High fire severity and frequency threaten the persistence of a widespread obligate-seeder Banksia in south-eastern Australia

Field survey design and sampling

The study area comprised low elevation areas in south-eastern Victoria, with a mean annual rainfall of approximately 800 mm, that burned in the 2019–2020 wildfires (Fig. 1). Sites were restricted to mixed eucalypt forests with a diversity of life-forms in the understorey (Tall Mixed Forest: Cheal 2010), which support the largest area of B. cunninghamii habitat in Victoria (DELWP 2022a). This minimised the effects of differing environmental conditions on recruitment outcomes. We selected known B. cunninghamii populations to ensure that there was the potential for regeneration from existing plants. These were a mixture of our monitoring sites for B. cunninghamii cone production between 2014 and 2016 (Muir et al. 2020), and records of B. cunninghamii from the Victorian Biodiversity Atlas (VBA: DELWP 2020b) with reliable locational accuracy (1990–2010) (Supplementary Table S1).

Sites represented fire regime attributes (interval and severity) that affect B. cunninghamii recruitment. Fire severity (vegetation consumption) was used as a proxy for fire intensity (energy output) (Keeley 2009). A two-factor stratified random sample approach was used to measure seedling recruitment of B. cunninghamii at sites that had experienced different fire intervals prior to the 2019–2020 wildfires and/or had been burned at various levels of severity during these fires (Table 1). Sites were chosen with fire intervals ranging from 5 to 37 years, using Victorian government spatial data layers (DELWP 2022a). Fire severity mapping for the 2019–2020 East Gippsland wildfires (DELWP 2022b) was used to identify sites ranging from <20% to >80% eucalypt canopy scorch (three categories), sites with eucalypt canopies burned (one category) or unburned (one category). Areas where timber harvesting occurred after the dates of the VBA records were excluded.

Sampling was carried out in East Gippsland at 14 sites in October 2020 (10 months after the fires) and 11 sites in May 2022 (29 months after the fires) (Table 1). Although some seedling mortality over the summer seasons between the sampling events would be expected (Lamont et al. 2007), this may have been limited by annual rainfall totals being above average across the entire sampling period from 2020 to 2022 (BOM 2022a).

Fire history for B. cunninghamii records was validated on site. The fire severity mapping was verified by checking the degree of scorch or burning of the eucalypt canopies. Where eucalypt canopies had been scorched, we observed that Banksia cones were present (Fig. 2a); whereas at sites where eucalypt canopies were burned, we could find no Banksia cones (Fig. 2b). The preceding fire interval (which determined the maximum age of B. cunninghamii plants) was checked by counting internodes, when they were present, on dead standing plants. Wills (2003) found that this method was accurate for B. marginata to within one year of site age for plants up to 21 years old, and we have confirmed during our previous monitoring that it is reliable for B. cunninghamii for the first two decades after germination.

Fig. 2.

Moderate and high fire severity effects on eucalypt canopies, B. cunninghamii plants, cones and monitoring tags: East Gippsland sites, November 2020. (a) Eucalypts with a mixture of scorched and unburned canopies, and dead B. cunninghamii plants with retained twigs and dried leaves; cone opened after fire with seedlings beneath; metal monitoring tag after fire. (b) Site with eucalypt canopies burned; B. cunninghamii stump burned at high severity; metal monitoring tag melted by fire.

Seedling recruitment was sampled in two ways. At sites where we had previously been monitoring cones on tagged B. cunninghamii plants (Muir et al. 2020), 5–10 circular plots were set up around individual tagged plants that had been killed by the wildfires, or at the coordinates where they had been previously recorded if they had been totally consumed by fire. The use of plots centred on individual plants had been devised to sample cone production on live plants because of the sparse and uneven spatial distribution of B. cunninghamii at our sites (Muir et al. 2020). At former VBA locations, we started at the geographical coordinates and searched 50 m in each cardinal direction, forming a 1 ha search area, and then set up 5–10 circular plots around recognisable dead plants. The plot radius was equal to the average height of the dead B. cunninghamii plants at that site plus 1 m. The rationale for restricting the plot area was based on field observations that almost all seedlings were observed germinating close to cones that had dropped below the adult plant, or below opened cones retained on the stems of the adult plants (Fig. 2a). We assumed wind dispersal was minimal based on these observations, as well as evidence for other Banksia species not being reliant on wind for seed dispersal because of the mechanism by which the woody separator comes out of the opened follicles (Lamont et al. 1991a). Within each of these plots, we recorded the numbers of B. cunninghamii seedlings. Where B. cunninghamii adult plants had been either consumed by fire or reduced to stumps and no seedlings were present, zero recruitment was recorded (Fig. 2b).

