Thylogale stigmatica), long-nosed potoroo (Potorous tridactylus), red veld rat (Aethomys chrysophilus), and white-tailed mongoose (Ichneumia albicauda) had significant positive effect sizes. The remaining species have 95% CIs overlapping zero. There was almost an even split between positive (4) and negative (5) responses in the species with significant responses. Across all effect sizes, there were 60 (52.2%) negative and 55 (47.8%) positive responses. Seven of 35 included studies suggested that direct mortality was responsible for any declines, and these were generally in high-severity wildfires. Most studies proposed that indirect or behavioral responses were behind changes through indirect mortality (e.g., prey species became easier for predators to locate), emigration from a burned area (such that individuals were not detected postfire), or immigration into burned areas (for easier hunting or fresh vegetation growth).
The top-ranked phylogenetic generalized least squares regression model included a negative impact of number of offspring per year and the species' geographic range, regardless of fire type, a positive impact of burrowing in prescribed fires, a positive impact of the average fire interval for wildfires, a negative impact of wildfire (magnitude significantly greater for species with fire listed as a threat), and a positive impact of fire as a

TABLE 2
Species traits and other variables included in phylogenetic least squares regression models of mammal response to fire as a function of these traits and variables and predicted association of variables with effect size of population response to fire Variable

Unit or category Predicted association
Direct impacts Female mass Grams Positive: Small-bodied species are likely to have greater difficulty fleeing a fire unless they can fly (i.e., bats). Average body mass of small mammals in Zambia decreased as time since fire increased, suggesting that smaller species had higher mortality than large species or emigrated from the site initially due to fire before gradually recolonizing (Namukonde et al., 2017). Small-bodied species can access some refuges (e.g., cracks in soil) more easily; this may counteract the expected effect of body mass.
Mobility 1, immobile, high fidelity to very small home range (<∼1 km 2 ); 2, small to medium home range (1-200 km 2 , relative to body size); 3, large home range with frequent forays (>100 km 2 , relative to body size); 4, nomadic or strong flyer Negative for 1 and 2, positive for 3 and 4: Mobile species generally have high survival during fire and can access resources, such as unburned patches, in a sea of burned habitat that smaller less mobile species cannot and take advantage of benefits of burned sites (Berry et al., 2015;Franklin et al., 2021;Peres et al., 2003).
Forage mostly in canopy 1, yes; 0, no Negative in severe fires, neutral otherwise: Radford et al. (2020) found the arboreal Australian brush-tailed rabbit-rat (Conilurus penicillatus) responded positively to low severity prescribed fire, whereas Chia et al. (2015) found that high-severity wildfire had a negative impact on several arboreal mammal species, but moderate severity fire had little impact. Canopy foraging may interact with time since fire because negative effect sizes may be recorded soon after moderate severity fires, but these effects are not expected to persist.
Use of hollows? 1, yes; 0, no Negative in severe fires, neutral otherwise: Species that shelter in hollows will avoid deleterious impacts of low-or moderate-severity fire, but hollows may become ecological traps in a high-severity fire.
Use of burrows? in Australia only persisted following a severe fire because of its ability to shelter in rocky crevices.

Indirect impacts
Young per year Average number of litters per year multiplied by average litter size Positive: The number of young per year will have a positive association with resilience because this trait is associated with a fast life history, which is associated with positive responses to some disturbances (Suraci et al., 2021).
Dietary trophic level 1, >50% grass or browse; 2, vegetation, seeds, flowers, invertebrate prey; 3, fruit, nectar, prey; 4, >50% invertebrate prey; 5, >50% vertebrate prey Positive for 1, 2, and 5, negative for 3, neutral for 4: Generalists (either through a regular food source or food switching [e.g., Radford, 2012]) and grazers (due to fast grass regrowth [e.g., Gigliotti et al., 2022;McHugh et al., 2020;Prada, 2001]) are likely to have food sources available to them soon after fire, but frugivores will have a scarcity of food sources for some time after fire (Barlow & Peres, 2006). Lack of shelter for prey species increases predators' hunting success and has a positive effect on their response to fire, at least in the short term (e.g., Hradsky et al., 2017;McGregor et al., 2015). Sociality 0, solitary; 1, pairs; 2, small (3-10 individuals) or scattered social groups across a territory (e.g., wolves); 3, large social groups, communal shelters (e.g., bats) Negative for 3, neutral otherwise: Highly social species require large tracts of habitat to support large groups, and partial mortality of a group may influence the survivors' ability to survive if dominant individuals are killed or group size is particularly important to their survival. Breeding congregation is part of the IUCN Red List criteria because it concentrates risk to a large proportion of populations at one time (IUCN, 2001).
(Continues)  ) Positive: Habitat generalization, represented here by geographic range, has a positive association with survival following fire. Geographic range is an indicator of habitat generalization, which is comparable across species. The larger a species' range, the more likely it is a habitat generalist (Eeley & Foley, 1999;Huang et al., 2021;Slatyer et al., 2013), a trait predicted to increase survival following disturbance, such as fire, compared with specialists (Fisher & Owens, 2004).
Forage mostly in leaf litter 1, yes; 0, no Negative: Food source is likely to have been lost and litter takes some time to accumulate and become shaded by canopy, so prey abundance can recover. Coleman and Rieske (2006) found that leaf litter arthropod abundance was severely affected by fire and recovery was slow.
Digs for food 1, yes; 0, no Positive: Food sources are likely to persist, except in some cases after fires of very high severity. For example, the proportion of hypogeous fungi consumed by northern bettongs did not change following a fire, and it was consistently the most prevalent food source for bettongs (Vernes et al., 2001).

