Pyric Herbivory and the Nexus Between Forage, Fire and Native and Introduced Large Grazing Herbivores in Australian Tropical Savannas

Earth’s tropical savannas typically support high biomass of diverse grazing herbivores that depend on a highly fluctuating resource: high-quality forage. An annual wet–dry cycle, fire and herbivory combine to influence forage quality and availability throughout the year. In the savannas of northern Australia, a depauperate suite of large native (marsupial) herbivores (wallaroos [Osphranter spp.] and the agile wallaby [Notamacropus agilis]) compete for resources with non-native large herbivores introduced in the late nineteenth century, particularly bovines (feral and managed cattle [Bos spp.] and feral water buffalo [Bubalus bubalis]) that now dominate the landscape. Anecdotal reports of recent population declines of large macropods and negative impacts of bovines highlight the need to better understand the complex relationship between forage, fire and abundance of native and introduced large herbivores. The pyric herbivory conceptual model, which posits complex feedbacks between fire and herbivory and was developed outside Australia, predicts that native and introduced large herbivores will both respond positively to post-fire forage production in Australian savannas where they co-occur. We used grazing exclosures, forage biomass and nutrient analyses and motion-sensor camera-trapping to evaluate the overall robustness of the pyric herbivory model in the Australian context, specifically whether forage quantity and quality are impacted by herbivory, season and fire activity, and which forage attributes most influence large grazing herbivore abundance. Forage quantity, as measured by live, dead and total herbaceous biomass and proportion of biomass alive, was higher inside herbivore exclosures, even at relatively low densities of herbivores. Forage quality, as measured by fibre content, was not affected by herbivory, however, crude protein content of live herbaceous biomass was greater outside herbivore exclosures. Recent fire was an important predictor of all measures of forage quantity and quality. Recent fire occurrence decreased overall quantity (biomass) but increased quality (decreased fibre content and increased crude protein content); late dry season fires resulted in forage with the highest crude protein content. The predictions of the pyric herbivory conceptual model are consistent with observations of the feeding behaviour of introduced bovines and some large macropods in northern Australian savannas, lending support to the global generality of pyric herbivory in fire-prone grassy biomes.

and feral water buffalo [Bubalus bubalis]) that now dominate the landscape. Anecdotal reports of recent population declines of large macropods and negative impacts of bovines highlight the need to better understand the complex relationship between forage, fire and abundance of native and introduced large herbivores. The pyric herbivory conceptual model, which posits complex feedbacks between fire and herbivory and was developed outside Australia, predicts that native and introduced large herbivores will both respond positively to post-fire forage production in Australian savannas where they co-occur. We used grazing exclosures, forage biomass and nutrient analyses and motion-sensor camera-trapping to evaluate the overall robustness of the pyric herbivory model in the Australian context, specifically whether forage quantity and quality are impacted by herbivory, season and fire activity, and which forage attributes most influence large grazing herbivore abundance. Forage quantity, as measured by live, dead and total herbaceous biomass and proportion of biomass alive, was higher inside herbivore exclosures, even at relatively low densities of herbivores. Forage quality, as measured by fibre content, was not affected by herbivory, however, crude protein content of live herbaceous biomass was greater outside herbivore exclosures. Recent fire was an important predictor of all measures of forage quantity and quality. Recent fire occurrence decreased overall quantity (biomass) but increased quality (decreased fibre content and increased crude protein content); late dry season fires resulted in forage with the highest crude protein content. The predictions of the pyric herbivory conceptual model are consistent with observations of the feeding behaviour of introduced bovines and some large macropods in northern Australian savannas, lending support to the global generality of pyric herbivory in fire-prone grassy biomes.

HIGHLIGHTS
The post-fire production of forage and response of some large grazing herbivores in northern Australian savannas are consistent with the pyric herbivory conceptual model. Introduced bovines are strongly attracted to recently burnt areas however the response of large native macropods is more variable. Although the short-term effect of fire is to decrease forage biomass it increases the nutritional quality of forage such that the biomass of high-quality forage increases after fire.

