Evidence that post‐fire recovery of small mammals occurs primarily via in situ survival

As climate change intensifies and wildfire frequency and scale increase, it is critical we develop a robust understanding of how species recover from these major disturbances. Here, we aim to determine whether source populations for recovery following large‐scale intense wildfires are derived from either in situ survival, or immigration from surrounding unburnt areas (ex situ). Secondly, we sought to determine whether habitat elements (e.g., logs) within the landscape facilitate in situ survival of small mammals during fires.


| INTRODUC TI ON
Fire shapes ecosystems around the world by consuming plant material and influencing the distribution and abundance of biota (Bond & Keeley, 2005;He et al., 2019). However, fire regimes across the world have changed dramatically over the past five centuries with the spread of colonialism and the displacement of Indigenous people (Liebmann et al., 2016;Nowacki & Abrams, 2008;Romanin et al., 2016). Fire regimes are also being pushed further from historical baselines by climatic changes (Mariani et al., 2018) and increased prescribed burning (Penman et al., 2011). In particular, climate change is resulting in longer fire seasons and larger and more intense fires in many fire-prone regions globally (Clarke et al., 2011;Jolly et al., 2015;Kirchmeier-Young et al., 2017).
Fire-induced mortality of wildlife leads to longer-term population reductions (Banks et al., 2011;Hale et al., 2016), which can increase the risk of local extinction. While mortality during fires is poorly documented (Friend, 1993;Koprowski et al., 2006), there is a growing body of evidence to suggest many small mammals can initially survive the immediate impacts of fire and the riskiest time is soon after fire (Leahy et al., 2015). Mortality after fire is associated with post-fire conditions, such as elevated rates of predation leading to population reductions Leahy et al., 2015;. Post-fire environments also have substantially altered vegetation structure and resource availabilities, which can affect survival and recruitment Sutherland, 1999). For instance, Sutherland (1999) found that post-fire food supplementation caused temporary increases in brown antechinus Antechinus stuartii abundance relative to areas without food supplementation.
Reversing the impacts of fire requires population recovery, and a key question for land managers and researchers is whether populations recover mainly from survivors within the fire ground via in situ survival and population growth (in situ recovery) or from individuals emigrating from nearby unburned areas (ex situ recovery) (Banks et al., 2017). In situ recovery may be spatially indistinct, whereby survival is dispersed and not related to specific landscape or habitat features, or nucleated, where populations grow from spatial refugia within the fire boundary where animals survived (Banks et al., 2017).
While animals may die during fire, certain habitat elements such as logs, burrows and rocks can buffer animals from the direct impacts of fire, thus facilitating their survival and providing potential nuclei for in situ recovery (Begg et al., 1981;Sutherland & Dickman, 1999).
This scenario of in situ survival is considered the most plausible mechanism for the recovery of mammals in landscapes which may be subject to natural disturbances such as wildfire (Lindenmayer et al., 2005) and has been suggested to be more important than ex situ recovery (Banks et al., 2011(Banks et al., , 2017Bradstock, 2008;Plavsic, 2014;Southgate & Masters, 1996), although there are exceptions. Puig-Gironès et al. (2018) found that while a small proportion of survivors were retained within burnt areas, the recovery process was also heavily assisted by recolonization from unburnt areas. The relative importance of in situ recovery may be dependent on the severity, intensity, or size of a fire event (Romme et al., 1998). Ecosystems exposed to large, high-intensity and relatively homogeneous fires may experience greater mortality and result in fewer small, unburned refuges (Collins et al., 2019), potentially leading to a greater reliance on an ex situ mode of post-fire recovery (Bradstock, 2008).
Recent mega-fires in Brazil, Australia, the USA and elsewhere point to a critical need to disentangle the relative importance of in situ and ex situ strategies in the post-fire recovery of mammal populations (Kganyago & Shikwambana, 2020;Nolan et al., 2020). Larger, more intense and homogeneous fires are becoming increasingly common (Clarke et al., 2011), leaving fewer small unburned refuges within fire-affected landscapes (Foster et al., 2017). These changes to fire regimes are combined with more frequent droughts (Cunha et al., 2019; Intergovernmental Panel on Climate Change 2013), which are known to drive severe reductions in mammal distributions within fire-prone landscapes, even in long unburnt areas (Crowther et al., 2018;Hale et al., 2016). Such interactions can have severe impacts on the rates of population recovery in post-fire landscapes (Hale et al., 2016).
Here, we use extensive sampling of mammal populations to test key questions relating to the mechanisms and drivers of post-fire population survival and recovery. The Grampians National Park in south-eastern Australia has experienced three large-scale (35,000-85,000 ha), high-intensity wildfires in recent years (2006,2013,2014), while also enduring highly variable climatic conditions, creating a unique opportunity to examine the post-fire recovery of small mammals under conditions mimicking those forecast by climate change models. Long-term small mammal research has also been undertaken in this region making it an ideal location to address the following two questions.

