Water, land, fire, and forest: Multi-scale determinants of rainforests in the Australian monsoon tropics

Abstract

The small rainforest fragments found in savanna landscapes are powerful, yet often overlooked, model systems to understand the controls of these contrasting ecosystems. We analyzed the relative effect of climatic variables on rainforest density at a subcontinental level, and employed high-resolution, regional-level analyses to assess the importance of landscape settings and fire activity in determining rainforest density in a frequently burnt Australian savanna landscape. Estimates of rainforest density (ha/km2) across the Northern Territory and Western Australia, derived from preexisting maps, were used to calculate the correlations between rainforest density and climatic variables. A detailed map of the northern Kimberley (Western Australia) rainforests was generated and analyzed to determine the importance of geology and topography in controlling rainforests, and to contrast rainforest density on frequently burnt mainland and nearby islands. In the northwestern Australian, tropics rainforest density was positively correlated with rainfall and moisture index, and negatively correlated with potential evapotranspiration. At a regional scale, rainforests showed preference for complex topographic positions and more fertile geology. Compared with mainland areas, islands had significantly lower fire activity, with no differences between terrain types. They also displayed substantially higher rainforest density, even on level terrain where geomorphological processes do not concentrate nutrients or water. Our multi-scale approach corroborates previous studies that suggest moist climate, infrequent fires, and geology are important stabilizing factors that allow rainforest fragments to persist in savanna landscapes. These factors need to be incorporated in models to predict the future extent of savannas and rainforests under climate change.

Full text

Ecology and Evoluon 2017; 1–13  
|
1
www.ecolevol.org
Received:2July2016
|
Revised:22November2016
|
Accepted:18December2016
DOI:10.1002/ece3.2734
ORIGINAL RESEARCH
Water, land, re, and forest: Mul- scale determinants of
rainforests in the Australian monsoon tropics
Stefania Ondei1 | Lynda D. Prior1 | Grant J. Williamson1 | Tom Vigilante2,3 |
David M. J. S. Bowman1
ThisisanopenaccessarcleunderthetermsoftheCreaveCommonsAribuonLicense,whichpermitsuse,distribuonandreproduconinanymedium,
providedtheoriginalworkisproperlycited.
©2017TheAuthors.Ecology and EvoluonpublishedbyJohnWiley&SonsLtd.
1SchoolofBiologicalSciences,Universityof
Tasmania,SandyBay,Tas.,Australia
2WunambalGaamberaAboriginalCorporaon,
Kalumburu,WA,Australia
3BushHeritageAustralia,Melbourne,Vic.,
Australia
Correspondence
StefaniaOndei,SchoolofBiologicalSciences,
UniversityofTasmania,SandyBay,Tas.,
Australia.
Email:stefania.ondei@utas.edu.au
Funding informaon
WunambalGaamberaAboriginalCorporaon;
BushHeritageAustralia.
Abstract
Thesmallrainforestfragmentsfoundinsavannalandscapesarepowerful,yetoen
overlooked,model systems to understand the controls of these contrasng ecosys-
tems.Weanalyzed the relaveeectofclimac variables on rainforestdensityata
subconnentallevel, and employedhigh-resoluon,regional-level analyses toassess
theimportanceoflandscapesengsandreacvityindeterminingrainforestdensity
inafrequentlyburntAustraliansavannalandscape.Esmatesofrainforestdensity(ha/
km2)acrossthe Northern Territory and Western Australia, derived from preexisng
maps,wereusedtocalculatethecorrelaonsbetweenrainforestdensityandclimac
variables. A detailed map of the northern Kimberley (Western Australia) rainforests
wasgeneratedandanalyzedtodeterminetheimportanceofgeologyandtopography
incontrollingrainforests,andtocontrastrainforestdensityonfrequentlyburntmain-
landandnearbyislands.InthenorthwesternAustralian,tropicsrainforestdensitywas
posivelycorrelated with rainfallandmoisture index, andnegavelycorrelated with
potenal evapotranspiraon. At a regional scale, rainforests showed preference for
complex topographic posions and more ferle geology. Compared with mainland
areas,islandshadsignicantlylowerreacvity,withnodierencesbetweenterrain
types.Theyalsodisplayedsubstanallyhigherrainforestdensity,evenonlevelterrain
wheregeomorphologicalprocessesdonotconcentratenutrientsorwater.Ourmul-
scaleapproachcorroborates previous studies that suggest moist climate, infrequent
res,and geologyareimportant stabilizing factorsthatallow rainforestfragmentsto
persistinsavannalandscapes.Thesefactorsneedtobeincorporatedinmodelstopre-
dictthefutureextentofsavannasandrainforestsunderclimatechange.
KEYWORDS
Australianmonsoontropics,re,geologicsubstrates,rainfallgradients,rainforests,topographic
reprotecon
1 | INTRODUCTION
The global extent of closed canopy tropical rainforests and savan-
nas is determined by climate, especially mean annual precipitaon
(Lehmannetal.,2014;Murphy&Bowman,2012).However,ataround
1,000–2,000mm/yearrainforestandsavannaformvegetaonmosa-
ics (Hirota, Holmgren,van Nes, & Scheer,2011; Staver,Archibald,
&Levin,2011a,2011b).Tropicalsavannasarecharacterizedbya low

2
|
ONDEI Et al.
treecoverandahighbiomassofC4grasses,whichsupportsfrequent
resinthedryseason(Bond,Woodward,&Midgley,2005;Homann,
Jaconis,etal.,2012). Bycontrast,tropicalrainforestshaveaspecies-
rich tree ora that formdense canopies, lile grass, and infrequent
reacvity.
