Fire and cattle disturbance affects vegetation structure and rain forest expansion into savanna in the Australian monsoon tropics

Abstract

Aims
To detect changes in area and vegetation dynamics of monsoon rain forests in relation to disturbance and an observed wetting trend.

Location
The Mitchell Plateau and the Bougainville Peninsula (north Kimberley, Australia).

Methods
Geo‐rectified aerial photographs acquired in 1949 and 1969 and a pre‐existing map from 2005 were used to detect changes in rain forest boundaries. To ground‐truth rain forest expansion, we established 20 transects running across rain forest‐savanna boundaries and recorded plant species, stand basal area, grass and rock cover, cattle impact, and canopy cover. Generalised linear models and Akaike's information criterion were used to detect differences in these variables between locations.

Results
On the Bougainville Peninsula average fire frequency was low (0.11 per year) and cattle entirely absent, while on the Mitchell Plateau average fire frequency was high (0.58 per year), and cattle were common and associated with lower seedling density in savannas. Rain forests expanded more on the Bougainville Peninsula (69%), where patches were bigger and more convoluted, than on the Mitchell Plateau (9%). Rain forest expansion was positively associated with rainfall and topographic complexity, and on level terrain it occurred only on the Bougainville Peninsula. Rain forests were floristically and structurally similar in the two locations, while savannas on the Bougainville Peninsula had denser vegetation and more abundant rain forest elements. The frequency distribution of canopy cover was bimodal on the Mitchell Plateau, signalling the presence of two distinct vegetation states, and unimodal on the Bougainville Peninsula, consistent with the blending of the two states.

Main conclusions
Wetting trends are likely strong drivers of rain forest expansion, but at a landscape scale their effect is probably modulated by fire activity and the presence of mega‐herbivores, which may also be pivotal in maintaining sharp floristic and structural distinctions between rain forests and savannas.

Full text

ORIGINAL ARTICLE
Fire and cattle disturbance affects vegetation structure
and rain forest expansion into savanna in the Australian
monsoon tropics
Stefania Ondei
1
|
Lynda D. Prior
1
|
Tom Vigilante
2,3
|
David M.J.S. Bowman
1
1
School of Biological Sciences, University of
Tasmania, Hobart, TAS, Australia
2
Wunambal Gaambera Aboriginal
Corporation, Kalumburu, WA, Australia
3
Bush Heritage Australia, Melbourne, VIC,
Australia
Correspondence
Stefania Ondei, School of Biological
Sciences, University of Tasmania, Hobart,
TAS, Australia.
Email: stefania.ondei@utas.edu.au
Editor: Jack Williams
Abstract
Aims: To detect changes in area and vegetation dynamics of monsoon rain forests
in relation to disturbance and an observed wetting trend.
Location: The Mitchell Plateau and the Bougainville Peninsula (north Kimberley,
Australia).
Methods: Geo-rectified aerial photographs acquired in 1949 and 1969 and a pre-
existing map from 2005 were used to detect changes in rain forest boundaries. To
ground-truth rain forest expansion, we established 20 transects running across rain
forest-savanna boundaries and recorded plant species, stand basal area, grass and rock
cover, cattle impact, and canopy cover. Generalised linear models and Akaike’s infor-
mation criterion were used to detect differences in these variables between locations.
Results: On the Bougainville Peninsula average fire frequency was low (0.11 per year)
and cattle entirely absent, while on the Mitchell Plateau average fire frequency was
high (0.58 per year), and cattle were common and associated with lower seedling den-
sity in savannas. Rain forests expanded more on the Bougainville Peninsula (69%),
where patches were bigger and more convoluted, than on the Mitchell Plateau (9%).
Rain forest expansion was positively associated with rainfall and topographic complex-
ity, and on level terrain it occurred only on the Bougainville Peninsula. Rain forests
were floristically and structurally similar in the two locations, while savannas on the
Bougainville Peninsula had denser vegetation and more abundant rain forest elements.
The frequency distribution of canopy cover was bimodal on the Mitchell Plateau, sig-
nalling the presence of two distinct vegetation states, and unimodal on the Bougain-
ville Peninsula, consistent with the blending of the two states.
Main conclusions: Wetting trends are likely strong drivers of rain forest expansion,
but at a landscape scale their effect is probably modulated by fire activity and the
presence of mega-herbivores, which may also be pivotal in maintaining sharp floris-
tic and structural distinctions between rain forests and savannas.
KEYWORDS
Australian tropics, cattle impact, climate change, fire, historical aerial photography, monsoon
rain forests, tropical savannas
DOI: 10.1111/jbi.13039
Journal of Biogeography. 2017;1–12. wileyonlinelibrary.com/journal/jbi ©2017 John Wiley & Sons Ltd
|
1

1
|
INTRODUCTION
Top-down disturbances, such as fire and mega-herbivory, have been
postulated to shape vegetation structure and in some cases cause
departure from climate-constrained potential vegetation (Bond,
2005). A prime example of this concerns the distribution of rain for-
ests and savannas. Globally, the geographic distribution of these
biomes is strongly controlled by climate, with rain forests, charac-
terised by high canopy cover and an absence of grass, dominant in
high rainfall areas, and savannas, distinguished by a continuous grass
layer and sparse trees, found in lower rainfall regions (Lehmann,
Archibald, Hoffmann, & Bond, 2011). However, in landscapes with
intermediate rainfall (1,000–2,000 mm year
1
), rain forests and
savannas can coexist in the same landscape, forming complex
mosaics (Staver, Archibald, & Levin, 2011). Contrasting fire regimes
are considered key factors influencing these vegetation types and
maintaining sharp boundaries between the two (Dantas, Hirota, Oli-
veira, & Pausas, 2016). Rain forests are rarely burnt, because the
dense canopy cover creates a moist microclimate and little flam-
mable grass is present (Little, Williams, Prior, Williamson, & Bowman,
2012), limiting the incursion of fire (Just, Hohmann, & Hoffmann,
2015). When fires do occur, they are typically mild surface fires in
leaf litter (Cochrane, 2003). By contrast, savannas are characterised
by high degree of disturbance, such as frequent fires (Huntley &
Walker, 2012). Compared to rain forest species, savanna plants are
better adapted to tolerate (e.g. thick bark) and recover (aerial buds)
from fire (Lawes, Midgley, & Clarke, 2013; Ondei, Prior, Vigilante, &
Bowman, 2015; Pausas, 2015). Dynamic global vegetation models
suggest that some tropical savannas have the climatic potential to
become forests in the absence of fire (Bond, Woodward, & Midgley,
2005). Several small-scale experiments have demonstrated substan-
tial changes in species diversity and stem density in response to fire
exclusion, although this was not necessarily associated with conver-
sion to rain forest (e.g. Bowman & Panton, 1995; Woinarski, Risler,
& Kean, 2004).
Another common disturbance is grazing by mega-herbivores. In the
Australian monsoon tropics, introduced cattle, pigs, and buffalo can
damage rain forest patches through trampling vegetation and wallow-
ing in moist soils (Petty et al., 2007; Russell-Smith & Bowman, 1992).
Experimental exclosures of medium and large herbivores in Mexican
and Brazilian rain forests resulted in higher seedling recruitment and
survival (Camargo-Sanabria, Mendoza, Guevara, Mart

