Quantifying the impact of severe bushres on biodiversity to inform conservation

The unusually severe 2019-2020 Australian bushre season destroyed large areas of habitat along the southeastern coast. We assess the differences between this re season and previous ones to understand their impacts and potential recovery. We used thermal satellite data to quantify the extent of eastern Australian bushres from 2012-2020. During the 2019-2020 anomalous re season, 134 mega-res, each over 10,000 ha, burned 62.5% of the total affected area, including critical forested landscapes. Previous re seasons were characterized by smaller, scattered res in mostly non-forested areas. The anomalous re season had direct negative impacts on wildlife including grey-headed ying foxes, which experienced substantial declines in immediately available foraging forest habitat, particularly key winter resources. We identied important areas for managing wildlife resources and supporting ecosystem health in the coming decades. Our research also identies key monitoring areas to improve our understanding of ecosystem recovery and resilience in changing re regimes.


Introduction
Fire is essential for maintenance of biodiversity across ecosystems globally 1 . Fire-adapted animal and plant species can thrive in nutrient-rich ash post re 2 and some are dependent on re for new generations 3 . Despite the vital role of res, extensive high-intensity res can have negative impacts on ecosystem health by reducing biodiversity 4,5,6 and causing declines in wildlife populations 7,8 .
Uncontrolled res can also damage human health 9 and property 10 . The negative impacts of re on ecosystems can be long-lasting 11 and new re regimes can delay or prevent the recovery of ecosystems altogether 12 . Climate change-induced heat and drought has extended the length of re seasons and augmented burnable area across the globe 13 . It is increasingly important to understand the impact of these emerging re regimes that reduce the resiliency of affected ecosystems to better manage them presently and mitigate future negative outcomes.
Australia is particularly vulnerable to climate-driven changes in re regimes. While re occurrence is a product of complex management practices, local conditions, history, and chance 14,15,16 , Australian wild res, often called bush res, are tightly linked with local weather conditions and climate oscillations 17,18 . The forested ecosystems found across Australia, namely Eucalypts (tribe Eucalypteae within Myrtaceae), are re-adapted and ower quickly following typical res 19 . However, high intensity res can damage epicormic growth beyond recovery 19 ; more frequent and/or severe res can prevent recovery of forests and promote domination of alternative ecosystems 20,21 . Both the intensity and frequency of bush res is predicted to increase with climate change 22 , raising questions of whether forested ecosystems will be able to persist and support endemic wildlife in shifting re regimes. In 2019-2020, local weather and climate conditions in eastern Australia gave rise to an unusually dry spring combined with an accumulation of fuels 23,24 . This created favorable conditions for anomalous, severe, devastating bush res that burned primarily in forested ecosystems and dominated international headlines for months 25 .
The location and extent of 2019-2020 bush res lead to initial estimates that one billion animals perished 26 and that there had been an increased extinction risk for some endemic species 27 , but the true impact of these res is still under investigation. In the aftermath of typical res, surviving animal populations decline in the face of higher predation risk and limited resources 7,8 . High intensity res often lead to steeper declines in species and have a longer-term effect on the landscape 28 . While bats have been found to suffer low direct mortality during res 29,30 , increasing evidence suggests that the extent and distribution of unburned habitat is key for predicting the post-re success of surviving animal populations. Not much is known about the consequences of bush res for nectar-and fruit-dependent ying foxes (Pteropus species) despite their role in vegetation recovery through pollination 31,32,33,34,35 and dispersal of seeds over long distances during seasonal movements 35 . Without regard to their importance within these ecosystems, these animals were legally culled historically due to their consumption of commercial fruits, nuisance (smell and noise) in urban environments, and fears of the emergence of the zoonotic pathogens they host, including Hendra and Lyssa viruses 35 . One species, the grey-headed ying fox (Pteropus poliocephalus),, is classi ed as 'Vulnerable' due to 30% population declines related to habitat loss and human disturbance 36,37,38 . Large res impact many animal and plant populations, but the responses of vulnerable and endangered populations are particularly critical to understand, especially for preserving biodiversity and maintaining ecosystem health and balance.
There is mounting evidence that the scale and intensity of 2019-2020 res were a consequence of climate change 39 , suggesting that conservation activities will need to incorporate these events into future planning. The short-and long-term population impacts post-re on wildlife caused by these res are yet to be determined. To quantify the recent anomalous re season (September 2019-March 2020), we used remotely sensed thermal data to compare the extent of burned area and distribution of res across eastern Australia of the anomalous bush res (2019-2020) to a previous bush re season with similar total burned area in the spring and summer of 2012-2013. We then compared the amount of forest cover burned during these re seasons across the heavily populated states of eastern Australia. Finally, we assessed how these res impacted the Vulnerable grey-headed ying fox by quantifying burned and remaining foraging habitats. This work prioritizes locations and species of vegetation for urgent surveillance, preservation, and restoration.  Table S2). The area burned in the 2012-2013 re season was the most similar to 2019-2020 re season (5% less than 2012-2013) and affected more than twice the area burned in most of the intervening years. The 2012-2013 re season, hereafter known as the previous focal re season, was used for comparison to the recent 2019-2020 re season, hereafter referred to as the anomalous re season. We assessed the similarities and differences in the distribution, extent, and impact of these two re seasons.