Themechanismsthatcontrolpaerningofrainforestandsavanna
mosaicsaredisputed,withdebatepolarizedbetweentheimportance
of reand soils. One view is that edaphic factors like soil nutrients
arethe maincontrolofrainforest–savannamosaics,and reisnot a
cause but rathera consequence of vegetaon paerns (Lloydetal.,
2008;Veenendaaletal.,2015).Althoughsavannasoilsmayhavesu-
cientnutrientstockstosupportrainforesttrees(Bond,2010;Vourlis
etal.,2015),rainforestsaregenerallyfoundonmorenutrient-richsoils
comparedwithsavannas(Dantas,Batalha,&Pausas,2013;Silvaetal.,
2013).Inferlesavannasoilsareknowntolimitexpansionofrainfor-
est(Silvaetal.,2013),whiledeeperandmoreferlesubstratesallow
rainforesttogrowindrierclimates(knownas“edaphiccompensaon”;
Ash,1988;Webb,1968). However,itisnot clearwhetherthesepat-
ternsresultfromadirectedaphiceectorfromlocalfeedbacks.Soils
underneathrainforestsareoenmorerichinnutrients,comparedwith
savannas, regardless ofthe inherent ferlity of soil parent material
(Dantasetal.,2013;Silvaetal.,2013),becauseofnutrientacquision
andcycling(Silvaetal.,2008).Treecanopycoverandcanopyproduc-
vityincreasesoilnutrient content(Paiva,Silva,&Haridasan,2015),
parcularly N concentraon and availability (Schmidt & Stewart,
2003). Consequently, there are substanal praccal dicules in
makingecologicallymeaningful measurementsofsoils ferlityvaria-
on,parcularlyacross rainforestecotones,whereforestboundaries
waxandwane(Silvaetal.,2013;Warman,Bradford,&Moles,2013).
Thealternaveviewisthatrainforestandsavannaare“bi-stable”in
regionswithintermediateproducvity,andtherealizaonofvegetaon
depends on landscape re history (Bond etal., 2005; Dantas, Hirota,
Oliveira, & Pausas, 2016; Homann, Geiger, etal., 2012; Murphy &
Bowman,2012;Staveretal.,2011a;Warman&Moles,2009).Thisview
isbasedonalternavestablestate(ASS)theorywherebystabilizingfeed-
backsholdrainforestorsavannainspecic“basinsofaracon”(Hirota
etal.,2011).Resolvingthe roleofedaphicfactors incontrollingrainfor-
est boundaries directly orindirectly via feedbacks is complex and de-
mandsmulplelinesofevidence,includingdirectmeasurementsofsoils,
modeling,andexperiments(Bowman,Perry,&Marston,2015).Analysis
ofremotesensingesmates ofcanopycoverata globalscalehasbeen
presentedasevidenceforthebimodaldistribuonofrainforestsandsa-
vannas(Staveretal.,2011b).Ithasbeenarguedthattheintensityofthe
bimodalitymaybeastascalarfactassociatedwiththeuseofregres-
siontree(CART)analyses,whichimpose disconnuiesinsatellitetree
cover esmates (Hanan,Tredennick, Prihodko, Bucini, & Dohn, 2014,
2015;Staver&Hansen,2015),althoughglobal canopyheightanalyses,
based on products derived fromLiDAR measurements, conrmed the
bimodalitydetectedthroughsatellitedata(Xuetal.,2016).
Regional-levelanalyses based on remote sensing have been em-
ployedinstudiesinvesgangtheenvironmentalcontrolsofdierent
types of vegetaon(Dahlin, Asner, & Field, 2014; Fensham,Fairfax,
& Archer, 2005; Murphy etal., 2010). However, there has been
surprisinglylimitedanalysisofrainforest–savannamosaicsataregional
level.In an important pioneering study,Ash (1988) synthesized data
fromtopographicmaps,aerialphotography,andelddatatocreatea
modelof the environmentalcontrolsofrainforestsandsavanna veg-
etaoninthe wet tropicsofNorth Queensland (Australia),to assess
therelaonshipbetweenrainforestlocaon andenvironmental char-
acteriscs. Ash (1988) concluded that the distribuon of rainforest
boundariescanbeempiricallypredictedbasedonwateravailabilityand
topography,and substrate ferlitymight allow rainforeststoexpand
intootherwiseunfavorableenvironments.Thisresearchwassupported
byFensham(1995),whoemployedaerialphotographyandsatelliteim-
agerytoinvesgatetherelaonbetweendryrainforestand environ-
mentalvariablesin NorthQueensland.Tothebest ofourknowledge,
thereareno other map-based analyses of rainforest–savannamosa-
ics at a regionalscale anywhere else in the tropics. These rainforest
patchesare known tobebiodiverse and importantfor abroadcross
seconoffauna(Price,2006;Tun,White,&Mackanga-Missandzou,
1997),yettheyhavebeenpoorlyresearchedcomparedwiththemore
extensivewetrainforests(Sánchez-Azofeifaetal.,2005).
NorthwesternAustralia is an aracvemodelsystembecause it
spansa widerainfall gradientat thedriestextreme oftheAustralian
tropicalrainforestestate(Bowman,2000).TheglobalanalysisofStaver
etal.(2011b)suggeststheregionisdeterminiscallysavanna;yet,ny
patches of rainforestexist, embedded in the savanna matrix. These
environmentsrainforestsaremoreexposedtoreduetotheirhigher
boundary/corerao;nonetheless,insomelocaonsrainforestexpan-
sionhasoccurred(Banfai&Bowman,2006;Bowman,Walsh,&Milne,
2001;Clayton-Greene&Beard,1985).StudiesfromnorthernAustralia
andelsewhere inthe tropicshaveidenedtheimportance ofland-
scapesengindeterminingrainforestdistribuoninareassubjectto
highreacvity.Forexample,rainforestscan bemoreabundant on
islandsthat havelowerreacvitythanadjacentmainlandsavannas
(Clayton-Greene&Beard,1985).Rainforestscan alsobe connedto
steepgullies orvalleys(Bowman,2000;Ibanezetal., 2013;Warman
&Moles, 2009)because ofthe reprotecon theyprovide(Murphy
&Bowman,2012),althoughaddionaleectsofhighernutrientand
wateravailabilitycouldalsobeimportant(Ash,1988).
