- Top of page
- Materials and methods
In this study we used a spatial model to examine the effects of a range of management strategies within semi-arid shrublands typical of inland south-eastern Australia. We examined whether the prescribed burning strategies successfully achieved management objectives addressed at biodiversity conservation. Specifically, we examined the effects of differing management strategies on characteristics of individual fires, fire regimes and the population dynamics of a prominent plant species, Callitris verrucosa (A. Cunn. ex Endl.) F. Muell.
In semi-arid mallee shrubland communities in southern Australia, C. verrucosa can codominate (as a shrub or small tree) with multistemmed eucalypts (Cheal, Day & Meredith 1979; Beadle 1981; Parsons 1994). It is a serotinous obligate seeder, retaining a seed bank within woody cones, and individuals are killed if their canopy is totally scorched, unlike the eucalypts which resprout (Cheal, Day & Meredith 1979; Bradstock & Cohn 2002a,b). In general, the former are highly sensitive to variations in fire regimes because their seed bank is usually exhausted following disturbance (Lamont et al. 1991; Pausas et al. 2004). In these communities C. verrucosa is usually the only species exhibiting this life history (Bradstock & Cohn 2002a). As a potential overstorey codominant, population changes may have a strong effect on general community structure and composition. For example, a number of endangered bird species are found in mallee (Woinarski & Recher 1997) and, in particular, the malleefowl Leipoa ocellata uses relatively old, dense patches of C. verrucosa (Woinarski 1989a, b; Benshemesh 1990). As such, the status of C. verrucosa can act as an indication of the consequences of differing management activities and resultant fire regimes.
The prevailing fire regime in semi-arid mallee landscapes typically consists of a 10–20-year cycle of often large fires (c. 105 ha) resulting from lightning ignitions (Bradstock & Cohn 2002a). These fires follow years of above-average rainfall (200–500 mm mean annual rainfall; Noble & Vines 1993; Bradstock & Cohn 2002a), resulting in high fuel continuity through above-average growth of ephemeral forbs and grasses (Noble 1989; Noble & Vines 1993). For example, in central and south-western New South Wales recent major fires of this kind affecting mallee shrublands occurred in 1957–58, 1974–75 and 1984–85 (Noble & Vines 1993; Bradstock & Cohn 2002a,b), representing an average interval of 13 years. Bradstock & Cohn (2002a) concluded that while there is little evidence that the current ‘natural’ (i.e. lightning sourced) regime is detrimental to any particular group or taxa, there is ongoing concern about the effects of large fires on the long-unburnt habitat of endangered birds. They (Bradstock & Cohn 2002a) noted that there is much scope for change in mallee fire regimes by an increase in both unplanned and prescribed fires.
Changes to rates of ignition could have a variety of effects. Keane, Cary & Parsons (2003) showed that landscape fire interval was sensitive to changes in both ignition rate and fire size using simulation models of forest landscapes. Fire interval decreased with increasing rates to a stable level, beyond which further increases in ignition rate had no effect. Heydon, Friar & Pianka (2000) showed that a large increase in rate of ignition (i.e. 150%) reduced the average size of fires and pixel age (time since last fire) in a spatial simulation model of arid Australian spinifex Triodia spp. landscapes. High ignition resulted in a higher proportion of younger age classes in the simulated landscape (Heydon, Friar & Pianka 2000). The consequences of these trends will depend on the life-history characteristics of resident species.
In mallee landscapes changes in land use and human activity may cause an increase in accidental or deliberate (e.g. arson) ignition rates. Thus variations in the rates of ignition in conditions likely to yield severe fire behaviour (wildfires) may reflect the outcome of such land-use changes and the subsequent consequences of any management attempts targeted at altering them.
Prescribed fire is commonly advocated as a means of controlling fuel levels, thereby limiting the incidence, spread, intensity and final size of subsequent unplanned fires (Fernandes & Botelho 2003). Increased levels of prescribed fire may therefore be beneficial to species that are perceived or known to be sensitive to the effects of high-intensity fires. Higher vertebrates that lack shelter and/or have limited dispersal ability may be disadvantaged by such fires. Species such as L. ocellata may fit into this category (Benshemesh 1990; Woinarski & Recher 1997), although data on survival of fires is limited or absent. On the other hand, an increase in ignition in general may result in a greater incidence of short-interval fires in the landscape, leading to localized decline and loss of species such as C. verrucosa that are sensitive to changes in the fire interval (Bradstock & Cohn 2002b).
Prescribed burning options in mallee landscapes are varied, with current levels being relatively low. There is scope to increase not only the overall level of prescribed burning but also to create differing spatial patterns of burning. The levels used in this study reflected this possible range. Options included the use of tracks and trails as ignition points, broad-scale multiple ignitions in either regular or random patterns, and the targeted use of prescribed fire to restrict the spread of wildfires into long-unburnt patches of vegetation.
