2.1 Study region and the 2019/2020 megafires

The study region for this paper covers the Victorian extent of the 2019/2020 Australian megafires as of 20 April 2020 (~1.5 million ha burn extent; Figure 1). The region has a temperate climate and consists of large tracts of contiguous forest, woodlands and alpine areas interspersed with patches of agricultural land. Wildfire is a regular and periodic disturbance affecting the region, with typical fire return intervals for the region generally varying depending on the part of the region and the forest type considered—from every 5–20 years in lower elevation dry eucalypt forest to a fire return interval over 100 years in more mesic eucalypt and rain forests (Murphy et al., 2013). Prior to 2000, the rolling 20-year average annual wildfire area oscillated around 100,000 ha per year (DELWP, 2020). As of 2020, this has tripled to approximately 300,000 ha (DELWP, 2020). This reflects the substantial amount of fire activity during the 2000–2020 period, during which there was a cumulative total of 6.2 million ha burned across eight major (>150,000 ha) wildfires; a quarter of this 20-year total is represented by the 2019/2020 fire season (DELWP, 2020).

In the summer of 2019/2020, Australia experienced an unprecedented fire season (Collins et al., 2021), with more than 6 million ha of temperate broadleaf forest and woodland burnt across the southeast of the continent (i.e., ~21% of this biome; Boer et al., 2020). Three large wildfire complexes accounted for most (>90%) of the area burnt in Victoria during the 2019/2020 fire season (Figure 1). The initial ignition for the large complex in the east of the state occurred in mid-November, with the two complexes in the north and west of this area igniting in late December (29 and 30, respectively). The fires burnt under a diverse range of weather conditions, including periods of severe—extreme fire weather, reaching a cumulative area of 1.5 million ha in Victoria, before being declared contained in late-March 2020. The fires made some large, rapid runs in native forest (e.g., >10,000 m hr⁻¹) under extreme fire weather conditions.

The megafires that affected eastern Australia were largely caused by record low fuel moisture levels, driven by an extreme and prolonged drought, coupled with regular occurrence of severe fire weather (Nolan et al., 2020). The drought conditions experienced were sufficient to facilitate fire encroachment into moist plant communities that typically provide fire refuge (e.g., rainforest, see Collins et al., 2019). Consequently, severe and potentially irrecoverable impacts are expected for many fire-sensitive plant communities of national and global significance (e.g., Gondwana Rainforest World Heritage Area; Nolan et al., 2020; Gallagher et al., 2021).

To undertake our analyses, we used the mapped fire extent—polygons of each fire boundary—to represent the full extent of the 2019/2020 fires (DELWP, 2020, Table S1). We also used fire severity mapping, generated using Sentinel-2 satellite imagery and a random forest classification procedure implemented in Google Earth Engine (full details of fire severity mapping method are outlined in Collins et al., 2020), to understand the spatial variation in fire effects on biodiversity (Figure 1; Table S1). For the purposes of this paper, we define ‘high severity fire’ as the two highest fire severity classes, which include areas where the canopy foliage has been either largely consumed or scorched leaving very little green foliage (>80% canopy scorch, Figure 1).

2.2 Quantifying fire overlap with species distributions

Many species that have had a large proportion of their habitat burnt by a megafire may be at a higher risk of local extinction, particularly those with restricted ranges, and/or are already threatened. To identify the species most affected (i.e., high proportion of their range within the fire extent), we quantified the spatial overlap between the 2019/2020 bushfires and native species habitat. While we recognize that the species in this study have highly variable responses to individual fires and the total regime, and responses are influenced by environmental context (Kelly et al., 2017), we consider this approach, paired with information on species’ fire vulnerability, appropriate for rapid identification of species of conservation concern. To do this, we used an existing set of species distribution models across Victoria for ~4,200 taxa of terrestrial vertebrate fauna and vascular plants (Table S1). These species distribution models were built using random forest classification with 26 environmental predictors, including satellite imagery, bioclimatic, terrain and soil-related variables and provide a measure of relative habitat suitability (probability of occurrence of suitable habitat) for each of the taxa (Liu et al., 2013, 2016). These models were previously built to inform various spatial conservation planning decisions in Victoria (Liu et al., 2013). Next, we calculated the percentage of each species’ habitat that fell within the boundary of the 2019/2020 bushfires. As the species distribution models provided a measure of relative habitat suitability, we weighted the percentage overlap by modelled habitat quality. We calculated the percentage of each species net habitat that was burned as:

where H_si is the habitat suitability value in a pixel i, and B_i indicates whether pixel i was burned (1) or not (0). Dividing the net amount of suitable habitat in the fire extent by the state-wide amount gives the proportion of state-wide net habitat burnt by the fire, therefore providing an indication of the level of impact of the fire on each species while accounting for relative suitability of the habitat burnt. Species were then categorized based on their conservation status and broad taxonomic/functional guild (Figure 2). We then repeated this analysis by calculating the proportion of habitat burnt by high severity fire (i.e., where B_i =1 only when a pixel was burnt by high severity fire), which allowed us to assess which species were affected most acutely by the fire (Figure 2). We also explored if species were disproportionally affected in areas of high-suitability habitat by calculating the proportion of habitat burnt by fire using high-threshold habitat distribution models (i.e., applying a new threshold to the habitat distribution models with only pixels with the highest values were retained; Supplementary Material).

2.3 Identifying species of most immediate concern and candidate recovery actions

We identified species with over 40% of their state-wide modelled distribution within the fire extent to identify which species might require short-term recovery actions immediately post-fire. We used 40% as a threshold because we considered that a species with this much of its distribution affected would likely have experienced a substantial impact due to the megafire. This initial list of species was then appended based on expert advice. This included the addition of species from taxa that did not have distribution models, including freshwater aquatic fauna and terrestrial invertebrates, where fire impact was assessed based on expert knowledge of species’ extent and fire sensitivity. In addition, the list of flora species of concern was further filtered using trait databases which included life history information (e.g., germination strategy, generation time, dispersal ability), genetic status (from the Genetic Risk Index developed by Kriesner et al., 2019) and vulnerability to fire (Tables S1 and S3). Species of immediate concern were those that were considered likely to experience local extinctions (indicated by a high proportion of habitat within the fire extent and relatively high vulnerability to fire) and species where those losses would have a strong negative impact on the long-term overall persistence of the species (i.e., those with limited or patchy distributions). See Figure S1 for the detailed criteria and process used.

Next, species trait databases were used to identify a range of short-term (0–6 months post-fire) recovery actions (Tables S1 and S3). Trait data were used to identify the potential mechanisms through which the megafire affected species of concern, and therefore the potential threats faced by these species and the types of interventions potentially required (Table 1). For instance, a trait database on species vulnerable to predation by feral cats and foxes was used to inform which species may be at risk from increased pressure from invasive predators post-fire, and therefore may benefit from fox and cat control. The resulting database outlined the number of species of concern which would potentially benefit from post-fire intervention (Table 1). These data were used to help identify candidate actions that may be required in the short-term post-fire. Where there was spatially explicit information available, actions were included in a spatially-explicit prioritization to inform decisions as to where best to implement them across the landscape.