| Question 1: Are source populations for recovery derived from in situ survival or via immigration from surrounding unburnt areas (ex situ) following large-scale, high-intensity wildfires?
If the main mechanism for recovery is ex situ immigration, we would expect to see a clear spatial pattern of post-fire population recovery, with sites closer to large patches of unburnt vegetation outside of the fire perimeter recovering more quickly than isolated sites in the interior of the burn. This is due to the larger distances individuals would have to travel through the landscape in order to recolonize interior burnt areas. If recovery is largely driven by in situ survivors, we would expect recovery to be unrelated to distance to external unburnt patches, as founding individuals would be derived from within the burn boundary and not from surrounding unburnt areas (Banks et al., 2011). 1.2 | Question 2: Do habitat elements (e.g., logs) facilitate in situ survival of small mammals in burnt areas?
If there is evidence of in situ survival after high-intensity fires, there are likely to be habitat features that have influenced survival by providing individuals with places to seek shelter during fire events and persist in the post-fire environment for a period of time. We expect to see more individuals surviving in areas with a higher abundance of particular habitat elements that could act as in situ refuges, such as large trees and logs, as animals may be able to readily seek shelter provided by these habitat elements both during and immediately after the fire.

| Study area
The Grampians National Park (168,000 ha) is located in central western Victoria,142.40) and has highly diverse flora and fauna (Enright et al., 1994;Parks Victoria, 2011). In 2006, approximately 50% of the park was burnt by a severe wildfire (85,000 ha) and subsequently 36 long-term study sites were established in 2008 to investigate the post-fire environment and mammal recovery.
In January 2013, the Victoria Valley in the west of the Grampians National Park experienced a 35,000 ha, high-intensity wildfire that burnt for approximately 2 weeks. Following this, in 2014, another extremely high-intensity wildfire burnt 55,000 ha of the northern Grampians over 5 days. Cumulatively, these fires burned 175,000 ha of the park in less than a decade, with some overlap occurring between fire boundaries, resulting in approximately 90% of the park being burnt. All three fires were characterized by near complete burning of all vegetation within the fire perimeter, and a very limited number of small unburnt patches. These fires provide a unique opportunity to investigate the spatial patterns of small mammal survival and recovery following fires, and whether those patterns relate to habitat elements that allow mammals to persist in burnt areas. We use multiple data sets to examine evidence of survival of small mammals and the effect of habitat elements on that survival as well as the drivers of recovery post-fire ( Figure 1). Infrared motion sensing ScoutGuard 550 cameras with a 2 GB memory card were used to monitor mammals. Cameras were programmed to take three five megapixel images over a 6 s period with a 30 s idle period between each set of images. At each camera point, a camera was fixed to a wooden stake at a height of approximately 1.3 m facing down to the ground (De Bondi et al., 2010). This downward facing camera technique results in less false triggers caused by vegetation and improves the ability to identify small mammal species (Rendall et al., 2014). To further improve our capacity to identify species and reduce the impact of moving vegetation, we cleared an area of approximately 1 m 2 of vegetation underneath each camera. Below each camera, a lure consisting of a highly perforated PVC pipe, was pegged to the ground. The lure was filled with polyester cushion stuffing soaked in a mixture of peanut butter, vanilla extract, linseed oil and tuna oil. In addition to the fixed oil-based lure, a handful of rolled oats was placed below the camera to encourage small mammals to remain in the field of view for an extended period of time.

