Severe, large-scale bushfire threatens metapopulation function of quokka (Setonix brachyurus) in south-western Australia

Study area

South-western Australia has a relatively intact fauna compared with other parts of Australia, having experienced unusually low rates of mammal extinction in the past 200 years (McKenzie et al. 2007). The area is characterised by high annual rainfall (800–1400 mm), dense vegetation and the occurrence of moist and stable ecosystems that have remained virtually unchanged for millions of years (Hopper 1979; McKenzie et al. 2007). These moist and stable ecosystems are among those most likely to be protected from fire by virtue of their differential moisture from the surrounding landscape (Robinson et al. 2013; Meddens et al. 2018).

The Northcliffe fire occurred in a large area of forest between the towns of Northcliffe and Walpole, Western Australia (Fig. 1). Forests in the region are dominated by jarrah (Eucalyptus marginata), marri (Corymbia calophylla) and karri (Eucalyptus diversicolor), with the karri growing up to 80 m tall. About 96% of the area that burnt in the fire is National Park or State Forest, vested in the Conservation Commission for the purpose of conservation and multiple use. The remaining area is private property. This project focused on the areas of the Northcliffe fire that are vested in the Conservation Commission, as well as a 1 km strip of conservation land adjacent to the fire scar (Fig. 1).

Fig. 1. 

Location of the study area in the southern forests of Western Australia, within the Northcliffe fire area.

Site selection

In total, 126 sites were surveyed for quokka occupancy and abundance. Site selection was based on:

Quantifying abundance and occupancy

Counting fresh faecal pellet groups has been demonstrated to be an effective method for rapidly determining a quantitative estimate of abundance for quokkas in the southern forest (Bain et al. 2014). The technique is widely used by land managers in this region due to the ability for the method to be applied at large spatial scales with a result that is highly comparable to the outcomes of more intensive methods, such as mark–recapture techniques (Bain et al. 2014).

At each site, two transects 1 km in length were surveyed for quokkas, and the occurrences of all fresh faecal pellet groups (deposited overnight) within 2.5 m of the transect were recorded (Bain et al. 2014). The transect surveys were repeated twice within a 2-month period for each site. Surveys were undertaken between February and March in 2016, 2018 and 2020.

Habitat assessment

A suite of habitat variables was measured based on previous work (Bain et al. 2015, 2016, 2020), including: number of vegetation layers (vegetation structure); understorey density; canopy cover; density of woody debris; sedge cover; grass cover; degree of canopy scorch/defoliation (fire severity); area damaged by feral pigs (Sus scrofa); distance to unburnt canopy; distance to fire edge; and distance to nearest creek line (Table 1). The variables were recorded every 200 m along each transect and averaged for each site for each year of the survey.

Analysis

To reduce the number of habitat variables to a suitable level, the selection procedure of Hosmer and Lemeshow (2000) was used with a stricter cut-off of P > 0.10. Three habitat variables were removed from further analysis as a result. Models were then developed in PRESENCE (Hines 2006) to estimate probabilities of occupancy (Ψ) and detection (P), and to evaluate the influence of covariates for each year of the survey (MacKenzie et al. 2017, Table 2). Encounter histories were generated for each site by converting survey counts to detection and non-detection data. Spatially and temporally replicated transects were treated as occasions. Akaike’s Information Criterion with a small sample size correction (AICc) was used for model selection, and models with delta AICc values ≤2 were considered to have strong support (Burnham and Anderson 2002). Pearson Chi-Square goodness-of-fit tests were used to assess the fit of the occupancy models to the data (MacKenzie and Bailey 2004).

To construct models, we first assessed whether detection probability was constant or varied throughout the survey season (Ψ(.), P(.) vs Ψ(.), P(t)) or with fire severity. Then we added site covariates to the model with the most support. We analysed each year separately based on pooled data for the transects encompassing each site. We opted for separate analyses over a multi-season framework because we were interested in the influence of covariate data on temporal occupancy patterns as the fire area recovered.