Fire extent between years
In the previous focal re season, 67,722 unique contiguous res burned 8.48 Mha, mostly in savannahs and open woodlands with low levels of canopy cover (<20% canopy cover, Figure 2c). In contrast, the anomalous re season burned 8.03 Mha but consisted of only 29,148 individual res, many burning areas that were an order of magnitude larger than the previous focal re season ( Figure 1). Fire size is a key predictor in the severity of res and determines the scale of consequences for animal and plant populations.
During the previous focal re season, individual bush res burned much smaller areas and were more numerous. Mega-res (>10,000 ha 40 ) only accounted for 17.1% of the total burned area. During the recent anomalous re season, the majority (62.5%) of the burned area was concentrated in a few mega-res ( Figures 1, 2e). The burned area calculations presented here re ect data collection at a spatial resolution of 440 meters, additional calculations at coarser resolutions to account for variation in the size of scan pixels are presented in the SI ( Figures S6-S8). Calculations using coarser resolutions lead to much larger estimates of burned areas but show a similar distribution of re sizes (Table S3, S4, Figures S6-S8).
The anomalous re season also coincides with greater rainfall de cits in the nine months preceding the peak of the re season, particularly on the coast between Sydney and Brisbane, and de cits covered a larger spatial extent preceding the anomalous re year (Figure 2b,d).
Forest fragmentation and impact on refugia after res Due to a combination of anthropogenic pressure and natural disasters, forests are steadily declining annually in eastern Australia (Figure 3a). Remaining forest area (de ned with a canopy cover of >20%) has decreased by 3.9% over the last decade. Forest area covered 12.4% of land in 2009, which declined to 11.9% in 2018 ( Figure 3a). Fragmentation accompanied much of this forest cover decline, resulting in an increased abundance of forest patches ( Figure 3a). In 2012, there were 28 forest patches that were greater than 100,000 ha in size, each with a high proportion of high-quality core forest habitat, de ned as forest area at least 100m from a forest edge 41 . These large forest patches accounted for 63.2% of total forest cover in eastern Australia in 2012. Although many of these patches are in protected areas ( Figure  S3), these forests had been fragmented and lost core habitat forest by 2018 (Figures 3b, S9).
The impact of res on forested habitat was drastically different between the previous focal re season and the anomalous re season. In the previous focal re season, the vast majority of total burned area (77.5%) was in non-forested areas, whereas in the anomalous re season, the majority of total burned area (66.2%) occurred in forested areas ( Figure 2e). The mega-res that accounted for the majority of the burned area occurred primarily in large, contiguous forests in New South Wales. Fires burned entire forest patches of all sizes, as demonstrated by the reduction in total number of forest patches ( Figure S10). Fires also effectively broke up large continuous forests and left a greater number of moderately sized forest fragments >1,000 ha, reducing the total unburned forest habitat area while increasing the number of patches (Figure 3c).
Fire impacts on grey-headed ying fox habitat We quanti ed burned habitat 42 from both the previous focal re season and anomalous re season in relation to roosts that were recently occupied (FF1 and FF2; Table 1). For each roost, we quanti ed total foraging habitat and winter habitat within a 50km radius to account for maximum foraging distances (average 20-30km) from individual roosts 43,44 . This area includes 87.5% of the total extant grey-headed ying fox habitat. Winter habitat, which is a subset of foraging habitat, includes the ve most annually reliable and productive nutritional resources during winter resource bottlenecks (Eucalyptus robusta, E. tereticornis, E. siderophloia, Melaleuca quinquenervia, and Banksia integrifolia) 42 . In the previous focal re season, the amount of foraging habitat burned within range of FF1 roosts ranged from 24 to 26,854 ha ( Figure S11, S12). Only 3.7% of the total grey-headed ying fox habitat across southeastern Queensland and New South Wales was burned. The median habitat burned within an individual roost's foraging radius was slightly lower, 2.89%; the most affected roost experienced 19.86% of its foraging habitat burned. At each roost, the percentage of total foraging habitat burned was highly correlated with the percentage of winter habitat burned, but winter habitat covers a much smaller spatial extent (Figure 4c-h).
The anomalous 2019-2020 re season burned 2.8 Mha (33.7%)of total foraging habitat, drastically reducing resources immediately available for grey-headed ying foxes in parts of their range (Figure 4b).
Of the 324 FF2 roosts, 62 had greater than one-fourth of their total foraging habitat (single roost Grey-headed ying fox roost occupancy before and after the previous focal re season We used a longitudinal survey of presence and abundance of ying fox species at roosts in the seasons before and after the previous focal re season to determine re impacts on grey-headed ying fox occupancy 45 . The presence of grey-headed ying foxes at the roosts identi ed in FF3 was signi cantly affected by total unburned area of winter habitat within a 50km foraging radius, meteorological season, co-occupation of black ying foxes, and geographic location. The best t mixed effects logistic regression showed that roosts with more abundant unburned winter habitat were more likely to be occupied by grey-headed ying foxes at any time of the year (adjusted odds ratio ( roost to burned area, total area burned in foraging radius, and habitat burned) were not included in the best t logistic regression.

