Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
1 Fire regime shifts affect bird species distributions 2 3 4 April Reside, Jeremy VanDerWal, Alex Kutt, Ian Watson and Stephen Williams (A) Abstract 5 (B) Aim: Fire is a major driver of ecosystem structure and process, and shifts in fire regimes are 6 implicated in the decline of many species. Shifting fire regimes have been documented around the 7 world, and fire frequency and extent is predicted to increase in many areas due to changes in both 8 climate and land management. Here we address the need for greater understanding of species’ 9 sensitivity to fire frequency, and the effect of increasing fire frequency on species distributions. 10 (B) Location: The tropical savannas of northern Australia. 11 (B) Methods: We developed distribution models for bird species using the modelling algorithm 12 Maxent. We included climate, fire frequency and the subset variable fire frequency late in the dry 13 season as environmental variables in the models. We investigated the effect of increasing fire 14 frequency, and increasing fire late in the dry season, on species distributions by projecting species 15 model algorithms onto scenarios of incrementally increased fire frequency. 16 (B) Results: The probability of presence for most species was higher when fire frequency late in the 17 dry season was low. Species showed a mixed response to an overall increase in fire frequency, with 18 one-third predicted to increase in distribution. However almost all species (98%) showed a decrease 19 in predicted range with increased late-dry season fire, and species distribution area was generally 20 negatively correlated with an increase in late-dry season fire. 21 (B) Main Conclusion: Our study highlighted the array of responses of species to increasing fire 22 frequency, and suggested that increased fire frequency late in the dry season is detrimental to most 23 savanna-restricted bird species. The understanding of individual species’ preferences for particular 24 fire frequencies is important for informed conservation planning. 25 (C) Keywords: Australian tropical savannas, birds, fire frequency, late-dry season fire, Maxent, 26 species distribution models 27 28 (A) Introduction 29 Shifting fire regimes are a critical conservation issue in many regions of the world, compounding the 30 pressures of climate change, intensifying land use and invasive species on biodiversity (Keith, 31 Williams et al. 2002; Stephens and Ruth 2005; Brook 2008). The frequency and extent of fire has 32 increased in many regions (Mouillot and Field 2005), and further increases are predicted in areas such 33 as the USA, Brazil, central Asia, south-eastern Europe, southern Africa and Australia (Liu, Stanturf et 34 al. 2010). Managing fire for optimal biodiversity outcomes is an essential conservation issue; however 35 our current understanding of the timing, frequency and scale of fire regimes that species require is 36 limited (Clarke 2008). Information on species’ sensitivity to fire, and the impact of increased fire 37 frequency on species, is therefore crucial to their conservation. 38 39 Observations suggest that fire patterns around the world are already shifting in response to changing 40 climate (Westerling, Hidalgo et al. 2006). Increasing wildfire activity in the western United States 41 has been associated with warmer springs, extended summer dry seasons and drier vegetation 42 (Westerling, Hidalgo et al. 2006). Increases in fire activity in Canada (Gillett, Weaver et al. 2004) 43 and the Neotropics (Barlow and Peres 2004) have also been attributed to climatic changes. Future fire 44 activity is predicted to vary considerably across regions in response to climate change, and land 45 management will continue to influence the extent of these changes (Liu, Stanturf et al. 2010). 46 Australia, already the world’s most fire-prone continent, is also predicted to face an overall increase in 47 fire (Beer and Williams 1995; Cary and Banks 2000; Williams, Karoly et al. 2001; Cary 2002; 48 Pitman, Narisma et al. 2007; Krawchuk, Moritz et al. 2009), though the spatial and temporal 49 realisation of fire increase across the landscape is uncertain. 50 51 Fire occurs in almost every biome, but since 1990 86% of fires globally occurred in tropical savannas 52 and grasslands (Mouillot and Field 2005). For tropical savannas worldwide, fire is a significant 53 determinant of landscape and biodiversity pattern (Davis, Peterson et al. 2000; Garnett and Brook 54 2007), and fire is essential for maintenance of these landscapes (Bond and Parr 2010). Fires shape 55 savanna vegetation (Woinarski, Risler et al. 2004), and fauna that inhabit these tropical savannas are 56 largely resilient to fire (Bond and Parr 2010). 57 58 Both the frequency and the seasonality of fire in savannas have important implications for 59 biodiversity (Garnett and Brook 2007; Woinarski, Armstrong et al. 2010). Frequent fires lead to 60 reduced tree density (Davis, Peterson et al. 2000) and decreased vegetation structural complexity 61 (Bowman 1998). Too-frequent fires are particularly detrimental to species favouring long-unburnt 62 patches, such as some mammals (Friend and Taylor 1985; Andersen, Cook et al. 2005), fire sensitive 63 plants (Bowman and Panton 1993), granivorous birds (Fitzherbert and Baker-Gabb 1988) and birds 64 requiring grassy refugia (Garnett and Bredl 1985; Rowley and Russell 1993; Jansen, Little et al. 65 1999). However where fires are sufficiently infrequent, the tree-grass ratio within the savanna could 66 increase (Roques, O'Connor et al. 2001) which can lead to lower bird diversity (O'Reilly, Ogada et al. 67 2006) and if excluded for long enough, a shifting of biomes from savanna to rainforest (Woinarski, 68 Risler et al. 2004; Accatino, De Michele et al. 2010). 69 70 Fire seasonality can be a major factor determining the impact of fire on biodiversity due to the 71 influence of season on the amount of fuel available to burn and the presence of weather conducive to 72 burning (Bradstock 2010). In tropical savannas, early dry-season fires tend to be of low intensity, and 73 their occurrence reduces the amount of herbaceous biomass available to burn later in the dry season. 74 In the absence of early-dry season fire, fuel build-up can lead to late-dry season fires that have greater 75 intensity and the capacity to burn larger areas (Williams, Griffiths et al. 2002). The late-dry season 76 fires remove unburnt patches from the landscape (Bird, Bird et al. 2008), homogenising the age-class 77 structure of large tracts of savanna (Bradstock, Bedward et al. 2005). 