2.4 Identifying cost-effective landscape-scale actions within the fire extent

To identify the most cost-effective location-specific actions to undertake within and adjacent to the burned area, we used a modified implementation of a dynamic spatial conservation action planning (SCAP) tool (Thomson et al., 2020). SCAP is a raster-based, landscape-scale tool developed for the State Government of Victoria to inform strategic conservation investment and planning. SCAP combines distribution models for 19 threats and the ~4,200 species (terrestrial fauna and flora), with structured expert elicitation and machine learning techniques to generate species-specific maps of the expected benefits (change in local persistence probabilities) of management actions (single and in combination) across Victoria. Total indicative costs of each action are also calculated at each location over 50 years. These inputs are then integrated in a modified Zonation conservation planning framework (Lehtomäki & Moilanen, 2013) to identify spatially explicit actions that collectively provide the most cost-effective outcome to minimizing the risk of species declines in Victoria over the next 50 years. A priority ranking of actions is produced across the landscape by iteratively removing actions that are the least cost-effective, until all actions have been removed. The last actions to be removed provide the highest return on investment for conservation funding, defined as the net change in summed persistence probabilities for all species per unit of cost action (Thomson et al., 2020). For development and implementation details of SCAP, refer to the Supplementary Material for a summary and to Thomson et al. (2020) for the full methods.

Immediately following the 2019/2020 Victorian bushfires, we modified SCAP to incorporate the impacts of fire and aid the state government's biodiversity response by identifying and prioritizing management actions that may benefit surviving or recolonizing native species. Fire impacts were incorporated into SCAP in two ways: (a) by reducing the expected net habitat available to each species in the absence of management actions by the proportion of their state-wide habitat burnt by severe fire and (b) by increasing the expected benefits of some actions within the fire extent to all species. The first adjustment effectively assigned greater net benefit to actions that benefit species with proportionally more burnt habitat, on the assumption that these species will have a greater reduction in state-wide persistence probability in the absence of management actions. It promoted any actions expected to benefit affected species, including actions outside of fire-affected areas (e.g., in unburnt areas adjacent to burnt areas and the fire perimeter).

The second adjustment was spatially explicit and assumed that many actions became more beneficial (lead to larger absolute increase in local persistence probability) in fire-affected areas than they were pre-fire, but that increase was inversely proportional to fire severity. In unburnt areas within the fire extent, the benefits of relevant actions (habitat protection and all feral animal and weed control actions) were assumed to be 50% higher, on the basis that it was now more beneficial to control threats in potential refugia within the fire extent than in those areas pre-fire. The increased benefit within the fire extent declined linearly with increasing fire severity, with benefits 40%, 30%, 20% and 10% higher in low, moderate, high and very high (canopy burn) severity categories, respectively. The spatial adjustments to expected benefits were applied to all species except hollow-dependent fauna. Expected benefits of post-fire management actions to hollow-dependent fauna were set to zero in severely burnt ash forest (and in clear fell and seed tree timber harvesting coupes), where hollow availability is expected to be severely limited over the next 50 years due to the effects of severe fire (Gibbons et al., 2000; Lindenmayer et al., 2016).

We acknowledge that these adjustments are somewhat arbitrary and that the benefits of post-fire management actions will in reality be highly variable within and among species. Nevertheless, the adjustments represent a general consensus that (a) loss of suitable habitat increases extinction risk (Crooks et al., 2017; Hanski & Gyllenberg, 1997), (b) native species may be particularly vulnerable to the impacts of feral animals (predators or herbivores) and weeds after fire (Hradsky, 2019) and (c) unburnt or less-severely burnt areas within fire-affected areas may be important refugia and sources of recovery following fire. These adjustments allowed a rapid assessment of the areas where post-fire management actions were mostly likely to be beneficial to biodiversity overall, and most cost-effective, accounting for the distributions of thousands of species. More detailed work is underway to refine estimates of the benefits of post-fire management, including more explicit expert elicitation, population modelling and, most importantly, field monitoring of actual outcomes from current actions across the state.

3.1 Quantifying the overlap of fires with biodiversity values

Our analyses indicate that the megafires overlapped with over 40% of the modelled habitat of 346 species in Victoria, including 45 listed as threatened under the Victorian Flora and Fauna Guarantee (FGG) Act 1988 (Figure 2). One hundred and two of these species had more than 40% of their modelled habitat affected by high severity fire. This included species that had almost all their range burnt, including betka bottlebrush (Callistemon kenmorrisonii), which had 93% of its modelled habitat within the fire extent. Although some species had relatively low amounts of modelled habitat in the fire extent, like the spotted tree frog (Litoria spenceri; 22% of modelled habitat in fire extent, 13% of which was burnt at high severity), the areas which were burnt included important habitat (e.g., key breeding sites).