Discussion
The anomalous 2019-2020 bush res were unprecedented in their size, impact on forest ecosystems, and destruction of wildlife habitat in eastern Australia. These analyses highlight the drastic reductions in foraging habitats, including key winter resources, for the already threatened grey-headed ying fox.
Monitoring wildlife population responses, quantifying re intensity, and understanding the resilience of vegetation to these extreme res will be essential for designing management and mitigation strategies in the coming months and years as re regimes shift with global climate change 46 .
Anomalous re characteristics and impact on forest habitats The anomalous size and location of res in 2019-2020 destroyed extensive areas of key forest habitat 25 . Mega-res, de ned as anything over 10,000 ha, are associated with higher intensity burns, longer recovery periods, major alterations of canopy structure, and higher direct and indirect mortality 47 . There were 67 mega-res across eastern Australia in the anomalous re season, with the largest re burning almost 1.6 Mha. Wildlife can often persist or relocate in smaller res, but these larger res are devastating in the short-and long-term for many species 5,47 . Here we highlight the signi cant amount of total and winter foraging habitat of grey-headed ying foxes that burned in the 2019-2020 re season. Flowering in Eucalypts commences earlier when re intensity is low and the full depth of bark is not scorched 28,48 . Epicormic resprouting along branches starts almost immediately after re and reduces the time to owering 19 . However, the extensive drought ( Figure 2d) and extreme heat conditions preceding the res 23 likely reduced or delayed owering of diet species in the short-term 49 . We expect a net decrease in oral resources for ying foxes following res, particularly from larger, high-intensity res.
Reductions in winter foraging habitats have been associated with increased adult mortality, low reproductive rates, young with low birth weight, and birth defects in grey-headed ying foxes 36 . Our analyses demonstrate that occupation of roosts is highly correlated with adjacent winter foraging resources. We propose the 2019-2020 res will have signi cant population impacts on grey-headed ying foxes through substantial reduction in habitat and feeding resources at speci c roosts. Although there are no other studies of ying fox responses to res, studies of other bat species have shown that res impact populations through the alteration of foraging and roosting habitat 50,51,52 , and other Australian bats (non-Pteropus) have been shown to decrease activity in local areas in the months following bush res and up to one-year after major res 29,52 . The anomalous season of 2019-2020 bush res signi cantly impacted foraging resources for grey-headed ying foxes, which will alter their feeding behavior, and in turn, could delay recovery of Eucalypts that bene t from their pollination.
Threats to wildlife and humans from anomalous res The eastern Australian forests, including Eucalypt dominated stands, are an important global biodiversity hotspot 53 . The ongoing anthropogenic loss of forest habitat throughout eastern Australia in the decades preceding these res puts additional constraints on the post-re recovery of wildlife populations. While forest fragmentation impacts populations variably, a decline in total amount of habitat negatively impacts many wildlife taxa 54,55,56 . Grey-headed ying foxes could theoretically respond to 60% forage habitat burning, as observed in New South Wales during the anomalous res, by migrating to unaffected (or less affected) roosts and foraging areas. But unaffected roosts are surrounded by highly fragmented habitat limited in extent and frequently overlapping with black ying foxes, which are resource competitors. Alternative roosts are also located in urban environments that are largely unaffected by annual re seasons. These areas contain a variety of mostly anthropogenic food resources, which that are lower-quality than their native food sources but more reliably available and may in uence roost occupancy 57 . Movement to alternative roosts with minimal burned habitat or urban sites may be further limited by nutritional status of individuals 58 . Extreme drought and few options for suitable alternative habitats put populations of grey-headed ying foxes in a precarious position to recover from these extreme events. Monitoring timing of resource owering, population movements and roost colonization will be essential for informing local conservation efforts to ensure their persistence, recovery of native Eucalypt forests, and safety for human populations.
Beyond wildlife populations, large wild res also have direct and indirect negative consequences for human populations. Loss of life and property is most frequently associated with large res 9, 59, 60 . Initial estimates from the 2019-2020 res resulted in 417 excess deaths, 3151 excess hospitalizations and 1305 emergency room visits 61 . There are also indirect negative human health impacts -the increase in air particulates damages respiratory tissues and increases susceptibility to respiratory infections, such as seasonal in uenza. In damaging lung tissue, the 2019-2020 res may have also increased susceptibility to infection with pandemic SARS-CoV-2, or increased the likelihood of severe symptoms with infection.