Threats
Conservation status IUCN Red List status (IUCN, 2021) Neutral: Whether a species is classified as least concern or critically endangered does not inherently predispose it to being vulnerable to fire.
Fire recorded as a threat?
1, yes; 0, no Negative: The inclusion of fire as a threat in IUCN Red List species accounts is typically based on a perception that a species is sensitive to fire based on nonexplicit assumptions, although detailed evidence is often lacking. For example, fire is listed as a threat to the puma (Puma concolor), despite the trend of fire impacts on the species being unknown or unrecorded, because of the potential contribution of particular fire regimes to habitat degradation (Nielsen et al., 2015). We predict this effect will be greater in studies of wildfires than prescribed fires and where fire severity is high.

Study variables Fire type
Prescribed or wildfire Negative for wildfire, neutral for prescribed: Wildfires are typically of a higher severity and larger extent than prescribed fires and cause high plant mortality, changing habitat structure (e.g., Barton, 2002). Conversely, prescribed fire usually burns smaller areas in a patchy manner, leaving behind much more unburned habitat (e.g., Legge et al., 2011).
Fire severity 1, low; 2, moderate; 3, high Negative: The higher the fire severity, the greater the negative impacts will be on mammals. For example, an increase in the frequency of extensive, high severity fires following the cessation of cool, patchy burns devastated rufous hare-wallaby (Lagorchestes hirsutus) populations (Bolton & Latz, 1978), possibly by increasing their susceptibility to feral predators (Short & Turner, 1994).

Average fire interval
Years Negative: Species in habitat with a longer fire return interval will be more negatively affected by fire than those in habitats with short fire return intervals (Nimmo et al., 2021), though there may be an interaction with fire severity because the benefits of a short fire return interval for species in fire-adapted habitats may be lost if fire is too severe and insufficient vegetation remains (e.g., McDonald et al., 2016). In systems where fire occurs at short intervals, species are more likely to possess adaptations that make them fire tolerant, whereas those in habitat where fire is rare, potentially with an average return interval greater than the life span of many species, will be fire sensitive (e.g., Blakey et al., 2019). (Continues)

Unit or category Predicted association
Maximum time since fire Years between earliest fire and latest data collection Positive: The longer the time since fire, the more opportunity habitat has had to recover and animals to recolonize (Radford et al., 2015). Hale et al. (2016) investigated the influence of fire history and climate on mammal abundance. They found each of the 8 Australian mammal species they assessed is negatively associated with recent fire and that older vegetation age classes are vital to mammal persistence (Hale et al., 2016). The effect of time since fire is weaker following a prescribed fire than after a wildfire and following low-severity fires, due to the presence of unburned habitat.

Mean annual rainfall
Millimeters Positive: High rainfall promotes postfire growth in some plants (e.g., Thomson et al., 2021;van Blerk et al., 2021), providing food and shelter. Postfire succession is slower in areas of lower rainfall, negatively affecting mammal resilience to fire (Prieto et al., 2009). High rainfall also encourages faster decomposition of fuel biomass, minimizing the amount available to burn (Palmero-Iniesta et al., 2017).
Fire in breeding season of species 1, yes; 0, no Negative: Fire during breeding season may result in failed breeding attempts and has a negative impact on breeding success in wild birds and amphibians and captive primates (Muñoz et al., 2019;Murphy et al., 2010;Powell, 2008;Rosche, 2018;Willson et al., 2021).

