Experimentally testing animal responses to prescribed fire size and severity

Deserts are often highly biodiverse and provide important habitats for many threatened species. Fire is a dominant disturbance in deserts, and prescribed burning is increasingly being used by conservation managers and Indigenous peoples to mitigate the damaging effects of climate change, invasive plants, and land‐use change. The size, severity, and patchiness of fires can affect how animals respond to fire. However, there are almost no studies examining such burn characteristics in desert environments, which precludes the use of such information in conservation planning. Using a before‐after control‐impact approach with 20 sampling sites, we studied the outcomes of 10 prescribed burns of varying size (5–267 ha), severity, and patchiness to identify which variables best predicted changes in small mammal and reptile species richness and abundance. Three of the 13 species showed a clear response to fire. Captures increased for 2 species (1 mammal, 1 reptile) and decreased for 1 species (a reptile) as the proportional area burned around traps increased. Two other mammal species showed weaker positive responses to fire. Total burn size and burn patchiness were not influential predictors for any species. Changes in capture rates occurred only at sites with the largest and most severe burns. No fire‐related changes in capture rates were observed where fires were small and very patchy. Our results suggest that there may be thresholds of fire size or fire severity that trigger responses to fire, which has consequences for management programs underpinned by the patch mosaic burning paradigm. The prescribed burns we studied, which are typical in scale and intensity across many desert regions, facilitated the presence of some taxa and are unlikely to have widespread or persistent negative impacts on small mammal or reptile communities in this ecosystem provided that long unburned habitat harboring threatened species is protected.


INTRODUCTION
Fire is a natural disturbance in many of the world's ecosystems (Bowman et al., 2009;He et al., 2019).It alters habitat structure and resource availability, ultimately shaping the occurrence and distribution of plants, animals, and many other organisms (Bond et al., 2004;González et al., 2022).Some species are most 2019).Such relationships typically emerge due to the coevolution of species with historical fire regimes (Pausas & Parr, 2018).
However, in recent centuries, fire regimes across the globe have become increasingly altered from historical conditions, as evidenced by increases and decreases in fire frequency, size, and severity (Bowman et al., 2020).These changes have been brought about by displacement of Indigenous peoples, land clearing, altered land use, proliferation of invasive plants, climate change, and the use of prescribed burning (Fletcher et al., 2021;Halofsky et al., 2020).A primary focus of prescribed burning is to protect human lives, infrastructure, and other assets by reducing fuel loads (Penman et al., 2011).Prescribed burning is also commonly used for the promotion of ecological values and maintenance of ecological processes (e.g., for promoting pyrodiversity [Parr & Andersen, 2006]).Other uses include enhancing agricultural and silvicultural production, flushing game animals for hunting, and cultural burning (Bliege Bird et al., 2013;Duff et al., 2020).
Studies from primarily temperate and Mediterranean ecosystems show that burn characteristics, such as fire size, severity, and patchiness, often affect how animal populations and communities respond to fire (Cohen et al., 2019;Hale et al., 2022;Shaw et al., 2021;Sommers & Flannigan, 2022;Wan et al., 2014;Watson et al., 2012).For instance, cavity nesting birds responded positively to severe burns in dry forests in the United States (Saab et al., 2022).In Australia, reptile species richness after fire in woodlands increased as the amount of unburned vegetation increased and some species responded positively to burn patchiness (Senior et al., 2023).Wood mouse (Apodemus sylvaticus) abundance in woodlands and shrublands in Spain was initially correlated negatively with distance from burn perimeter postfire and later with distance from internal unburned patches (Puig-Gironès et al., 2018).
Hot deserts cover 19% of Earth's land surface (Dinerstein et al., 2017), and fire is a dominant force for biotic change in many deserts, including in regions that have a long history of burning by Indigenous peoples for cultural purposes (Bird et al., 2016).Proposed or current objectives of prescribed burning in deserts include reducing shrub encroachment (Daryanto et al., 2019), promoting pyrodiversity (Greenwood et al., 2021), protecting and maintaining cultural sites (Moorcroft et al., 2012), maintaining foraging and hunting resources (Tarlka Matuwa Piarku Aboriginal Corporation, 2015;Yununijarra Aboriginal Corporation, 2012), creating firebreaks to protect long unburned habitat (Mcdonald et al., 2012), and improving the productivity of food plants for threatened wildlife (Southgate et al., 2007).However, relative to temperate and Mediterranean ecosystems, there is limited knowledge regarding how burn characteristics influence animal responses to fire in deserts.Almost all previous studies have focused on simple contrasts of burned and unburned areas (including pre-and postfire) and did not consider characteristics, such as fire size, severity, or patchiness (Day et al., 2019;Killgore et al., 2009;Kirkpatrick et al., 2002;Letnic, 2003;Masters, 1996;Pastro et al., 2011;Sharp Bowman et al., 2017).Although these studies reveal the relative impacts of burning at a coarse level, they provide limited information about the finer-scale effects of prescribed fire (but see Hulton VanTassel et al., 2015).Such information is vital for determining the spatial characteristics of prescribed burning targets.More broadly, understanding the role of prescribed burning in deserts is important because these ecosystems are typified by extreme temporal and spatial variation in resource availability that ultimately results in boom-bust dynamics of consumer populations (Greenville et al., 2016;Holmgren et al., 2006).Desert fauna are adapted to these dynamic environments, where wildfires can commonly exceed 100,000 ha (Verhoeven et al., 2020;Wright et al., 2021).This means these species may respond to prescribed burning somewhat differently from species in less dynamic mesic systems, although this remains to be determined.
There is concern that climate change will lead to more expansive and frequent wildfires in some deserts by causing higher temperatures and increased frequencies of droughts interspersed with extreme rainfall events that will increase ground fuel loads (Chen & Wang, 2022;van Etten et al., 2022;Verhoeven et al., 2020).Invasive grasses and shrubs, which are common in some arid areas, can also fuel more frequent and extensive fires and create a positive feedback loop that favors invasive plants and disadvantages native plants and animals (e.g., McDonald & McPherson, 2013;Setterfield et al., 2010;Underwood et al., 2019).Such increases in fire size and frequency pose a threat to animal species that rely on long unburned habitat (Santos et al., 2022;Von Takach et al., 2022).As such, there is concern among conservation managers, Indigenous peoples, and other stakeholders about the impacts of altered fire regimes on desert ecosystems, and, consequently, there is a growing interest in how prescribed burning can be most effectively implemented as a management tool (Aslan et al., 2021;Ruscalleda-Alvarez et al., 2023).However, there is a substantial gap between this growing interest and the state of knowledge regarding how different fires may affect animal populations.This information is crucial for guiding practitioners wishing to undertake prescribed burns to meet ecological objectives.
To this end, we experimentally tested how prescribed burning affects small mammal and reptile populations in a desert ecosystem.Using a before-after control-impact (BACI) approach with 20 sites in total, we studied 10 burns varying in size, patchiness, and severity to assess whether changes in species richness and relative abundance could be predicted by burn status (burned or unburned), burn size (a proxy for burn severity in this study), burn patchiness, and habitat structure.We integrated our findings with those from other desert studies to advance the knowledge base of prescribed burning in desert ecosystems.