ınez-Ramos, &
Dirzo, 2015; Fleury, Silla, Rodrigues, Do Couto, & Galetti, 2015), due to
the removal of the direct negative effects of browsing on plant biomass
and the indirect effects of trampling (Fleury et al., 2015). While her-
bivory can affect woody vegetation floristics and structure (Midgley,
Lawes, & Chamaille-Jammes, 2010), it is unclear if mega-herbivore dis-
turbance substantially influences rain forest-savanna boundaries at a
landscape-level (Petty et al., 2007). Furthermore, a strong interaction
between megaherbivores and fire has been posited by palaeoecologi-
cal studies. Following the extinction of Australian megafauna, the con-
sequent increase in fire activity could have led to a shift from broadleaf
to more flammable vegetation (Rule et al., 2012).
Factors affecting tree growth rates, particularly water availability
and nutrient availability, influence the rate of forest expansion (Mur-
phy & Bowman, 2012), because fast-growing plants require shorter
disturbance-free time to grow tall enough to resist the negative
effects of fire disturbance (Hoffmann et al., 2012). A trend in
increasing precipitation, possibly amplified by enhanced CO
2
, has
been postulated as the cause for the expansion of rain forests into
savannas in northwestern Australia (Banfai & Bowman, 2006; Bow-
man, Murphy, & Banfai, 2010). Significant soil fertility gradients are
also found across most rain forest-savanna boundaries (Dantas,
Batalha, & Pausas, 2013). Indeed, a controversial alternative perspec-
tive is that disturbance-based feedbacks on rain forest-savanna
boundaries are a consequence of the established vegetation type,
rather than a primary determinant (Lloyd & Veenendaal, 2016; Vee-
nendaal et al., 2015). Consequently, controlling for edaphic factors
and climatic trends is essential when evaluating the effect of fire
and megaherbivore disturbance on rain forest-savanna dynamics.
Classical experiments designed to determine the effects of exclu-
sion of fire and herbivores on rain forests and savannas at the land-
scape level are often impractical, given the spatial and temporal
scales involved and the difficulty in having adequate replication
(Andersen et al., 1998). A realistic alternative to tackle large-scale
ecological problems is to take advantage of conditions naturally
occurring in a landscape by performing natural experiments (Dia-
mond, 1983). This approach requires careful site selection, to ensure
locations share a common or known environmental history (Johnson
& Miyanishi, 2008), and some treatment combinations can be absent,
due to the rarity of naturally long-unburnt areas (Andersen et al.,
1998). Nonetheless, natural experiments can successfully test the
effects of disturbance on vegetation in fire prone environments (e.g.
Vigilante, Bowman, Fisher, Russel-Smith, & Yates, 2004; Woinarski,
Risler, & Kean, 2004).
We used a natural experiment to investigate rain forest expan-
sion and vegetation structure in two extensive locations of the north
Kimberley (Western Australia): the Bougainville Peninsula (hence-
forth “the Peninsula”) and the Mitchell Plateau (henceforth “the Pla-
teau”). These two locations share similar climate and geologic
substrate, but experience strikingly different levels of disturbance:
the Plateau has a high frequency of extensive savanna fires and a
large, unmanaged population of cattle, whereas the Peninsula is
rarely burnt and is cattle-free. This contrast allows us to determine
the combined effects of fire and cattle on rain forest-savanna
dynamics while controlling for climate and geology. We used histori-
cal aerial photography (1949, 1969, and 2005), which provided time
depth in forest boundary dynamics across the entire study areas.
We then ground-truthed these analyses using transects that
recorded variation in tree species populations across selected bound-
aries. Specifically, we hypothesised that rain forest expansion has
occurred in the north Kimberley in response to the wetting trend in
northern Australia over the last century (Bowman et al., 2010),
together with increasing atmospheric CO
2
, but that the expansion is
strongly influenced by the combined effects of fire and megaherbi-
vores. We also predicted that areas subject to high disturbance are
2
|
ONDEI ET AL.