Latitude
Absolute decimal degrees at study site Negative: Those species at lower latitudes are better adapted to fire than those at higher latitudes due to the high fire frequency in widespread low-latitude vegetation types, such as savannas and tropical woodlands (Yang et al., 2014). The dominance of fire-sensitive rainforest at very low latitudes may counteract this effect to some extent.

Longitude
Absolute decimal degrees at study site Neutral threat for prescribed fires, but negative for wildfires (Table 3; Figures 2-4). There were 8 other models with ΔAICc < 2, suggesting no clear best model, however. The covariates included were generally consistent with the top-ranked model, with different interaction terms, and with nonsignificant impacts of rainfall and use of hollows present in some models (Table 4). Nonetheless, alternative interpretations of the best-supported models are possible. We compared the top-ranked model with and without phylogenetic signal with an ANOVA, and it performed significantly better with phylogenetic signal included (p < 0.01).

Publication bias
We found no evidence for publication bias toward significant effect sizes. The funnel plot showed a mostly symmetrical spread of effect sizes and variance, and our rank correlation test supported this (Kendall's tau = −0.014, p = 0.796) (Appendix S3).

DISCUSSION
Our results confirm that some life-history traits, ecological traits, and fire characteristics can predict mammal response to fire, although most species (>92%) in our analyses returned nonsignificant responses to fire. Mammal populations are declining worldwide for myriad reasons, and fire is a common reported threat to mammals. If managers can predict impacts of fire on species, they can tailor fire management to best suit species of interest. A statistically significant negative effect of the average number of offspring per year was included in the top-ranked model: mammals with faster life histories were disadvantaged regardless of fire type. The highest average number of young per year in a species with a significant positive response was 3.1, whereas 4 of the 5 negatively affected species had more offspring per year than this on average. Reproductive rate reflects life-history strategy in mammals because it trades off strongly with life span, generation time, growth, and survival . Although mammals with slow life histories are expected to show low resilience to chronic threats, such as hunting, persecution, and invasive predators, because they cannot compensate for ongoing mortality through a high reproductive rate (Clements et al., 2017;Fisher & Owens, 2004), slow-breeding species with long life spans may have an advantage when breeding success is low in the season after a fire. For example, European birds with short life spans coped poorly with variable annual conditions, which decreased food and breeding success in some years (Jiguet et al., 2007). A species' geographic range showed a significant, negative association with response to fire for both prescribed and wildfires. Species with a small geographic range tend to be habitat specialists and more sensitive to changes or loss of habitat (Fisher & Owens, 2004); however, our model suggests these species responded more positively to fire than widespread species. Similarly, species with fire as a threat responded significantly positively to prescribed fire, compared with species that are not considered threatened by fire. Many published studies of the effects of cool fires on mammals (6 out of 14 studies of prescribed fire included in this analysis) involve burning for ecological goals or to benefit focal species (ecological fire), where managers may take the distribution and threat status of species into account (Radford et al., 2020). This representation of strategic fire regimes, potentially tailoring fire to the needs of restricted-range and fire-threatened species, may explain these relationships.
Wildfires had a significant negative impact relative to prescribed fires; statistically significant interactions with fire return interval (positive) and whether fire is listed as a threat to a species (negative); and a nonsignificant negative interaction with burrowing. The benefit of prescribed fire (relative to wildfire) to mammals was consistent with recent research (Legge et al., 2019;Radford et al., 2020), although this pattern is not always observed (e.g., Pastro et al., 2011). Prescribed burning is implemented with strategic timing and tends to be of lower severity (cooler) and patchier than wildfire (Legge et al., 2011). Although fire-threatened species responded positively to prescribed fire compared with those not threatened by fire, the inverse is true for wildfires. This finding supports the designation of fire as a threat to these species by experts.
Average fire return interval had a significant, positive association with mammal response to wildfires and no impact in prescribed burns. This finding indicates a negative impact of wildfires at short return intervals, whereas prescribed fire is likely carried out at intervals that land managers identify as suitable for the persistence of at least some species in an area. Burrowing had a significant positive effect on response to prescribed fires and no significant effect in wildfires. The lower severity and increased patchiness of prescribed fires likely allow both the immediate survival of burrowing species and ongoing food or habitat availability to promote resilience in a postfire landscape. High-severity wildfires are unlikely to kill most burrowing individuals (O'Brien et al., 2006), although a lack of resources may limit their resilience after fire.