INTRODUCTION
Savanna ecosystems occupy large areas of the tropics, representing approximately 22% of Earth's total land area (Ramankutty and Foley 1999). Their formation and maintenance are based, in part, on highly fluctuating dynamics between fire and herbivory (Van Langevelde and others 2003; Archibald and others 2005;Bond and Keeley 2005). Both fire and herbivory consume herbaceous biomass, thereby indirectly affecting each other in space and time. The interplay and feedbacks between fire and herbivory have been termed pyric herbivory (Figure 1;Fuhlendorf and Engle 2001). According to the pyric herbivory conceptual model, consumption of herbaceous biomass by fire reduces the amount of available forage (edible herbaceous biomass) but increases the quality of regrowth, or 'green pick', attracting herbivores to the area. At the same time, the effects of herbivory on vegetation, that can be dependent on the intensity of grazing, affects fuel loads and fire spread thereby creating feedbacks between fire spatio-temporal patterns, forage quality and subsequent focal grazing (Fuhlendorf and Engle 2004;Archibald and others 2005). This conceptual model has never been evaluated outside North America and southern Africa, although may be generally applicable to a much wider range of Earth's grassy biomes.
In the tropical savannas of northern Australia, European colonists introduced domestic herds of a suite of large-bodied, bulk-feeding ruminant herbivores throughout the nineteenth century that, in turn, established feral populations. Immediately prior to the arrival of Europeans, these landscapes were occupied by a depauperate herbivore guild consisting entirely of macropods (that is, members of the marsupial family Macropodidae, including kangaroos and wallabies) with more selective feeding strategies, and completely lacked megaherbivores following a wave of late-Pleistocene extinctions (all extant macropod species are < 100 kg). Previous research suggests that at least three of the introduced herbivores, water buffalo (Bubalis bubalis), banteng (Bos javanicus) and cattle (Bos taurus), fill a feeding niche left vacant by the Pleistocene extinction of the marsupial megafauna (Bowman and others 2010; Reid and others 2020a); however, there is evidence of competition between introduced bovines and large native herbivores (Reid and others 2020b). Both in Australia and elsewhere, competition between introduced livestock and native herbivores is expected to intensify in the future, with trends in agricultural land-use showing a net redistribution towards the tropics (Mishra and others 2002;Stewart and others 2002;Young and others 2005;Foley and others 2011;Ogutu and others 2011;Kutt and others 2012), highlighting the importance of understanding how native and introduced herbivores respond to fluctuating forage conditions. This is especially true when native and introduced herbivores are ecotaxonomically dissimilar, as in Australia (ungulate vs. macropod), in contrast to African and North American savannas where both groups are ungulates and have strong functional similarities (Veblen and others 2016). There are few places within the Australian tropical savannas that do not have substantial numbers of introduced herbivores and given the evidence of competition between the herbivore groups, the more relevant question is not how native and introduced herbi-vores behave in absence of one another but whether the pyric herbivory model functionally predicts behaviour in the context of both groups.
The pyric herbivory concept has been developed and tested on ungulates in grass-dominated ecosystems of North America and southern Africa (Fuhlendorf and Engle 2004;Archibald and others 2005;Fuhlendorf and others 2009;Allred and others 2011a;Kimuyu and others 2017), but fire is thought to have been used to manage forage nutrients and habitat mosaics for the maintenance of populations of macropods in tropical savannas by Australian Aboriginal people for thousands of years (Bowman and others 2001;Vigilante and others 2009). Based on the above research in North America and Africa and traditional ecological knowledge recorded in Australia, we expect that abundance of all large herbivores in Australian savannas will respond positively to recently burnt areas.
In the more diverse southern African herbivore assemblages containing all ungulates, a body size effect on the intensity of response to burnt areas has been observed with smaller-bodied herbivores responding more strongly than larger-bodied herbivores (Kimuyu and others 2017;Odadi and others 2017). Applying these observations to Australian savannas we might expect to find that native macropods have a stronger response than introduced bovines. However, recent research has found an unexpected negative association between large macropods and recently burnt areas where feral cattle are well-established but remain at relatively low densities compared to northern Australian commercial pastoral operations (Reid and others 2020b). The antagonistic relationship between cattle and macropods may signify that disrupted natural herbivore assemblages and altered fire regimes can lead to complex and unpredictable native herbivore responses to burning. For example, in Africa domestic cattle have been shown to influence habitat preference, selectivity and diet composition of native ungulate herbivores (Fritz and others 1996).
In comparison to elsewhere in the world, Australia's tropical savannas have nutrient-poor soils that support lower diversity and biomass of large herbivores, including introduced large herbivores that are evolutionarily distinct from the native large herbivores (Calaby 1980;Mott and others 1985;Freeland 1990). There has been substantial research on relationships between forage quantity and quality, fire and herbivores in grassy landscapes in North America and southern Africa, but not in Australian savannas. Testing whether the pyric herbivory model holds across the Earth's major fire-prone grassy biomes is critical to understanding the generality of this important ecological concept. Hence, we examined the nexus between forage, fire and herbivory, in order to evaluate the applicability of the pyric herbivory model -summarised as a conceptual diagram in Figure 1-to Australian tropical savannas. Combining methods from the fields of plant and wildlife ecology (herbivore exclosures, forage chemical analyses and remote camera trapping), we tested key linkages in the conceptual model. In particular, we focused on the hypotheses that: (1) fire causes a short-term increase in the nutritional quality of forage; (2) despite decreasing total forage biomass, fire causes a short-term increase in the biomass of high-quality forage; and (3) large herbivores (both native and introduced) are strongly attracted to recently burnt areas to graze.