Recruitment analysis

Time since previous fire ranged from 5 to 37 years across the 25 sites (Table 1). This represented the range in age of the plants and hence likelihood of their having seed at the time of the fires. Fire severity was categorised as either having the eucalypt canopy scorched (moderate severity) or the eucalypt canopy burned (high severity) (Table 1). It was expected that B. cunninghamii cones would be impacted by fire severity and hence the availability of seed for germination. Presence or absence of seedlings, rather than abundance, in relation to the fire variables was used for the analysis. This was due to the combination of plot data (numerical) and site data (presence/absence) from the monitoring and VBA sites.

A logistic model in a Bayesian framework was used to model whether seedlings were present or absent at site i (X_i), in relation to fire severity and fire interval (Kéry 2010; Korner-Nievergelt et al. 2015). X_i follows a Bernoulli distribution, where X_i is 1 when seedlings are present and 0 if seedlings are absent. The minimum reproductive interval (MRI) for B. cunninghamii was estimated as a threshold parameter within the model. The model gave different values depending on whether the time-since-fire (TSF) was above or below the MRI threshold. The probability p of seedlings being present at a location (μ_i) was estimated under four scenarios: eucalypt canopy scorched and interval below MRI, eucalypt canopy scorched and interval above MRI, eucalypt canopy burned and interval below MRI, eucalypt canopy burned and interval above MRI (Eqn 1).

The recruitment analysis was conducted using the statistical software program R (R Core Team 2022). The Bayesian logistic model was constructed in JAGS using the package R2jags (Su and Yajima 2021). The Bayesian models used weakly informed priors for all probability parameters (pi,i∈1,2,3,4) and a naïve uniform prior [U(0,30)] for the MRI parameter. Weakly informed beta-distributed priors were chosen to reflect the understanding that recruitment is unlikely when the canopy is burned, or the time-since-fire is less than the minimum reproductive interval [β(1,9) prior was used for p₁, p₃ and p₄]; but likely otherwise [β(9,1) prior was used for p₂]. A stronger prior could have been used to reflect the knowledge that recruitment is rare after the canopy is burned or before the MRI is reached, which would have reduced the estimates for those scenarios closer to zero. However, a strong prior may have overwhelmed the posterior (results) given the relatively small number of observations, and we considered it was better to allow the observations to influence the posterior and show the uncertainty in the relationship. Four chains were run through 12,000 iterations with a 2000 iteration burn-in and no thinning. For all analyses, trace plots for each parameter in the models indicated that the Markov chains were well mixed. All parameters had Gelman-Rubin statistic (Gelman and Rubin 1992) values less than 1.05, indicating convergence of the chains and reliable samples for posterior inference.

Spatial analysis

Results from the recruitment analysis were the precursors to the spatial analysis. To identify areas of low survival and recruitment of B. cunninghamii we used the estimates from the Bayesian model for fire severity and time since fire to predict zero B. cunninghamii recruitment.

A habitat distribution model (HDM) was used to represent the maximum geographical extent of B. cunninghamii in south-eastern Victoria (East Gippsland), which was then combined with spatial data on fire severity and fire intervals. The habitat distribution model for B. cunninghamii had previously been developed using ‘a multi-objective regression-tree analysis of plant quadrat data to jointly model flora species using a suite of climate, terrain and remotely sensed environmental variables’ (Muir et al. 2020).

Fire history spatial data in the Victorian Government Corporate Spatial Data Library (DELWP 2022a) was interrogated using R package FAMEFMR (ver. 0.5.0; Amos 2021). We calculated fire intervals between 1960 and 2020, for all overlapping recorded wildfires and planned burns. Fire history spatial data is more comprehensive after 1980, but we wanted to include major fires from 1960.

Fire severity mapping for large fires was used to identify areas burned at high fire severity (FFMV 2016; DELWP 2022b). This dataset presents aggregated historic fire severity classification from 1998 onward for wildfires, so we could map previous fire events as well as the most recent fires. Categories derived from remote sensing were used to determine equivalent classifications used at different times for the highest severity burns (i.e. canopy burned equals high severity burn) (see Fig. S1).