Study area
Our study system was the spinifex (Triodia spp.) grasslands that cover much of arid and semiarid Australia.Spinifex is the dominant functional component of the vegetation for ∼37% of the continent (Smyth et al., 2012).We conducted this study at 2  1).Matuwa is part of a national park (formerly an Indigenous Protected Area) comanaged by Martu People and the Western Australian Department of Biodiversity, Conservation and Attractions.Jundee is a pastoral lease 40 km west of Matuwa and is under the management of Northern Star Resources, who operate an underground gold mine in the eastern part of the property (<5% of the property's area).A small number of cattle are grazed at Jundee, but they do not frequent the spinifex grasslands.There are no cattle at Matuwa.Feral camels (Camelus dromedarius) are present at both properties.
The climate of the study area is arid, with mean maximum temperatures of 30 • C in summer and mean annual rainfall of ∼260 mm, most of which falls from December to May (Bureau of Meteorology, 2023).The 2 dominant vegetation types are spinifex grasslands on sandplains and mulga (Acacia aneura) woodlands on stony plains and clay-dominated soils.We focused on spinifex grasslands, where spinifex grows as hummocks (mounds), which eventually mature into large rings-sometimes comprised of multiple plants if left unburned for decades (Figure 1).Spinifex is considered a foundation species because it is the dominant vegetation cover of the study area and broader region, and it provides habitat for a wide range of plant and, especially, animal species (Bell et al., 2021;Dickman & Pavey, 2023;Downey & Dickman, 1993;Verdon et al., 2020).
Matuwa has a predator control program that primarily involves aerial baiting of the property once annually with Eradicat baits that contain 1080 poison (Lohr & Algar, 2020).The control program primarily targets feral cats (Felis catus), although dingoes (Canis familiaris dingo) and introduced red foxes (Vulpes vulpes) are also killed by the baits.Jundee does not have a property-wide predator control program, although ad hoc trapping and removal of cats is conducted around the mine infrastructure in the eastern part of the property, 17-36 km from our study sites.
Our original intention was to assess the effects of fire in the presence and absence of intense cat control due to evidence that very recently burned areas can be preferred hunting grounds of feral cats (Doherty et al., 2023;McGregor et al., 2016).However, 2 factors prevented us from achieving this.First, cat activity was very low during the study period due to multiple years of below average rainfall prior to and during the study, which depletes the prey base for predators in arid Australia and leads to reduced predator abundance (Letnic & Dickman, 2010).Second, although we intended for the burning to be similar between the 2 properties, the burns at Matuwa were of lower severity and were smaller and patchier than those at Jundee because a more cautious burning strategy was ultimately implemented at Matuwa.Nonetheless, this unplanned outcome gave us the unique opportunity to experimentally test the effects of burn size and severity on small mammals and reptiles in a desert environment.Although the cat control program also differed between the 2 properties, cat activity was similarly low at both properties throughout the study, except for the final 2 (of 5) trapping sessions when cat activity was higher at Jundee than Matuwa but still relatively low.On average, cats were detected on each camera on fewer than 1 in every 100 days across both properties (Appendix S1).