For additional descriptions of other biodiversity values affected by the 2019/2020 megafires, such as ecological vegetation classes and areas of high biodiversity value that were not explicitly considered in this framework, refer to the Supplementary Material.

3.2 Identifying species of immediate concern and candidate actions

Species of immediate concern included 154 flora and 67 vertebrate fauna species, including 38 species listed under the Federal Government's Environment Protection and Biodiversity Conservation (EPBC) Act 1999 and 74 species listed under the Victorian Flora and Fauna Guarantee (FFG) Act 1988.

Our trait analyses for these species identified 11 threats related to the megafire and 21 candidate recovery actions that could be enacted to ameliorate these threats (Table 1). In particular, direct mortality of individuals, loss of critical habitat features and a change in importance of other populations were identified as threats for more than 60 species of concern (Table 1). Similarly, loss of food resources was expected to impact at least 39 species of concern. Impacts of invasive species through predation, herbivory and competition were also identified as important threats requiring action (Table 1).

3.3 Identifying cost-effective landscape-scale actions within the fire extent

Before the fires, the SCAP identified that controlling deer, predators (feral cats and foxes) and weeds were expected to be highly cost-effective actions across much of the fire-affected region, relative to all other actions across the whole state (Figure 3a,b). Feral horse control also ranked as highly cost-effective in many alpine areas (Figure 3c). Our preliminary post-fire analysis, based on initial expert advice, suggests that these actions have become even more beneficial (and hence cost-effective) in fire-affected regions (Figures 3b and 4), because feral animals and weeds may severely inhibit post-fire recovery for many native species. For example, deer control (~800,000 ha), weed control, cat control and fox control (all ~600,000 ha) and pig control (~500,000 ha) are likely to be cost-effective over large tracts of the burned area. Importantly, SCAP also identifies where within and adjacent to the burned areas these actions are likely to be most cost-effective, depending on where the biodiversity values represented will benefit most per unit cost (Figures 3c and 4). For example, the unburned areas within and adjacent to the megafire extent and low severity burned areas were identified as high priority for action (e.g., deer control, weed control and predator control) according to the analyses.

4.1 Application to 2019/2020 Black Summer

At least 346 species had more than 40% of their state-wide habitat burnt, and some of these species had a much higher proportion of their habitat burnt. For example, over 87% of the nationally vulnerable large brown tree frog's (Litoria littlejohni) state-wide modelled habitat was burnt, 47% at high severity. The majority of species heavily affected by the megafires were flora, including the nationally endangered betka bottlebrush which had almost all of its state-wide modelled habitat burnt (93%; 71% at high severity). High severity fires have had substantial conservation implications for individual species. For example, a 2014 megafire in California, North America significantly reduced the occupancy of the spotted owl (Strix occidentalis occidentalis; Jones et al., 2016). Similarly, a high severity fire in Tasmania initiated the collapse of a population of a long-lived conifer (Athrotaxis cupressoides) by triggering recruitment failure (Holz et al., 2015). Therefore, identifying and implementing appropriate short- and long-term conservation responses for heavily affected species is vital to assist their recovery.

We used the impact assessment in conjunction with existing spatial conservation planning tools and threat databases to rapidly identify species-specific actions and landscape-scale actions within and adjacent to burned areas. We identified 21 candidate recovery actions that may reduce the likelihood of extinction for fire-affected species, that ranged from species-specific (e.g., translocation, seed banking) to landscape-scale (e.g., invasive species management, protection of unburned areas). Species-specific actions aimed to both increase survival of populations within the fire area and to spread the risk of future megafires across multiple populations. By using these complementary approaches (SCAP and impact assessment), we were able to ensure the post-megafire needs of all affected species were considered in the planning process.