Eight site covariates were included in the models: (1) pig damage; (2) vegetation structure; (3) canopy cover; (4) distance to unburnt canopy; (5) distance to edge of fire; (6) distance to closest creek; (7) sedge cover; and (8) fire severity. For the 2020 model set (year five), the additional variable of number of years of pig damage was included.

We estimated the abundance of quokkas in the surveyed habitat by multiplying the faecal pellet count for each site by the detection probability generated by the highest ranked occupancy model. These abundance estimates are not intended to be extrapolated outside of the surveyed area.

Factors influencing occupancy

In all years surveyed, fire severity, sedge cover, vegetation structure and pig damage had the strongest influence on the probability of occupancy of post-fire habitat by quokkas (Table 2). Distance to unburnt canopy and distance to riparian systems were also important variables for the first and third years following the fire. In the first year following the fire, distance to fire edge was important.

Fire severity

The probability of occupancy of habitat by quokkas in all years of the survey was dependent on the severity of fire that the sites had experienced. Sites with some canopy left unburnt were occupied by quokkas more frequently and earlier than areas with scorched or defoliated canopy (Table 3). Sites within the fire scar where canopy had been retained were, on average, occupied within 2.5 years, and by 2020 had an occupancy estimate of 0.85. Sites with a mix of scorched canopy and unburnt canopy were, on average, occupied within 2.8 years, and by 2020 had an occupancy estimate of 0.65. Sites where the canopy had been completely scorched were, on average, occupied within 3.8 years, and by 2020 had an occupancy estimate of 0.64. Sites defoliated by the fire were occupied within 4.7 years, but the occupancy estimate for these sites in 2020 was only 0.24 (Fig. 2).

Table 3. Post hoc means comparison tests for habitat variables found to be significant in predictive models for occupancy of quokkas in the southern forests.

Variable	Year	P	t	Occupied		Unoccupied
				Mean (s.e.)	Maximum	Mean (s.e.)	Maximum
Fire severity	1	0.00	6.2	1.1 (0.2)	4.0	3.1 (0.2)	6.0
	3	0.00	5.5	1.7 (0.2)	6.0	3.4 (0.3)	6.0
	5	0.00	2.9	2.0 (0.2)	6.0	3.1 (0.3)	6.0
Distance to unburnt canopy	1	0.00	4.3	12.6 (14.2)	183.2	1237.9 (204.8)	11 249.8
	3	0.00	4.7	219.9 (79.7)	530.5	1485.1 (274.3)	11 249.8
	5	0.00	3.2	428.7 (121.6)	4826.8	1334.0 (290.0)	11 249.8
Distance to fire edge	1	0.00	4.3	528.9 (148.3)	4334.3	2362.8 (297.4)	12 706.2
	3	0.04	2.1	1291.4 (251.1)	12 706.2	2173.1 (351.8)	11 750.5
	5	0.41	0.8	1549.8 (264.6)	12 706.2	1907.2 (357.3)	11 750.5
Distance to riparian	1	0.00	3.3	57.7 (12.6)	126.7	126.0 (28.8)	2358.8
	3	0.06	1.9	73.9 (8.6)	323.4	148.6 (40.8)	2358.8
	5	0.38	0.9	163.7 (13.2)	541.2	221.9 (74.2)	2358.8
Sedge cover	1	0.00	−4.7	22.7 (1.2)	43.7	15.6 (0.9)	29.4
	3	0.00	−8.9	39.2 (1.4)	59.4	17.6 (2.0)	55.3
	5	0.00	−16.4	44.6 (1.2)	61.3	12.5 (1.6)	45.8
Vegetation structure	1	0.00	−9.3	3.4 (0.1)	4.80	2.1 (0.1)	4.4
	3	0.00	−8.1	3.7 (0.1)	5.1	2.7 (0.1)	4.8
	5	0.00	12.5	4.5 (0.1)	6.8	2.5 (0.1)	3.9
Pig damage	1	0.06	1.8	1.2 (0.5)	22.1	8.6 (2.9)	140.0
	3	0.01	2.6	12.1 (5.5)	150.3	253.1 (93.6)	4604.6
	5	0.00	4.0	3.5 (1.3)	81.9	481.0 (118.1)	4718.2

Mean (standard error), Maximum, t-test and P-values are presented for occupied vs unoccupied sites. N = 126.