Study design
We established 10 monitoring sites at each property.Sites within properties were separated by 1.7−9.4km each (mean = 4 km).Sites were chosen that had mature spinifex (13 to over 20 years since the last fire), which was determined using fire scar mapping and on-ground verification.We ensured that the sites at the 2 properties had similar vegetation structure and composition (mature spinifex-dominated grasslands), sandy soil texture, and flat terrain.Each site had a 1-ha pitfall trapping grid situated 200 m off track and comprised of 16 traps in a 4×4 grid at 30-m spacing.Traps were PVC pipes (60 cm deep by 15 cm wide) and each trap was intersected by an aluminum flywire drift fence (6 m long by 25 cm high) designed to intercept animals moving on the ground surface and guide them toward the traps.Traps also had a flywire floor to prevent animals from escaping by digging.
We used a BACI design, with 5 sites at each property subjected to burning, whereas the other 5 sites at each property remained unburned (Figure 1).Choice of sites for burning was informed by logistical and cultural considerations, and we interspersed burned and unburned sites as much as possible.Burning was initially attempted in June 2021, but cool conditions and low wind speed hampered ignition.As a result, only small areas were burned at all treatment sites in June and full burning was undertaken in August 2021 when conditions were more appropriate.The size of the burned area at each trapping site ranged from 4.9 to 12.8 ha at Matuwa (mean = 9.1 ha) and 17.2 to 267.3 ha at Jundee (mean = 112.7 ha).The burns at each of the 10 treatment sites were independent of one another and did not join up (Figure 2).We refer to the burns at Jundee as high severity because most vegetation was consumed within the fire boundaries at those sites, whereas less vegetation was consumed by the fires at Matuwa and more unburned patches remained within the fire boundaries.In general, larger burns were hotter and consumed more vegetation.

Data collection
The preburn survey was conducted in May and June 2021 (3 months prefire), and postburn surveys were conducted in October 2021 (2 months postfire), March and April 2022 (8 months postfire), October 2022 (14 months postfire), and March and April 2023 (20 months postfire).Each site was surveyed for 4−7 nights per session, which represents slight variation in sampling effort within and between sessions.Traps were checked each morning, and any captured animals were identified to species, weighed, measured, marked, and sexed when possible.Mammals were individually marked using ear punching.Reptiles were given a temporary, mark (not unique) on their underside with a nontoxic paint pen, which persisted within but not between sessions.All individuals were released near the point of capture after processing.Sampling methods were approved by The University of Sydney Animal Ethics Committee (project 2021/1859) and the Western Australian Department of Biodiversity, Conservation and Attractions (license BA27000391).
We quantified habitat structure at each pitfall trapping grid during each survey.Within a 3-m radius of each trap, we visually estimated the percentage cover of spinifex and measured the width of the widest clump.For the width measurement, the clump was allowed to be composed of multiple plants form-ing continuous spinifex cover.In the first postfire survey, we also recorded the percent area burned within 3 m of each trap.Percent cover estimates were performed by a single observer (T.S.D.) to reduce observation error (Nguyen et al., 2015).
We used very high-resolution satellite imagery to map burned areas (Planet Team, 2017).We selected PlanetScope imagery (3 m resolution) from January 2021 (preburn) and January 2022 (postburn) because January falls during the peak growing season when reflectance by the vegetation is highest.We identified burned areas based on changes in true color images and 2 indices (the normalized difference vegetation index and biological soil crust index) between the 2 years (Levin et al., 2012).We manually traced around burned and unburned areas to create a binary map and validated this in the field.We converted the vector layer to a raster and calculated the total burn size (in hectares) and a measure of patchiness (conditional entropy [Nowosad & Stepinski, 2019]).Conditional entropy is a measure of configurational complexity; a value of 0 represents an area with only 1 land cover type (e.g., unburned sites) and higher values represent more heterogenous and finer-scaled patterns of multiple land cover types (Nowosad & Stepinski, 2019).We calculated conditional entropy (hereafter burn patchiness) within a 25-ha area centered on the trapping grid because this is the area that was the focus of burning efforts.We calculated the total burn size as the area burned within and beyond the 25-ha area.Appendix S2 contains a summary of the spinifex and fire variables for each site.