characterized by smaller and more compact rain forest patches, sepa-
rated from the surrounding savanna by sharp boundaries, whereas in
cattle- and fire-free locations we expected to find bigger and more
convoluted patches, with wider ecotones containing a mix of rain
forest and savanna species, signs of ongoing rain forest expansion.
This natural experiment therefore contributes to the broader debates
about tropical savanna-forest boundaries, by testing the role of fire
and megafauna disturbance as drivers of rain forest change in time
and space.
2
|
MATERIALS AND METHODS
2.1
|
Study area
The study was conducted on the Wunambal Gaambera (WG)
Country in the north Kimberley, Western Australia, which occupies
an area of 9,144 km
2
. It is defined by the Wanjina Wunggurr
Uunguu Native Title Determination, and represents the traditional
lands of the Wunambal Gaambera Aboriginal people. The geology
of the region is characterised by deeply weathered sandstones
and basaltic base rocks of Precambrian age, often capped by Cain-
ozoic laterites (Beard, 1976). Rainfall occurs almost entirely during
the summer wet season (November to April), while the rest of the
year is almost rain-free (Beard, 1976). Current average annual
rainfall across the region ranges from 1,200 to 1,400 mm. A wet-
ting trend has been detected since the beginning of the 20th cen-
tury, with an increment in average annual rainfall of 40–50 mm
every 10 years since the late 1940s (Bureau of Meteorology,
2016). The vegetation is predominantly biodiverse tropical savan-
nas: Eucalyptus tetrodonta–E. miniata savannas are found on the
laterite mesas and hills, while E. tectifica–E. grandifolia savannas
are common on deeper, clay soils on the plains. Small patches of
semi-deciduous rain forests are interspersed in the savanna, typi-
cally found in fire-protected locations (Beard, 1976; Vigilante
et al., 2004).
The study was centred on two of the locations with the highest
density of rain forests in the north Kimberley: the Plateau and the
Peninsula (Figure 1). These locations have similar geology (basalt
and laterite) and mean annual rainfall (range 1,300–
1,400 mm year
1
), but contrasting management histories. The Pla-
teau (754 km
2
) has a high fire frequency (average times burnt:
0.5 year
1
; data from North Australia Fire Information (NAFI), based
on the 15-year time period from 2000–2014). Conversely, the
Peninsula (298 km
2
), connected with the mainland only by a narrow
strip of sand, has a much lower fire frequency (average times burnt:
0.08 year
1
). The fire regime of the entire study area has undergone
dramatic shifts in the last 100 years. Wunambal Gaambera people
practiced landscape scale burning up until the 1940s and 1950s,
when they moved off their country to settlements. The next
50 years were dominated by unmanaged wildfires moving in from
adjacent areas or started by lightning. Since 2009, Aboriginal land
management programs have initiated prescribed burning programs.
While pastoral leases have been established in parts of the north
Kimberley since the 1950s, the study area has never been formally
used for pastoral purposes. A 1976 biological survey of the Plateau
did not record any cattle in the area (Wilson, 1981). However, a bio-
logical survey of Kimberley rain forests in 1987 found evidence of
cattle on the Plateau, but not on the Peninsula (Mckenzie & Belbin,
1991).
2.2
|
Rain forest-savanna boundary change
Digitized aerial photographs of the study locations were obtained for
the years 1949 (black and white; 1:50,000) and 1969 (black and
white; 1:83,300) and georectified, using the spline tool in ArcGIS 10
Georeferencing Toolbox to correct for obvious misalignments. Rain
forests were mapped at a common scale of 1:2,500, adapting the
method described in Bowman, Walsh, and Milne (2001), based on
grid cell classification. We overlaid a 30 930 m lattice grid to the
aerial photographs and, for each investigated year, grid cells were
manually classified as “rain forest”or “other”, based on the vegeta-
tion type occupying the highest proportion of the cell. To compare
the historic rain forest extent with a more recent distribution (dry
season of 2005), we used a pre-existing, validated map of the north
Kimberley rain forests produced using the same methods and
resolution (30 930 m) by Ondei, Prior, Williamson, Vigilante, and
Bowman (2017) (Figure 2).
2.3
|
Patch characteristics and location
A map of the rain forest patches was produced for the years 1949,
1969, and 2005, obtained by merging the contiguous cells classified
as “rain forest”. For each patch we calculated area, perimeter, dis-
tance from the coastline, and topographic position index (TPI). The
latter was calculated as per Ondei et al. (2017), which classifies the
land in four different topographic categories: valleys, slopes, ridges
and flat areas. We analysed the effect on expansion of both patch
size and shape, and defined the degree of patch convolution by cal-
culating patch fractal dimension (2 9ln(0.25 9perimeter)/ln(area)),
which ranges from 1 (compact) to 2 (highly convoluted) (Hargis,
Bissonette, & David, 1998).
2.4
|
Correlates of rain forest expansion
To investigate the environmental drivers of rain forest expansion in
the time periods 1949–1969 and 1969–2005, we selected all grid
cells located within 60 m (equivalent to two times the map resolu-
tion) of the patch perimeter at the first year of each time period
(1949 or 1969). For each selected grid cell, we calculated distance
from the coastline, TPI, and the distance from and the size, convolu-
tion, and location of the nearest patch. We also calculated aspect,
obtained from a 30-m Digital Elevation Model (DEM), and the bear-
ing angle, defined as the angular direction of the cell relative to the
patch of origin. Both bearing angle and aspect were decomposed
into their North–South (cosine of angle) and East–West (sine of
angle) components.
ONDEI ET AL.
|
3

2.5
|
Vegetation structure
To determine vegetation structure and composition across the rain
forest-savanna transition zones, we established a total of 20 tran-
sects, 10 placed on the Plateau and 10 on the Peninsula. Transects
were all located on different patches and at a minimum distance of
500 m. Each transect was 300 m long, and consisted of five
10 920 m plots placed at regular intervals along it. The first plot
was located 60 m inside the rain forest, to measure structure and
floristic composition of the forest core; the second was at the patch
boundary, capturing, when present, the characteristics of the eco-
tone, while the remaining three were placed in the vegetation out-
side the rain forest. The patch boundary was visually determined as
by Hennenberg, Goetze, Kouam
e, Orthmann, and Porembski (2005),
based on discontinuities in floristic composition and canopy cover.
For each plot we recorded grass cover, rock cover, canopy cover,
number of seedlings, and signs of cattle presence. Canopy cover was
measured by taking hemispherical photographs using a fish-eye lens
(Nikon AF Fisheye NIKKOR 10.5 mm), taking pictures at 1 m height
and calculating the percentage of closed canopy using the software
CAN-EYE (http://www6.paca.inra.fr/can-eye/). The number of seed-
lings was counted on a 1 920 m sub-plot. The presence of cattle
was ranked on a qualitative scale, from “0”indicating no sign of cat-
tle, to “3”, high cattle impact, based on the presence of tracks and
excrements on the ground and partially browsed plants. For each
adult tree, operationally defined as having a diameter at breast
height (DBH) >5 cm and being taller than 2 m, we recorded species
name, height, and DBH. Total basal area was calculated for each
plot. We also recorded the species identity of every tree and shrub
in the plot. When species identification was not possible in the field,
voucher samples were collected and identified at the Northern Terri-
tory Herbarium, where they have been lodged. To determine
whether a species was a rain forest element we relied on the floristic
classification of Kenneally, Keighery, and Hyland (1991) for the north
Kimberley rain forests. Each plot was then assigned to one of the
following vegetation classes based on the mapping for 1949, 1969,
and 2005: “stable rain forest”plots were those mapped as “rain for-
est”during all the investigated years; “converted to rain forest in
1969”were plots mapped as “savanna”in 1949 and “rain forest”in
1969; “converted to rain forest in 2005”plots were mapped as “sa-
vanna in 1969”and “rain forest”in 2005; “stable savanna”plots
were mapped as “savanna”during the entire length of the study. Fire
frequency was calculated for each plot based on 15-year data
obtained from NAFI (available at www.firenorth.org.au), based on
MODIS products displayed at a resolution of 250 m.
2.6
|
Statistical analyses
All analyses were performed using the software R Studio ver.
1.0.136 (R Core Team, 2013). To assess variation in patch size and
convolution we employed generalized linear models (GLMs), using
the gamma family of distribution (link =“log”). Patch size and convo-
lution were response variables and year and location were explana-
tory variables. For each location we calculated differences in the
proportion of rain forests located on different topographic settings
in the years 1949, 1969, and 2005.
To evaluate how environmental variables affected the conversion
from savanna to rain forest in each time period, we randomly
selected a subset of 5,000 grid cells within 60 m of a rain forest
patch (selected as described above), stratified for vegetation type.
FIGURE 1 Location of the two study
sites in the north Kimberley. The Mitchell
Plateau extends in latitude from 14.457°S
to 14.893°S and in longitude from
125.722°E to 125.949°E. The Bougainville
Peninsula stretches from 13.897°Sto
14.146°S in latitude and from 125.975°E
to 126.225°E in longitude. Dashed lines
represent isohyets (mm year
1
). The inset
shows the location of the study sites
within Australia
4
|
ONDEI ET AL.