Study Area
This study was undertaken in the tropical savannas of northern Australia with field sites in two regions: (1) North Kimberley bioregion, Western Australia; and (2) Arnhem Land, Northern Territory. showing relationships between forage, fire and large grazers in a fire-prone grassy biome (Fuhlendorf and Engle 2004;Archibald and others 2005). Fire and herbivory reduce the quantity of forage on the landscape (linkages 2 and 5) while increasing the quality of forage (linkages 1 and 4). Large grazers and landscape fire indirectly influence each other, and their relationship is mediated by forage. Large grazers can reduce fire frequency by reducing available fuel (linkages 5 and 6). Landscape fire improves the post-burn forage quality thus attracting more grazers to the area (linkage 3). Solid lines indicate positive relationships and dashed lines indicate negative relationships.

The North Kimberley
The North Kimberley (NK) sites were in the Uunguu Indigenous Protected Area (Uunguu IPA; approximately 8000 km 2 ), declared in 2011 and managed by Wunambal Gaambera Aboriginal Corporation (WGAC,Kalumburu,Australia;Figure 2a), encompassing diverse savannas overlaying both fertile and infertile substrates, primarily derived from volcanic rocks and siliceous sediments, respectively. The climate is defined by a monsoonal wet-dry cycle with mean annual rainfall of 1100-1600 mm across a steep latitudinal gradient, approximately 90% of which falls during the 5month wet season (December-April; Bureau of Meteorology 2018).
The mean fire return interval during the project (2014-2017) was 2.5 years with 73% of fires occurring in the early dry season (before 1 August) based on a satellite-derived (MODIS) fire history (NAFI 2018). Significant changes to the prehistorical fire regime of patch burning throughout the year occurred in the mid twentieth century in what is now the Uunguu IPA because the Wunambal Gaambera people, the Aboriginal people of the region, moved to nearby settlements. As a result, traditional fire management was replaced by large, frequent wildfires in the latter part of the dry season. Traditional fire regimes have become better replicated in recent decades with the establishment of fire management programs controlled by local Aboriginal people (Vigilante andothers 2004, 2017).
The large grazing macropods present at the North Kimberley study site are the common wallaroo (Osphranter robustus), antilopine wallaroo (O. antilopinus) and agile wallaby (Notamacropus agilis). Pastoral leases adjacent to the Uunguu IPA were established in the early 1900s, becoming a source of feral cattle (Bos taurus) dispersal, which have now reached higher densities than native macropod herbivores (Reid and others 2020b).

Arnhem Land
Arnhem Land (AL) covers a large region overlaying mainly sandy, infertile substrates (c. 97,000 km 2 ) in the north-eastern corner of the Northern Territory and was declared an Aboriginal reserve in the 1930s, subsequently owned and managed by an Aboriginal land trust. Field sites were located near Kolorbidahdah on the Cadell River, an isolated Aboriginal outstation (very small settlement, typically consisting of a single extended family group), with a continuous history of Aboriginal occupation, except for a few years in the 1950s (Yibarbuk and others 2001). Mean annual rainfall is 1,080 mm, with approximately 90% falling during the 5month wet season (December-April; Bureau of Meteorology 2018).
The mean fire return interval was 2.7 years with 48% early dry season fires for the 100 km 2 area surrounding Kolorbidahdah (NAFI 2018). Kolorbidahdah is managed in a traditional manner with small patches being ignited by hand throughout the dry season, mainly by the Aboriginal extended family residing at the outstation. The suite of grazing macropods in Arnhem Land is the same as in the North Kimberley with the addition of the black wallaroo (O. bernardus), which is endemic to the Arnhem Plateau (Telfer and others 2008). The main introduced herbivore is the water buffalo (Bubalus bubalis), originally introduced to the Northern Territory mainland in 1827, dominating the sub-coastal plains across the Northern Territory since the 1880s (Letts 1962).