78 79 Understanding the influence of fire frequency and seasonality on savanna biodiversity is a primary 80 conservation objective in tropical savannas (Williams, Griffiths et al. 2002; Uys, Bond et al. 2004). In 81 Australia, shifts away from traditional indigenous land management have seen an end to small scale 82 early-dry season fires towards extensive, high intensity late-dry season fires. The increase in late-dry 83 season burning is thought to be a main cause of declines in many vertebrate species populations in the 84 tropical savannas of northern Australia (Franklin 1999; Woinarski, Brock et al. 1999; Woinarski, 85 Milne et al. 2001). Experimental and localised studies on fire frequency and fire seasonality and 86 biodiversity in this system have elucidated species responses to fire events and regimes (Andersen, 87 Cook et al. 2005; Kutt and Woinarski 2007; Murphy, Legge et al. 2010). However, species-level 88 studies and natural experiments should also be complemented by simulation models to investigate 89 landscape scale fire patterns and its effect on biodiversity (Driscoll, Lindenmayer et al. 2010). In this 90 study we focus on Australian tropical savanna birds to investigate 1) species sensitivity to fire 91 frequency across their range, and 2) the effect of increased fire frequency and in particular increased 92 late-dry season fire frequency (herein “late-fire frequency”) on bird species distributions. To address 93 the first objective we examine species occurrences in relation to fire frequency and late-fire 94 frequency, and then estimate the probability of presence for each species according to fire frequency 95 and late-fire frequency. For the second objective, distribution models for Australian tropical savanna 96 birds built using current conditions of fire and climate were projected onto spatial surfaces of 97 simulated fire frequency increases. Understanding species sensitivities to fire frequency, and how 98 their distributions will be affected with fire frequency increases (a likely scenario under climate 99 change predictions (Krawchuk, Moritz et al. 2009)) is an important step in developing the mitigation 100 strategies for bird species conservation (Driscoll, Lindenmayer et al. 2010). 101 102 (A) Methods 103 (B) Study Area 104 Tropical savannas stretch almost continuously across Australia north of c. 23oS (Franklin, Whitehead 105 et al. 2005), occupying nearly one quarter of the Australian continent (Williams, Carter et al. 2005). 106 Savannas are characterised by a discontinuous stratum of trees above a mostly continuous layer of 107 grasses (Williams, Griffiths et al. 2002). Rainfall in the Australian tropical savannas is highly 108 seasonal, with most rain falling in the wet season between December and March (Felderhof and 109 Gillieson 2006). There is a climatic gradient across the Australian savannas, with a trend of 110 decreasing rainfall and increasing inter-annual rainfall variability with distance from coast (Mott, 111 Williams et al. 1985). 112 113 (B) Bird Data 114 We focussed our study on 44 bird species that are largely restricted to the northern tropical savanna 115 woodlands (see Appendix S1 in Supporting Information for full list of species), excluding waterbirds 116 and rainforest species that may occur intermittently in savanna regions, and species with ranges that 117 extend substantially into other biogeographic regions. We only included species with distributions 118 almost entirely in the study area as less confidence can be placed in models that are restricted to a 119 small subset of species’ ranges (Elith and Leathwick 2009). Despite this caveat, we undertook 120 preliminary investigations for a further 163 species that occur within the savanna but also in other 121 parts of the Australian continent to see if the pattern of responses found for the core 44 species are 122 reflected in a larger sample of species occurring in the region. However, the analysis focuses on the 123 44 species which are more restricted to the savanna. Occurrence records of all 207 bird species were 124 collated from the Birds Australia Atlas (Blakers, Davies et al. 1984; Barrett, Silcocks et al. 2003), the 125 Queensland Governmental atlas WildNet (Environmental Protection Agency 2004) and CSIRO 126 (protocol as in (Reside, VanDerWal et al. 2010)) for the period from 1997 to 2008. 127 128 (B) Environmental Data 129 The climate data used for modelling included gridded spatial layers of annual mean temperature, 130 temperature seasonality (the mean coefficient of variation of temperature over an annual period), 131 annual precipitation, maximum temperature of the warmest period, annual precipitation, precipitation 132 seasonality and precipitation of the driest quarter, estimated using Anuclim 5.1 software (McMahon, 133 Hutchinson et al. 1995) and a ~250 m resolution DEM (GEODATA 9 Second DEM Version 2; 134 Geoscience Australia). The fire data included maps of fire frequency, number of years land was burnt 135 in the period from 1997 to 2008; and late-fire frequency (a subset of fire frequency), the number of 136 years an area was burnt late in the dry season between 1997 and 2008 (Northern Australia Fire 137 Information, NAFI http://138.80.128.152/nafi2/). The late dry-season is defined as the period 138 between August and November (Felderhof and Gillieson 2006). The fire layers were derived from the 139 National Oceanic and Atmospheric Administration’s Advanced Very High Resolution Radiometer 140 (NOAA-AVHRR) and other images from the Bushfire Council of the Northern Territory. 141 142 (B) Environmental Space Plots 143 Environmental space plots were produced to visually inspect the species’ occurrence against fire 144 frequency, regardless of geographic space using the statistical package “R” version 2.9.0 (www.r- 145 project.org). Annual rainfall, fire frequency and late-fire frequency variables for every 1o pixel of the 146 study area were extracted, and then plotted with rainfall on the x-axis and either fire frequency or late 147 fire frequency on the y-axis. This provided the “background” available environmental space. The 148 values of annual rainfall and fire frequency or late-fire frequency corresponding to each species’ 149 location were plotted on top of the background. This gave an indication of the selectivity of the 150 species, whether a species was selecting areas with particular fire frequencies, particular annual mean 151 rainfall values, or both. Both the background environmental space and the species’ occurrence were 152 weighted to frequency of occurrence, so that more frequently recorded combinations of fire and 153 rainfall are represented by different colours than those less frequently recorded. 