Statistical analyses
We created plots of capture rates over time for species that were captured in at least 10% of site-survey combinations, but we only fitted statistical models to species captured in at least 20% of site-surveys.We used generalized linear mixed models to test relationships between animal capture rates, species richness, and the predictor variables.Capture rates were calculated as the number of individuals captured at a site divided by the number of trap nights in each session at that site and then multiplied by 100 (excluding within-session recaptures).Species richness was calculated separately for mammals and reptiles.We fitted the species richness models with mean-parametrized Conway−Maxwell Poisson regression because the response variables were underdispersed (Huang, 2017).We included an offset term representing the number of trap nights per site and session in the richness models to account for varying sampling effort.We fitted the capture rate models with the Tweedie distribution, which is appropriate for zero-inflated continuous data (Jorgensen, 1997;Tweedie, 1984).We fitted the following 6 subglobal models and all lower order model combinations: treatment (burned and unburned) × time (before and after) × property; burned_trap_scale × time; total_burn_size × time; patchiness × time + property; spinifex_cover × time + property; spinifex_width × time + property.
We did not include property in models 2 and 3 because property perfectly predicted trap-scale burned area and total burn size.We included property as a main effect only in models 4−6 because we did not expect animal relationships with burn patchiness and the spinifex variables to differ between the 2 properties.We allowed a maximum of 1 fire or spinifex variable to feature in any model.The models also included random effects of site and session to account for repeat sampling over time.Additionally, for the reptile models, we included mean maximum daily temperature for each site in each survey as a covariate because air temperature can influence reptile capture rates (Read & Moseby, 2001;Spence-Bailey et al., 2010).Temperature data were sourced from the Wiluna Aero weather station ≤128 km from the study sites (Bureau of Meteorology, 2023).We deemed our modeling approach to be the most appropriate considering the confounding between property and fire (e.g., rather than conducting separate analyses across properties), which we detail further in the discussion.We scaled and centered continuous predictor variables prior to modeling.We ranked the candidate models with the Akaike Information Criterion corrected for small sample size (AICc) and considered a model to be well supported if the AICc value was within 2 units of the top-ranked model (Burnham & Anderson, 2002).We based inferences on the well-supported models and considered model terms statistically influential if the 95% confidence inter-vals around the estimates did not include zero.We inspected residual plots to ensure that models provided an adequate fit for the data.We conducted model fitting, selection, and verification with glmmTMB (Brooks et al., 2017), MUMIn (Bartoń, 2022), and DHARMa (Hartig, 2022) packages in R 4.2.2 (R Core Team, 2022).Some species were not captured in at least 1 combination of time crossed with treatment or property, which meant that certain models did not produce sensible parameter estimates due to the problem of complete separation (Albert & Anderson, 1984).To counter this, we excluded the prefire data for 5 species (Ctenophorus nuchalis, Ctenotus grandis, Menetia greyii, Varanus gouldii, Sminthopsis hirtipes) and thus also excluded the time variable (before and after) from the models for those species.Although this approach did not allow us to statistically test the full BACI design, it did allow us to test the effects of fire with a controlimpact design.An additional step was required for 2 species because Ctenophorus nuchalis was never caught at unburned sites at Matuwa and S. hirtipes was never caught at unburned sites at Jundee (i.e., all values were zero).For those 2 treatment combinations, we replaced a single zero value with a value equal to 1% of the minimum nonzero value for that species across all sites.This approach produced sensible parameter estimates and selected influential variables that aligned with our understanding of how these 2 species responded to the burns.Additionally, we excluded the site random effect from the models for Ctenotus grandis, S. hirtipes, and V. gouldii because it accounted for almost zero variance, and doing so allowed either all candidate models to converge properly or produced more appropriate residual plots (depending on the species).
The mean area burned within 3 m of each trap was 84.6% (SD 24.5) at Jundee (range = 0−100%) and 26.4% (26.4) at Matuwa (0−90%).At Jundee, mean spinifex cover around traps at the burned sites decreased from 58.6% prefire to 11.4% at 2 months postfire and then increased to 24.3% by the end of the study (Figure 3).At Matuwa, mean spinifex cover at the burned sites decreased from 47.7% prefire to 32.4% at 2 months postfire and remained similar at the end of the study (31.4%) (Figure 3).Spinifex cover at all unburned sites remained relatively constant throughout the study (Figure 3).Mean maximum spinifex width around traps at the Jundee burned sites decreased from 228.4 cm prefire to 76.2 cm at 2 months postfire, recovering to 129.0 cm by the end of the study (Figure 3).Fire had little effect on maximum spinifex width at the Matuwa burned sites (range 191.2−224.3cm throughout the study) (Figure 3).Spinifex width increased at the unburned sites at both properties throughout the study (Figure 3).