Site Establishment
Monitoring sites (n = 14; Figure 2b,c) were selected to span geological formations that yield comparatively fertile soils (n = 7; Carson Volcanics) and relatively infertile soils (n = 7; King Leopold Sandstone and Colluvium and Alluvium in the North Kimberley and Marlgowa Sandstone in Arnhem Land) based on geological classifications (Geoscience Australia 2012; DMPWA 2010). Vegetation at the field sites is dominated by eucalypt (Eucalyptus and Corymbia spp.) savanna with a physiognomy of woodland and open forest (as defined by Specht 1970). The fertile sites are characterized by gently undulating to hilly terrain with shallow stony soils dominated by a mixture of perennial and annual tussock tall grasses; infertile sites are characterized by gently undulating sandstone terrain with sandy soils of variable depth dominated by a mixture of hummock grasses in the endemic Australian genus Triodia and perennial and annual tussock grasses (DAFWA Undated;DENRNT 2000). Sites were selected based on fire management and utilisation by both large macropods and feral bovines. Aboriginal Traditional Owners located sites in recently burnt and unburnt areas known to be historically good macropod habitat and areas were searched for macropod and bovine dung to confirm presence of both herbivore groups. Dingo abundance at monitoring sites, as measured by dingoes per camera trap night, was similar for both regions (0.011 ± 0.003 SE at Arnhem Land sites and 0.008 ± 0.001 SE at North Kimberley sites; Reid 2019).  inclusive, in the North Kimberley. Arnhem Land sites were sampled 1-2 times per year between July and October and North Kimberley sites were sampled 2-3 times per year between June and November after establishment (four North Kimberley sites were not established until 2016) for a total of 89 forage sampling occasions across all sites. Forage cover, height, moisture, crude protein (CP) and fibre (amylase and sodium sulphite treated neutral detergent fibre [aNDF]) were measured at each sampling occasion for both standing dead and live herbaceous biomass (Table 1). Standing biomass was measured at the final sampling occasion each dry season by clipping all herbaceous vegetation inside each quadrat and separating dead and live biomass. Standing biomass measurements were significantly related (p < 0.001) with forage volume (forage height multiplied by forage cover) thus we were able to estimate standing biomass for sampling occasions throughout the dry season (''Appendix 1''). Annual production, proportion of biomass alive and an index of dead to live herbaceous biomass were calculated for the standing biomass in each quadrat (Table 1).

Forage Quantity and Quality
Forage moisture and nutrients (CP and aNDF) were sampled from the areas directly surrounding quadrat sets throughout the dry season so as not to reduce biomass from inside the quadrats. A minimum of 10 g each of standing dead and live forage were clipped and placed separately into sealed plastic bags per monitoring site to represent the species of grass and herbs inside the quadrats. At the end of each dry season when biomass clipping occurred, a subsample was kept for moisture and nutrient measurements. Samples were weighed in the field to provide wet weights and later dried in an oven at 60˚C for 48 h then reweighed. Dried samples were milled to pass through a 1 mm sieve. Crude protein was determined by combustion (AOAC Official Method 990.03. 2005) with a CN628 Carbon/Nitrogen Determinator. To provide a measure of the fibrous bulk of the forage, amylase and sodium sulphite treated neutral detergent fibre (aNDF) was assessed with ANKOM Technology Method 6 (ANKOM Technology, Macedon, NY, USA) using solutions as in Van Soest and others (1991). Forage samples collected in 2015 were also analysed for phosphorus (CEM, Undated) but it was found to be significantly correlated with crude protein (r = 0.84, p < 0.05) so subsequently only crude protein was analysed (''Appendix 2'').