Body mass was not predictive, consistent with Geary et al.'s (2020) findings. Smaller species may survive fire better than predicted by their general life-history strategy by hiding in small cracks or crevices and may have recolonized burned areas at a faster rate than large species (Sensenig et al., 2010). Sensenig et al. (2010) found a negative correlation between body size and use of burned areas in a suite of African herbivores, and this relationship is possibly driven by resource requirements; there may not be enough plant biomass after fire to support largebodied species. Large-bodied animals often have large home ranges and are therefore more likely to have a home range that incorporates unburned parts of the landscape, which they may preferentially use after fire (Tucker et al., 2014). Mobility also had no association with response to fire. Although mobile species may be better able to avoid mortality, they are also likely to be able to access, and may preferentially use, unburned parts of the landscape after fire for foraging or shelter, resulting in a decrease in abundance or occupancy at burned sites (Berry et al., 2015;Franklin et al., 2021).
Dietary trophic level did not influence mammal resilience to fire. We predicted grazers would respond more positively than species whose food source takes some time to recover, such as frugivores. The strong negative effect size of R. lutreolus, whereas most grazers responded slightly positively to fire, may FIGURE 1 Mean effect size and 95% confidence interval for mammal species response to fire (red, species with statistically significant effect size; numbers in parentheses, number of effect sizes included for each species). have dampened the modeled response of grazing in our analyses. There was a lack of true frugivores in this study, however, meaning our data set was not suitable to properly assess fire impacts on them. Most fruit-eating species in this data set had flexible, omnivorous diets, so fire may not alter their survival through a mechanism of fruit shortage (Roberts et al., 2015;Zwolak et al., 2012).
We predicted both digging for food (positive) and foraging in leaf litter (negative) would influence species' resilience to fire, although neither prediction was supported. Both traits were poorly represented in the analyzed data set; we had just 15 effect sizes for species that forage in leaf litter and 16 for species that dig for food out of 161 total effect sizes. The average effect size for species that dig for food was positive, and it was negative for those that forage in the leaf litter. However, there was too much uncertainty around these values to make any inferences.
The response of a population to fire is not always negative; many species respond positively to fire. The family Macropodidae (kangaroos, wallabies, and allies), for example, is represented by 5 species in our analyses, all of which showed a positive response to fire, including 1 significant effect size (T. stigmatica). These almost certainly represent behavioral responses to fire, rather than changes in population size. Macropods are known to be attracted to fresh grass growth in recently burned areas (Chard et al., 2022;Styger et al., 2011). Indigenous Australians burned patches of vegetation to encourage kangaroos and wallabies to feed in the area and increase hunting success (Yibarbuk et al., 2001). Macropods are highly mobile and can avoid a fire front, providing it is not of too high severity. Some have been documented doubling back to burnt patches behind a fire front to safely wait out the fire (Garvey et al., 2010). We can assess differences between other closely related species in the forest plot ( Figure 1). Most groups showed consistent responses between species; all bats returned effect sizes very close to zero, for example. Mustelids are 1 exception; M. foina showed a strong negative response to fire. This species was represented by only 1 study that compared sites burned in a high-severity wildfire to unburned control sites. Martes foina was not detected in the burned sites (Birtsas et al., 2012).
The 95% CIs around effect sizes for most species overlapped zero. The traits in the top-ranked models may predict the direction of a species response to fire, but most mammal species in our study appeared minimally affected by fire. Managers are likely to use caution when threatened species or those considered fire sensitive are present, as would be recommended, but many mammal species seem somewhat resilient to fire (Andersen, 2021; Radford et al., 2020). For every fire, there will be species that benefit and others that are disadvantaged, despite this general resilience, and anticipating which species fall into which category can help to guide conservation actions (Andersen, 2021; Radford et al., 2020). Optimal fire regimes have been proposed for some fauna species (e.g., Department of Environment & Heritage Protection, 2017), although these are rarely implemented.
In a recent meta-analysis, Geary et al. (2020) found no consistent trends in predators' responses to fire and no traits that FIGURE 2 Model-predicted effect sizes with 95% confidence intervals for the observed range of natural-log-transformed fire return intervals for all combinations of species for which fire is (1) or is not (0) considered a threat and those that do (1) or do not (0) use burrows for wildfires and prescribed fire.

FIGURE 3
Model-predicted effect sizes with 95% confidence intervals for the observed range of natural-log-transformed species geographic ranges for all combinations of species for which fire is (1) or is not (0) considered a threat and those that do (1) or do not (0) use burrows for wildfires and prescribed fire.