We employed GLMs, considering the response variable as binary: 1
—savanna converted to rain forest, or 0—savanna remained
savanna, and using binomial models (link “logit”). We tested the rela-
tionship between the response variable and TPI, distance from the
coastline, location, area, and convolution of the nearest patch, dis-
tance from the nearest patch, and North–South and East–West com-
ponents of both aspect and bearing angle. Spatial autocorrelation
was assessed by plotting semi-variograms of model residuals. As only
few cells underwent a conversion from rain forest to savanna, no
analyses on that vegetation change were performed.
To determine differences in fire activity between the two study
locations, we employed GLMs and the Gaussian distribution. Differ-
ences in cattle impact between vegetation classes located on the Pla-
teau were tested using ordinal logistic regression (package “MASS”;
Venables & Ripley, 2002) and the influence of cattle impact on the
number of seedlings was investigated using the Poisson distribution.
GLMs were also used to test whether the disturbance levels and
environmental and vegetation characteristics recorded in each plot
and listed below varied between locations (Plateau or Peninsula) for
plots within the same vegetation class (e.g. “stable savanna”). For
these analyses the two classes “converted to rain forest in 1969”
and “converted to rain forest in 2005”were merged due to the lim-
ited number of plots. When required, data were log transformed to
meet the assumption of normality. Within the GLMs, the Poisson
distribution (log link) was used for count data such as the number of
seedlings, number of trees, species richness. The binomial distribu-
tion (logit link) was used to assess differences between locations in
the proportion of rain forest and savanna species per plot, while the
Gaussian distribution (identity link) was used for the response vari-
able basal area. Differences in grass cover and rock cover were
assessed using ordinal logistic regression.
For all the modelling we employed complete subset regression
and Akaike’s information criterion (AIC; Burnham & Anderson, 2002)
to evaluate models. To do this, candidate sets were constructed with
models containing all possible combinations of explanatory variables,
without interactions. Akaike weights (w
i
) were calculated for each
FIGURE 2 Example of aerial
photography of the Mitchell Plateau and
Bougainville Peninsula used in the
analyses. Polygons obtained from merging
grid cells classified as “rain forest”are
shown. Percentage of change between
each pair of photographs is indicated
ONDEI ET AL.
|
5

model to indicate the probability that a given model is the best in
the candidate set (Burnham & Anderson, 2004). The importance of
single variables (w+) was calculated as the sum of w
i
of the models
within the set in which the variable occurred. Variables were consid-
ered important predictors if w+exceeded 0.73, as per Murphy et al.
(2010). Summaries of all analyses are reported in Tables S1.3 and
S1.4 in Appendix S1.
To compare variation in canopy across the rain forest-savanna
boundary in the Plateau and the Peninsula, we calculated the aver-
age canopy cover profile for each location, as described in Fig. S2.1
in Appendix S2. To detect the presence of distinct vegetation states
we also tested differences in canopy cover modality employing
latent class analysis (Hirota, Holmgren, Van Nes, & Scheffer, 2011).
To do this, we analysed the frequency distribution of canopy cover
data, pooled for all plots within the vegetation transects at each
location. We compared the fit of models with 1, 2 or 3 modes using
the Bayesian information criterion. More details are given in
Table S2.6 in Appendix S2.
3
|
RESULTS
3.1
|
Rain forest expansion/contraction
During the 20-year time period 1949–1969, the north Kimberley
experienced expansion of rain forest cover. The extent of expan-
sion varied depending on the location: rain forests expanded by
52%, equivalent to 4.25 km ha
1
, on the Peninsula, and only by
9%, corresponding to 0.12 km ha
1
, on the Plateau (Figure 3a,b).
During the following 36-year period (1969–2005), the expansion
continued more slowly on the Peninsula, reaching an overall
expansion of 69%, whereas on the Plateau rain forest extent
remained stable. On the Peninsula, areas of contraction were
recorded during both time periods, although always compensated
by an overall higher proportion of expanded rain forest. At both
sites rain forests consistently preferred slopes, valleys and ridges
(Figure 4).
3.2
|
Patch characteristics
In 1949 a total of 1,668 patches, covering 3% of the land, were
mapped, and this area increased during the 20-year period to
1969, with a smaller increase between 1969 and 2005 (Table 1).
There were marked differences in trends between the Plateau and
the Peninsula in terms of patch density, individual patch size and
convolution. At all three observation times, patches on the Plateau
were smaller (w+=1.00) and more compact (w+=1.00) than
those on the Peninsula. Patch size and density increased on the
Peninsula, while on the Plateau there was no substantial change in
average patch size over the entire 56-year time frame (Figure 5a,
b), but a small increase in patch density was detected between
1949 and 1969. Rain forests on the Peninsula were bigger in
1969 and less convoluted in 2005 compared with 1949
(Figure 5b,c).
3.3
|
Correlates of rain forest expansion
Model selection showed that between 1949 and 1969, the local
variables that affected rain forest expansion were: convolution and
area of the original patch, distance from patch edge, and TPI
(Table 2). Areas on the Peninsula, and those close to the edge of
bigger and more convoluted patches, were more likely to convert
from savanna to rain forest. The same variables affected the expan-
sion of rain forests between 1969 and 2005. During this period, rain
forests were also more likely to establish on the northern side of
already existing patches and close to the coast (Table 2). On the
Peninsula there was substantial expansion on flat locations during
both periods, but this was not evident on the Plateau (Figure 4). The
best models for both periods explained 22% of the deviance.
3.4
|
Disturbance
No sign of cattle was found in any of the plots on the Peninsula. On
the Plateau, cattle presence was detected in 62% of the plots, with
no substantial differences in the intensity of cattle impact among
vegetation classes (w+ =0.51; Table S1.1 in Appendix S1). The
intensity of cattle grazing was negatively correlated with tree seed-
ling density in savannas (w+ =1.00), but not in rain forests
(w+ =0.63). Fire frequency was lower on plots located on the
Peninsula (w+ =1.00).
FIGURE 3 Rain forest extent over time on the Mitchell Plateau
and Bougainville Peninsula, shown as (a) rain forest area during
1949, 1969 and 2005, expressed as proportion of total land, and (b)
proportion of change in rain forest extent, relative to the baseline
year 1949
6
|
ONDEI ET AL.