Remote Camera Trapping
Remote (motion-activated) camera traps were used as an indication of large herbivore abundance at monitoring sites in the time surrounding forage sampling occasions. Five RECONYX PC800 Hyperfire cameras (RECONYX, Inc.) were deployed at each monitoring site at various periods between 2015 and 2017 ranging from 31 to 294 days (Figure 2b,c). Approximately 25 m 2 areas around the cameras were cleared with a whipper snipper and cameras were attached to robust trees 1 m off the ground for fire protection and to reduce false triggers due to vegetation blowing in the wind. Clipping the grass was necessary but likely to stimulate localised new growth if moisture conditions were suitable, however, we assume that a small area within a larger unburnt area would not be drawing in animals that were not already in the vicinity. Cameras were set to trigger mode with motion sensor on, medium/high sensitivity, 3 photos per trigger with a 1 s interval between photos and a 1 min quiet period between triggers. Images were classified by species (cattle, water buffalo, agile wallaby, wallaroo) and number of animals. Due to the difficulty of positively identifying antilopine, common and black wallaroos in the night-time black and white photos these large-bodied species were lumped together and are referred to as ''wallaroos.''

Herbivory and Forage Attributes
Generalized linear mixed modelling (GLMM) was used to evaluate the effect of herbivory on measures of forage quantity (live, dead and total standing biomass, proportion of biomass alive) and quality (live and dead forage moisture, CP, aNDF). Due to the destructive sampling process explained above, forage quality measurements for both grazed and ungrazed quadrats could only be made at the end of the dry season. Forage measurements were modelled by plot location (inside or outside exclosure) and compared to null models; models contained exclosure as a random variable. Model fit was evaluated using second order Akaike's Information Criterion for large and small sample sizes (AICc). When the null model had the lowest AICc or fell within the DAICc < 2 criteria, the conclusion was ''no effect of herbivory.''

Environment and Forage Attributes
Linear modelling was used to evaluate the influence of seasonality (time since end of wet season) and fire (time since fire and fire season) on forage quantity and quality; substrate fertility and savanna region were included in models as nuisance variables. Dependant variables were transformed as needed to achieve a normal distribution of residuals. Model fit was evaluated using AICc and in the case of multiple models being within DAICc < 2 model averaging was conducted using Akaike weights and the significant variables within these models were graphically presented. Model visualization graphs were produced with 'visreg' package in R (Breheny and Burchett 2017). For these analyses, the end of wet season was defined by the date 95% of annual rainfall was received following the previous dry season. Fire season was catego-rized into early dry season fire and late dry season fire (before or after 1 August respectively), and unburnt in the dry season of sampling. Total annual production (g m -2 ) and proportion of biomass alive were selected as measures of forage quantity and CP (%) and aNDF (%) of live biomass for forage quality. Since forage quality measures required destructive sampling, our forage nutrient dataset (CP and aNDF) only contains full dry season records for forage outside the exclosures. However, due to the previously mentioned relationship between forage volume and biomass, our data set has complete records of forage quantity from protected quadrats and we were able to evaluate quantity in the absence of herbivory utilising only inside quadrat data. Forage measurements were averaged by monitoring site for each sampling occasion and principal components analysis (PCA) was conducted to identify independent forage quantity and quality variables.

Large Herbivore Abundance and Forage Attributes
The number of bovines, wallaroos and agile wallabies recorded by camera trap for the 10 days before and 10 days after each forage sampling occasion (including the day of forage sampling) were calculated as a measure of large herbivore abundance. The log of camera trap days was used as an offset in zero-inflated Poisson models to account for camera traps that malfunctioned during the 20 days surrounding forage measurements.
In contrast to the models above where annual production and proportion of biomass alive were modelled, we selected total standing biomass and the ratio of dead and live biomass as predictor variables of forage quantity. Total standing biomass at time of sampling is more likely to influence large herbivore abundance than annual production because this is a direct measure of the available pasture resources. Similarly, dead:live biomass index gives a better measure of the disparity between dead and live than the proportion of biomass alive.
Crude protein of live biomass was significantly correlated with dead:live biomass index (p < 0.05) so live forage moisture (significantly correlated with crude protein of live biomass but not dead:live biomass index) was used alternatively along with aNDF of live biomass to represent measures of forage quality. Substrate fertility and savanna region were included in models as nuisance variables. Bootstrapping was used to generate confidence intervals for significant variables in the best model for each species. For each significant variable, data was sampled with replacement and the model run 1,000 times, and the median, 5 th and 95 th percentiles for the variable of interest were calculated from the model predictions, with all other variables held constant at their mode value.