Some potential species-specific actions, however, required further work to assess the need and feasibility for each identified species. For instance, 66 species were identified as being vulnerable to direct mortality of individuals post-fire, for which emergency extraction of surviving individuals was identified as a potential recovery action. Emergency extraction, however, is a conversation tool only appropriate for species where a large proportion of habitat has been burnt and individuals or populations are unlikely to persist post-fire. Therefore, it is unlikely to be beneficial for the large majority of the 66 species identified. Further assessments were undertaken involving consultation with species experts and land managers to determine if emergency extraction was needed and was undertaken for the eastern bristlebird and multiple freshwater fishes. Some of these species-specific actions have been beneficial in previous megafires. For example, after the 2009 Black Saturday fires in Victoria, a number of freshwater fishes were extracted and placed in temporary housing to prevent the loss of entire populations to post-fire debris flows (Ayres et al., 2012). In North America, over US$3 million is spent annually on emergency reseeding of forest ecosystems following megafires (Peppin et al., 2011). While large conservation gains can be made with actions tailored to species highly affected by megafires, our analyses demonstrate that species-specific actions need to be undertaken in combination with landscape-scale actions that benefit multiple species.

We identified locations within the fire-affected area where invasive species management (deer, red foxes, feral cats, weeds and feral pigs) would be most cost-effective. For instance, deer, horse, weed and predator control were identified as high priority actions in areas which experienced low severity burns or that were adjacent to the fire extent. This is unsurprising as unburned patches or patches burned at low severity are important refuges for fauna post-fire, particularly for species vulnerable to predation by foxes and feral cats or modifications to habitat structure caused by feral deer and horses (Nimmo et al., 2019; Shaw et al., 2021). These landscape-scale actions will be essential to assist recovery of vulnerable native species post-fire and reflect existing literature detailing the importance of invasive species management post-fire (Brown et al., 2016; Hradsky, 2019). In addition to prioritising actions post-fire, the spatial conservation action planning tool guided threat management prior to the fire, which is likely to further increase species' chances of recovery. For example, implementing ongoing landscape-scale predator management programs before fires occur will reduce the risk of predation for vulnerable mammals in burned landscapes, as fewer predators may be present in areas under sustained control (Hradsky, 2019). Similarly, native mammals in northern Australia recovered more quickly from high severity fire where invasive herbivores had been removed (Legge et al., 2011).

4.2 Pre- and during-fire planning tools

Maintaining up-to-date, spatially explicit biodiversity data can also help to prioritize species populations for specific management actions before or during extended fire events, and as these priorities evolve. This requires integrating known and estimated genetic, trait and important population data to identify feasible candidates for intensive emergency management or extraction and plan for operational requirements such as access routes, capture and transport methods, and receiving centres. Given these complexities, in future, species could be identified prior to predicted high-risk fire seasons and prioritized for active protection from fire, or potentially extracted earlier to reduce the operational risks. The integration of weather forecasting with forest fuel dynamics and fire behaviour modelling has provided fire suppression agencies with the capacity to model risks to life and property while a wildfire is in progress (Duff et al., 2018). This could be used to also model the risks of ongoing fires to biodiversity and justify targeted fire suppression efforts in high biodiversity value locations and/or identify species that may require emergency extraction.

Climate change means that conservation managers now need to routinely build disturbance into their decision support tools. For example, future decision support products need to explicitly integrate what we know of disturbance history (e.g., previous fire extents and severity, and related disturbances associated with timber harvesting and clearing) when predicting species distributions, population viability or identifying cost-effective actions (Nitschke et al., 2020), especially for actions with costs and benefits that might be different in burned landscapes (e.g., invasive species control).