| Site selection and description
Each site was surveyed for eight nights. This duration was selected as it has been found to confidently determine absences of species at the study site (De Bondi et al., 2010). cover abundance of large rocks (>30 cm diameter); and (c) number of logs (>10 cm diameter). These variables were used as they have been shown to provide refuge to mammal species both during (e.g., protection from fire) and post-fire (e.g., shelter) (Robinson et al., 2013). The productivity of each site was calculated using 6 years (prior to the 2006 wildfire) of satellite data from the summer months (the most water stressed period of the year in southeastern Australia) and then averaged to produce a single NDVI value for each site using a 50 m buffer (Hale et al., 2016). The cover abundance of large rocks was measured within a 10 m quadrat centred on each camera location. The number of logs within the quadrat was also counted (McCarthy et al., 1999).

| Dependent and independent variables
In 2014, at all of the burnt sites (n = 119) we measured four independent variables at each camera location: (a) large rock cover abundance (>30 cm diameter); (b) canopy cover; (c) the number of large trees (>40 cm diameter at breast height); and (d) prior productivity (NDVI). Rock cover abundance was measured in the 10 m quadrat centred on each camera location and canopy cover was estimated by observing the proportion of sky obscured by the canopy above the quadrat. Prior productivity for the site was measured as described above. We used a slightly different set of variables in 2014, because after 2013 we determined that measuring the number of large trees and canopy cover may provide a better metric of tree-based survival options than log cover on the ground.

| Statistical analysis
2.4.1 | Source of recovery studies (long-term small mammal trapping data) To examine the source of recovery of small mammals, we modelled the effects of time since fire, distance to unburnt vegetation, rainfall and productivity on the occurrence of six small mammal species using generalized additive mixed models (GAMMs). The response variable was the presence/absence of the species across the four nights of trapping conducted annually at each site, modelled as a binomial distribution of errors. We used mixed models as they allow for the inclusion of a random factor that accounts for repeated measurements (Zuur et al., 2009), which was required due to the repeated sampling of the same sites each year following fires. As our primary interest was in the role of distance to unburned vegetation on post-fire recovery, we considered three models, all of which included rainfall and NDVI in addition to (a) a time since fire x distance interaction, to examine whether distance to unburned vegetation modified species' recovery trajectories, cies and year with main effects of the habitat variables, which we scaled and centred prior to analysis. We also included a detection covariate of survey night (1-8) to account for potential changes in lure attractiveness over time. We present model coefficients and 95% confidence intervals. We excluded Antechinus mimetes for 2014 because the model did not converge.

| RE SULTS
3.1 | Question 1: Are source populations for recovery derived from in situ survival or via immigration (ex situ) from surrounding unburnt areas following large-scale, high-intensity wildfires?
There was little evidence of a spatial pattern in the persistence or recovery of native mammals in relation to distance to unburned vegetation: the model which included time since fire only was the most parsimonious model for every species, except for R. rattus which also included an additive effect of distance to unburnt vegetation (Table 1). Time since fire was significant for all species (Table 2), and all species aside from M. musculus were least common immediately after fire and increased in their probability of occurrence thereafter ( Figure 4). Rainfall was positively associated with M. musculus occurrence, and NDVI was positively associated with A. agilis, P.

| Question 2: Do habitat elements (e.g., logs) facilitate in situ survival of small mammals in burnt areas?
Camera trapping within the fire perimeters of both the 2013 and 2014 wildfire revealed evidence of survival of small numbers of small mammals (Table 3). The species with the most detections were arboreal, including Antechinus agilis (present at 12.5%-16.6% of cameras), Antechinus flavipes (25.8%-29.4%) and Trichosurus vulpecula (10.4%-25.6%), or the invasive Rattus rattus (13.8%-22.1%) and M. musculus (13.1%-39.9%), whereas terrestrial native rodents were detected in low numbers (1.2%-7.5%; Table 3). For most species, there was little difference in presence/absence ratios between sites that were burnt and the very small unburnt patches within the fire boundary (Table 3). The native rodent P. shortridgei was more likely to be present in unburnt patches than in burnt sites in 2013 (Table 3). There was also a weak signal for a similar trend towards higher presence in unburnt patches for A. flavipes in 2014 (Table 3). A different trend occurred in the invasive M. musculus with it being more likely to be present in burnt areas than unburnt patches (Table 3). Surviving individuals were also detected at our long-term trapping sites soon after they were burnt during the 2013 (n = 6 burnt) and 2014 (n = 4 burnt) wildfires (e.g., A. agilis and P. shortridgei) (Appendix : Table A1).
For 2013, post-fire occupancy of A. flavipes at burnt sites was positively related to rocks and negatively to logs, whereas R. rattus occurrence was positively related to logs (Table 4). For 2014, occurrence of A. agilis was positively related to large trees, M. musculus responded negatively to canopy cover, and R. rattus responded positively to both rocks and NDVI (Table 5). There were no effects for the remaining combinations of species and habitat variables (Table 5).