Species' and community responses to fire over time
Neither mammal nor reptile species richness varied predictably with fire treatment over time, except that mammal species richness was higher at burned than at unburned sites at Jundee during 2 of the postfire sessions (Appendix S4).Temporal changes in capture rates of D. blythi and N. ridei were independent of burning, whereas S. hirtipes was more common at burned sites at Jundee 8−20 months postfire (Figure 4).N. alexis showed a similar response to fire as S. hirtipes but had higher uncertainty and less distinction between burned and unburned sites (Figure 4).Sminthopsis youngsoni was more com-mon at burned than at unburned sites at Jundee in 2 of 4 postfire sessions (Figure 4).
Most reptile species showed little response to burning; temporal changes in capture rates were broadly similar at burned and unburned sites (Figure 5).An exception to this was the large increase in captures of Ctenophorus nuchalis at burned sites at Jundee in all postfire sessions, but there was no corresponding change at Matuwa (Figure 5).Ctenotus calurus was more common at unburned than burned sites in some, but not all, postfire sessions (Figure 5).M. greyii was rarely caught at burned sites at Jundee postfire (Figure 5).

DISCUSSION
Our prescribed burning experiment showed that most species either responded positively to the burns or were seemingly unaffected, and burn status and trap-scale burn coverage were the only influential fire variables.We assumed that, because the study ran for 20 months postfire, all the major and immediate effects of the fires were detected.Of 3 species that showed clear responses to fire, 2 were captured more frequently at burned sites and 1 was captured less frequently.There were also 2 other species that were more common at some burned sites postfire, but they were captured too infrequently to warrant formal analysis.Captures of an additional species showed a weak negative response to fire, but the cover of spinifex grass (which declined in response to fire) was a better predictor rather than fire itself.Captures of 2 other species were also positively associated with spinifex cover or width, but there was no evidence that fire drove those trends.