3.5
|
Floristics and vegetation structure
We identified 82 species belonging to the rain forest flora in plots
on the Peninsula, and 71 in those on the Plateau. Most records were
of species commonly found in Kimberley rain forests, with the
exception of Pouteria richardii (F. Muell.) Baehni, found on the Penin-
sula and never recorded before in Western Australia. Species com-
position was uniform within rain forest patches; of the species
sighted at least five times, only four rain forest species were found
exclusively on the patch edge (Bridelia tomentosa Blume, Ficus acu-
leata Miq., Flueggea virosa (Willd.) Voig, and Trema tomentosa (Roxb.)
Hara), and only two were limited to the patch core (Diospyros mar-
itima Blume and Meiogyne cylindrocarpa (Burck) Heusden). The
savanna flora had similar richness in the Peninsula and Plateau (29
and 31 species respectively). However, while all the savanna species
identified on the Plateau are common throughout the north Kimber-
ley, on the Peninsula we detected uncommon species recorded in
only a few sites in Western Australia, such as Eucalyptus oligantha
Schauer, and Xanthostemon psidioides (Lindl.) Peter G.Wilson &
J.T.Waterh. The latter is considered near threatened in Western
Australia. We also recorded two species thought not to grow in the
north Kimberley: Acacia drepanocarpa subsp. latifolia Pedley and
Vachellia ditricha (Pedley) Kodela (Atlas of Living Australia website at
http://www.ala.org.au; Western Australian Herbarium, 1998). Some
of the savanna elements detected only on the Peninsula, such as
acacias (Acacia hemignosta F.Muell. A. stigmatophylla Benth,
FIGURE 4 Proportion of landscape occupied by rain forest in slopes, valleys, flat areas, and ridges on the Mitchell Plateau and Bougainville
Peninsula over time, from 1949 to 2005
TABLE 1 Total number of rain forest patches, rain forest extent, and proportion of land covered by rain forest for the years 1949, 1969
and 2005 on the Bougainville Peninsula (BP), Mitchell Plateau (MP) and the entire study area
Year
Number of patches Rain forest extent (ha) Rain forest cover (ha km
2
)
BP MP Total BP MP Total BP MP Average
1949 759 776 1,668 2,504 968 3,786 8.4 1.3 3.5
1969 908 831 1,891 3,770 1,053 5,230 12.6 1.4 4.8
2005 989 869 2,053 4,085 1,054 5,713 13.7 1.4 5.2
ONDEI ET AL.
|
7

A. drepanocarpa sub. latifolia Pedley), are known to be sensitive to
frequent fires (Russell-Smith, Yates, Brock, & Westcott, 2010).
Abrupt changes in vegetation characteristics were detected
across rain forest-savanna boundaries on the Plateau, while on the
Peninsula the differences were less apparent (Table S1.2 in
Appendix S1; Figure 6). The floristic composition and vegetation
structure of stable and expanded rain forests were very similar at
the two study locations, with the exception of higher tree and seed-
ling density on the Peninsula (w+=0.92 and w+=1.00 respectively;
Table S1.5 in Appendix S1). Stable rain forests also had more seed-
lings (w+=0.84) and a higher proportion of rain forest species
(w+=1.00) compared with recently expanded rain forests. For the
savannas, there were clear differences between the two location,
with signs of rain forest invasion on the Peninsula. Indeed, every
savanna quadrat on the Peninsula contained rain forest species in
the understorey, compared with only 27% on the Plateau. Conse-
quently, savannas on the Peninsula displayed greater species rich-
ness (w+=1.00), due to more rain forest species per plot
(w+=1.00). The Peninsula’s savannas also had higher densities of
adult trees and seedlings (w+=0.88 and w+=1.00 respectively),
and greater grass cover (w+=0.87).
The invasion of rain forest species into the savanna on the
Peninsula resulted in contrasting profiles of tree canopy cover
across the rain forest-savanna boundaries at the two locations.
There was a gradual change in canopy cover across boundaries on
the Peninsula compared with abrupt boundaries on the Plateau
(Fig. S2.1 in Appendix S2). Furthermore, frequency distribution of
canopy cover on the Plateau transects displayed the bimodality
characteristic of two distinct vegetation states (Table S2.6; Fig. S2.2
in Appendix S2). By contrast, the unimodal canopy cover model, cor-
responding to blending of vegetation, was the best model for the
Peninsula.
4
|
DISCUSSION
Our natural experiment in the north Kimberley, based on two geo-
graphically similar areas with contrasting disturbance regimes,
revealed significant differences in vegetation structure and in the
rate, magnitude, and environmental correlates of rain forest
FIGURE 5 Patch characteristics on the Bougainville Peninsula and Mitchell Plateau in the years 1949, 1969, and 2005, delineating (a) patch
density, expressed as number of rain forest patches normalized by the location area, (b) average patch size, and (c) average convolution of rain
forest patches. Error bars represent standard errors
TABLE 2 Explanatory variables included in the models assessing
the likelihood of a savanna cell, located within 60 m to a rain forest
patch, to convert in a rain forest cell on the Bougainville Peninsula
and Mitchell Plateau. The importance of each variable is expressed
as w+, the probability of the factor to be included in the best model.
The direction of the effect is indicated within brackets: +positive,
or negative for continuous variables; for categorical factors the
categories positively affecting the conversion to rain forest are
shown. Variables with w+values >0.73 (in bold) are considered
important predictors
Variable
w+
Time period
1949–1969
Time period
1969–2005
Location (Bougainville Peninsula) 1.00 1.00
Convolution (+)1.00 1.00
Patch area (+)1.00 0.98
Distance from rain forest edge ()1.00 1.00
TPI (slopes, ridges, and valleys) 1.00 1.00
Distance from the coastline () 0.27 1.00
Expansion North–South (North) 0.40 0.97
Expansion East–West (West) 0.44 0.32
North–South component
of aspect (North)
0.31 0.40
East–West component
of aspect (West)
0.30 0.28
8
|
ONDEI ET AL.

expansion in time. Specifically, the study design combined historical
aerial photography and field measurements to address three predic-
tions: (1) rain forest expansion has occurred in the north Kimberley
concurrently with the trend of increasing precipitation and/or atmo-
spheric CO
2
(2) this trend was locally influenced by the combined
effects of fire and megaherbivores and (3) areas subject to high dis-
turbance have different boundaries and patch shapes from less dis-
turbed areas. Below we discuss our findings with respect to the
current theories on rain forest-savanna dynamics.
The two study locations have similar climates, with mean annual
precipitation towards the lower limit of rain forests in Australia and
globally (Bowman, 2000). In the area, annual rainfall increased from
an estimated 1,080 to 1,280 mm during the period 1949 and 2005,
consistent with increasing precipitation and longer wet seasons
observed in north-western Australia in the past decades (Bureau of
Meteorology, 2016; Feng, Porporato, & Rodriguez-Iturbe, 2013). The
positive correlation between wetting trends and rain forest expan-
sion in the north Kimberley is consistent with findings of previous
studies in the Australian tropics, which associated increased rainfall
with rain forest expansion (Banfai & Bowman, 2006; Bowman et al.,
2010) or savanna woody thickening (Lehmann, Prior, & Bowman,
2009). However, we used space-for-time substitution and data for
current rain forest cover in monsoonal Australia (Ondei et al., 2017)
and estimated that this increase in rainfall corresponds to a 41% rel-
ative increase in the 95th percentile of rain forest cover (Fig. S3.3 in
Appendix S3). It is therefore improbable that wetting trends are
solely responsible for the 69% increase in rain forest cover observed
on the Plateau. Elevated CO
2
has also been associated with
enhanced tree growth and recruitment (Kgope, Bond, & Midgley,
2010) and resprouting of seedlings after defoliation, potentially con-
tributing to shift savannas towards a tree-dominated state (Bond &
Midgley, 2000; Buitenwerf, Bond, Stevens, & Trollope, 2012; Hoff-
mann, Bazzaz, Chatterton, Harrison, & Jackson, 2000). We also can-
not rule out relaxation of disturbance regimes contributing to the
rain forest expansion.
Our experiment was designed to compare locations with similar
climate and geology, and thus amounts of nutrients derived from the
parent material. Yet, remarkable differences in rain forest expansion
were detected between the Plateau and the Peninsula, suggesting
that rain forest distribution is not determined by resources alone.
Furthermore, our results showed that on the Peninsula, rain forest
expanded into infrequently burnt savannas across all landscape set-
tings, while on the Plateau rain forest expansion was constrained to
slopes, valleys and ridges. These differences highlight the importance
of disturbance history in modifying the expansion of rain forest. The
frequently burnt savannas on the Plateau had structurally abrupt and
FIGURE 6 Average plot values, recorded on the Mitchell Plateau and Bougainville Peninsula, of (a) proportion of rain forest plants amongst
the adult trees, (b) grass cover, (c) rock cover, (d) number of adult trees, (e) number of seedlings, and (f) species richness in the two study
locations for each vegetation class: (St RF) stable rain forests, (Exp 69) converted to rain forest in 1969, (Exp 05) converted to rain forest in
2005, (St SAV) stable savannas. Error bars represent standard errors
ONDEI ET AL.
|
9