| D ISCUSS I ON
We sought to determine the relative importance of in situ survival and ex situ emigration to small mammal population recovery following wildfires and the influence of habitat variables in facilitating survival. We did not find a relationship between distance to the continuous unburnt vegetation and the occurrence of any native species, suggesting that in situ survival is a more likely mechanism for the recovery of post-fire mammal populations than immigration from the continuous unburnt habitat. We acknowledge that this research did not begin until 2 years after the 2006 wildfire and that small mammals could have moved long distances in this period. However, as discussed in detail below, the system was in extreme long-term drought conditions and small mammal abundances were extremely low across the landscape and vegetation recovery was extremely slow (Hale et al., 2016) and we saw the same recovery trend in sites TA B L E 1 Model selection for response of mammal species to time since fire (TSF), Time since fire and distance to continuously unburnt vegetation (distance) and the interaction between time since fire and distance to continuously unburnt vegetation The occurrence of a further two high-intensity wildfires provided the opportunity to investigate survival of small mammals soon after the fires. While we could not commence field surveys until almost 3 months after the fire, we consider it unlikely native small mammals had rapidly invaded the burnt landscape, as most of the small mammal species in the system have a requirement for structurally complex ground vegetation which was missing immediately after the fires (Hale et al., 2016). The potential for extremely rapid colonization of the burnt landscape by animals from the unburnt vegetation at the perimeter of the fire cannot be discounted as a possible explanation of the patterns observed, but given the habitat requirements of the species being studied we consider it unlikely.
We found no evidence for continuous unburned habitat being the primary source of small mammal population recovery postfire, with species occurrence being similar between sites near and far from continuous unburned vegetation (spanning distances 0.1-13 km from the burn edge). The lack of a spatial signal in small mammal population recovery across three high-intensity, large-scale fires across the landscape suggests that this mesic ecosystem is largely recovering internally via in situ survival rather than externally through emigration from unburned areas (Banks et al., 2017;Lindenmayer et al., 2008;Recher et al., 2009). Our results support the prediction of Banks et al. (2011) that residual populations made up of animals that survived the wildfires will be more important than emigration for population recovery of many species following wildfires. However, as mentioned previously, we note that our study did not commence until 2 years post-fire; therefore, evidence for recov-