The chief aim of this study was to see whether there are levels and spatial patterns of prescribed ignitions and rates of unplanned ignitions that produce an optimal solution for fire management in mallee landscapes. Such a solution would require a significant reduction in size of unplanned fires coincident with severe weather, with concurrent provision of a landscape-level fire interval distribution that maintained a viable population of C. verrucosa. Topographic variations, which affect the continuity of fuel elements and the resultant probability of fire propagation, were also considered (Bradstock & Cohn 2002a).
Materials and methods
- Top of page
- Materials and methods
A two-dimensional cellular model (CAFÉ; Bradstock et al. 1996; Bradstock et al. 1998) was developed to simulate effects of varied ignition levels/rates, spatial ignition patterns and topographic variations on populations of C. verrucosa. The model simulates the spread of fires using a flammability parameter governing the probability of propagation between neighbouring cells. This parameter can be adjusted to reflect the effects of fuel accumulation with time since fire and weather on propagation. The model also simulates population processes of plants such as survival (during and between fires), fecundity and seed bank accumulation, seedling establishment and dispersal. The model is an occupancy type (Akçakaya & Sjören-Gulve 2000) that represents changes to populations on the basis of cells occupied (percentage of landscape). Specific changes to the model used to represent the nature of fires in mallee landscapes and life-history features of C. verrucosa are described below.
A square landscape composed of 104 square cells was used in all simulations. Notionally the cells represent an approximate scale of 1–10 ha. Fire and plant dynamics were simulated in alternative flat and dune landscapes. The latter consisted of elliptical patterns of parallel dunes, represented by cells in three classes (dune crest, dune slope and swale). A swale is the depression at the base of the slope. The relative width and spacing of dunes as represented by the arrangement of these differing categories of cells was similar to that found in the field in mallee landscapes (R. A. Bradstock, M. Bedward & J. S. Cohn, unpublished data).
modelling of fire management strategies
Two non-random approaches to prescribed burning were contrasted with a random approach. The latter involved selection of ignition points at random and the subsequent spread of fire to occur up to a predetermined limit (Fig. 1a). The non-random approaches were based on the creation of two perpendicular lines of cells (one cell width) bisecting the modelled landscape, intended to represent access trails. The first approach (NRA) used the lines as an ignition point but allowed fires to spread freely from these lines (Fig. 1b). The second non-random approach (NRB) involved the creation of buffer strips containing low fuels by the concentration of prescribed burning in segments of contiguous cells within these lines (Fig. 1c).
Figure 1. Differing spatial patterns of prescribed ignition: (a) random ignition source; (b) non-random (NRA) ignition source located on an internal grid (shaded cells) with fires allowed to spread randomly; (c) non-random (NRB) ignition source located on an internal grid with fires confined to cells within the grid.
Download figure to PowerPoint
A range of differing levels (percentage of landscape targeted for burning each year) was specified for each prescribed burning strategy (Table 1). Fires lit under both the random and the NRA strategy were allowed to spread to size limits specified for each level. In some cases restrictions imposed by flammability/terrain interactions extinguished these fires (see Modelling of fire spread).
Table 1. Ranges of input variables used in landscape simulations of the effects of fire regimes on Callitris verrucosa: (a) chance of single ignition capable of burning 100% of landscape; (b, c) target burn area each year achieved by discrete ignitions, each 100 cells (1% of landscape); (d) area achieved by burning segments of 200 target cells contained within fixed buffer strips (see text). All burning targets were dependent on the availability of flammable cells
|Fire type||Fire pattern||Annual burning rate*/level†|
|(a) Unplanned|| ||0·05, 0·10, 0·20 ignitions*|
|(b) Prescribed||Random||0%, 1%, 5%, 10%, 20% landscape†|
|(c) Prescribed||NRA||0%, 1%, 5%, 10%, 20% landscape†|
|(d) Prescribed||NRB||0%, 1%, 2% landscape†|
modelling of fire spread
Fire propagation was modelled in a two-stage probabilistic manner by including a provision for setting the direction and shape of fires and for generation of spot fires, as dictated by the effects of wind. This enabled fire propagation to overcome barriers created by fuel discontinuity. Initially, the chance of a cell receiving fire from an ignited donor cell was estimated as a function of distance and direction of ignited donors. The status and condition of the cell (e.g. topographic position, time since fire) then determined its chance of ignition. The first step allowed the effect of wind and spotting on fire propagation to be varied according to weather conditions. The second step allowed the internal condition of the cell, in interaction with weather, to influence its flammability.