Future of vulnerable species and ecosystem health under changing re regimes
The loss of habitat during res may also drive grey-headed ying foxes and black ying foxes into urban environments, increasing the risk for spillover, or cross-species transmission, of pathogenic viruses 62 . For example, these Australian ying foxes carry and transmit Lyssa virus and Hendra virus are examples of viruses, which can sicken or kill humans and animals 62 . Previous studies have proposed preserving and restoring critical resources to draw ying foxes out of human settlements to reduce risk of viral transmission 63 . Implementation of these proposed management strategies is even more critical now to help restore the large loss of native vegetation and prevent pathogenic viruses from transmitting across species 63 . Areas of unburned winter habitat near heavily impacted roosts identi ed here (blue, Figure 4ch) need to be monitored closely and preserved, as these are now lifelines for displaced populations. Loss of this vulnerable vegetation will have direct negative impacts on biodiversity, particularly for birds, mammals, and invertebrates that rely on Eucalypts for food, refuge and breeding sites 64 . Restoration of forests generally should focus on increasing overall area while limiting negative impacts of edge-effects on populations by joining existing fragmented patches 65 . Strategic restoration of foraging habitat may offer some resilience to mitigate impacts of bush res on ying fox populations and other wildlife that are dependent on native forests.
Increases in the size and frequency of large res have been documented in North America 66,67,68 .
Climate change is expected to increase extreme re conditions in Australia 22 and burnable area in Europe 69 . Prolonged drought and higher temperatures are leading to the drying out of new landscapes and burning of ecosystems not adapted to res 70 . Empirical evidence from the US also demonstrates that, with climate change, post-re weather conditions show reduced capacity for forest recovery 71 . The recent large bush res affected Eucalypt forest communities that are resilient to high severity wild re 72 , but less is known about their response to high frequency, high severity res. Increased re frequency is expected to promote shrub recruitment and shift ecosystems away from tree-dominated landscapes in Australian alpine environments 17 and enable encroachment of wild re into re refugia 73 . Species in typical re refugia, including many in the genus Melaleuca, were much more likely to burn in the 2019-2020 anomalous re season than the previous focal re season. The encroachment of wild re into less reresilient vegetation communities increases as extreme drought dries fuels below critical thresholds within re refugia 73 . The recovery of vegetation from these res in already fragile ecosystems will depend on the severity and frequency of re and the post re climate. Post re climate can promote recovery or facilitate additional res that might push these systems beyond resilience 40 .