Weemployedamacroecologicalapproachtodeterminetheeect
ofclimacandgeomorphologicalfactors(topographyandgeology)on
rainforestabundance ata large spaalscale.Geology wasused as a
proxyforthenutrientstockprovidedbytheparentmaterial,toexclude
theeectofvegetaononsoilferlity.Toassessthecorrelaonsbe-
tweenclimate and rainforestdistribuon in the enrenorthwestern
Australian monsoon tropics, we analyzed exisng subconnental-
scale vegetaonmaps. We then assessed the importance of topog-
raphyandgeologyataregionalscale,as theeectsofthese factors
onrainforestdistribuon aredetectable atthis scale,comparedwith
climate(Murphy&Bowman,2012).Todoso,wegeneratedadetailed
map of rainforests in the northern Kimberley (Western Australia),
whichis characterizedbyalimited rainfallrange(200mm/year),and
avarietyofgeologiesandtopographicsengs.Withinthisregion,we
undertookalocal-scale“naturalexperiment”comparingtheinuence
oftopographyandreacvityonrainforestdensityonmainland and

|
3
ONDEI Et al.
adjacent islandswith similar rainfall, geology, and distance fromthe
coastline.Weaddressedthefollowinghypotheses:
• Atasubconnentalscale,factorsassociatedwithwateravailability
arethemainclimacdriversdeterminingrainforestdensity;
• At a regional scale, topography and geology aect rainforest
distribuon;
• At a local scale, the importance of insularity and topography is
directlyrelated to re acvity. In locaons with high reacvity
(mainland),rainforestsarepredominantlyconnedtore-sheltered
sengs,whereasin areaswithlowerreacvity(islands), rainfor-
estswillalsobeabletogrowinmoreexposedsengs.
Collecvelythisstudyinvesgatesthedriversofrainforestdistribu-
onacrossmulple spaal scalesinnorthernAustralia, therebyillumi-
nang the capacity forclimate change and re management to aect
rainforestcoverageand providing insightsforboth theorecalecology
andappliedlandmanagement.
2 | METHODS
2.1 | Geographic context
The Australian monsoon tropics are characterized by a pronounced
wet and dry seasons associated with the Australian summer mon-
soon(Bowman etal.,2010).This regionincludesthe wholeofnorth-
ernAustraliaexcepttheAustralianwet tropics in North Queensland
(Bowman, 2000; Figure1a-c). In contrast to the wet tropics, where
tropicalrainforestsdominate,themonsoontropicssupportvasteuca-
lyptsavannas(Bowman,2000;Figure2a,c).Embeddedinthesesavan-
nasareverysmallpatchesofmonsoontropicalrainforest,rangingfrom
afewtreesto100hainarea(McKenzie,Belbin,Keighery,&Kenneally,
1991). These rainforests have orisc and biogeographic anies
with wet tropical rainforests in both Asia and Australia. They have
been intensively studied given their unusual biogeography and ecol-
ogy,parcularlytheirabilitytopersistinahighlyammabletropical
savannaenvironment(Bowman,2000).Somerainforestsareknownto
growonaquifers(Kenneally,Keighery,&Hyland,1991;Russell-Smith,
1991),whichinsulatethepatchesfromregionalclimate,butourmap-
pingcouldnotdierenatethesetypesfromthemorewidespreadand
drought-adaptedrainforests(Bowman,Wilson,& McDonough,1991;
Russell-Smith, 1991). The locus of the subconnental study was the
Australian monsoon tropics west of the Carpentarian Gap biogeo-
graphicdivide,whichseparatesthebiotaoftheNorthernTerritoryand
Western Australia from Cape York Peninsula (Bowman etal., 2010).
Annual rainfall in this area varies from approximately 1,900mm in
thenortheast to700mmin thesouthwest (Figure1c), thatwould be
expectedto exert astronginuence on theabundanceof rainforest.
Thisanalysiswasmadepossiblebycombiningvegetaonmapspro-
ducedbytheNorthernTerritory andWesternAustraliangovernment
landmanagementagencies, nong that the border betweenthetwo
statesbroadlyalignwiththeOrdAridIntrusion,amajorbiogeographic
boundarythatseparatesthebiotaoftheKimberleyregionofWestern
Australiafromthatofthe“TopEnd”oftheNorthernTerritory(Figure1;
Bowmanetal.,2010;Eldridge,Poer,&Cooper,2011),andthatlikely
aectsrainforestspeciesdiversity.Inaddiontothiscoarse-scalesub-
connentalstudy, weundertook amore detailedanalysisoftherain-
foreststothe west of the Ord AridIntrusion.This region, located at
theextremeendoftheprecipitaongradient whererainforestoccurs
innorthernAustralia,haslimitedspaal variability in rainfall (1,200–
1,400mm),whichallowedustoidenfyecologicalfactors,otherthan
precipitaon, that shape rainforest distribuon. This was based on
ne-scalemappingofthetradionallandsoftheWunambalGaambera
people,henceforthcalledtheWunambalGaamberaCountry.
FIGURE1 Themonsoonrainforest
domaininnorthwesternAustralia.
(a)Thegrayarearepresentsthemonsoon
rainforestdomaininthenorthofWestern
Australia(WA;Kimberetal.,1991)and
theNorthernTerritory(NT).TheOrdArid
Intrusion,themainbiogeographicbarrier
betweenthetwostates,isindicated.
Dashedlinesindicaterainfallisohyets(mm).