Burn size and severity
A range of studies from forests, woodlands, shrublands, and grasslands across the world has shown that fire characteristics, such as burn size, severity, and patchiness, can be important predictors of how animal species respond to fire (Cohen et al., 2019;Hale et al., 2022;Shaw et al., 2021;Sommers & Flanni-gan, 2022;Wan et al., 2014;Watson et al., 2012).We compared 4 different fire variables and found that only burn status (burned or unburned) and trap-scale burn coverage were influential predictors across all species.Contextualizing our results is difficult because at this time of this study, there were almost no other studies assessing how desert animals respond to fire characteristics beyond simple contrasts of burned and unburned areas of varying times since fire.An exception to this is a study from the Mojave Desert that showed small mammal abundance and species richness tended to be highest in large areas of unburned vegetation and lowest in burned areas lacking small unburned patches and far from the burn boundary (Hulton VanTassel et al., 2015).However, the burned areas in that study were 6 and 17 years postfire at the time of sampling (Hulton VanTassel et al., 2015).In chaparral shrublands, small mammal community composition differed between burned and unburned areas at 13−43 months postfire, but fire severity and distance from burn edge had weak effects (Diffendorfer et al., 2012).We did not find that total burn size was an influential predictor for any species, although total burn size was moderately and positively correlated with the area burned around traps, which was an influential predictor for some species.This indicates that small-FIGURE 4 Capture rates of small mammal species (Dasycercus blythi, Ningaui ridei, Sminthopsis hirtipes, S. youngsoni, Notomys alexis) at burned and unburned sites on the study properties (Jundee and Matuwa) (points, mean capture rate; error bars, 95% confidence intervals; vertical dashed lines, time burning was conducted).
scale burn coverage measured in the vicinity of the traps was a more informative metric than the total area burned both within and beyond the trapping grid (in some instances more than 2 km away).
It is noteworthy that we observed clear positive and negative responses to fire at Jundee but not Matuwa.The fires at Jundee were much larger and less patchy than those at Matuwa.This suggests that there may be a threshold of fire size or severity that triggers responses to fire by certain species.Mean postfire spinifex cover at burned sites was 11% at Jundee compared with 32% at Matuwa.Ctenophorus nuchalis and S. hirtipes are known to strongly favor open habitats due to the favorable foraging and thermoregulatory opportunities these areas provide (Daly et al., 2007;Haythornthwaite & Dickman, 2006a), so the reduction in spinifex cover at Matuwa may have been insufficient to create favorable habitat for these 2 species.By the same token, the burns at both properties may have been of insufficient size or severity to negatively affect the capture rates of Ctenotus grandis and C isolepis, both of which have shown a negative response to fire in previous studies (How & Dell, 2004;Masters, 1996).Low-severity, patchy fires are often promoted as a favorable management approach because they can retain shelter and other resources that allow animals to either persist in recently burned areas or survive until they emigrate to more suitable habitat (Lees et al., 2022;Silveira et al., 1999).There is some evidence to support this for small mammals in tropical savannas (Leahy et al., 2015;Shaw et al., 2021) and temperate woodlands (Swan et al., 2016), but corresponding evidence from arid grasslands is lacking.A study on the tjakura (great desert skink) (Liopholis kintorei) in spinifex grasslands showed that their burrow systems had higher occupancy and higher breeding activity if they were burned by a patchy rather than clean (thorough) burn (Moore et al., 2015).Burn patchiness was not an influential predictor for any of our study species.However, the range of patchiness seen across our sites may have been insufficient to detect a response (i.e., the patchiness variable may have been more influential if more burned sites were almost completely devoid of vegetation).Alternatively, burn patchiness may not be an important determinant of how our study species respond to fire.Rodent abundance was extremely low during this study, which meant that we could only fit statistical models to 3 marsupial species of the mammal community.We would expect the spinifex hopping mouse and the sandy inland mouse to be relatively common during periods of higher rainfall, but it remains to be determined whether burn patchiness may be influential for them, as has been seen for other small mammals elsewhere (Leahy et al., 2015;Shaw et al., 2021;Swan et al., 2016).

Small mammals
Small mammal species richness increased with the area burned around traps, which was driven by increased captures of the spinifex hopping mouse (N.alexis) and 2 dunnart species (Sminthopsis spp.) at burned sites at Jundee.In contrast, fire in the Great Basin desert (United States) reduced rodent abundance, species richness, and diversity at 1 to 31 months postfire (Sharp Bowman et al., 2017).In the Mojave Desert (United States), rodent abundance was unaffected by fire, but richness and diversity were higher on burned plots at 4 months postfire only, with full sampling spanning 1−34 months postfire (Sharp Bowman et al., 2017).We did not record increased captures of small mammals until 8 months postfire and after summer rains, which may reflect the time required for animals either to breed and rear young or for animals to emigrate to burned sites from surrounding areas.There is some evidence that N. alexis favors burned areas 2−5 years postfire (Bennison et al., 2018;Letnic & Dickman, 2005;Masters, 1993), although other studies show little response of N. alexis to fire (Letnic, 2003;Letnic et al., 2005;Southgate & Masters, 1996).Both the lagged response of N. alexis to fire and the incongruous results across studies may be due to variability in the amount of rainfall postfire that is needed for grasses to regenerate and set seed, which is a key food source for rodents, along with invertebrate prey (Murray & Dickman, 1994).During our study, the amount of rainfall was similar between the 2 properties (513 mm at Matuwa [Lorna Glen weather station] and 493 mm at Jundee [Wiluna Aero weather station]) (Bureau of Meteorology, 2023).As such, it seems most likely that the differing response of N. alexis to fire at the 2 properties is due to the nature of the fires, rather than other environmental factors.
The responses to fire we detected were likely to be partly driven by behavioral, rather than demographic, processes.Some terrestrial desert fauna, particularly highly mobile mammal species, can move long distances in response to changing resource availability and environmental conditions (Haythornthwaite & Dickman, 2006b;Letnic, 2001).Increased movement by resident animals within the fire-affected areas, as well as immigration by animals from further afield, may have increased capture likelihood (Driscoll et al., 2012).However, we captured small numbers of juvenile Ctenophorus nuchalis, S. hirtipes, and N. alexis at burned sites.Due to their relative size and immaturity, it is less likely these animals immigrated from outside the burned areas, indicating that demographic processes may have also played a role in elevating capture rates.