Herbivory and Forage Attributes
Dead, live and total standing herbaceous biomass and the proportion of biomass alive were higher in ungrazed (inside exclosures) than grazed quadrats (Table 2; Figure 3a,b). Forage quality at the end of the dry season, as measured by forage moisture and fibre content of live and dead biomass, were best explained by the null model, showing no effect of herbivory. Mean estimates for aNDF content of live (71.0% ± 0.6 SE; range 56.0-88.6%) and dead biomass (71.3% ± 0.5 SE; range 51.8-86.5%) were high and almost identical by the end of the dry season while moisture contents of live (53.1% ± 1.2 SE; range 22.7-76.9%) was much greater than dead biomass (6.3% ± 0.9 SE; range 0.4-47.3%). However, crude protein content of live biomass was higher in grazed (7.4% ± 0.2 SE; range 2.6-15.1%) than ungrazed quadrats (7.0% ± 0.2 SE; range 2.4-13.2%; Table 2; Figure 3c) while there was no effect of herbivory on crude protein content of dead biomass, which was comparatively much lower (3.0% ± 0.1 SE; range 1.4-7.3%). Grazing by bovines alone, and by both bovines and macropods together, left less standing dead biomass than macropods alone. Grazing stimulated grass growth primarily occurred in areas with only bovine grazing (''Appendix 3'').

Environment and Forage Attributes
All analysed measures of forage quantity and quality were significantly associated with fire activity, but seasonal progression (time since end of wet season) was only related to forage quality (Table 3, Figure 4). Burnt sites had the lowest annual production but the highest proportion of live biomass, with no distinction between early and late dry season fires (Table 3a, Figure 4a,c). Fibre content of live biomass was lower at burnt sites with no distinction between early and late dry season fires (Table 3b, Figure 4d). However, sites with late dry season fires had the highest crude protein content of live biomass while unburnt sites had the lowest (Figure 4e). The proportion of biomass alive was highest immediately following the wet season, declining throughout the dry season with changing climatic conditions (Table 3a, Figure 4b).

Large Herbivore Abundance and Forage Attributes
A total of 2,171 animals were identified in camera trap images, including cattle (1,753), water buffalo (110), wallaroo (158) and agile wallaby (150). These figures would have included instances where the same individual was detected multiple times.
The influence of forage attributes on large herbivore abundance is complex, as indicated by the best-supported models for bovine and combined wallaroo abundance including all predictor variables: fibre and moisture of live biomass; total standing biomass; and dead:live biomass index (Table 3c, Figure 5). The number of introduced bovines (cattle at the North Kimberley sites and water buffalo at the Arnhem Land sites) was highest at sites with forage of low fibre content, low standing biomass and high forage moisture (corre-  (Figure 5a-c). There was also a positive relationship with dead:live biomass index which could relate to patchily burnt areas with remaining dead forage or unburnt areas (Figure 5d).
Similar to introduced bovines, albeit with weaker relationships, wallaroo abundance was higher in areas with high quality forage (low fibre and high moisture content of live biomass) but in contrast, wallaroos were associated with higher total standing biomass and negatively related to dead:live biomass index (Table 3c, Figure 5e-h). Agile wallaby abundance was best described by a single variable, total standing biomass, with higher abundance at sites with greater total biomass, consistent with wallaroos (Table 3c, Figure 5i).

DISCUSSION
The pyric herbivory model, developed in North America and southern Africa, is a key conceptual model to explain how forage, fire and herbivory interact to shape fire regimes, herbivore communities and associated biodiversity in fire-prone grassy biomes through maintenance of spatiotemporally variable fire and grazing mosaics (Fuhlendorf and Engle 2004;Archibald and others 2005;Fuhlendorf and others 2006;Engle and others 2008;Fuhlendorf and others 2010). We have demonstrated that the pyric herbivory model is clearly applicable to northern Australian savannas, where the current dominant herbivores are native macropods (kangaroos and wallabies) and introduced bovines. We found that: (1) fire causes a short-term increase in the nutritional quality and relative abundance of high-quality forage, despite reducing total forage biomass; (2) grazing herbivores decrease the quantity but increase the quality of forage; and (3) introduced bovines are strongly attracted to recently burnt areas to graze, while for macropods the relationship was not as clear. Our study provides evidence to support all the linkages, except one, in our conceptual diagram (Figure 1; Table 4). The last linkage, that was not examined here (grazing reduces the probability of fire) is well-supported by the existing Australian literature (Sharp and Whittaker 2003;Sharp and Bowman 2004;Lehmann and others 2008).