Theinsetsshow(b)thestudyareawithin
Australiaand(c)elevation(minimum,0m,
white;maximum,960m,black)

4
|
ONDEI Et al.
TheWunambalGaambera Countryoccupies anarea of9,144km2,
dominatedbybiodiversetropicalsavannasoccurringon deeplyweath-
ered sandstones and basalc base rocks of Precambrian age, oen
cappedbyCainozoiclaterites.Inthisregion,averageannualrainfalloc-
cursalmost enrelyduring thesummerwetseason(November–April),
whiletherestoftheyearisalmostrain-free(Beard,1976).Thelandscapes
areshapedbygeology;thedominantsubstratesareinferlesandstone,
where the Holocenesea-levelrise has created rugged coastlines, and
themoderatelyferlebasaltcountry,characterizedbygentleslopesand
hills(Beard,1979;Specketal.,2010;Figure2b).Thevegetaonispre-
dominantlyeucalypt savanna.Eucalyptus tetrodonta–Eucalyptus miniata
savannasarefoundonthelateritemesasandhills,whileEucalyptus tec-
ca–Eucalyptus grandifoliasavannasarecommonondeeper,claysoils
onplains.Smallpatchesofsemi-deciduous rainforestsareinterspersed
inthesavanna(Figure2c,e),typicallylocatedinre-protectedlocaons
(Vigilante,Bowman,Fisher,Russel-Smith,&Yates,2004).
Fireregimes in the northern Kimberleyare stronglyshapedbyan-
thropogenicignions and have been forover 40,000years(O’Connor,
1995).This ancient tradion ofAboriginal remanagementis likely to
havemaintainedbiodiverseopensavannahabitatsand protectedsmall
isolatedrainforest fragments(Mangglamarra,Burbidge, & Fuller,1991;
Trauernicht, Brook, Murphy, Williamson, & Bowman, 2015; Vigilante,
Murphy,&Bowman,2009).ThecessaonofAboriginalremanagement
in many northern Australian environments has been associated with
degradaonof somerainforestsand otherre-sensive plant commu-
nies(Russell-Smith&Bowman,1992;Trauernicht,Murphy,Portner,&
Bowman,2012),althoughinrarelyburntareastherecanbeexpansionof
rainforest(Bowman&Fensham,1991;Clayton-Greene&Beard,1985).
2.2 | Rainforest mapping and analyses
2.2.1 | Subconnental scale—climac drivers of
rainforest density
The distribuon and areal extent of the rainforests in the north-
western Australian monsoon tropics was determined by blending
exisng vegetaon maps. Total coverage of rainforest in Western
Australiaand Northern Territorywas calculated fora lace ofgrid
FIGURE2 (a)ExtentofthemappedareainthenorthernKimberley;(b)themaingeologytypesinthearea,and(d)rainforestdensity,
expressedashaperkm2ofland.Inthisarea,rainforeststypicallyoccuras(c)smallpatches(green)surroundedbysavanna(brown),with(e)sharp
boundariesbetweenthetwovegetationtypes
(a)
(d)
(b) (c)
(e)

|
5
ONDEI Et al.
cells 50×50km in area. The Western Australia map (1:200,000)
wasderivedfromKimber,Forster,andBehn(1991),whousedsemi-
automatedclassicaonofLandsat imagery taken in 1986 and did
not dierenate oriscs or structure variaon among rainforests.
The Northern Territory vegetaon data (1:80,000), based on in-
terpretaon of aerial photography classied according to Russell-
Smith (1991), were supplied by the Department of Land Resource
Management,© NorthernTerritoryof Australia.In calculangrain-
forest coverage in the Northern Territory lace cells, we selected
bothdryandwetrainforest typesbecausetheyarestructurally and
oriscally similar to the Western Australian rainforests (Bowman,
1992;Kenneallyetal.,1991).Wecombined the Western Australia
and Northern Territory data to create a map of the northwestern
rainforestdomain,extendingfrom11.00°Sto18.00°Sinlatudeand
from122.14°Eto 138.00°E in longitude (Figure1).Thisresulted in
192and63gridcellsintheNorthernTerritoryandWesternAustralia,
respecvely. For each grid cell, mean annual rainfall, precipitaon
seasonality(coecientofvariaonofmonthlyrainfallexpressedasa
percentage),potenalevapotranspiraon, moistureindex(meanan-
nual precipitaon over potenal evapotranspiraon; Thornthwaite,
1948),andannualmeantemperaturewerecalculated forthecenter
pointofeachcell.Rainfall,precipitaonseasonality,andtemperature
datawereobtained from WorldClim Global ClimateData(Hijmans,
Cameron,Parra, Jones,& Jarvis,2005),and moistureindex andpo-
tenalevapotranspiraonweredownloadedfromtheGlobalAridity
and PET database (Zomer, Trabucco, Bossio, & Verchot, 2008).
Minimum, maximum, and median values of the climac variables
werecalculated separately for theWesternAustralia and Northern
Territorygridcells.
2.2.2 | Regional scale—regional drivers of
rainforest density
We generated a map of the rainforests in the northern Kimberley,
covering the enre Wunambal Gaambera Country and expanding
theanalysis totheadjacent coastalareas (totalsurface12,572km2),
asfollows. Orthophotos (scale 1:8,000)takenduring the dryseason
(May–August)oftheyears2005–2007wereusedcreateamapofthe
rainforestpatcheslocatedinthestudyarea.Alaceof30×30mcells
wasoverlaidontheorthophotos,andeverycell wasmanuallyclassi-
edas“rainforest,”“savanna,”or“other.”Thevegetaontypeofeach
cellwas considered tobethe one occupyingthehighest proporon
ofthecell.Amapoftherainforestpatcheswasproducedbymerging
theconguouscellsclassiedas“rainforest”(Figure2d).Ahelicopter
surveywasconductedtovalidatethemap.Theight path,designed
toincludelocaonswithbothhighandlowrainforestcover,included
coastalandinlandareasaswellasislands.Itcoveredthemaingeologic
substrates,inparcularbasalt,sandstone,andlaterite.Weewalong
theselectedpathatanaverageheightof300mabovethegroundfor
atotallengthof550kmon22September2013.Waypoints,collected
every10s,were visually idened as “rainforest”or“savanna.” The
pointswerethen buered 30m and intersected with the rainforest
map. A confusion matrix was generated to calculate map accuracy,
omissionandcommission errors,andkappacoecientof agreement
(Congalton,1991).