4.3 Future development needs

Key to informing rapid analyses of the impact of megafires on biodiversity is having a sound information base from which analyses can be performed, and building this can be done now by conservation managers to better prepare for future megafires. This includes collated spatial information on where biodiversity is distributed (e.g., species distribution models, polygon-based representations or even occurrence records), trait databases that explicitly link fire vulnerability, other proximal threats, and potential management actions, and, where practicable, estimates of the relative effectiveness of conservation actions post-fire. Without these data a priori, quantifying the possible effects of fires on biodiversity and identifying species of concern and candidate management actions is difficult, particularly in an emergency setting. Once collated, these data and information can also be applied to other conservation decisions and challenges. For example, the SCAP tool used in this analysis is applied to a range of decision contexts, including forest management, invasive species control and threatened species conservation. To that end, we outline key datasets and information products that can be built by conservation managers in fire-prone biomes across the globe to assist in preparing conservation responses to future megafires (Table 2). While our framework uses all of these datasets, application of a subset (e.g., mapped distributions of biodiversity values, trait data) can help to guide conservation responses to megafires and other contexts (Bosso et al., 2018; Legge et al., 2020).

Species’ vulnerability to fire and expected responses to conservation actions are important pieces of information for identifying post-megafire actions. To facilitate our approach used here, we had to apply existing fire vulnerability data and generic rules of thumb of the benefit of conservation actions in burnt landscapes, in lieu of more robust data. These represent important uncertainties in our approach and avenues for future research; in particular, there are few empirical studies of the relative benefit of conservation actions in burnt landscapes (e.g., invasive species control, supplementary feeding, emergency extraction). The benefit of invasive species control for some species is a key source of uncertainty in the SCAP tool (Thomson et al., 2020) and is likely to be even more uncertain in a post-megafire context. Similarly, species responses to fire can vary substantially depending on context and fire severity (Fontaine & Kennedy, 2012; Kelly et al., 2017), and there are considerable gaps in the knowledge of species’ responses to megafires. To address this, we are currently quantifying experts’ uncertainties about the benefits of post-fire actions, which will be used in value-of-information analyses to prioritize future research into post-fire management (Thomson et al., 2020). New data currently being collected, including further expert opinion, population modelling results and field monitoring of actual outcomes, will help to provide estimates of species persistence in burnt areas with and without various management actions, accounting for both fire history and predicted future fire regimes.

Our framework will also be further advanced by a better understanding of the costs and benefits of the more bespoke actions in Table 1, which are not currently represented in the SCAP tool. This is an area of active improvement and such estimates will inevitably be uncertain, and those uncertainties should be quantified and incorporated into decision making and research prioritization. Future responses to megafires can be better targeted by addressing these knowledge gaps, and integrated monitoring programs that evaluate the success of post-fire actions for a broad range of species will provide important information in improving conservation responses (Southwell et al., 2019). The dynamic nature of SCAP that allows it to be rapidly updated with new data demonstrates that spatial conservation planning tools are valuable assets for informing responses to megafires, as well as other landscape-scale perturbations.

Moving towards representations of currently occupied habitat or temporally dynamic habitat suitability models, rather than distribution models informed by a-temporal covariate data, will help guide post-fire conservation actions to locations that will have the greatest benefit. This is particularly true for species experiencing ongoing declines and/or whose abundance is mediated by disturbance (Burns et al., 2020), as well as actions that are highly influenced by site-scale disturbance history. These could then be used to inform dynamic maps of important ecological refuges (Reside et al., 2019), with both short-term (e.g., remaining long-unburnt habitat) and long-term (e.g., climate change refuges) considerations to facilitate better-targeted actions.

4.4 Concluding remarks

Climate change is intensifying multiple stochastic disturbances and threats to biodiversity that overlap in time and space, and altered fire regimes are emerging as a key threatening process for species across the globe (Kelly et al., 2020). Conservation managers increasingly require the capability and capacity to quantify the effects of actions, rapidly re-analyse options, and to redirect operational effort adaptively. Climate change requires managers to balance implementing rapid and reactive emergency responses to significant individual megafires, and considering how the consequences of recent and future events should influence long-term conservation planning. In the face of future megafires, rapid implementation of conservation actions may be required to give biodiversity the best chance of recovery. Standardized frameworks such as ours provide conservation managers with an approach that can be applied to future megafires in any ecosystem globally.