154 155 (B) Distribution Models 156 Species distribution models were run using the presence-only modelling program Maxent (Phillips, 157 Anderson et al. 2006). Maxent uses species presence records to statistically relate species occurrence 158 to environmental variables on the principle of maximum entropy. All default settings were used, and 159 models were run at a spatial resolution of 0.05 degrees (c. 5 x 5 km). Model performance (defined as 160 the models’ consistency and ability to identify species actual presence and actual absence (Ling, 161 Huang et al. 2003)) was evaluated by the area under the receiver operator curve (AUC): an AUC 162 score of 1 is a perfect fit of the data, 0.5 is no better than random (Elith, H. Graham et al. 2006; 163 Phillips, Anderson et al. 2006). Models for each species were screened for low AUC (<0.75) so that 164 underperforming models were not included in further analyses. 165 166 The percentage contribution of each environmental variable to the model was used to determine their 167 proportional influence. The Maxent output response curves were then used to further examine the 168 influence of the fire variables on the model. The response curves show the logistic probability of 169 presence using only one variable at a time so that the contribution of individual variables can be 170 examined when there is autocorrelation in the environmental layers. This was important in this case, 171 as the late-fire frequency variable is a subset of fire frequency and therefore is highly correlated. 172 173 (B) Increased Fire and Late-Fire Frequency Projections 174 The impact of increasing fire frequency and late-fire frequency was predicted by projecting the model 175 algorithm for each species onto a series of spatial surfaces, each consisting of a fire layer with 176 artificially increased fire or late-fire frequency, and all climate variables remaining unchanged. These 177 spatial surfaces were created by increasing each cell of the fire frequency raster layer by an increment 178 of one (therefore a cell that was burnt five times between 1997 and 2008 then had a fire frequency of 179 six). This process was repeated six times for both fire frequency and late-fire frequency. A total of 180 thirteen spatial surfaces were generated: the first with current levels of fire frequency, late-fire 181 frequency and climate, then six surfaces with increasing fire frequency, and six with increasing late- 182 fire frequency. 183 184 The Maxent default cumulative output value was converted to a binary presence/absence for each 185 species using the threshold that balances training omission rate, predicted area and logistic threshold 186 value (Liu, Berry et al. 2005). This threshold value provides realistic predictions of species 187 distributions (Prates-Clark, Saatchi et al. 2008; Vanderwal, Shoo et al. 2009). With the binary output, 188 area could be denoted as “suitable predicted area” or “unsuitable predicted area” for each species. 189 This enabled the calculation of the proportion of the whole study area that was suitable. We could 190 then predict the percentage of the study area (Australian tropical savannas) that was suitable under the 191 current conditions, and how this proportional suitability shifted when fire frequency or late-fire 192 frequency increased incrementally six times. The change in suitable area under the extreme case of a 193 six-increment increase in fire frequency or late-fire frequency was calculated by subtracting the 194 percentage suitable under the current conditions from that calculated for a six-increment increase. 195 196 In addition to the change between current and the extreme case, we examined the change in 197 percentage suitable area for each species with each increment of increasing fire frequency and late- 198 fire frequency. We calculated the Pearson correlation between increasing fire and increasing late-fire 199 frequency from current to a maximum increase of six years burnt, with the percentage of suitable area 200 predicted by the model for each species. The Pearson correlations (ρ) for all species are displayed in a 201 histogram. All analyses were conducted using the statistical package “R” version 2.9.0 (www.r- 202 project.org). Images of the suitable area, unsuitable area, and area that changed in suitability 203 depending on fire frequency as predicted by the Maxent model were produced for each species. Four 204 exemplar species with a range of fire sensitivities are presented as case studies to demonstrate the 205 results. 206 207 (A) Results 208 (B) Environmental Space Plots 209 Birds in Australian tropical savannas show a broad range of responses to existing fire frequencies 210 (Fig. 1). The environmental space plots (Fig. 1a c, e, g) show that species were recorded in a smaller 211 fire frequency envelope than the entire range of fire frequencies. The highest concentration of 212 species’ occurrence points generally clustered around the lower fire frequencies. Some species, for 213 example Gouldian finch Erythrura gouldiae, were found fairly broadly across fire frequencies (zero to 214 ten years out of twelve burnt) but were not found in areas with late-fire frequency greater than six 215 years out of twelve (Fig. 1a). In contrast, red-backed fairy-wren Malurus melanocephalus (Fig. 1c) 216 showed broad tolerance for fire, being recorded at every fire frequency and most frequencies of late- 217 dry season fire. A few species were less tolerant, such as the white-streaked honeyeater Trichodere 218 cockerelli (Fig.1g) which was recorded almost exclusively in locations with fire frequencies less than 219 five years out of 12. 220 221 (B) Probability of Presence Curves 222 The logistic probability of species presence depending on the fire frequency calculated by Maxent is 223 shown in Fig. 1(b), (d), (f) and (h). Gouldian finch (Fig. 1b) and red-backed fairy-wren (Fig. 1d) 224 showed slight increases in the probability of presence as fire frequency increases, but decreasing in 225 probability of presence as late-fire frequency increases. In contrast, star finch Neochmia ruficauda 226 (Fig. 1f), and white-streaked honeyeater (Fig. 1h) both show decreases in probability of presence with 227 increases in both fire frequency and late-fire frequency. White-streaked honeyeater is predicted to not 228 occur where fire frequency or late-fire frequency exceed eight or seven years respectively. The 229 response curves are generally consistent the environmental space plots. For example, Gouldian finch 230 is frequently recorded at fire frequencies between two and nine (Fig. 1a). The Maxent probability of 231 presence peaks between fire frequencies of seven and ten (Fig. 1b). Gouldian finch is recorded only 232 where late-fire frequency is between zero and six, and the probability of presence decreases after late- 233 fire frequency exceeds three. On the other hand, red-back fairy-wren (Fig. 2c) is recorded at every 234 fire frequency, and correspondingly, fire frequency has little effect on their probability of presence. 