Patch size, distance from the coastline, and distance from the
nearestdrainagelinewerecalculatedforeveryrainforestpolygonon
theregionalscalemap.Rainforestdensitywascalculatedashaofrain-
forestperkm2ofland,basedonagridof1×1kmsizecellsforcom-
putaonalreasons. Foreach cell,wealso calculated:(1)the geology
category,based on the predominant geology type in each 1×1km
cell;and(2)thetopographiccategory,basedonthepredominanttopo-
graphic posion index (TPI;Jenness, 2006) in the cell. The TPI was
calculatedforeverypixelinthemappedareafroma30-mresoluon
digital elevaon model (DEM), based on the dierencein elevaon
betweeneachpixelandtheaverageelevaonof theeight neighbor-
ingpixels;valueslowerthan −1wereclassiedas “valley,”andvalues
higherthan+1as“ridge.”Intermediatevalueswereclassiedas “at”
or“slope”dependingon theslopeofthepixel(≤4° foratareas,>4°
forslopes),obtainedfromthe30-mDEM.(3)Eachgridcellwasfurther
classiedashaving “complex”or“level” terrain,nong that complex
terrainisoenassociatedwithrockiness.Cellsinwhichthecategories
“valley”+“ridge”+“slope”occupiedmore than50%ofthecellwere
classied as “complex,” the others as “level,” The average rainforest
densityin thenorthern Kimberleywasthencalculated foreach geo-
logicsubstrateandTPIbasedontherainforestdensitygrid.
2.2.3 | Local- scale natural experiment–Mainland
versus Islands
Weexpected there wouldbe dierences inre acvity andrainfor-
estdistribuon between islandsandmainland, because islandshave
been subject to fewer human ignions due to infrequent visitaon
inrecentmes(Vigilante etal.,2013)andtheseaprovidesanatural
rebreakfromsurroundinglandscaperes.Totestthis,wecompared
rainforestdensitygridcellsonislandsandthemainland.Weselected
areasthatweregeographically,oriscally,andecologicallysimilarby
extracngfromthe“regionalscaledataset”onlygridcellswiththefol-
lowingaributes:meanannualrainfallbetween1,250and1,382mm/
year,distancefromthe coastline <5km (equivalent to the radiusof
the biggest island hence the maximum distance from the coastline
onislands),andgeologydevelopedonbasalt,laterite,orcoastalsedi-
ments. Islands and coastal areas of the northern Kimberley are o-
riscally similar, with only a very small group of taxa recoded only
fromislands(Lyons,Keighery,Gibson,&Handasyde,2014).Gridcells
locatedon the BougainvillePeninsula were includedinthe category
“islands,”dueitsnarrowneckwhichmakesitfunconallyequivalent
toanislandintermsofisolaonfromthemainland.
Fireacvitywascalculatedfroma15-yearrehistorymap(2000–
2014),createdatapixelresoluonof250mbasedon MODISsatel-
lite imagery,accessed via North AustralianFire Informaon website
(hp://www.renorth.org.au/na3/).Due tothe coarse resoluonof
there historymap,itwasimpossibleto accuratelylocateeveryre
scar,sothedatawereusedtoprovidecoarse-scaleinformaonabout
dierencesinreacvitybetweenthemainlandandislands.Forevery
cell of the rainforestdensity grid, the area-weighted proporon of

6
|
ONDEI Et al.
yearsburntwas calculated bydividingtheaverage number ofyears
inwhich thecell wasburnt by15, thetotalnumberofyearsinves-
gated.The averagereacvity peryear andrainforestdensitywere
calculatedforcellsclassiedas“island”or“mainland”and,withineach
category,“complex”or“level”terrain.
2.3 | Stascal analyses
Atasubconnentalscale,weemployedthePearsonproductmoment
correlaoncoecienttoexaminecorrelaonsamongrainforestden-
sityandtheclimacvariables,andpresentedtheresultsinaconstel-
laondiagram.Forpresentaon(butnottheanalysis),weaggregated
thegridcellsinto200-mmmeanannualrainfallbinsandcalculatedthe
averagerainforestdensityforeachbin.
Ataregionalscale,we rsttested forspaal autocorrelaonin
rainforest density and assessed minimum sampling distance, es-
matedbyplongthesemi-varianceasafunconofdistance,using
thesowareR(RCoreTeam,2013)andtheRpackagegeoR(Ribeiro
&Diggle,2001;AppendixS1).Wethentestedwhetherthefactors
terrainandgeologyarerelatedtorainforestdensity.Wealsochecked
whetherrainforestdensitywasassociatedwithgeologywithinlevel
and complex terrain types. Todo this, we used generalized linear
models(GLMs)andcompletesubsetregressionandmodelselecon
basedonAkaike’sinformaoncriterion(AIC;Burnham&Anderson,
2002),calculatedusing theRpackage “MuMIn”(Bartoń,2009).We
used the compound Poisson-gamma distribuon, included in the
tweediefamilyofdistribuons,whichallowsregressionmodelingof
zero-inatedposiveconnuousdata(Rpackages“tweedie”[Dunn,
2014] and “statmod” [Smyth,Hu, Dunn, Phipson, & Chen, 2015]).