The habitat distribution model for B. cunninghamii in East Gippsland was then overlaid with mapped fire intervals and or areas burned at high severity, to identify all areas with High Fire Severity or with one or more fire intervals of less than 12 years (best estimate from the Bayesian model; Fig. 3). This was done in R (R Core Team 2022) using packages terra and sf (Pebesma 2018; Hijmans 2022). Areas were calculated on a 25 m raster.

Fig. 3.

Probability of presence of Banksia cunninghamii seedlings as a function of time since previous fire (years) and fire severity (eucalypt canopy burned or 20–80% scorched). The points are the median values, and the lines represent the 95% credible interval.

To validate the spatial data, the areas predicted to no longer support B. cunninghamii populations were checked against VBA records (DELWP 2020b) in the years following short fire intervals. VBA records of the species in the years after a fire interval of less than 12 years would indicate that these populations had persisted.

Recruitment analysis

The recruitment model showed that there was a clear difference in the probability that a site would support recruitment based on fire severity experienced in the 2019–2020 wildfires or fire interval prior to these fires (Fig. 3). The minimum reproductive interval (MRI) for B. cunninghamii was estimated to be 11.5 years (95% credible interval from 10.7 to 18.2 years). At sites with a burnt eucalypt canopy (high severity) or where the time-since-fire was below the MRI for B. cunninghamii (11.5 years), recruitment was low (generally <25%). At sites with a scorched (20–80%) eucalypt canopy and a time-since-fire greater than 11.5 years the chance of recruitment was high (generally >75%).

Note that the estimates shown in Fig. 3 for sites below the MRI are above zero even though no site with a fire within the last 12 years had any seedlings. This discrepancy is because we used a weakly informed prior that did not preclude recruitment after high severity fires, because there was not a large amount of data available.

Spatial analysis

Sixty-seven percent (135,380 ha) of the modelled distribution of B. cunninghamii in East Gippsland was mapped with at least one fire interval of 12 years or less since 1960 and/or had been burned at high severity (eucalypt canopies burned) since 1998 (Figs 4 and 5). This comprised 127,069 ha burned at less than 12-year intervals (1960–2020) and 27,476 ha burned at high severity (1998–2020), with an overlap of 19,165 ha. The probability of recruitment is low below either of these thresholds of fire interval or severity.

Fig. 4.

Past fire intervals and severity mapping, superimposed over the extent of the modelled distribution of Banksia cunninghamii in East Gippsland. (a) Blue: areas of B. cunninghamii HDM never burned below a 12-year interval. Beige, tan, brown, red: areas burned below a 12-year interval for the first time between 1960 and 2020. Black: areas burned at high severity since 1998. (b) Blue: areas of B. cunninghamii HDM never burned below a 12-year interval. Beige, tan, brown, red: the most recent fires between 1960 and 2020 where the preceding intervals were less than 12 years, including areas burned below this threshold previously. Black: areas burned at high severity since 1998.

Fig. 5.

Area of B. cunninghamii HDM burned at fire interval of less than 12 years and/or mapped as high severity fire.

Substantial areas were burned in the 2019–2020 wildfires below the minimum reproductive interval identified for B. cunninghamii. Areas burned multiple times below a 12-year interval have a decreasing likelihood of B. cunninghamii persistence. In the 2019–2020 wildfires, 44,404 ha were burned below a 12-year interval that had also previously been burned below this threshold (Figs 4b and 5), compared to 23,000 ha burned below this threshold for the first time in 2019–2020 (Figs 4a and 5).

However, when the dates of VBA records of B. cunninghamii were checked, 44% were still recorded as present after the year that the area was mapped as burned below 12 years (Fig. 6; red dots). This is most likely due to the patchiness of burns that had been mapped as having complete coverage, and consequent survival of some stands of B. cunninghamii prior to the 2019–2020 wildfires. These plants would have been older than 12 years at the time of the fires and hence had accumulated canopy seed banks.

Fig. 6.

Victorian Biodiversity Atlas records of Banksia cunninghamii, compared with fire intervals above and below 12 years (1960–2020). Data sources: DELWP (2020b, 2022a). Grey shading is the B. cunninghamii HDM, and dots are VBA records of B. cunninghamii (presence at one point in time). Black dots represent sites not burned twice in an interval of less than 12 years. Blue dots are sites recorded before a fire interval of less than 12 years. Red dots are sites recorded after a fire interval of less than 12 years.