A template of potential donor cells that could transfer fire to a neighbouring target cell was defined (Fig. 2). Each donor cell within this template was assigned a probability of transmission according to its position in the neighbourhood and the weather conditions coincident with the fire (i.e. severe weather for unplanned fires, moderate weather for prescribed fires). Spot fires resulted from a template of cells wider than that of the eight contiguous neighbours of the recipient. Thus the template for unplanned fires was composed of a larger radius of potential donor cells (i.e. allowing potential spotting; Fig. 2a) than that for prescribed fires (Fig. 2b,c).
Figure 2. Flammability templates used to simulate the influence of neighbourhood effects on intercellular propagation of fires. The probability (% chance) of propagation from burning donor cells (light shading) to a neighbouring recipient cell (black shading, centre cell in each diagram) is illustrated. Probabilities for donor cells not contiguous with the recipient cell represent potential fire propagation through spotting (a). Effects of different ignition sources/patterns and weather conditions: (a) unplanned (random), (b) prescribed (random), (c) prescribed (non-random, NRA).
Download figure to PowerPoint
The direction and shape of fires was determined by the values for probability of transmission assigned within each template according to the type of ignition. For unplanned fires in particular (Fig. 2a), the template produced initial fire spread patterns that were elliptical in shape, moving from left to right (west to east) within the landscape. Templates for prescribed fire produced fires more irregular in shape and direction. For the random strategy (Fig. 2b) the template was symmetrical, producing an even probability of fire spread in any direction. Irregular fire shapes arose solely through the influence of the condition of cells on flammability. For the NRA strategy the template was varied to allow fires to spread in a diagonal manner relative to the lines of cells representing trails and ignition points (Fig. 2c). Such burn patterns potentially created low-fuel strips that cut across the potential direction of spread of unplanned fires (see above). Prescribed fires lit under NRB strategies were confined to predetermined segments of cells. Again, in certain instances, some fires of this type did not burn all cells in the target segment because of flammability constraints.
In general, cell flammability was specified to increase with time since last fire towards an asymptote (Fig. 3) consistent with known patterns of litter accumulation (Bradstock 1990). Effects of topography on fire spread were achieved by specifying a separate flammability schedule for discrete topographic classes (flat, dune crest, dune slope and swale). Differences in flammability between topographic classes were selected to represent known differences in the density of eucalypts (the primary source of litter fuel) and the grass Triodia scariosa (Bradstock 1989; Bradstock & Gill 1993; Bradstock & Cohn 2002a). Dune slopes and flats were the most flammable element because of relatively high densities of eucalypts and T. scariosa. The flammability of swales and dune crests was represented as relatively low because of a lower density of perennial grasses and eucalypts. A separate flammability schedule was prepared for swales to represent the effects of above-average rainfall and consequent herbage growth. This raised the overall level of flammability of swales closer to that of dune slopes until burning occurred (Fig. 3). Annual probability of occurrence of such conditions was fixed at 0·05 in accordance with the frequencies postulated by Noble & Vines (1993).
Figure 3. Schedules of flammability, reflecting the influence of weather and fuel condition (time since last fire) on the probability of intercellular propagation of fires. Values (% probability) represent the chance of a cell igniting once a fire has been transferred from a neighbouring donor cell (see Fig. 2). Effects of topography and weather associated with alternative ignition sources (a) prescribed and (b) unplanned are indicated by differing symbols: flat and dune slope (before/after rain, squares); dune crest (before/after, triangles) and swale (before, triangles); swale (after, diamonds).
Download figure to PowerPoint
Flammability schedules were also varied to account for the effects of weather on fire spread, in interaction with fuel age and site effects (Fig. 3). Schedules for moderate (prescribed fire) and severe weather (unplanned fire) were specified with lower time since fire thresholds for fire spread in the latter case. Fire size limits also varied between ignition/weather types. Unplanned (severe weather) fires were unrestricted in size, with the potential to burn the whole landscape given the availability of fuel. In contrast, prescribed (moderate weather) fires were restricted to a maximum size of 100 cells per ignition (1% of the landscape), depending on fuel availability. Unplanned fires were ignited randomly in space.
modelling of plant dynamics
Schedules of survival, maturation and fecundity used to model C. verrucosa populations are given in Table 2 (Bradstock & Cohn 2002b). While C. verrucosa does not possess any capacity for vegetative recovery following fire, bark thickness at the base of stems and the basal height of the foliage are sufficient to allow some older plants to survive particular fires (Bradstock & Cohn 2002b). Thus a multistage schedule of fire survival was specified. Seeds of Callitris species, while winged, are relatively heavy (Bowman & Harris 1995) and dispersal is known to be over a short range (< 50 m) in species such as Callitris glaucophylla (Lacey 1973). Dispersal in the model was therefore confined to a single cell radius, as was the case for other shrub species simulated in earlier applications of the model (Bradstock et al. 1996, 1998). A strong level of serotiny is exhibited in C. verrucosa and establishment appears to be confined to the immediate post-fire period (Bradstock & Cohn 2002b), in a similar manner to that found for cohabiting mallee eucalypts (Wellington & Noble 1985). Therefore, in the model, seeds only became available for germination in the year immediately following fire, provided any given cell was occupied by a mature adult or else seed dispersal had occurred from a neighbouring source.