To assess the importance of each variable,we calculated Akaike
weights(wi),which representtheprobabilitythat a given modelis
thebestinthecandidateset(Burnham&Anderson,2004).Wethen
calculatedvariableimportance(w+) asthesummed wiofthe mod-
els in which thevariable occurs. w+ values higher than 0.73 were
considered to indicatethat the variable is a stascally important
predictor (Murphyetal., 2010). Model summaries are provided in
AppendixS2.
When comparing mainland versus islands, we examined dif-
ferences in rainforest density and re acvity between locaons,
tesng for the factors insularity (island or mainland) and terrain
(complex or level).Todo so, we employed GLMs, using the com-
poundPoisson-gammadistribuon for both rainforestdensityand
reacvity,completesubsetregression,andmodelseleconbased
on AIC as described above.Variable importance was assessed by
FIGURE3 Comparisonoftherainforest
domaininWesternAustralia(WA)and
NorthernTerritory(NT),showing(a)
averageannualrainfall,(b)moisture
index,(c)meanannualtemperature,
(d)precipitationseasonality,and(e)
potentialevapotranspiration(PET).Boxes
indicatemedianvaluesandupperand
lowerquartiles,barsthe10thand90th
percentiles

|
7
ONDEI Et al.
calculangw+, asdescribedabove.Modelsummaries areprovided
inAppendixS2.
3 | RESULTS
3.1 | Subconnental scale
ThenorthwesternAustralianrainforestdomaincoveredan areaof
640,000km2,ranging fromthe coastlinetoamaximumof 350km
inland.RainforestdensitywaslowerwestoftheOrdAridIntrusion:
inWestern Australiarainforest densityrangedfrom0to 8.7haof
rainforestper km2ofland (average1.1±0.2ha/km2),while inthe
NorthernTerritory the rangewas0–19.0ha/km2 of land(average
1.4±0.2ha/km2). The Northern Territory showed higher median
valuesandabroaderrangeofbothmean annualrainfallandmois-
tureindex(Figure3a,b).Meanannualtemperatureandprecipitaon
seasonalityshowedhighermedianandmaximumvaluesinWestern
Australia and minimum in the Northern Territory (Figure3c,d),
while annual potenal evapotranspiraon had a similar range in
thetwo statesbut highermedianvalues inthe NorthernTerritory
(Figure3e).
Therewasaposivecorrelaonbetweenrainforestdensityand
bothmeanannualrainfallandmoistureindex(Figures4and5),which
were also posively correlated. Potenal evapotranspiraon was
negavely correlated with rainforest density, moisture index, and
rainfall,whileprecipitaonseasonalitywasnegavelycorrelatedwith
rainfalland moisture index. Meanannual temperaturewasnot cor-
relatedwithanyoftheclimac variablesinvesgatednorwithrain-
forestdensity.
3.2 | Regional scale
Intotal,2,902points were assessed during the aerial survey. There
wasastrongconcordancebetweentherainforestmapandtheaerial
assessment,witharesulngoverallmapaccuracyof93%(Kappacoef-
cient.78; AppendixS3).A highaccuracywas obtainedforsavanna
points (95% for both producer’s and user’s accuracy), meaning few
savannapoints weremistaken forrainforest. Weaributethe lower
producer’s and user’s accuracy for rainforests (83% and 82%, re-
specvely)tothe orisccomposionof themonsoonvine thickets,
where semi-deciduous species can dominate (Beard, 1979), making
poronsof theforest patchesundetectablefromorthophotostaken
duringthedryseason.
Savannawasbyfarthemostcommonvegetaon,covering98.9%
ofthe area.Wedetectedatotal of6,460rainforest patchescover-
ing10,300ha,equivalentto 0.82% of the mapped land. Patchsize
rangedfrom 0.1to 220ha andaveraged1.6ha±0.1(SE).Seventy-
vepercentof patchesweresmallerthan1ha,and only2.5%were
larger than 10ha (Figure6a).More than 40% of the mapped rain-
forestpatcheswerelocatedwithin1kmofthecoastline(Figure6b),
butpatchesweredetectedupto47kminland(average4.7km±0.1
[SE]).Asimilarpaernwasidenedfordistancefromdrainagelines,
with64%ofthepatcheslocatedwithin1kmofthenearestdrainage
line(Figure6c),but someup to32kmdistant(average1.7km±0.0
[SE]).
Rainforest density was strongly dependent on both terrain
(w+=1.00) and geology (w+=1.00); averagerainforest density was
higheronrelavelyferlesubstrates(laterite,coastalsediments,and
basalt),and lower onalluviumand colluvium andinferlesandstone
(Figure7a). Average rainforest density was also higher in complex
terrain such as ridges, slopes, and valleysand lower on level areas
(Figure7b).Themodelincludinggeologyandterrainexplained32.1%
of the deviance.The preference for relavelynutrient-rich geology
was independent on terrain, as on both level and complexterrains
rainforest density was strongly associated with geology (w+=1.00
inboth cases;AppendixS4). Geologyexplained13% ofdevianceon
complexterrainand10%ofdevianceonlevelterrain.
FIGURE4 Constellationdiagramshowingthestrength
anddirectionofcorrelationsamongrainforestdensityandthe
climaticvariablesaverageannualrainfall,moistureindex,potential
evapotranspiration(PET),precipitationseasonality,andmeanannual
temperatureinthemonsoonrainforestdomaininnorthwestern
Australia.Positivecorrelationsarerepresentedbyblacklines,
negativecorrelationsbygraylines.Correlationsstrongerthan.4or
−.4areindicated;widerlinesindicatestrongercorrelations,narrower
linesweakercorrelations
FIGURE5 Averagerainforestdensitybyaverageannualrainfall,
calculatedwithintherainforestdomainintheAustralianmonsoon
tropics.Thenumbersaboveeachcolumnrepresentthenumber
ofgridcellincludedinthatrainfallinterval.Errorbarsrepresent
standarderror

8
|
ONDEI Et al.