Fire effects on B. cunninghamii recruitment

Seedlings were only found at sites where B. cunninghamii plants were 12 years or older at the time of the 2019–2020 wildfires. The effects of fire intervals on recruitment outcomes are primarily related to seed resources, i.e. there was limited cone production on plants that were less than 12 years old at the time of the wildfires. The results from this study generally support our previous results on B. cunninghamii cone production in Victoria, where we found that it took 14 years since the last fire for 75% of plants to have viable cones (Muir et al. 2020) and 20 years since fire to reach 80% of seed production (Muir et al. 2014). The shorter time frame in our study reflects the presence/absence nature of the model used, and measurement of recruitment rather than cone production. The model also does not take into account seed bank accumulation or fire intervals needed for peak seedling recruitment.

Our results are in accord with many studies where the relationship between fire interval and the persistence of obligate-seeder Banksia species is well established (Bradstock and O’Connell 1988; Morrison et al. 1995; McCarthy et al. 2001; Wooller et al. 2002; Lamont et al. 2007). Reproduction of these species is dependent on canopy-stored seed and the quantity of seed that develops is a function of the age of a plant since the last fire (Whelan 1995). Fire interval effects on recruitment have also been well documented for serotinous obligate-seeding plants globally (Buma et al. 2013; Kraaij et al. 2013; Fernández-García et al. 2019).

Regardless of preceding fire interval, recruitment of B. cunninghamii was very low in areas of high fire severity where eucalypt canopies were burned (recruitment observed in only one of the eight sites). Lack of recruitment of B. cunninghamii in areas of high fire severity was interpreted as being due to destruction of seeds. In the areas where eucalypt canopies had been completely burned, adult Banksia plants were absent or reduced to stumps and no cones were observed on dead plants or the ground (Fig. 2b).

We could find only one published research article on extreme fire temperatures and/or long residence times resulting in incineration of Banksia cones: a study of B. spinulosa plants burned in the 2019–2020 wildfires, where most cones were reduced to ash and no seedlings were detected (Whelan and Ayre 2022). A recent laboratory study on two species of Hakea and Banksia found that sustained heat for long durations led to significant reduction in seed survival (Tada et al. 2024). A field experiment on a Hakea species measured high seed mortality when temperatures were high enough for shrub crowns to burn (Bradstock et al. 1994). Decreased seed viability in Pinus (another serotinous genus) has also been measured if flame residence time of a crown fire is prolonged (Habrouk et al. 1999; Alexander and Cruz 2012).

There are few studies on the effects of fire severity on Banksia persistence, especially at the very high severity levels experienced in the 2019–2020 wildfires (Morgan and Neild 2011; Whelan and Ayre 2022). Generally, higher fire intensity causes enhanced opening of Banksia cones to release seed (Clarke et al. 2010) and therefore leads to greater rates of seedling recruitment (Bradstock and Myerscough 1981) because seeds are protected by the woody cones from a relatively short burst of high temperature as a fire front passes (Whelan 1995; Huss et al. 2019), whereas low-intensity fires may not cause as much follicle opening (Enright and Lamont 1989; Clarke et al. 2010). Canopy-consuming fires have been associated with lower recruitment of other serotinous genera such as Pinus (Maia et al. 2012; Fernández-García et al. 2019). Our findings suggest an upper threshold past which fire severity reduces rather than increases seed availability and subsequent recruitment of B. cunninghamii.

Germination rates are, of course, affected by factors other than fire interval and severity, such as the amount and timing of rainfall before and after fires (Enright et al. 2014). Rainfall deciles in the 36 months preceding the 2019–2020 East Gippsland wildfires were the ‘lowest on record’ or ‘very much below average’ in areas where B. cunninghamii occurs (BOM 2022b). Drought conditions may have resulted in lower seed availability at the time of the fires, either due to reduced seed production, as has been observed in studies on other Banksia species (Burrows and Middleton 2016) or reduced seed bank size from spontaneous follicle opening and seed release (Enright et al. 1996). Post-fire germination and survival of Banksia seedlings has been found to be dependent on soil moisture levels (Lamont et al. 1991b; Cochrane et al. 2014) and seedling recruitment is expected to be less successful in years where dry weather follows wildfires or planned burns. However, following the 2019–2020 wildfires, there was above-average rainfall between 2020 and 2022 (BOM 2022a) across the whole study area, and where fire interval and severity were appropriate, seedlings were abundant.