Table 2. Demographic characteristics of Callitris verrucosa used to model population responses to fire regimes
|Initial population size||5000 cells occupied (50% landscape)|
|Initial population age||5–15 years|
|Survivorship probability||0–200 years, 100%; > 200 years, 0%|
|Fecundity probability||0–12 years, 0%; > 12 years, 100%|
|Seed dispersal probability||20%|
|Fire mortality (prescribed)||0–49 years, 100%, 50–99 years, reduces by 1% each year; > 99 years, 50%|
|Fire mortality (unplanned)||0–49 years, 100%, 50–90 years, reduces by 20% each decade; > 90 years, 10%|
Combinations of unplanned ignition rate, prescribed fire level and pattern were simulated for both flat and dune field landscapes. Simulation duration was 1000 years, with initial occupancy (population size) of C. verrucosa set at 50% of the cells in the landscape. The initial population was evenly spread across a range of age classes and the initial time since fire distribution in the landscape was similarly configured. Ten replicates of all treatments were simulated.
While data on fire size (number of continuous cells) and population size (number of cells occupied) were compiled from the whole matrix, fire interval data were collected (years elapsed between fires) from a point source (from an individual cell at or near the centre of the landscape but distant from cells subjected to non-random prescribed burning; Fig. 1). All these data were compiled for each topographic position (flat, swale, dune slope, dune crest).
The effects of different treatments (alternative combinations of unplanned ignition rate, level and spatial pattern of prescribed ignitions) on size of unplanned fires, fire interval and size of C. verrucosa populations (occupancy of cells in landscape) were examined (Table 3).
Table 3. Summary of statistical analyses (see the Materials and methods). Means are arranged in ascending order and italic indicates no significant difference between the means. Factors include strategy (s, random, NRA), prescribed burning level (p, 0%,1%, 5%,10%, 20% per year), unplanned ignition rate (up, 0·05, 0·1, 0·2 per year), topography (t), flat landscape (f), dune landscape (d). Heteroscedastic data were square-root transformed†or assessed at P < 0·01‡. Probability levels (P) < 0·05*, < 0·01**, < 0·001***, NS, not significant
|Response variables|| ||Test||Factors||Significant results||F-ratio (d.f.) or t||P|
|1||Unplanned fire size|| || || || || || |
|a||Random and NRA||anova||[s + p + up + t]†||p||20%, 10% < 5% < 1%, 0%|| 54·7 (4533)||***|
|b||NRB||anova||[p + up + t]||–||–||–||NS|
|c||NRB vs. random/NRA||t-test||p = 0%||–||–|| 0·25 (178)||NS|
|d||NRB vs. random/NRA||t-test||p = 1%||–||–||−0·90 (178)||NS|
|2||Fire interval|| || || || || || |
|a||Random and NRA||anova||[s + p + up + t]†||s.p.up.t||See Fig. 5|| 2·1 (24,3757)||**|
|b||NRB||anova||[p + up + t]†||up||0·2 < 0·1 < 0·05|| 32·4 (2321)||***|
|3||Callitris population|| || || || || || |
|a||Random and NRA||anova||[s + p + up + t]‡||s.t.p||See Fig. 6|| 62·5 (4300)||***|
| || || || ||s.p.up||See Fig. 6|| 2·5 (8300)||**|
| || || || ||t.p.up||See Fig. 6|| 14·7 (8300)||***|
|b||NRB||anova||[p + up + t]||up.t||0·05d, 0·05f < 0·1f, 0·2f,0·1d < 0·2d|| 33·1 (2,72)||***|
Because the levels of prescribed ignitions were not orthogonal for all strategies (Table 1), separate analyses were undertaken (Table 3). Four-factor analyses of variance (anova) examined the effect of unplanned ignition rate, topography, level and spatial pattern of prescribed ignitions (random vs. NRA) on each dependent variable (fire size, fire interval and population size). Separate three-factor anovas examined the effects of the first three factors on the same dependent variables under the NRB strategy. All factors were regarded as fixed, allowing testing of significance of all higher order interactions. Tukey tests were used for post-hoc comparisons. To eliminate heterogeneity of variance, as indicated by Cochran's test, data were either square-root transformed prior to analysis or results were assessed at a more conservative probability (P < 0·01; Underwood 1981). The two-sample t-test compared unplanned fire size of the pooled data for random and NRA strategies with NRB strategy at 0% and 1% prescribed burning levels.
The last 10 fire intervals and fire sizes recorded in a single simulation (1000 years), representative of each treatment combination, were used for data analysis. The final population size recorded for 10 simulations of each treatment combination were used for data analysis.