235 236 (B) Distribution Models 237 Model performance across the core 44 savanna bird species was high, with all species’ model AUC 238 scores above 0.75 (range 0.76 – 0.99). The environmental factors of fire frequency and late-fire 239 frequency had a small but meaningful mean contribution in the models: 3.9% (range 0.6 – 23.1%) and 240 3.1% (range 0.04 – 11.9%) respectively (Table 1). Climate provided most of the explanatory power 241 to the distribution models, which is to be expected for models at this scale (Pearson and Dawson 242 2003). Annual precipitation showed the highest average contribution (37.8%), followed by 243 temperature seasonality (24.3%). 244 245 (B) Increased Fire and Late-Fire Frequency Projections 246 Predicted changes in the percentage area suitable for the core 44 species under a six-increment 247 increase in fire frequency and late-fire frequency are shown in Table 2. Two-thirds of these species 248 faced a decrease in suitable area with increased fire frequency, and the mean change in area was 249 slightly greater for the species with decreased ranges (9.4% of suitable area lost, versus 4.4%) than the 250 species that increased in predicted range. Many more species showed a decrease in suitable area with 251 increased fire in the late dry season, with 98% of species showing decreased suitable area. The mean 252 change in suitable area was also more dramatic for the species facing a decrease in suitable area, at 253 7.2% of area lost versus 0.8% of area gained for the one species (long-tailed finch 254 Poephila acuticauda) that showed increases in suitable area. 255 256 The effect of increasing fire frequency and late-fire frequency at each of the increments is shown in 257 detail for the four exemplar species (Fig. 2). Species showed a variety of responses to increased fire 258 frequency. Gouldian finch showed a slight increase with increasing fire frequency (Fig. 2a). Red- 259 backed fairy-wren showed little change (Fig. 2c), whereas star finch (Fig. 2e) and white-streaked 260 honeyeater (Fig. 2g) decreased. All four species showed a trend of decreasing suitable area associated 261 with an increase in late-fire frequency. Star finch showed a particularly strong decline with late-fire 262 frequency (Fig. 2f). White-streaked honeyeater showed strong declines with both increasing fire and 263 late-fire frequencies (Fig. 2g and 2h). 264 265 The relationship between increasing fire frequency or late fire frequency and the percentage suitable 266 area across the 44 species is shown with Pearson correlations (ρ) in Fig. 3(a). Two-thirds of species 267 were negatively correlated between increased fire frequency and the percentage suitable area, and 268 one-third positively correlated. In contrast, the majority of species show a strong negative correlation 269 between increasing late-fire frequency and the percentage suitable habitat. The ρ are also shown for 270 all 207 of species that also occur in Australian tropical savannas (Fig. 3b). The same pattern is shown 271 across the 207 species as that seen for the core 44 savanna specialist species – a mixed response to 272 increasing fire frequency. However, under increasing late-fire frequency most species show a strong 273 decline in suitable area. 274 275 Some species may face substantial distribution loss with increases in fire frequency. In the extreme 276 event of a six-increment increase in fire frequency, 11 species are predicted to lose more than 10% of 277 their range within the tropical savannas (Appendix S1). These same 11 species are also predicted to 278 lose more than 10% of their range within the study area in the case of a six-increment increase in late 279 fire frequency. Under both scenarios, three species are predicted to lose more than 20% of their 280 distribution: yellow honeyeater Lichenostomus flavus, black-throated finch Poephila cincta and 281 oriental cuckoo Cuculus saturatus. A six-increment increase in fire frequency may prove beneficial 282 for some species, although predicted increases in suitable area are comparably small. Thirteen species 283 showed very slight increases, with the most substantial increase shown by long-tailed finch Poephila 284 acuticuada, with an increase of 10%. Long-tailed finch was the only species to show an increase in 285 area due to a six-increment increase in late fire frequency; however this increase was minimal (Table 286 1). 287 288 (A) Discussion 289 290 This study suggests that on broad bioclimatic scale Australian tropical savanna birds have a low 291 sensitivity to fire frequency and fire frequency increase, though this sensitivity increases with 292 increased late frequency fires. All methods employed within this study: the environmental space 293 plots, the probability of presence curves and the simulations of increased fire frequency; all show that 294 most savanna bird species are tolerant of all but the highest fire frequencies. In fact, one-third of the 295 44 savanna bird species examined in our study showed a positive correlation between suitable habitat 296 area and fire frequency increase if the increases were not confined to the late-dry season. Many 297 species within Australian tropical savannas are known for their resilience to fire, and this is likely due 298 to the long history of coexisting with fire (Andersen and Hoffman 2010). Birds within this biome 299 have high mobility, a key attribute in resilience to fire (Moretti and Legg 2009). Early-dry season 300 burns in particular may have a beneficial effect on particular species. Many species are attracted to 301 recently burnt areas as fire may make food resources, such as seeds and small animals, more 302 accessible due to removal of the grass layer (Braithwaite and Estbergs 1985; Braithwaite and Estbergs 303 1987; Dean 1987; Woinarski 1990). However the main benefit of early-dry season burns may be that 304 