Fig. 2. 

The effect of fire severity within the Northcliffe Fire area on habitat availability and occupancy estimates for quokkas in relation to (a) the relative proportion of sites within the fire scar for which each of the fire severity categories was recorded and (b) the average number of years taken for recolonisation of areas with different fire intensities. Occupancy estimates reported (Ψ) are for the fifth year post-fire and have been corrected using the detection probability estimated during data modelling. Sites with more than one category listed contained patches of both fire intensities. Canopy refers to areas where the overstorey canopy remained unburnt during and following the fire. Crown scorch is defined as the browning of leaves in the overstorey canopy that occurred as a result of heat generated from the fire. Defoliation refers to situations where the overstorey canopy experienced a complete loss of leaves during the fire.

Feral pig damage

In total, 47 sites sustained pig damage in the first 5 years following the fire, with damage ranging from isolated diggings on less than 1 m² through to extensive diggings up to 4718 m². Quokkas were present in areas where pigs were active, but sites were occupied by quokkas only where the damage was less than 150 m² (Table 3).

Feral pig activity was greatest 3 years post-fire, with a total of 15 479 m² damaged across 28 sites compared with 766 m² of damage across 26 sites in the first year (Fig. 3). Pig damage detected in the third year was an extension of damage recorded in the first year for 68% of sites. Damage was also recurrent within 77% of sites in the fifth year post-fire (Fig. 3).

Fig. 3. 

The area of feral pig damage (m²) caused by digging and wallowing, and the number of new and recurrent sites affected by pig damage within the Northcliffe fire area in the first, third and fifth years post-fire.

Spatial and temporal occupancy

Prior to the Northcliffe fire, 92 sites within the fire scar were known to be occupied by quokkas, based on high confidence records from the 3 years prior to the fire. Of these sites, 38 were located within 1 km of the edge of the fire and 54 were located more than 1 km from the edge. In the first year following the fire, 17 (18%) of these sites were occupied by quokkas and only five of these sites were more than 1 km from the edge of the fire (Figs 4 and 5). Occupancy estimates increased most between years 1 and 3 for burnt sites both near the edge and in the core of the fire scar (Fig. 4). In the third year following the fire, 54% of burnt sites (core and edge) were occupied, increasing slightly to 59% in the fifth year. Occupancy remained stable over time at the 34 unburnt sites near the fire edge (Fig. 4).

Fig. 4. 

Number of sites occupied by quokkas within and adjacent to the Northcliffe fire area in the first, third and fifth year post-fire. Core sites were internal refuge areas >1 km from the edge of the fire (N = 54), burnt edge sites were within 1 km of the edge of the fire (N = 38) and unburnt edge sites were sites adjoining the fire scar (N = 34). Occupancy estimates (Ψ) presented have been corrected using detection probabilities estimated during data modelling.

Fig. 5. 

Occupancy of sites by quokkas in the first year, third year and fifth year following the Northcliffe fire. The year each site was first occupied by quokkas is shown, as well as significant riparian corridors mapped from aerial imagery and field truthing.

The marked increase in occupancy of burnt sites in the third year post-fire resulted from 33 additional sites becoming occupied, all within 12.7 km from the fire edge, and all within 7 km of sites that had been occupied in the first year post-fire. In the fifth year only five additional sites were occupied, and all were within 5 km of sites that were occupied in the third year. In all cases, newly occupied sites were linked by riparian corridors to previously occupied sites (Fig. 5). Within the fire scar, a total of 41 sites remained unoccupied in the fifth year post-fire and all of these sites were located in the core, southern and western parts of the fire, between 5 and 30 km from the closest occupied site (Fig. 5) in areas that had been severely affected (defoliated) by the fire.