Conclusion
Using remotely sensed active re data, we rapidly mapped and quanti ed estimates of burned area in the anomalous 2019-2020 bush re season in eastern Australia. Compared to a previous bush re season (2012)(2013), we show that the res were much larger and more concentrated across the landscape, burning much larger swaths of forest than previous years. This study highlights the substantial impact on foraging habitat of grey-headed ying foxes and identi es the most affected roosts, which will be important to monitor in the coming months and years. We also identify key areas of unburned habitat that need to be preserved to limit loss of biodiversity. The severity of the 2019-2020 re season has been linked to climate change and such seasons are expected to become more frequent in the coming years. Understanding the response of vegetation and wildlife to these extreme events will improve our ability to help build resilience into these systems as climate change alters the frequency and intensity of res in Australia and other ecosystems globally.

Methods
This study focuses on the eastern Australian states of Queensland (QLD), New South Wales (NSW), Australian Capital Territory (ACT) and Victoria (VIC) (Figure 2a). This region of Australia has the nation's highest human population density, highest biodiversity and largest tracts of remaining forest habitats 53 . Acquisition and processing of spatial re and forest data We retrieved re incidence data from NASA's Fire Information for Resource Management System (FIRMS) Visible Infrared Imaging Radiometer Suite (VIIRS) Active Fire and Thermal Anomalies for September 1 to March 1 of each re season from 2012 to 2020 74 . For example, we used VIIRS data from September 1, 2012 to March 1, 2013 to quantify burned area in the previous focal re season. VIIRS Active Fire and Thermal Anomaly data are collected every 12 hours and anomalies are based on a dual gain high saturation temperature mid-infrared channel, band M13, and single gain thermal infrared channel, band M15 75 (see SI). An algorithm classi es pixels as thermal anomalies or "hotspots" 75 . We converted re anomaly points to a raster to generalize burned area at the highest spatial resolution of the FIRMS data in Australia, 440 meters. We also describe analysis at a coarser spatial resolution to address variation in scan and pixel size in the SI. We additionally calculated total burned area with the incorporation of small 'unburned' gaps enclosed in the raster to prevent underestimating burned area missed by satellite passes (SI). We used the resulting raster of burned area in each year to summarize key metrics: including total burned area, number of unique res (as number of unconnected polygons), proportion of area burned by 'moderately sized' res (>100 ha), and proportion of area burned by 'mega-' res (>10,000 ha).
Spatial and temporal data on forest cover comes from a global dataset of 30m resolution tree canopy cover estimates derived from LANDSAT imagery 76 . For the purposes of this analysis, we de ne forested areas as tree cover of 20% or more in a 30m x 30m pixel. Tree cover in 2012 and 2018 was calculated based on gain and loss products provided in the Global Forest Change dataset v1.6 (see SI for details; original dataset in 76 ). We used 30m resolution data from 2009-2018 to examine forest fragmentation and forest loss in this period. Forest patch sizes, edges, and core-area indices were calculated for forest patches (considered when pixel edges, but not vertices, are adjacent). To measure the degree of fragmentation in forests across the region, we estimated core area by creating a 100m inner buffer for each forest patch >100,000 ha in 2012 and compared these metrics to the largest remnant forest patch in 2018 (2018 was used as a best approximation for forest cover in 2019 at the start of the re season). We then resampled the forest area using maximum combined area selection to 440m resolution to compare with re extent, again using a 20% cut-off for forest and non-forest pixels (see SI).