they reduce the extent and impact of late-dry season fires (Williams, Griffiths et al. 2002; Bird, Bird 305 et al. 2008). Studies have shown that the proximity to unburnt area is important for species to survive 306 the actual fire event (Murphy, Legge et al. 2010). In addition, fire may facilitate access to foraging 307 resources, but be detrimental to nesting sites (Woinarski 1990), and species benefit most with access 308 to both burnt and unburnt areas within their home range (Murphy, Legge et al. 2010). Wet season 309 burns are also likely to be beneficial. They have higher bird abundance than unburnt sites immediately 310 post-fire, and maintain similar vegetation structure and bird assemblages as unburnt sites several years 311 following the burning event (Valentine, Schwarzkopf et al. 2007). Of the species predicted to 312 experience increases in suitable range due to increases in fire frequency in our study, some are 313 associated with vegetation close to water such as long-tailed finch, purple-crowned fairy-wren 314 Malurus coronatus, Gouldian finch, northern rosella Platycercus venustus, silver-crowned friarbird 315 Philemon corniculatus and crimson finch Neochmia phaeton. These species may be more resilient to 316 frequent fires due to their proximity to water courses. Previous studies have found that black-tailed 317 treecreeper in particular has been associated with frequent burning (Woinarski, Brock et al. 1999). 318 Species responses to fire predicted by our study appeared to be independent of functional groups, as 319 granivorous, insectivorous and foliage insectivore/nectarivore species showed both positive and 320 negative relationships with increasing fire frequency in our study. 321 322 Our study demonstrated that despite the resilience of many species in tropical savannas to regular fire, 323 increasing late-dry season fire has a detrimental effect on the majority of species. These findings are 324 consistent with other studies which highlight the impact of late-dry season fire on bird species in the 325 tropical savannas (Bradstock, Bedward et al. 2005). High late-fire frequency can influence bird 326 composition and assemblage pattern by changing habitat structure through the decline in fire sensitive 327 vegetation (Bowman, Wilson et al. 1988; Russell-Smith and Bowman 1992; Price and Bowman 328 1994), increasing tree mortality (Williams, Griffiths et al. 2002), and a reduction in structural 329 diversity (Woinarski 1990; Valentine, Schwarzkopf et al. 2007). Other studies have shown late-dry 330 season fires to be detrimental to tropical savanna bird species such as black-tailed treecreeper, 331 Climacteris melanura (Garnett and Crowley 1995), red-backed fairy-wren (Murphy, Legge et al. 332 2010), lemon-bellied flycatcher Microeca griseoceps and Australian owlet-nightjar Aegotheles 333 cristatus (Andersen, Cook et al. 2005). 334 335 Fire frequency and late-fire frequency had a relatively small proportional contribution to the 336 distribution of most species, as was expected as species distributions are predominantly governed by 337 climate at large spatial scales (Pearson and Dawson 2003). For some species, fire frequency 338 contributed substantially to the model (e.g. >20% for the northern rosella), though percentage 339 contributions for individual variables can be misleading when variables are highly correlated, as with 340 fire frequency and late-fire frequency, or mean annual temperature and maximum temperature of the 341 warmest period. Despite the relatively small contribution of the fire variables, simulating changes in 342 fire frequency resulted in substantial changes in the distribution models for many species. Thorough 343 understanding of the effects of fire frequency and seasonality on individual species will require 344 detailed mechanistic studies (Driscoll, Lindenmayer et al. 2010); these would determine the 345 importance of interactions of local factors e.g. burnt patch size, proximity to unburnt refugia or habitat 346 heterogeneity to species resilience to fire (Murphy, Legge et al. 2010). 347 348 For many species likely to lose suitable habitat as fire frequency and late fire frequency increases, 349 decreases mostly occurred in the south parts of their distribution. The southern section of the tropical 350 savannas is largely semi-arid, with generally a much lower fire frequency than in the north (Felderhof 351 and Gillieson 2006). Fire can have a severe impact on these habitats, as lower and less consistent 352 rainfall can slow the post-fire regeneration process (Noble 1989). Many semi-arid plant species are 353 fire-killed, such as acacia (e.g. Acacia aneura), and spinifex (e.g. Triodia spp), and high fire 354 frequency coupled with the lack of post-fire rain can prevent resprouting (Allan and Southgate 2002). 355 This severe impact of fire on the vegetation in the semi-arid savannas could explain the lack of 356 resilience of bird species to high fire frequencies on the southern edge of the savannas, as birds are 357 highly responsive to vegetation structure (Martin and Possingham 2005; Sirami, Seymour et al. 2009). 358 It could also explain why our study found detrimental effects of late fire on some species, for example 359 blue-winged kookaburra (Dacelo leachii) and forest kingfisher (Todiramphus macleayii), while in the 360 northern mesic savannas of the Northern Territory experimental studies have found no effect of fire 361 on these species: in the mesic savannas the late fire effects are not as severe as in the southern, semi- 362 arid savanna. 363 364 While our study is likely to reflect real-world sensitivities of species to fire, it is not without 365 limitation. One issue is that the spatial scale – 5 km pixels – may miss the detail of fine burnt and 366 unburnt mosaics that are likely to be influential to bird species. However given the mobility of birds, 367 many species are operating on a spatial scale of hundreds of kilometres (Griffioen and Clarke 2002), 368 therefore although a 5 km x 5 km scale may be missing some detail, it is still fine enough to be 369 relevant to the species studied. Another limitation is the bird data and the fire data are from the same 370 period, therefore this study might not be capturing the lag-effect of past fire influences on habitat 371 structure and therefore the birds. However the broad patterns of fire frequencies across the savanna 372 have been consistent over the last 20 years (Russell-Smith 2002), therefore the fire history prior to our 373 study is likely to be broadly reflected in the vegetation patterns. 