Fire Decreases Quantity but Increases Quality of Forage
Fluctuations in forage quantity and quality were closely related to fire activity. Recent fires reduced overall forage availability but improved the quality of forage (via resprouting). Resprouting grasses had lower fibre content, which is inversely related to herbivore forage intake (Freudenberger and Hume 1992), and higher crude protein content than grass at unburnt sites. Crude protein is a critical nutrient, Figure 3. Predicted dead, live and total herbaceous biomass (t ha -1 ), proportion of biomass alive (%) and live crude protein content (%) ± standard error in quadrats ungrazed (grey) and grazed (white) from vertebrate herbivores across fertile and infertile savanna substrates in the North Kimberley and Arnhem Land (estimates derived from models in Table 2). required for maintenance, lactation, growth and reproduction of mammalian herbivores (Schaefer 1946). These results represent the mechanism underpinning pyric herbivory and align with previous research showing the increased quality of forage shortly after fire Allred and others 2011a;Powell and others 2018). High fibre content of live herbaceous biomass has been linked to the seasonal shift from consumption of grasses (in the wet season and early dry season) to forbs and/or browse (in the late dry season), by both wallaroos and feral cattle (Reid Fire Season-early dry season fire (before 1 August), late dry season fire or unburnt in the dry season of sampling; Fertility-fertile or infertile soils; Region-North Kimberley or Arnhem Land; TSF-time since fire; TSEWS-time since end of wet season as defined by the date 95% of annual rainfall was received following the previous dry season; Live aNDF-fibre (amylase and sodium sulphite treated neutral detergent fibre) content of live herbaceous biomass; Live moisture-moisture content of live herbaceous biomass; Total standing biomass-combined biomass of standing dead and live herbaceous vegetation; Dead:live index-the ratio of dead and live standing herbaceous biomass. Figure 4. Predicted effect of time since end of wet season (TSEWS; the date 95% of annual rainfall was received following the previous dry season) and time in fire season on measures of forage quantity and quality including annual forage production, proportion of biomass alive, percent amylase and sodium sulphite treated neutral detergent fibre (aNDF) and percent crude protein (CP) in live biomass for forage plots in the North Kimberley and Arnhem Land. Plots show mean prediction and 95% confidence bands (shaded grey). and others 2020a), highlighting the importance of fire in prolonging the availability of preferred forage for herbivores throughout the year. Abundant palatable high-protein grass has previously been linked to high densities of the common wallaroo (Taylor 1984), and we found that crude protein content was highest in grass regrowth following late dry season fires. This suggests a disproportionate importance of late dry season fires for providing high-quality forage in a landscape dominated by dry, senescent grasses, and highlights the importance of maintaining temporally heterogeneous fire regimes to provide high-quality green pick throughout the dry season. Such a temporal staggering of burning through the dry season is likely similar to pre-European Aboriginal fire regimes in Australian tropical savannas (Lewis 1985;Bowman and others 2001;Vigilante 2001; Vigilante and others 2009).

Grazing Herbivores Decrease Quantity but Increase Quality of Forage
Despite Australian savannas having relatively low biomass of native (Calaby 1980) and introduced large herbivores compared to African savannas (Freeland 1990), our results show that large herbivores can significantly impact forage quantity (negatively) and quality (positively). Grazing by both native and introduced large herbivores, even at low densities, significantly reduced herbaceous biomass and increased crude protein content of live biomass, consistent with other studies where nutrient uptake by roots in grazed grasses was higher (Chaneton and others 1996;Mbatha and Ward 2010), yet fibre content was unaffected by herbivory. Grazing of tropical tussock grasses by introduced bovines during the early wet season reduces grass productivity (due to a lack of replacement buds and inability to maintain a positive carbon balance after defoliation), with subsequent growth only partially compensating for plant tissue lost to herbivory (Ash and McIvor 1998). Hence, the differences in forage quantity and quality detected between grazed and ungrazed quadrats at the end of dry season could be a residual effect of grazing earlier in the year, as well as differential grazing pressure during the dry season relating to fire activity (Fuhlendorf and Engle 2004).