FIGURE 4
Model-predicted effect sizes with 95% confidence intervals for the observed range of natural-log-transformed offspring per year for all combinations of species for which fire is (1) or is not (0) considered a threat and those that do (1) or do not (0) use burrows for wildfires and prescribed fire.
could predict the effect of fire on species. Geary et al. (2020) analyzed effect sizes in predatory reptiles and birds, as well as mammals, and found most species exhibited effect sizes with confidence intervals overlapping zero (Geary et al., 2020). Our results suggest that mammals, across guilds, are likely to be more consistent in the traits that affect their response to fire than all predators (across mammals, birds, and reptiles). Geary et al. (2020) assessed some of the same traits we included, although none of the traits significantly affected effect size according to our model. The lack of effect of mass, dietary trophic level, and IUCN Red List status, for example, is consistent between our study and Geary et al.'s (2020) meta-analysis.
Species with multiple reported effect sizes from different studies often returned variable effect sizes. For example, 4 effect sizes were included for the gray fox (Urocyon cinereoargenteus), 2 positive and 2 negative. All 4 of these studies were of wildfires with similar fire severity, time since fire, and habitat. One of the negative effect sizes came from a study that included fire during the breeding season of U. cinereoargenteus, but otherwise there were no clear differences that could explain the contrasting findings. Conversely, 5 effect sizes were extracted from studies for the coyote (Canis latrans), all of which were negative. The inconsistent responses from U. cinereoargenteus to fire, along with other species with multiple effect sizes, emphasize the difficulty in predicting response across different studies and that there may be subtle study-or fire-specific variables that have an influence on the effect size for some species. Similar work has been conducted for plants. Traits such as the ability to resprout following fire, thick bark to insulate a tree's living tissue from most of the heat of a fire, and a seedbank triggered to germinate by the heat of fire are found in many fire-adapted plants (Keeley et al., 2011;Pausas & Bradstock, 2007;Russell-Smith et al., 2012). However, Vivian et al. (2010) showed that different populations of the same plant species can respond differently to fire and that the potential for intraspecific variation must be considered when predicting response to fire. Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, MacGregor, Russell-Smith, et al. (2010) identified a dearth of knowledge around the effects of different components of fire regimes on biodiversity. One of the key knowledge gaps is an understanding of species responses to fire regimes, and many species are too rare, or have too low detectability, to accurately observe and quantify their response to fire (Driscoll, Lindenmayer, Bennett, Bode, Bradstock, Cary, Clarke, Dexter, Fensham, Friend, Gill, James, Kay, Keith, MacGregor, Russell-Smith, et al., 2010). In the absence of empirical data for many species, modeling takes on increased relevance, despite the inherent uncertainty.
Our study had limitations. Time since fire was included as a covariate in candidate models, and although it was not included in any of the supported models, species responses are likely to appear very different immediately after fire than 5 years after fire, for example. Most included studies (73.1%) (Table 1) showed responses within 3 years of fire, so our analyses were most relevant to short-term responses. The lack of effect sizes from species with some traits (e.g., frugivores or leaf litter foragers) also introduces difficulty in applying this approach to these guilds. Fire extent or patchiness could not be assessed explicitly because it is rarely reported. We relied on fire severity as a covariate to show impacts relating to fire extent (although severity does not always correlate with extent). Despite these limitations, method is an additional tool with which conservation practitioners can concentrate their efforts.
The variables that we found to be important in predicting mammal response to fire will help managers and researchers predict how a fire will affect a species. Our modeling approach can predict an effect size for a population where relevant species traits and site and fire characteristics are known and can be used to refine simulation models, for example, to predict how a species may respond to different fire regimes. Where practitioners expect wildfire to be a threat to local mammal species, they can prioritize prescribed fire. In the event of a wildfire, conservation efforts can prioritize those species predicted to be most affected, that is, those that are considered threatened by fire with high annual reproductive output. Modeling will not produce data as informative as empirical, experimental studies, but we provide an alternative approach that can be implemented quickly, along with methods, such as expert elicitation. These techniques can provide predictions of the responses of a whole suite of species, allowing practitioners to focus their efforts and resources efficiently.

C.A.P. was supported by an Australian Government Research
Training Program scholarship. This project was supported by the Australian Government's National Environmental Science Program (NESP) through the Threatened Species Recovery Hub.
Open access publishing facilitated by The University of Queensland, as part of the Wiley -The University of Queensland agreement via the Council of Australian University Librarians.