F I G U R E 4
Response curves of small mammal recovery following high-intensity wildfires area in the 2 years post-fire. Further, our camera trapping studies clearly show that multiple species can persist in burnt areas in the months following fires, with the more common species present at 7.5%-39.9% of sites. Survival of animals was documented in both small unburnt patches as well as burnt sites, suggesting both types of areas may be important for facilitating population recovery. The primary invasive disturbance specialist in the system M. musculus had a different response to most native species, with higher presence than expected in burnt sites than in the small unburnt patches, suggesting a rapid invasion of the burnt landscape, possibly to take advantage of seed resources in the post-fire environment. Additionally, after both the 2013 and 2014 wildfires, ten of our long-term sites were burnt, and a number of small mammal species (e.g. A. agilis and P. shortridgei) were captured at these sites within 3 months of the fire (Appendix : Table A1). This further demonstrates that some individuals are able to survive in the immediate post-fire environment, with some continued survival throughout these sites 1 year post-fire (Appendix : Table A1). Given this evidence, we are confident that our findings suggesting the low relative importance of emigration from the continuously unburnt habitat for mammal population recovery versus the important role of in situ survival are robust.
In the initial post-fire environment, there is survival of a number of species which are able to maintain small populations, particularly arboreal species including A. agilis, which supports previous studies (Banks et al., 2011). Small unburnt patches within the burn perimeter facilitate some survival of species, but given the similar presence/ absence ratios in the burnt areas, small mammals are clearly able to survive fire events. In the burnt areas, habitat elements that are able to withstand fire, such as logs, rocks and large trees, provide sources of shelter for species, allowing them to survive the fire as it moves through an area (Begg et al., 1981;Ford et al., 1999). However, the presence of these initial survivors is likely to decrease following the TA B L E 3 Percentage of burnt or unburnt sites that were occupied and the site occupancy for all sites combined, following the 2013 and 2014 wildfires  Where "n" equals the number of sites species were present at. Chi-square analysis establishes whether there is a difference in the ratio of presence/absence of a species and whether the site was burnt or unburnt. Abbreviation: ID, Insufficient Data.
a Species not native to Australia. fire due to a lack of resources  or elevated predation rates within the burn area (Leahy et al., 2015;Russell et al., 2003). While the Grampians is subjected to long-term intense fox baiting, both foxes and cats still have relatively high occupancy across the entire landscape (Geyle et al., 2020) suggesting that predation pressure likely remains high in burnt areas. A fruitful area for future research is to measure the post-fire survival and population recovery of small mammals in areas with and without effective predator control.
Previous studies have shown that post-fire population recovery of bush rats occurred from topographic drainage lines (Banks et al., 2017); Antechinus agilis abundance increased in gullies immediately post-fire (Swan et al., 2016); and elephant shrews in in South Africa survived fire by sheltering in termite mounds (Yarnell et al., 2008). We found mixed support for particular habitat elements influencing the occurrence of small mammals in the months following wildfire. The cover of rocks and logs was important for the invasive black rat, whereas the marsupial A. flavipes showed a negative response to logs and A. agilis a positive response to the number of large trees. The lack of response of the remaining species suggests that the measured habitat variables are not key to their postfire persistence, at least in the context of this study.
It is essential to take into account the state of the landscape prior to a fire event when considering the timeframe and method of recovery. Post-fire recovery of small mammals in mesic systems has often been considered a product of waves of immigration from unburnt areas and having a strong link with time since fire and vegetation recovery (Fox, 1982). However, for our mesic study system, rainfall has a more prominent role in recovery than time since fire alone (Hale et al., 2016). During periods of below average rainfall, systems become depleted of mammals (Crowther et al., 2018;Hale et al., 2016;Meserve et al., 2011) and resources  leaving a smaller pool of animals in the environment that have the potential to survive fire. The interplay between fire and underlying climatic conditions, exacerbated by climate change, may substantially alter how species recover post-fire, resulting in an increasingly dynamic system.
With small mammal recovery varying with the prevailing climatic conditions (Hale et al., 2016;Recher et al., 2009), it may be that the mode of recovery will shift as climatic conditions across the globe continue to change. Many species are likely to recover through in situ survival, at least during the initial stages of recovery, especially species that are less fire sensitive (e.g., A. agilis) as they are able to survive more readily in the post-fire environment (Banks  et al., 2017). The more sensitive species (such as R. lutreolus) may initially disappear from burnt areas and therefore rely more heavily on immigration for recovery once landscape conditions become more suitable (Banks et al., 2017). Following this logic, it may be that recovery occurs in a step-wise nature with initial recovery a result of individuals surviving within the burnt landscape which eventually build up populations as conditions improve (e.g. increased rainfall, available habitat). As these populations increase in size and burnt areas begin to recover, it is likely that individuals will subsequently disperse into the broader landscape.
The changes in fire regimes and rainfall observed in the study area are predicted to become increasingly "normal" under future climate change scenarios, with more intense and frequent wildfires paired with periods of below average rainfall punctuated by flooding events (Clarke et al., 2011;Intergovernmental Panel on Climate Change 2013). Even areas of the landscape which are long unburnt have suppressed animal populations during drought periods (Hale et al., 2016). We present strong evidence for post-fire recovery being driven largely by in situ survival. In situ survival is facilitated by small unburnt patches and key habitat elements in burnt areas and presumably prior populations (Hale et al., 2016).
Given that south-eastern Australia and many other parts of the world are seeing increasingly frequent large-scale wildfires due to climate change, the capacity for species to survive fires via the mechanism of in situ survival and then become the foundation of population recovery will mitigate some of the critical impacts of climate change.

ACK N OWLED G EM ENTS
We thank and acknowledge the traditional owners and custodians, past, present and emerging of the Gariwerd landscape. We gratefully acknowledge funding and field support that was provided

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13283.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.