Recovery to pre-fire abundance

Although occupancy estimates provide an indication of the spatial pattern of recolonisation, in many instances burnt sites were occupied by less than five individuals, and for this reason, a measure of abundance was also established for each of the sites based on Bain et al. (2014) and corrected for detection probability. In the first 12 months following the fire, 97 quokkas were estimated to be present within the Northcliffe fire scar (burnt sites only), and by the fifth year post-fire, this estimate had increased to 272 individuals (Fig. 6). It is estimated that approximately 594 individuals were present within the fire scar prior to the fire, based on the number of known occupied sites and the average density of quokkas in this area. Quokka recovery within the Northcliffe fire scar had subsequently reached approximately 46% of pre-fire abundance levels by the fifth year post-fire.

Fig. 6. 

Linear projection of increasing quokka abundance based on data collected in the first, third and fifth year post-fire.

A linear projection based on the 5 years of data suggests that the abundance of quokkas will reach pre-fire levels 12 years post fire (Fig. 6). At year five, a linear relationship was the best fit for the data; however, quokka recolonisation/recovery patterns are likely to be affected by changes in the rate of recovery of the vegetation and the continuing availability of suitable habitat and habitat connectivity.

Spatiotemporal recolonisation patterns

For the first year, few quokkas within the fire scar were greater than 1 km from the fire edge. In fire-prone forests burnt under extreme fire conditions, the presence of fire refugia has been shown to be limited to sheltered parts of the landscape with extremes in topography and wetness (Robinson et al. 2013; Leonard et al. 2014; Berry et al. 2015). In the Northcliffe fire area, the few occupied refugia coincided with deep, mesic riparian systems low in the profile, or perched wetland systems retaining high levels of summer moisture. These areas burnt less intensely than the surrounding forest and contributed to small habitat patches where canopy and midstorey vegetation and tiny pockets of understorey vegetation had been left unburnt by the fire. In most cases, these patches were less than 0.5 ha in size, and in all cases the abundance of survivors was low.

The abundance and spatial distribution of survivors strongly influences post-fire recovery following a major disturbance, in conjunction with the effect of the fire on the distribution of resources that are important to the species and the sensitivity of the species to the disturbance (Robinson et al. 2013; Banks et al. 2017; Meddens et al. 2018). The distinction between recolonisation from in situ survivors and recolonisation from the fire edge as starting points for recovery therefore has implications for understanding ongoing demographic and metapopulation processes. The first 3 years of population recovery were predominantly restricted to topographic refugia. Spatial recolonisation patterns were likely to have been limited by the low availability of refugia within the fire scar and the low abundance of individuals persisting in these. Although recovery from in situ subpopulations was possible, post-fire recolonisation from the fire edge was more likely contributing to the bulk of the observed recovery patterns, given the connection to occupied sites around the eastern and southern fire boundary, where likely source animals were located (Fig. 5).

The presence of fire refugia contributed to the availability of resources for animals residing within and moving through the fire scar. These refugia were even more valuable when considered in the context of the relatively small distances moved from unburnt canopy by quokkas in the post-fire environment. Despite quokkas being able to move significant distances (up to 10 km in a night) in the southern forest (Bain et al. 2020), the average distance from refugia (unburnt canopy) at which quokkas were detected was 13 m (maximum 183 m) in the first year following the fire. These small distances are likely to be associated with perceived safety, stress levels and/or predator pressure, and are consistent with previous studies for this species in the southern forest (Bain et al. 2016). The increase in the average distance from refugia in the fifth year to 429 m (maximum 4826.8 m) corresponds with an increase in the availability of shelter vegetation in the surrounding landscape.

The sites that remained uncolonised were those that had been defoliated during the fire. These sites were still mostly absent of green vegetation in the first year, and the understorey vegetation recovery was slower than in sites that had been less intensely burnt, possibly due to increased soil temperatures and lower availability of seed and lignotuber material in the soil (Auld and Bradstock 1996; Penman et al. 2008; Nicholson et al. 2017). The midstorey and overstorey components of living vegetation were completely lost in these areas, and in the karri forest, these elements of the vegetation structure may take more than 10 years to re-establish (Bradshaw and Rayner 1997).