3.3 | Mainland versus Islands
Thegridcellsonbasalt,laterite,andcoastalsubstratesandwithin5km
ofthe coastcovered anarea of332.4km2 onislands and693.1km2
onthemainland. Thetotalareacoveredby rainforestswas47.0km2
ontheislands,comparedwith13.6km2onthemainland,sothatrain-
forestdensitywassevenmeshigherontheislands(Table1).Islands
weremoretopographicallycomplexthanthemainland(Table1).There
wasstascal supportfor aninuence ofboth insularity(w+=1.00)
andterrain(w+=1.00)on rainforestdensity(Table2),andthemodel
includingboth explained35%of deviance.There waslessre acv-
ityonislands(average0.061±0.003mesburntperyear)thanon
themainland(average0.266±0.004mesburntperyear;w+=1.00),
andinsularityaloneaccountedfor39.1%ofdeviance.Contrarytoex-
pectaons,therewasnostascalsupportforaneectofterrainon
reacvity(w+=0.48;Table2).
4 | DISCUSSION
WefoundthatinnorthwesternAustralianmonsoontropicsrainforest
patchesare ny andscaeredacross a vastsavannamatrix. Due to
theirsmallsize, these rainforest fragments areessenallyundetect-
ableattheresoluonemployedbyglobal-levelassessments(Murphy
&Bowman,2012;Staveretal.,2011b).Atasubconnentalscale,the
strongcorrelaonbetweenrainforestdensityandannualrainfall,po-
tenalevapotranspiraon,andmoistureindexhighlightedtheprimacy
ofwatersupplycomparedwithmeanannualtemperatureandprecipi-
taonseasonality,andsupportedourrsthypothesis.Thiscorrelaon
iscongruouswith the observaon that atrendof increasing rainfall
innorthernAustraliasince the 1940s is the key driver of rainforest
patch expansion (Banfai & Bowman, 2007; Bowman etal., 2001).
These ndings are also consistent with the global trend of increas-
ingproporonofrainforest(anddecreasingsavanna)asmeanannual
precipitaonincreases(Hirotaetal.,2011;Murphy&Bowman,2012).
InthedrierlandscapeswestoftheOrdAridIntrusion,rainforestspe-
ciesdiversityisalsolowerthantotheweereastandmostofthe
FIGURE6 FrequencydistributionofrainforestpatchesinthenorthernKimberleyregionaccordingto(a)size,(b)distancefromthecoastline,
and(c)distancefromthenearestdrainageline.Notethelogarithmicscaleforthexaxes
FIGURE7 Rainforestdensityinrelationto(a)geologicsubstrate,
and(b)topographicpositioninthenorthernKimberley(regional
analysis).Rainforestdensitywashighestoncoastalsediments,basalt,
andlateriticsubstrates.Itwasalsohigheronridges,slopes,and
valleys,andalmostabsentinflatareas.Errorbarsrepresentstandard
errors

|
9
ONDEI Et al.
speciesinWestern Australia are assubsetof those intheNorthern
Territory(Kenneallyetal., 1991). However, thepresenceof rainfor-
estsinthenorthwesternAustralianmonsoontropicsshowedthatthe
regioncannotbedenedasdeterminiscallysavannabasedsolelyon
climate(Murphy&Bowman,2012).Similarly,rainforestpatchesexist
throughout much of the tropics globally, which suggests that in all
butaridtropicalregions,climate aloneisnottheonlyfactorcontrol-
lingrainforestdistribuon (Staver etal., 2011b). InBrazil,forexam-
ple, small patches of deciduous and semi-deciduous rainforests are
interspersedin a matrix ofsavannaplants (Cerrado) or thornscrubs
(Caanga)andarerestrictedto slopes and moist, nutrient-rich sites
(Leal, da Silva, Cardoso, Tabarelli, & Lacher, 2005; Oliveira-Filho &
Raer,2002).Likewise,inIvoryCoastthedominantsavannavegeta-
onisscaeredwithsmallpatchesofdryrainforest(Goetze,Hörsch,
&Porembski,2006).
Wealsofoundsupportforoursecondhypothesisthattopography
and geologyaect rainforest distribuon. Theinuence of topogra-
phyonrainforestdensitywasmanifestinthehigherrainforestdensity
oncomplex comparedwith levelterrain.Rainforestdensitywasalso
higheronnutrient-richbasaltcomparedwiththenutrient-poorsand-
stone,despitethehigherrefrequencyandintensityrecordedonba-
salt(Vigilanteetal.,2004).Thispaernisconsistentwiththeedaphic
compensaonhypothesis (Ash,1988;Webb&Tracey,1981), whose
underlyingmechanismisprobablythe eect ofincreasedferlity in
enhancingplantgrowth,allowingtreestoreachthethresholdsizethat
triggersthe switchfrom savannato rainforestthroughgrass shading
(Homann,Geiger, etal., 2012; Murphy& Bowman, 2012). It is im-
portant to note that geology and terrainare typically not indepen-
dent.Forexample, inthenorthern Kimberleyroundedhillsaremore
commononbasalt,whilesteepgorgesarefrequentlyfoundonsand-
stone(Vigilanteetal., 2004). However,in ouranalysis wewereable
todemonstrateaneectofgeologyalonebycomparingareasonthe
sameterrain, showingthat thereare morerainforestson basaltthan
onlessferlegeologies.
There was onlyparal support for our third hypothesis that in-
sularity and topographic eects are directly related to re acvity.