The Contrast Between Responses of Large Native Herbivores and Introduced Bovines
Our results suggest that pyric herbivory strongly influences the feeding behaviour of introduced bovines and, to a lesser extent, only some large native herbivores in northern Australian savannas. The greatest abundance of introduced bovines (water buffalo and feral cattle) was associated with low forage quantity (that is, grass biomass) but high forage quality (that is, low fibre content and high moisture content which is correlated with crude protein content). These forage attributes are characteristic of recently burnt areas. In contrast, macropods were generally associated with the opposite forage quantity attributes (that is, high quantity) and similar high quality forage attributes, albeit with much weaker relationships, raising the question of whether native and introduced large herbivores might compete for forage resources. Our results support a recent landscape-scale analysis of the distribution of feral cattle and macropods in the North Kimberley, that showed that cattle were more abundant in recently burnt areas while large macropods were more abundant in unburnt areas (Reid and others 2020b   Low forage quantity reduces fire risk 6 (Sharp and Whittaker 2003, Sharp and Bowman 2004, Lehmann and others 2008 The introduction of large non-native herbivores, combined with replacement of spatially finegrained traditional Aboriginal fire regimes with large late dry season fires, may have forced macropods to rely more upon sub-optimal food resources. Wallaroo in the North Kimberley were more abundant on unburnt fertile sites and infertile burnt sites (Reid and others 2020b), possibly reflecting the ability of small-mouthed, selectivefeeding macropods to better utilise the limited green pick in infertile savannas. In contrast, fertile sites, with a greater flush of post-fire vegetation, may be dominated by bulk-feeding introduced bovines. Our results show that wallaroo abundance was highest when the proportion of living biomass was higher or equal to dead biomass, even though the live biomass forage was of lower quality than post-fire green pick. By contrast, introduced bovine abundance was higher in patchily burnt areas with a high proportion of dead to live biomass. Cattle and water buffalo are large bulk feeders so they benefit from forage mosaics of highly nutritious green pick and an abundance of bulk dry forage that are created by patchier fires. Similarly, Wilsey (1996) suggested that larger ungulate species in Africa try to maximize energy intake by feeding on both burned sites with low quantity but highquality forage and unburned sites with high forage biomass but poor quality.
There is some limited evidence that patterns of foraging by macropods differ between the North Kimberley and Arnhem Land, possibly due to differences in the dominant introduced bovine (cattle vs. water buffalo, respectively), length of time since herbivore introduction and differences in seasonal fire activity. Cattle and water buffalo have greater dietary flexibility in response to low-quality forage than macropods, the mean proportion of grass in the diet of cattle is greater than buffalo (Bowman and others 2010; Reid and others 2020a). By contrast, wallaroos are dependent on high-quality forage, typically exclusively consuming grass (Ellis and others 1977;Croft 1987;Telfer and Bowman 2006;Murphy and others 2007;Reid and others 2020a). Therefore, it is possible that wallaroos are experiencing greater competition for preferred resources against cattle than water buffalo. Arnhem Land field sites were characterised by a finer-scale fire mosaic and a more even distribution of fire throughout the dry season, with more fire in the late dry season than the North Kimberley field sites. A more even distribution of fire could be responsible for providing a constant source of forage to wallaroos throughout the dry season. Further research is required to confirm the above argument as our studies were unreplicated in the North Kimberley and Arnhem Land (Reid 2019).

Implications for Fire Management Approaches
Abundance of introduced bovines and large macropods, wallaroos specifically, is affected by a contrasting set of forage attributes affected by fire with bovines most abundant on burnt areas and macropods most abundant on unburned areas. The feeding behaviour of native large herbivores may be affected by competition for limited high-quality food resources with introduced herbivores that have similar food preferences. Fire activity that is spread throughout the dry season (rather than concentrated at the start) is important for providing continual access to low-fibre, high-protein forage for macropods. Strategic early dry season burning is important to break up the fuel matrix (inter-annual and intra-annual) and ensure large wildfires do not remove all forage and shelter across an entire landscape. However, targeted patchy late dry season fires, set under appropriate weather and moisture conditions, are important to achieve such optimal 'forage' mosaics for macropods. The maintenance of fine-scale fire mosaics across space and time are important for the management of large macropods. Fine-grained fire mosaics benefit large macropods as they provide access to highquality forage throughout the dry season while large-scale fire benefits introduced bovines as they can consume large amounts of the high-quality forage and then switch to other food sources. Large-scale fires across landscapes may be limiting macropod populations as they encounter food shortages for some period of every dry season.
Our findings show that the pyric herbivory conceptual model is applicable to Australian tropical savannas, contributing to understanding of the underlying ecological principles of traditional Aboriginal burning/patch mosaic burning. More broadly our study provides insights into how fire management can sustain high-quality forage throughout the dry season in tropical savannas.  Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/li censes/by/4.0/.