Reptiles
Given that reptile responses to fire only comprised 1 strong positive response (Ctenophorus nuchalis) and 2 weak negative responses (M.greyii, Ctenotus calurus), it is unsurprising that reptile species richness was not related to the fire variables.The small number of species that responded to fire is surprising, however, because previous studies have documented strong successional patterns triggered by fire (Letnic et al., 2004;Masters, 1996;Pianka & Goodyear, 2012).For instance, Ctenophorus nuchalis and Rhynchoedura spp.were more common in recently burned areas at multiple locations across Australia's deserts, whereas Ctenotus calurus, Ctenotus pantherinus, Ctenotus grandis, M. greyii, and Ctenophorus isolepis favored longer unburned areas (Letnic et al., 2004;Masters, 1996;Pianka & Goodyear, 2012).Our results support these past findings for some species but not all.A recent meta-analysis found that fire, on average, reduced mammal abundance and species richness, whereas herpetofauna (reptiles and amphibians) did not show a consistent positive or negative response (González et al., 2022).In fire-prone ecosystems across the globe, it is likely that the generally overall higher species richness of reptile communities relative to mammals, combined with strong turnover in community composition, masks overall changes in reptile species abundance and richness because late successional specialists are quickly replaced by early successional species.Such results highlight the risks of focusing only on community-level metrics and emphasize the importance of quantifying individual species' responses, along with changes in species richness and diversity.

Study limitations and future research
A limitation of our study is that fire size and severity were confounded with property, which could make it difficult to attribute species responses to the effects of fire.However, there are several reasons why we believe that the fire responses we recorded are robust.First, the responses of S. hirtipes, Ctenophorus nuchalis, and M. greyii to fire all align with existing knowledge of these species (Daly et al., 2007;Haythornthwaite & Dickman, 2006a;Letnic et al., 2004;Masters, 1993Masters, , 1996;;Pianka & Goodyear, 2012), so our results make sense ecologically.Second, all the species that showed some response to fire were recorded at both properties, which means they were potentially exposed to the full range of fires at our study sites, and thus available to respond.Additionally, our camera trap records from the same sites (which are based on sampling over longer periods and larger areas) showed similar fire responses to those revealed by pitfall trapping (T.S.D., unpublished data).Third, the study sites at the 2 properties were separated by <100 km, meaning the 2 properties were ecologically and bioclimatically very similar, plus they experienced similar rainfall during the study period (Bureau of Meteorology, 2023).We would have reservations making similar conclusions if the 2 properties were much farther apart.We believe that fire is the most logical and appropriate explanation for the responses we recorded.Nonetheless, future studies using a similar design would benefit from having a comparable range of fires at both properties, as was originally intended here.
Australian deserts are characterized by boom-bust dynamics, whereby long periods of low rainfall and low resource availability (bust phase) are punctuated by temporary spikes in rainfall and resources that inflate animal population densities (boom phase) (Dickman et al., 2010;Jordan et al., 2017;Pavey & Nano, 2013).Our study was conducted during a bust phase with low animal population densities, particularly of rodents and large mammalian predators (cats and dingoes).It is possible that the fire responses we recorded may vary during boom times when there is higher resource availability, small mammal diversity and abundance (especially of rodents), and predator abundance (Greenville et al., 2017).Theoretically, higher resource availability could help buffer any negative effects of fires on animal populations, but this could be offset by increased predation pressure.There is some evidence that feral cats are attracted to very recently burned areas due to the ease of hunting when vegetation cover is reduced (Doherty et al., 2023;McGregor et al., 2016).Evidence suggests that the suppressive effects of predators on small mammal populations in arid Australia are strongest during the "decline" phase, when the system is transitioning from boom to bust (Letnic & Dickman, 2006;Letnic et al., 2011;Spencer et al., 2017), but it is unclear how prescribed burning might alter this dynamic.We recommend that further research examine how desert fauna respond to both prescribed burning and wildfire during boom times when resource availability is higher, but so are predator densities.This is particularly important given that climate change is likely to change the relative frequencies of boom versus bust years (Chen & Wang, 2022;Cobon et al., 2019).Prescribed burning could possibly be strategically targeted to occur when predator facilitation is minimized.