374 (A) Conclusion 375 376 Australian tropical savannas have “fire weather” conducive to burning for most of the dry season, but 377 burning is limited by the number of ignitions and fuel reduction by previous fires or intensive grazing 378 (Ash, McIvor et al. 1997; Bradstock 2010). Therefore a regime of moderate fire frequency early in 379 the dry season may be the best way to prevent the large destructive late-dry season fires (Williams, 380 Griffiths et al. 2002; Bird, Bird et al. 2008). Our study has shown that frequent fires late in the dry 381 season are likely to be detrimental to many bird species within savannas. In addition, we have 382 identified species that are particularly sensitive to high fire frequency, and those that benefit from 383 frequent fire when it occurs outside the late-dry season. Our study could therefore aid design of fire 384 regimes tailored to the management of species, particularly threatened or endemic species that are fire 385 sensitive (Garnett and Crowley 2002; Perry, Fisher et al. In Press). The relative congruence in 386 species’ sensitivities to the range of fire frequencies as demonstrated by our three modelling 387 approaches (environmental occurrence, their probability of presence, and change in suitable area 388 under hypothetical fire frequency increases) gives us confidence that our methods show meaningful 389 patterns. We suggest that this approach may be useful for identifying at risk species in other biomes 390 facing changes in fire regimes. 391 392 (A) References 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 Accatino, F., C. De Michele, et al. (2010). Tree-grass co-existence in savanna: Interactions of rain and fire. Journal of Theoretical Biology 267, 235-242. Allan, G. E. & R. I. Southgate (2002). Fire regimes in the spinifex landscapes of Australia. Flammable Australia: The Fire Regimes and Biodiversity of a Continent. (ed. by R. A. Bradstock, J. E. Williams and A. M. Gill), pp. 145-176. Cambridge University Press. Andersen, A. & B. D. Hoffman (2010). Conservation value of low fire frequency in tropical savannas: ants in monsoonal northern Australia. Austral Ecology. Andersen, A. N., G. D. Cook, et al. (2005). Fire frequency and biodiversity conservation in Australian tropical savannas: implications from the Kapalga fire experiment. Austral Ecology 30, 155167. Ash, A. J., J. G. McIvor, et al. (1997). Building grass castles: integrating ecology and management of Australia's tropical tall grass rangelands. Rangeland Journal 19, 123-144. Barlow, J. & C. A. Peres (2004). Ecological responses to El Niño–induced surface fires in central Brazilian Amazonia: management implications for flammable tropical forests. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences 359, 367-380. Barrett, G., A. Silcocks, et al. (2003). The New Atlas of Australian Birds, Melbourne, Royal Australasian Ornithologists Union. Beer, T. & A. L. Williams (1995). Estimating Australian forest fire danger under conditions of doubled carbon dioxide concentrations. Climatic Change 29, 169-188. Bird, R. B., D. W. Bird, et al. (2008). The fire stick farming hypothesis: Australian Aboriginal foraging strategies, biodiversity, and anthropogenic fire mosaics. Proceedings of the National Academy of Sciences, -. Blakers, M., S. J. J. F. Davies, et al. (1984). The Atlas of Australian Birds, Melbourne University Press: Melbourne. Bond, W. J. & C. L. Parr (2010). Beyond the forest edge: Ecology, diversity and conservation of the grassy biomes. Biological Conservation 143, 2395-2404. Bowman, D. M. J. S. (1998). The impact of Aboriginal landscape burning on the Australian biota. New Phytologist 140, 385-410. Bowman, D. M. J. S. & W. J. Panton (1993). Decline of Callitris intratropica R. T. Baker & H. G. Smith in the Northern Territory: Implications for pre- and post-European colonization fire regimes. Journal of Biogeography 20, 373-381. Bowman, D. M. J. S., B. A. Wilson, et al. (1988). Response of Eucalyptus forest and woodland to four fire regimes at Munmarlary, Northern Territory, Australia. Journal of Ecology 76, 215-232. Bradstock, R. A. (2010). A biogeographic model of fire regimes in Australia: current and future implications. Global Ecology & Biogeography 19, 145-158. Bradstock, R. A., M. Bedward, et al. (2005). Which mosaic? A landscape ecological approach for evaluating interactions between fire regimes, habitat and animals. Wildlife Research 32, 409423. Braithwaite, R. W. & J. Estbergs (1987). Fire-birds of the Top End. Australian Natural History 22, 299-302. Braithwaite, R. W. & J. A. Estbergs (1985). Fire patterns and woody vegetation trends in the Alligator Rivers region of northern Australia. Ecology and Management of the World's Savannas. (ed. by J. C. Tothill and J. J. Mott). Australian Academy of Science, Canberra. Brook, B. W. (2008). Synergies between climate change, extinctions and invasive vertebrates. Wildlife Research 35, 249-252. Cary, G. J. (2002). Importance of a changing climate for fire regimes in Australia. Flammable Australia: The Fire Regimes and Biodiversity of a Continent. (ed. by R. A. Bradstock, J. E. Williams and A. M. Gill). Cambridge University Press. Cary, G. J. & J. C. Banks (2000). Fire regime sensitivity to global climate change: an Australian perspective. Advances in Global Change Research: Biomass Burning and its Inter- 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 Relationships with the Climate System. (ed. by J. L. Innes, M. Beniston and M. M. Verstraete), pp. 233-246. Springer Netherlands. Clarke, M. F. (2008). Catering for the needs of fauna in fire management: science or just wishful thinking? Wildlife Research 35, 385-394. Davis, M. A., D. W. Peterson, et al. (2000). Restoring savanna using fire: impact on the breeding bird community. Restoration Ecology 8, 30-40. Dean, W. R. J. (1987). Birds associating with fire at Nylsvley Nature Reserve, Transvaal. Ostrich 58, 103-106. Driscoll, D. A., D. B. Lindenmayer, et al. (2010). Fire management for biodiversity conservation: Key research questions and our capacity to answer them. Biological Conservation 143, 1928-1939. Elith, J., C. H. Graham, et al. (2006). Novel methods improve prediction of species' distributions from occurrence data. Ecography 29, 129-151. Elith, J. & J. R. Leathwick (2009). Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics 40, 677-697. Environmental Protection Agency (2004). WildNet (Database). Environmental Protection Agency). http://www.derm.qld.gov.au/wildlife-ecosystems/wildlife/wildlife_online/index.html, Brisbane. Felderhof, L. & D. Gillieson (2006). Comparison of fire patterns and fire frequency in two tropical savanna bioregions. Austral Ecology 31, 736-746. Fitzherbert, J. C. & D. J. Baker-Gabb (1988). Australasian grasslands and their avifauna. Ecology and Conservation of Grassland Birds. (ed. by P. D. Goriup). International Council for Bird Preservation, Cambridge. Franklin, D., P. Whitehead, et al. (2005). Geographic patterns and correlates of the decline of granivorous birds in northern Australia. Wildlife Research 32, 399-408. Franklin, D. C. (1999). Evidence of disarray amongst granivorous bird assemblages in the savannas of northern Australia, a region of sparse human settlement. Biological Conservation 90, 53-63. Friend, G. R. & J. A. Taylor (1985). Habitat preferences of small mammals in tropical open-forest of the Northern Territory. Australian Journal of Ecology 10, 173-185. Garnett, S. & R. Bredl (1985). Birds in the vicinity of Edward River Settlement. Part II. Discussion, references, list of passerines. The Sunbird 15, 25-40. Garnett, S. & G. M. Crowley (1995). The decline of the Black Treecreeper Climacteris picumnus melanota on Cape York Peninsula. Emu 95, 66-68. Garnett, S. T. & B. W. Brook (2007). Modelling to forestall extinction of Australian tropical birds. Journal of Ornithology 148, S311-S320. Garnett, S. T. & G. M. Crowley (2002). Recovery Plan for the golden-shouldered parrot Psephotus chrysopterygius 2003-2007. Report to Environment Australia). Queensland Parks and Wildlife Service, Brisbane, Canberra. Gillett, N. P., A. J. Weaver, et al. (2004). Detecting the effect of climate change on Canadian forest fires. Geophysical Research Letters 31, L18211. Griffioen, P. A. & M. F. Clarke (2002). Large-scale bird-movement patterns evident in eastern Australian atlas data. Emu 102, 99-125. Jansen, R., R. M. Little, et al. (1999). Implications of grazing and burning of grasslands on the sustainable use of francolins (Francolinus spp.) and on overall bird conservation in the highlands of Mpumalanga province, South Africa. Biodiversity and Conservation 8, 587-602. Keith, D. A., J. E. Williams, et al. (2002). Fire management and biodiveristy conservation: key approaches and principles. Flammable Australia: The Fire Regimes and Biodiversity of a Continent. (ed. by R. A. Bradstock, J. E. Williams and A. M. Gill). Cambridge University Press. Krawchuk, M. A., M. A. Moritz, et al. (2009). Global pyrogeography: the current and future distribution of wildfire. PLoS ONE 4, e5102. Kutt, A. S. & J. C. Z. Woinarski (2007). The effects of grazing and fire on vegetation and the vertebrate assemblage in a tropical savanna woodland in north-eastern Australia. Journal of Tropical Ecology 23, 95-106. 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 Ling, C. X., J. Huang, et al. (2003). AUC: a better measure than accuracy in comparing learning algorithms. Advances in Artificial Intelligence. (ed. by Y. Xiang and B. Chaib-draa). Springer-Verlag, Berlin Heidelberg. Liu, C., P. M. Berry, et al. (2005). Selecting thresholds of occurrence in the prediction of species distributions. Ecography 28, 385-393. Liu, Y., J. Stanturf, et al. (2010). Trends in global wildfire potential in a changing climate. Forest Ecology and Management 259, 685-697. Martin, T. G. & H. P. Possingham (2005). Predicting the impact of livestock grazing on birds using foraging height data. Journal of Applied Ecology 42, 400-408. McMahon, J. P., M. F. Hutchinson, et al. (1995). ANUCLIM user's guide, Version 1). Centre for Resource and Environmental Studies, Australian National University, Canberra, Australian Capital Territory, Australia. Moretti, M. & C. Legg (2009). Combining plant and animal traits to assess community functional responses to disturbance. Ecography 32, 299-309. Mott, J. J., J. Williams, et al. (1985). Australian savanna ecosystems. Ecology and Management of the World's Savannas. (ed. by J. C. Tothill and J. J. Mott). Australian Academy of Science, Canberra. Mouillot, F. & C. B. Field (2005). Fire history and the global carbon budget: a 1°× 1° fire history reconstruction for the 20th century. Global Change Biology 11, 398-420. Murphy, S. A., S. M. Legge, et al. (2010). The effects of early and late-season fires on mortality, dispersal, physiology and breeding of red-backed fairy-wrens (Malurus melanocephalus). Wildlife Research 37, 145-155. Noble, J. C. (1989). Fire studies in mallee (Eucalyptus spp.) communities of western New South Wales: the effects of fires applied in different seasons on herbage productivity and their implications for management. Australian Journal of Ecology 14, 169-187. O'Reilly, L., D. Ogada, et al. (2006). Effects of fire on bird diversity and abundance in an East African savanna. African Journal of Ecology 44, 165-170. Pearson, R. G. & T. P. Dawson (2003). Predicting the impacts of climate change on the distribution of species: are bioclimate envelope models useful? Global Ecology & Biogeography 12, 361371. Perry, J., A. Fisher, et al. (In Press). Status and habitat of the Carpentarian grasswren (Amytornis dorotheae) in the Northern Territory. Emu. Phillips, S. J., R. P. Anderson, et al. (2006). Maximum entropy modeling of species geographic distributions. Ecological Modelling 190, 231-259. Pitman, A. J., G. T. Narisma, et al. (2007). The impact of climate change on the risk of forest and grassland fires in Australia. Climatic Change 84, 383-401. Prates-Clark, C. D. C., S. S. Saatchi, et al. (2008). Predicting geographical distribution models of high-value timber trees in the Amazon Basin using remotely sensed data. Ecological Modelling 211, 309-323. Price, O. & D. M. J. S. Bowman (1994). Fire-stick forestry: a matrix model in support of skilful fire management of Callitris intratropica R. T. Baker by north Australian Aborigenes. Journal of Biogeography 21, 573-580. Reside, A. E., J. J. VanDerWal, et al. (2010). Weather, Not Climate, Defines Distributions of Vagile Bird Species. PLoS ONE 5, e13569. Roques, K. G., T. G. O'Connor, et al. (2001). Dynamics of shrub encroachment in an African savanna: relative influences of fire, herbivory, rainfall and density dependence. Journal of Applied Ecology 38, 268-280. Rowley, I. & E. Russell (1993). The Purple-crowned Fairy-wren Malurus coronatus. II. Breeding Biology, Social organization, Demography and Management. Emu 93, 235-250. Russell-Smith, J. (2002). Pre-contact Aboriginal, and contemporary fire regimes of the savanna landscapes of northern Australia: patterns, changes and ecological processes. Australian Fire Regimes: Contemporary Patterns (April 1998-March 2000) and Changes Since European Settlement. (ed. by J. Russell-Smith, R. Craig, A. M. Gill, R. Smith and J. Williams). Australia State of the Environment Second Technical Paper Series (Biodiversity), Department of the Environment and Heritage, Canberra. 