Grey-headed ying fox roost responses to re and habitat loss
To examine how wild res may impact wildlife populations, we utilized longitudinal monitoring data of ying fox (Pteropus species) roost sites in the states of Queensland and New South Wales. Roost counts are conducted approximately quarterly, location and occupancy data (count estimates in Queensland and presence/absence in New South Wales) are publicly available through the National Flying Fox Monitoring Program 45 . The ying fox roost surveys were subset into three datasets to quantify the area of habitat burned from both the previous focal re season (FF1) and the anomalous re season (FF2) re years and to evaluate 2012-2013 re season responses in southeast Queensland (FF3) ( Table 1).
For each roost, we quanti ed total foraging habitat, winter habitat, and burned area within a 50km radius to account for maximum foraging distances (average 20-30km) from individual roosts 43,44 . We de ne winter habitat as areas that contain at least one of ve species (Eucalyptus robusta, E. tereticornis, E. siderophloia, Melaleuca quinquenervia, and Banksia integrifolia).. While there are 13 grey-headed ying fox diet species that can ower during winter, the selected ve species represent the most annually reliable and productive species during resource bottlenecks in winter months 42 . To evaluate differences between re seasons, we only quanti ed habitat around roosts that were occupied in the preceding ve years - To explore evidence of re impacts on roost occupancy, we focused on modelling roost dynamics in the four seasons preceding and following the 2012-2013 wild re season (from September 2011 to March 2014) ( Figure S4). Observations recorded during the re season (September 2012-March 2013) were excluded. Presence and absence of grey-headed ying foxes were t with logistic mixed effects models, with roost identities were included as random effects to account for temporal autocorrelation. The dataset was truncated to only include camps that had a minimum of four observations in pre-and postre periods, resulting in 92 roosts in Queensland, referred to here as dataset FF3 (Table 1). Univariate and multivariate logistic regressions were evaluated with several predictor variables described below, including: presence of another species of ying fox, area of grey-headed ying fox habitat, area of winter habitat, distance to closest burned area, proportion of habitat burned, proportion of winter habitat burned, latitude, and season (Tables S1, S4). Logistic mixed effects models were created and evaluated in R version 3.6.1 using lme4 78,79 .

Declarations
Author Contributions K.B., C.L.F., and N.B. designed the research; K.B. analyzed re extent; C.L.F analyzed forest fragmentation and impact of re; K.B. and C.L.F analyzed impact of re on grey-headed ying fox habitat; all authors contributed to writing the paper.

Data availability statement
Data analyzed in this study are available in public repositories (see Methods) and the scripts to analyze these data will be made available on GitHub (https://github.com/cfaustus/aus20192020_ res_ffcurrently private, will be made public).

Funding
This work was funded by the National Science Foundation (grant no. CNH-L: 1716698), the Defense Advanced Research Projects Agency PREEMPT program (Cooperative Agreement #D18AC00031), and the Huck Institutes of the Life Sciences at The Pennsylvania State University. The content of the information does not necessarily re ect the position or the policy of the U.S. government, and no o cial endorsement should be inferred.