Management considerations
From a fire management point of view, it is encouraging that only 2 species showed any sign of a negative response to prescribed burning, and the response by 1 of those species was evident only at 2−8 months postfire.This suggests that the nature of burning studied here is unlikely to have widespread or persistent negative impacts on the small mammal and reptile community.One may expect more species to show negative responses to fire where burns more closely resemble the size of wildfires in this region (often exceeding 1000 ha).Given that large wildfires are inevitable in this ecosystem, particularly following extreme rainfall events (Verhoeven et al., 2020), prescribed burning could play a key role in protecting areas of long unburned spinifex, which is critical habitat for threatened species in the region, including the tjakura, greater bilby Macro-tis lagotis, and night parrot Pezoporus occidentalis (Hamilton et al., 2017;Lohr et al., 2021).Prescribed burning can also increase the diversity of fire age classes in the landscape (i.e., pyrodiversity [Greenwood et al., 2021]), potentially leading to a higher abundance of native predators, such as the dingo and sand goanna (Bliege Bird et al., 2013Bird et al., , 2018)).
The spatial scale of burning is frequently mentioned in prescribed fire studies but rarely examined quantitatively (Mason & Lashley, 2021).Additionally, prescribed burning studies often place multiple survey sites within a single burn scar, whereas we examined 10 independent burns varying in size and severity.This approach allowed us to reveal that small, low-severity burns (in this case ∼5-13 ha at Matuwa) can be insufficient for triggering positive responses by species that favor more open habitats.This has consequences for fire management programs informed by ideas around patch-mosaic burning and pyrodiversity because it is often assumed that low-severity, patchy fires are beneficial for biodiversity (Jones & Tingley, 2022;Parr & Andersen, 2006).This can certainly be the case (e.g., Shaw et al., 2021), but in this instance, we would describe the lowseverity, patchy fires at Matuwa as being benign, rather than beneficial, at least over the 20 months we surveyed postfire.Given the increasing focus on prescribed burning as a management strategy, we encourage more detailed assessments of how wildlife responds to burns of varying size, severity, and patchiness.The increasing availability of very high-resolution imagery obtained by satellites and drones now makes it possible to quantify fine-scale spatial characteristics of burns for use in predictive models.We also recommend that fire management and species conservation plans be informed by a quantitative synthesis of available evidence to identify generalities and context dependencies, because it appears that the outcomes of prescribed burning can be highly variable, even within a single ecosystem type.

ACKNOWLEDGMENTS
We acknowledge the Martu People as the traditional custodians of the land on which this research was conducted.We thank Martu Rangers from the Tarlka Matuwa Piarku Aboriginal Corporation (TMPAC) for their support for the project, assistance with fieldwork, and sharing their knowledge of Country with us.We also thank Dorian Moro (TMPAC) for logistical support; the Department of Biodiversity, Conservation and Attractions for logistical and in-kind support at Matuwa; Northern Star Resources for logistical and in-kind support at Jundee; and the many volunteers and research assistants who assisted with fieldwork.We thank Planet Labs for access to highresolution satellite imagery.We thank 3 anonymous reviewers for their helpful feedback.This project was funded by the Australian Research Council (DE200100157) and The University of Sydney.
Open access publishing facilitated by The University of Sydney, as part of the Wiley -The University of Sydney agreement via the Council of Australian University Librarians.

FIGURE 1
FIGURE 1 (a, b) Locations of study sites and distribution of vegetation types on the study properties (Jundee and Matuwa), (c) unburned spinifex grassland, and (d) spinifex grassland after a prescribed burn (inset map, location of the study area within Australia).

FIGURE 3
FIGURE 3 (a) Mean spinifex cover and (b) mean maximum spinifex width within 3 m of pitfall traps at burned and unburned sites on the study properties (Jundee and Matuwa) (points, mean; error bars, 95% confidence intervals; vertical dashed line, time burning was conducted).

FIGURE 6
FIGURE 6Relationships between 5 predictor variables and mammal and reptile species richness and capture rates (see Appendices S5 & S6 for model rankings) (points and solid lines, mean effects; error bars and shading, 95% confidence intervals).For conciseness, plots of time (before or after) for reptile species richness and treatment (burned or unburned) for Ctenophorus nuchalis are not shown, but mean values are provided in the text.