floristically distinct rain forest-savanna boundaries, consistent with
the view that fire maintains a sharp transition between these two
vegetation types (Dantas et al., 2013, 2016). Elevated fire activity
may also explain the very restricted patches and their limited convo-
lutions. By contrast, on the Peninsula the lower fire activity resulted
in large, convoluted patches that occurred across a range of land-
forms. Grass cover was higher in savannas on the Peninsula than the
Plateau, and a striking feature of the Peninsula’s savannas was the
admixture of savanna and rain forest trees. Because of the low fire
frequency in the savannas on the Peninsula, rain forest trees, includ-
ing some species not found elsewhere in the Kimberley, were able
to grow even where the canopy was not yet closed sufficiently to
exclude grasses (Lawes, Murphy, Midgley, & Russell-Smith, 2011).
The rarity of anthropogenic ignitions on the Peninsula and its fire-
protected position have possibly caused this anomaly. Here, the
main limitation to rain forest tree growth appears to be competition
with savanna trees and grasses for resources (Bond, 2008). By con-
trast, on the Plateau, frequent disturbance also imposes a constraint
on rain forest tree growth.
The invasion of cattle on the Plateau during the time period
1969–2005 is likely to have accentuated the differences in rain for-
est expansion between the Plateau and the Peninsula, through direct
and indirect effects. Cattle may have limited the expansion of
patches on the Plateau, because trampling and browsing of juveniles
likely contributed to lower seedling density. Cattle may also have
increased grass cover by opening up understories on rain forest
boundaries, thereby increasing fire activity (Camargo-Sanabria et al.,
2015; Fleury et al., 2015; Mckenzie & Belbin, 1991).
Because our natural experiment was based on only two sites, we
cannot rule out the possibility that factors other than those we
tested contributed to differences in rain forest structure, distribution,
and expansion. Nonetheless, by removing climatic and geological dif-
ferences from our study design, we showed that fire activity proba-
bly has a primary role in driving vegetation dynamics (Hoffmann
et al., 2012; Lawes et al., 2011; Murphy & Bowman, 2012), with cat-
tle herbivory perhaps playing a subsidiary role (Figure 7). On fertile
substrates and when freed from disturbance, rain forests can occupy
a larger range of landscapes than is typically observed. We found
that the expansion of rain forests into unburnt savannas resulted in
ecotonal vegetation that blends the floristic and structural elements
of both vegetation types. This highlights the importance of fire
regimes in shaping vegetation structure and floristic composition in
regions where savanna and rain forest co-exist (Dantas et al., 2013).
It is possible that, prior to Aboriginal colonisation and the marsupial
megafaunal extinctions, there would have been a less pronounced
dichotomy between savannas and rain forests, and long-lasting tran-
sitional states such as those that now occur on the Peninsula may
have been present. Analysis of pollen, charcoal and Sporormiella
records in north-east Australia have been interpreted as showing
that the extinction of marsupial megafauna following human coloni-
sation led to a transition from relatively open and mixed rain forest-
sclerophyll forest to pure sclerophyll vegetation and an increase in
fire activity (Rule et al., 2012). Aboriginal fire management possibly
sharpened boundaries between savanna and rain forests, maintaining
fire sensitive vegetation in a matrix of flammable savanna
(Trauernicht, Brook, Murphy, Williamson, & Bowman, 2015).
FIGURE 7 Feedbacks regulating the presence of rain forests or savannas. The existing vegetation type, rain forest or savanna, is the results
of a complex network of interactions, involving bottom-up controls (resource) and top-down controls (consumers). Vegetation acts as a
consumer of resources such as water and nutrients, and as a resource for the consumers of biomass (megaherbivores and fire). Top-down and
bottom-up controls influence each other via direct effect or by facilitating the establishment of one vegetation type or the other. The
approximate strength of each relationship is indicated by the width of the arrows. (Adapted from Murphy & Bowman, 2012)
10
|
ONDEI ET AL.