Vegetation structure is important for quokkas in this region (Bain et al. 2015). Multiple vegetation layers provide quokkas shelter from the elements and protection from predators. Vegetation structure is likely to vary in the post-fire environment as a result of topography, soil temperature, weather and the varying growth rates of floristic species present (Lentile et al. 2007; Littlefield 2019). In the southern forest, the re-establishment of vegetation structure occurs most rapidly in areas with topographic moisture, such as within riparian systems. Riparian systems also provide greater ground-level shelter resources for quokkas in the post-fire environment due to the presence of unburnt vegetation and shading associated with rocks, steep banks and fallen trees. Sedges within these systems are associated with moister conditions and are often the first element of the vegetation to recover following fire, rapidly forming a dense understorey layer that provides increased cover.

The rapid regeneration of a dense sedge understorey layer also reduces the availability of exposed substrate to feeding and wallowing behaviours of the introduced feral pig. The wallowing, digging and feeding behaviours of this species contributed to a total of 26 044 m² of area with altered soil structure, physical disturbance of vegetation that was recovering from the fire and removal of propagules. This disturbance was sustained across multiple years and in most cases was an extension of damage recorded in the first year. This suggests that the first year post-fire was the most important in allowing the pigs access to the freshly disturbed areas.

Landscape-scale management of metapopulation structure

This project has demonstrated the complexity of recovery of a threatened species in an environment that has been impacted by an intense and large bushfire. The rate of recovery of quokkas in this landscape has been slow, and despite 5 years of vegetation recovery, the quokka subpopulation has yet to reach pre-fire abundance and distribution. The disruption to the metapopulation is likely to be a conservation and management issue for at least 12 years post fire and will need careful consideration during planning to ensure that any additional disturbance activities do not contribute to additional habitat fragmentation.

There is pressure for the land manager to undertake prescribed burning surrounding and within the Northcliffe fire scar to ensure that a bushfire of the scale and severity of the 2015 fire does not occur again. Prescribed burning is necessary to reduce the risk of such broadscale and intense bushfires; however, the current management paradigm in the southern forest of block-scale prescribed burning adjacent to bushfire scars within 2–3 years has the potential to be counter-productive to the recovery of the quokka and other species through the exacerbation of habitat fragmentation, displacement of threatened species, loss or disruption of metapopulation function and landscape-level population declines. This can be mitigated through explicit management of important corridors for movement, retention of patches of habitat in a matrix of burnt and unburnt vegetation that creates feeding, refuge and movement opportunities, proactive management of threats such as introduced predators and feral pigs and planning for spatial and/or temporal separation of burnt areas that facilitate effective recolonisation and recovery at a landscape scale prior to introducing additional disturbance.

In addition, prioritising the protection of critical habitat elements (e.g. potential fire refugia and important movement corridors) during a bushfire emergency may help to mitigate the detrimental effects of large fires (Lindenmayer et al. 2013; Robinson et al. 2013; Meddens et al. 2018). Statistical landscape models that predict the occurrence of potential fire refugia from landscape level variables, such as vegetation type, seral stage, fuel loads, climatic conditions and topography, may help land managers to identify and protect areas of the landscape of high conservation value (Mackey et al. 2012; Leonard et al. 2014).

Identification and protection of critical habitat elements, including managing their spatial and temporal configuration, composition and structure, is becoming increasingly important for effective conservation of threatened species under predictions of climate change (Gibson et al. 2010) and the associated predicted increase in frequency of large bushfires (Clarke et al. 2011; Abram et al. 2021). In this context, there is also an increasing need for a formalised approach to fire management and species conservation across the continent. In particular, there is a need for collaboration and consistency in fire management for ecological outcomes, the spatiotemporal management of habitat elements and the approach to monitoring and evaluation of management effectiveness across the geographical extent of those groups of species likely to be most vulnerable to fire regime effects.