Current and future persistence of B. cunninghamii populations

Sixty-seven percent of the distribution of B. cunninghamii in East Gippsland was mapped as likely to have low recruitment (at least one fire interval of less than 12 years since 1960 or a high severity burn since 1998). Given that longer fire intervals may be needed for adequate seed bank accumulation (Muir et al. 2020), the areas where persistence is at risk may be greater. Across the Australian distribution of B. cunninghamii, the East Gippsland populations represent approximately one third of its habitat (Wilson et al. 2022). Furthermore, the 2019–2020 wildfires overlapped with 40% of records of B. cunninghamii Australia-wide (Gallagher et al. 2021). Combined, these findings suggest that this once-common species may no longer be able to persist across a significant part of its former range.

An important caveat to our spatial model of likely recruitment (and hence persistence) is that there is a finer scale of burn heterogeneity than is mapped, meaning that our estimate is an upper bound. A proportion of areas mapped as having been burned at less than 12-year intervals between 1960 and 2020 may have remained unburned due to patchiness of fires within mapped fire boundaries, and thus some individuals in these areas would have persisted prior to the 2020 wildfires and could have reached reproductive maturity. In addition, mapping of high fire severity from 1998 to 2020 is likely to have overestimated the loss of B. cunninghamii and it would have been present in some areas where our spatial modelling predicted plants were not likely to persist.

Regardless of the spatial resolution uncertainties in mapping the persistence of B cunninghamii prior to the 2019–2020 wildfires, 80% of its recorded occurrences in East Gippsland were burned in these fires (Muir et al. 2020), and plants that germinated after this event are not expected to reach reproductive maturity for at least a decade (early 2030s). If fire returns in the next decade to areas burned in 2019–2020, the remaining populations of B. cunninghamii would be under threat of recruitment failure.

Management to protect B. cunninghamii

The extensive and rapid declines of B. cunninghamii and other plant species in the aftermath of the 2019–2020 wildfires require multiple management approaches to maintain their populations and prevent declines towards local extinctions (Gallagher et al. 2021; Le Breton et al. 2022). For B. cunninghamii this includes: protecting regenerating populations from fire until they have accumulated canopy seed banks, planning burns at appropriate intervals and weather conditions to optimise recruitment, adaptive management experiments to test the effects of fire severity on seed cones, conservation of genetic diversity across its range and the possibility of seed banking and restoration.

A key consideration is to manage and monitor exposure to fire of regenerating B. cunninghamii populations. A large proportion of its habitat in eastern Australia was burned in the 2019–2020 wildfires, and plants resulting from seedlings in areas where prior fire intervals and severity were suitable will not themselves bear seeds for at least another decade. Therefore, the most pressing action needed for B. cunninghamii is for all populations burned in 2019–2020 to be protected from fire until at least after 2032. This will enable sufficient seed banks to accumulate for successful recruitment and persistence of populations (Enright et al. 1996, 2014).

In practical terms, this could be achieved by incorporating recovery of populations of B. cunninghamii into planned burning programs by government agencies. Several steps are needed to implement this: locating and mapping all remaining populations; monitoring plants to determine when sufficient canopy seed banks have developed to reliably reproduce after a fire; burning only when most plants in a population have seed cones; delaying burns when forecast weather is hot and dry to maximise the chances of seedling germination and survival; promoting patchiness of burns so that not all plants in a population are burned at once; and ongoing monitoring to follow trajectories of populations.

Ground truthing predictions about the presence or absence of B. cunninghamii from our spatial analysis would be an important step in refining the model for future management use. Adaptive management experiments would be useful to investigate seedling recruitment rates at varying fire severity, including its influence on cone opening to release seed and reduction in ground layer competition.

Conservation of genetic diversity across its range important, as a large proportion of B. cunninghamii populations were affected by the 2019–2020 wildfires. The southern populations of B. cunninghamii (including our study area) are genetically distinct from the northern populations (Wilson et al. 2022), and two geographically defined genetic clusters have been identified in Victoria (Hopley and Simmons 2023).

Collection of seed and ex situ conservation may need to be considered for the species in areas where it fails to recover. Restoration of the species by direct seedling or replanting in areas where it has become locally extinct could also be considered.