Across northern Australia, contemporary disturbance regimes,
which include frequent landscape burning and widespread cattle
grazing, are thought to be contributing to the decline of many small
to medium sized mammals due to loss of shelter and food resources
(Legge, Kennedy, Lloyd, Murphy, & Fisher, 2011; Woinarski,
Burbidge, & Harrison, 2015). In this context, the cattle-free and
rarely burnt Bougainville may be important for biodiversity, by offer-
ing more long-unburnt habitats for large numbers of threatened
small to medium sized mammals and fire-sensitive floristic elements
rarely found elsewhere in northern Australia (Radford, Gibson, Corey,
Carnes, & Fairman, 2015; Woinarski, Risler, & Kean, 2004).
ACKNOWLEDGEMENTS
We thank the Uunguu Rangers, Wunambal Gaambera Traditional
Owners, Carly Ward and Phil Docherty for their support in the field
work components of this study, and Ian Cowie and Nick Cuff at North-
ern Territory Herbarium for their help with species identification.
CONFLICT OF INTEREST
The authors declare that they have no conflict of interest.
AUTHOR CONTRIBUTIONS
All authors conceived the idea; S.O. performed remote sensing
analyses; S.O. and T.V. collected field data; S.O. and L.D.P. con-
ducted data analyses under the direction of D.M.J.S.B.; all authors
contributed to writing the manuscript.
REFERENCES
Andersen, A. N., Braithwaite, R. W., Cook, G. D., Corbett, L. K., Williams,
R. J., Douglas, M. M., ... Muller, W. J. (1998). Fire research for con-
servation management in tropical savannas: Introducing the Kapalga
fire experiment. Australian Journal of Ecology,23,95–110.
Banfai, D. S., & Bowman, D. M. J. S. (2006). Forty years of lowland mon-
soon rainforest expansion in Kakadu National Park, Northern Aus-
tralia. Biological Conservation,131, 553–565.
Beard, J. S. (1976). The monsoon forest of the Admirality Gulf, Western
Australia. Vegetatio,31, 177–192.
Bond, W. J. (2005). Large parts of the world are brown or black: A differ-
ent view on the ‘Green World’hypothesis. Journal of Vegetation
Science,16, 261–266.
Bond, W. J. (2008). What limits trees in C4 grasslands and savannas?
Annual Review of Ecology, Evolution, and Systematics,39, 641–659.
Bond, W. J., & Midgley, G. F. (2000). A proposed CO2-controlled mecha-
nism of woody plant invasion in grasslands and savannas. Global
Change Biology,6, 865–869.
Bond, W. J., Woodward, F. I., & Midgley, G. F. (2005). The global distri-
bution of ecosystems in a world without fire. New Phytologist,165,
525–537.
Bowman, D., Murphy, B. P., & Banfai, D. S. (2010). Has global environ-
mental change caused monsoon rainforests to expand in the Aus-
tralian monsoon tropics? Landscape Ecology,25, 1247–1260.
Bowman, D. M. J. S. (2000). Australian rainforests: Islands of green in a
land of fire. Cambridge: Cambridge University Press.
Bowman, D. M. J. S., & Panton, W. J. (1995). Munmarlary revisited:
Response of a north Australian Eucalyptus tetrodonta savanna pro-
tected from fire for 20 years. Australian Journal of Ecology,20,526–531.
Bowman, D. M. J. S., Walsh, A., & Milne, D. J. (2001). Forest expansion
and grassland contraction within a Eucalyptus savanna matrix
between 1941 and 1994 at Litchfield National Park in the Australian
monsoon tropics. Global Ecology and Biogeography,10, 535–548.
Buitenwerf, R., Bond, W. J., Stevens, N., & Trollope, W. S. W. (2012).
Increased tree densities in South African savannas: >50 years of data
suggests CO2 as a driver. Global Change Biology,18, 675–684.
Bureau of Meteorology (2016). Climate Data: Australia (CD). Climate Ser-
vices, Bureau of Meteorology, Kent Town. Retrieved from http://
www.bom.gov.au/climate/
Burnham, K. P., & Anderson, D. R. (2002). Model selection and multimodel
inference: A practical information-theoretic approach. New York:
Springer.
Burnham, K. P., & Anderson, D. R. (2004). Multimodel inference under-
standing AIC and BIC in model selection. Sociological Methods &
Research,33, 261–304.
Camargo-Sanabria, A. A., Mendoza, E., Guevara, R., Mart

ınez-Ramos, M.,
& Dirzo, R. (2015). Experimental defaunation of terrestrial mam-
malian herbivores alters tropical rainforest understorey diversity. Pro-
ceedings of the Royal Society of London B: Biological Sciences,282,
20142580.
Cochrane, M. A. (2003). Fire science for rainforests. Nature,421, 913.
Dantas, V. D. L., Batalha, M. A., & Pausas, J. G. (2013). Fire drives functional
thresholds on the savanna–forest transition. Ecology,94, 2454–2463.
Dantas, V. D. L., Hirota, M., Oliveira, R. S., & Pausas, J. G. (2016). Distur-
bance maintains alternative biome states. Ecology Letters,19,12–19.
Diamond, J. M. (1983). Ecology: Laboratory, field and natural experi-
ments. Nature,304, 586–587.
Feng, X., Porporato, A., & Rodriguez-Iturbe, I. (2013). Changes in rainfall
seasonality in the tropics. Nature Climate Change,3, 811–815.
Fleury, M., Silla, F., Rodrigues, R. R., Do Couto, H. T. Z., & Galetti, M.
(2015). Seedling fate across different habitats: The effects of her-
bivory and soil fertility. Basic and Applied Ecology,16, 141–151.
Hargis, C. D., Bissonette, J. A., & David, J. L. (1998). The behavior of
landscape metrics commonly used in the study of habitat fragmenta-
tion. Landscape Ecology,13, 167–186.
Hennenberg, K. J., Goetze, D., Kouam
e, L., Orthmann, B., & Porembski, S.
(2005). Border and ecotone detection by vegetation composition
along forest-savanna transects in Ivory Coast. Journal of Vegetation
Science,16, 301.
Hirota, M., Holmgren, M., Van Nes, E. H., & Scheffer, M. (2011). Global
resilience of tropical forest and savanna to critical transitions. Science,
334, 232–235.
Hoffmann, W. A., Bazzaz, F. A., Chatterton, N. J., Harrison, P. A., & Jack-
son, R. B. (2000). Elevated CO2 enhances resprouting of a tropical
savanna tree. Oecologia,123, 312–317.
Hoffmann, W. A., Geiger, E. L., Gotsch, S. G., Rossatto, D. R., Silva, L. C.
R., Lau, O. L., ... Lloret, F. (2012). Ecological thresholds at the
savanna-forest boundary: How plant traits, resources and fire govern
the distribution of tropical biomes. Ecology Letters,15, 759–768.
Huntley, B. J., & Walker, B. H. (2012). Ecology of tropical savannas. Ber-
lin-Heidelberg-New York: Springer Science & Business Media.
Johnson, E. A., & Miyanishi, K. (2008). Testing the assumptions of
chronosequences in succession. Ecology Letters,11, 419–431.
Just, M. G., Hohmann, M. G., & Hoffmann, W. A. (2015). Where fire
stops: vegetation structure and microclimate influence fire spread
along an ecotonal gradient. Plant Ecology,217, 631–644.
Kenneally, K. F., Keighery, G. J., & Hyland, B. P. M. (1991). Floristics and
phytogeography of Kimberley rainforests. In N. L. Mckenzie, R. B.
Johnston, & P. G. Kendrick (Eds.), Kimberley rainforests of Australia
(pp. 93–131). Chipping Norton, NSW: Surrey Beatty and Sons.
ONDEI ET AL.
|
11