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 Russell-Smith, J. & D. M. J. S. Bowman (1992). Conservation of monsoon rainforest isolates in the Northern Territory, Australia. Biological Conservation 59, 51-63. Sirami, C., C. Seymour, et al. (2009). The impact of shrub encroachment on savanna bird diversity from local to regional scale. Diversity and Distributions 15, 948-957. Stephens, S. L. & L. W. Ruth (2005). Federal forest-fire policy in the United States. Ecological Applications 15, 532-542. Uys, R. G., W. J. Bond, et al. (2004). The effect of different fire regimes on plant diversity in southern African grasslands. Biological Conservation 118, 489-499. Valentine, L. E., L. Schwarzkopf, et al. (2007). Burning season influences the response of bird assemblages to fire in tropical savannas. Biological Conservation 137, 90-101. Vanderwal, J., L. P. Shoo, et al. (2009). Abundance and the environmental niche: environmental suitability estimated from niche models predicts the upper limit of local abundance. The American Naturalist 174, 282-291. Westerling, A. L., H. G. Hidalgo, et al. (2006). Warming and earlier spring increase western U.S. forest wildfire activity. Science 313, 940-943. Williams, A. A. J., D. J. Karoly, et al. (2001). The sensitivity of Australian fire danger to climate change. Climatic Change 49, 171-191. Williams, R. J., J. Carter, et al. (2005). Carbon accounting, land management, science and policy uncertainty in Australian savanna landscapes: introduction and overview. Australian Journal of Botany 53, 583-588. Williams, R. J., A. D. Griffiths, et al. (2002). Fire regimes and biodiversity in the savannas of northern Australia. Flammable Australia: The Fire Regimes and Biodiversity of a Continent. (ed. by R. A. Bradstock, J. E. Williams and A. M. Gill). Cambridge University Press. Woinarski, J. C. Z. (1990). Effects of fire on the bird communities of tropical woodlands and open forests in northern Australia. Australian Journal of Ecology 15, 1-22. Woinarski, J. C. Z., M. Armstrong, et al. (2010). Monitoring indicates rapid and severe decline of native small mammals in Kakadu National Park, northern Australia. Wildlife Research 37, 116-126. Woinarski, J. C. Z., C. Brock, et al. (1999). Response of birds and reptiles to fire regimes on pastoral land in the Victoria River District, Northern Territory. Rangeland Journal 21, 24-38. Woinarski, J. C. Z., D. J. Milne, et al. (2001). Changes in mammal populations in relatively intact landscapes of Kakadu National Park, Northern Territory, Australia. Austral Ecology 26, 360370. Woinarski, J. C. Z., J. Risler, et al. (2004). Response of vegetation and vertebrate fauna to 23 years of fire exclusion in a tropical Eucalyptus open forest, Northern Territory, Australia. Austral Ecology 29, 156-176. 590 Figure 1. Environmental space plots and Maxent variable response curves for four exemplar species 591 within the Australian tropical savannas. The species shown are Gouldian finch Erythrura gouldiae 592 ((a) environmental space plot and (b), Maxent response curves), red-backed fairy-wren Malurus 593 melanocephalus (c, d), star finch Neochmia ruficauda (e, f) and white-streaked honeyeater Trichodere 594 cockerelli (g, h). The environmental space plots (a, c, e, g) show the available fire frequency and late- 595 fire frequency versus annual rainfall. Results for the whole ATS region in greyscale. Darker grey 596 represents greater prevalence. The values of fire frequency and late-fire frequency versus annual 597 rainfall for which the species have been recorded is shown in yellow-to-red scale. For the species, the 598 lighter colour (yellow) represents the most records, where the red represents fewer. The response 599 curves (b, d, f, h) show the probability of presence of each species depending on the fire frequency or 600 late-fire frequency, assuming that each of these variables in turn are the only variable in the model. Figure captions: 601 602 Figure 2. The proportional change in suitable area for species from the current levels of fire 603 frequency (a, c, e, g) and late-fire frequency (b, d, f, h), to a six-increment increases in fire frequency 604 variables, for the same four species within Australian tropical savannas as in Fig 1. In the maps, the 605 dark blue represents area that is always suitable, the dark grey is always unsuitable, and the light blue 606 is the area that has changed in suitability. For the plots, the Y-axis represents the proportion of the 607 entire Australian tropical savanna region that is suitable, depending whether the fire frequency is at 608 current levels (0) or increasing by each factor 1 – 6. 609 610 Figure 3. Histogram representing the Pearson correlation (ρ) for the relationship between increasing 611 fire frequency or late-fire frequency and the proportion of suitable habitat for each species. Figure 2a 612 shows the ρ for 44 species largely confined to the northern Australian savannas, whereas 2b shows the 613 ρ for 207 species (inclusive of the original 44), which incorporates species found outside the savanna. 614 Model results that do not include the full range of species’ environmental tolerances should be treated 615 with caution. 616 617 618 619 Table 1. The mean, standard deviation and range of percentage contributions that each variable made 620 to the Maxent models for the 44 species. Variable Mean ± Standard Deviation Range Fire Frequency 3.9 ± 4.5 0.6 – 23.1 Late-Fire Frequency 3.1 ± 2.7 0.04 – 11.9 Mean Annual Temperature 7.9 ± 10.1 0.3 – 61.5 Temperature Seasonality 24.3 ± 17.2 0.9 – 67.5 8.9 ± 8.3 0.8 – 38.2 Annual Precipitation 37.8 ± 22.1 2.6 – 78.3 Precipitation Seasonality 10.6 ± 11.5 0.3 – 50.7 3.3 ± 3.4 0.1 – 15.0 Max Temp Warmest Period Precipitation Driest Quarter 621 622 623 624 625 Table 2. The number (and percentage) of the 44 modelled species predicted to face an increase or 626 decrease in suitable area as a result of a factor of six increase in both fire frequency and late-fire 627 frequency. The mean change in suitable area (either increase or decrease) and standard error are 628 shown. 629 Fire Frequency Mean Change in Late-fire Mean Change in Increase Suitable Area Frequency Increase Suitable Area # increasing species 15 (34%) 4.4% ± 3.2 1 (2%) 0.8% ± 0 # decreasing species 29 (66%) 9.4% ± 7.4 43 (98%) 7.2% ± 6.2