Kgope, B. S., Bond, W. J., & Midgley, G. F. (2010). Growth responses of Afri-
can savanna trees implicate atmospheric [CO2] as a driver of past and
current changes in savanna tree cover. Austral Ecology,35, 451–463.
Lawes, M. J., Midgley, J. J., & Clarke, P. J. (2013). Costs and benefits of
relative bark thickness in relation to fire damage: A savanna/forest
contrast. Journal of Ecology,101, 517–524.
Lawes, M. J., Murphy, B. P., Midgley, J. J., & Russell-Smith, J. (2011). Are
the eucalypt and non-eucalypt components of Australian tropical
savannas independent? Oecologia,166, 229–239.
Legge, S., Kennedy, M. S., Lloyd, R., Murphy, S. A., & Fisher, A. (2011).
Rapid recovery of mammal fauna in the central Kimberley, northern
Australia, following the removal of introduced herbivores. Austral
Ecology,36, 791–799.
Lehmann, C. E., Prior, L. D., & Bowman, D. M. (2009). Decadal dynamics
of tree cover in an Australian tropical savanna. Austral Ecology,34,
601–612.
Lehmann, C. E. R., Archibald, S. A., Hoffmann, W. A., & Bond, W. J.
(2011). Deciphering the distribution of the savanna biome. New Phy-
tologist,191, 197–209.
Little, J. K., Williams, S. E., Prior, L. D., Williamson, G. J., & Bowman, D.
M. J. S. (2012). Fire weather risk differs across rain forest-savanna
boundaries in the humid tropics of north-eastern Australia. Austral
Ecology,37, 915–925.
Lloyd, J., & Veenendaal, E. M. (2016). Are fire mediated feedbacks burn-
ing out of control? Biogeosciences Discussions,2016,1–20.
Mckenzie, N. L., & Belbin, L. (1991). Kimberley rainforest communities:
Reserve recommendations and management considerations. In N. L.
Mckenzie, R. B. Johnston, & P. G. Kendrick (Eds.), Kimberley rain-
forests of Australia (pp. 453–468). Chipping Norton, NSW: Surrey
Beatty and Sons.
Midgley, J. J., Lawes, M. J., & Chamaille-Jammes, S. (2010). Savanna
woody plant dynamics: The role of fire and herbivory, separately and
synergistically. Australian Journal of Botany,58,1–11.
Murphy, B. P., & Bowman, D. M. J. S. (2012). What controls the distribu-
tion of tropical forest and savanna? Ecology Letters,15, 748–758.
Murphy, B. P., Paron, P., Prior, L. D., Boggs, G. S., Franklin, D. C., & Bow-
man, D. M. J. S. (2010). Using generalized autoregressive error mod-
els to understand fire-vegetation-soil feedbacks in a mulga-spinifex
landscape mosaic. Journal of Biogeography,37, 2169–2182.
Ondei, S., Prior, L. D., Vigilante, T., & Bowman, D. M. J. S. (2015). Post-
fire resprouting strategies of rainforest and savanna saplings along
the rainforest–savanna boundary in the Australian monsoon tropics.
Plant Ecology,217, 711–724.
Ondei, S., Prior, L. D., Williamson, G. J., Vigilante, T., & Bowman, D. M. J.
S. (2017). Water, land, fire, and forest: multi-scale determinants of
rainforests in the Australian monsoon tropics. Ecology and Evolution,
7, 1592–1604.
Pausas, J. G. (2015). Bark thickness and fire regime. Functional Ecology,
29, 315–327.
Petty, A. M., Werner, P. A., Lehmann, C. E. R., Riley, J. E., Banfai, D. S., &
Elliott, L. P. (2007). Savanna responses to feral buffalo in Kakadu
National Park, Australia. Ecological Monographs,77, 441–463.
R Core Team (2013). R: A language and environment for statistical com-
puting. URL http://www.R-project.org/. [Accessed 03/03/2016].
Radford, I. J., Gibson, L. A., Corey, B., Carnes, K., & Fairman, R. (2015).
Influence of fire mosaics, habitat characteristics and cattle distur-
bance on mammals in fire-prone savanna landscapes of the northern
Kimberley. PLoS ONE,10, e0130721.
Rule, S., Brook, B. W., Haberle, S. G., Turney, C. S. M., Kershaw, A. P., &
Johnson, C. N. (2012). The aftermath of megafaunal extinction: Ecosys-
tem transformation in Pleistocene Australia. Science,335, 1483–1486.
Russell-Smith, J., & Bowman, D. M. J. S. (1992). Conservation of mon-
soon rainforest isolates in the Northern Territory, Australia. Biological
Conservation,59,51–63.
Russell-Smith, J., Yates, C. P., Brock, C., & Westcott, V. C. (2010). Fire
regimes and interval-sensitive vegetation in semiarid Gregory National
Park, northern Australia. Australian Journal of Botany,58, 300–317.
Staver, A. C., Archibald, S., & Levin, S. (2011). Tree cover in sub-Saharan
Africa: Rainfall and fire constrain forest and savanna as alternative
stable states. Ecology,92, 1063–1072.
Trauernicht, C., Brook, B. W., Murphy, B. P., Williamson, G. J., & Bow-
man, D. M. J. S. (2015). Local and global pyrogeographic evidence
that indigenous fire management creates pyrodiversity. Ecology and
Evolution,5, 1908–1918.
Veenendaal, E., Torello-Raventos, M., Feldpausch, T., Domingues, T., Ger-
ard, F., Schrodt, F., ... Ford, A. (2015). Structural, physiognomic and
above-ground biomass variation in savanna–forest transition zones
on three continents–how different are co-occurring savanna and for-
est formations? Biogeosciences,12, 2927–2951.
Venables, W. N., & Ripley, B. D. (2002). Modern applied statistics with S
(4th ed.). New York: Springer.
Vigilante, T., Bowman, D. M. J. S., Fisher, R., Russel-Smith, J., & Yates, C.
(2004). Contemporary landscape burning patterns in the far North Kim-
berley region of north-west Australia: Human influences and environ-
mental determinants. Journal of Biogeography,31, 1317–1333.
Western Australian Herbarium (1998). FloraBase–the Western Australian
Flora. URL https://florabase.dpaw.wa.gov.au/. [Accessed on 15/2/
2016].
Wilson, B. (1981). Biological survey of Mitchell Plateau and Admiralty Gulf,
Kimberley, Western Australia. Western Australian Museum, Perth.
Woinarski, J. C. Z., Burbidge, A. A., & Harrison, P. L. (2015). Ongoing
unraveling of a continental fauna: Decline and extinction of Aus-
tralian mammals since European settlement. Proceedings of the
National Academy of Sciences USA,112, 4531–4540.
Woinarski, J. C. Z., Risler, J., & Kean, L. (2004). Response of vegetation
and vertebrate fauna to 23 years of fire exclusion in a tropical Euca-
lyptus open forest, Northern Territory, Australia. Austral Ecology,29,
156–176.
BIOSKETCH
Stefania Ondei is interested in vegetation ecology and conservation
of threatened communities/species. This study forms part of her
doctoral dissertation in the Environmental Change Biology Labora-
tory at the University of Tasmania led by Professor David Bowman.
A key focus of the group is understanding the impacts of global
environmental change on Australian landscapes and their fire
regimes.
SUPPORTING INFORMATION
Additional Supporting Information may be found online in the sup-
porting information tab for this article.
How to cite this article: Ondei S, Prior LD, Vigilante T,
Bowman DMJS. Fire and cattle disturbance affects vegetation
structure and rain forest expansion into savanna in the
Australian monsoon tropics. J Biogeogr. 2017;00:1–12.
https://doi.org/10.1111/jbi.13039
12
|
ONDEI ET AL.