Clearly,thereweremorerainforestsonislands,wheretherewasalso
lessreacvitycomparedwiththemainland.Theimportanceofrein
restricngrainforestshasbeendemonstratedbyrainforestexpansion
inothersavannalandscapeswhererehasbeenexcluded,innorthern
Australia(Fensham& Butler,2004; Scoetal., 2012)and elsewhere
(Bond, Midgley,Woodward, Homan, & Cowling, 2003). Rainforest
speciesaretypicallylessretolerantthansavannaspeciesduetothin-
ner bark and less developedpost-rerecovery mechanisms (Lawes,
Midgley, & Clarke,2013; Ondei, Prior, Vigilante, & Bowman, 2015;
Pausas,2015). However,we failed to detect acorrespondingdier-
enceinreacvitybetweenterraintypesonthemainlandandfound
onlyaminordierenceonislands.Therearetwopossiblereasonsfor
this lack of correspondence,which are not mutually exclusive. One
isthat terrain, orassociated rockiness,did exertaninuence onre
acvity,butthiswas obscuredbythecoarsescaleofthe gridcellsin
ouranalysis(1×1km).Anotherpossiblereasonisthatthehigherrain-
forestdensity oncomplexterrain is theresultofwaterand nutrient
accumulaon (Ash, 1988; Daws, Mullins, Burslem,Paton, & Dalling,
2002),ratherthantopographicreprotecon.Nonetheless,thepres-
enceofrainforestonlevelterrainonislands,butnotonthemainland,
suggeststhat reis animportantcontrollerof rainforestdistribuon
intheregion.
Wesuggestthatrainforestdensityisdeterminedbythe interplayof
reacvityandplantgrowthrates(Figure8a).Fireacvityisshapedby
insularity and possiblyterrain complexity, while plant growth ratesare
knowntobecontrolledbywateravailabilityand thenutrientstockpro-
videdbythegeologicalsubstratecontrol(Murphy&Bowman,2012),with
aneectofterraininenhancingwaterandsoilaccumulaon(Ash,1988).
Growthratesaectthecapacityofrainforesttreestogrowrapidlyandes-
capethe“retrap,”therebydevelopingaclosedcanopywhichshadesout
grassbiomass,reducingrefrequencywhichinturnreinforcesrainforest
TABLE1 Extent,rainforestcover,averagerainfall,andextentof
thegeologicsubstratesandterraintypeontheselectedgridcells
usedtocomparerainforestdensityonislandsandthemainlandin
thenorthernKimberley
Islands Mainland
Totalland(km2)332.4 693.1
Rainforestarea(km2)47.0 13.6
Rainforestdensity(ha/km2)14.1 2.0
Averagerainfall(mm/year) 1348±1 1307±1
Geologytype(%)
Basalt 64.4 81.2
Coastalsediments 0.5 2.7
Laterite 35.1 16.1
Terraintype(%)
Complex 62.8 55.6
Level 37.2 44.4
TABLE2 Averagereacvity,measuredasmesburntperyear,andaveragerainforestdensity,measuredasha/km2,forcomplexandlevel
terrainslocatedontheselectedgridcellsonislandsandmainlandinthenorthernKimberley
Level terrain Complex terrain w+
Mainland Islands Mainland Islands Terrain Insularity
Fireacvity(averagemesburntper
year±SE)
0.26±0.01 0.08±0.01 0.27±0.01 0.05±0.00 0.48 1.00
Rainforestdensity(ha/km2±SE)0.30±0.06 6.81±0.73 4.29±0.42 19.69±0.88 1.00 1.00
The w+indicatesthestascalsupportforthetermsterrainandinsularitywhencomparingmainlandandislands(fullresultsarepresentedinAppendixS2).

10
|
ONDEI Et al.
expansion(Dantasetal.,2013;Homann,Geiger,etal.,2012;Murphy&
Bowman,2012).OurndingsaresummarizedinFigure8b,whichshows
characteriscpaernsofrainforestfragmentsinthelandscapeandhow
these fragments are likelyto expand in response to a weng climate
undercontrasngferlityandreregimes.Rainforestexpansionshould
beproporonally greaterin lower-rainfallareas thatcurrentlyhavelow
rainforestdensity,like thenorthernKimberley, becausetherearemore
landscapenichesavailableforoccupancy,suchasnutrient-richandre-
protectedsites.Aprediconofourworkisthat,underthecurrentweng
trend,therewillbeconnuingrainforestexpansion inthe Kimberley,as
has been observed elsewhere in the Australiantropics (Russell-Smith,
Stanton,Edwards,&Whitehead,2004).
FIGURE8 SynthesisoftheenvironmentaldeterminantsofrainforestfragmentsinthenorthernKimberley.(a)Diagramshowingthepositive
(+)andnegative(−)effectsofenvironmentalfactorsonrainforestdensity.Dottedlinesrepresentprobableeffects.(b)Toprow:obliqueaerial
photographsshowingexamplesofthedensityofrainforestfragmentsonsiteswithcontrastinggeology(sandstonevs.basalt)andfireactivity
inthenorthernKimberley.Secondrow:3Drenderingsofrainforestdistribution(darkgreen)onsandstone(nutrient-poor)andbasalt(nutrient-
rich)landscapeundercurrentdryclimate.Bottomrow:3Drenderingsofplausiblerainforestdensityunderaclimateaswetascoastalregionsof
theNorthernTerritory.Inthisstudy,wedemonstratedthatinsularsiteshavesubstantiallylowerfireactivitythanenvironmentallycomparable
mainlandsavannaareas.Underthecurrentclimate,rainforestdensityishighestonfertileinfrequentlyburntareas,andinfrequentlyburnt
landscapesisconfinedtotopographicallyfire-protectedsettings,particularlyonnutrient-poorgeology.Underawetterclimate,weexpectthe
rainforestpatchestoexpandandnewpatchestoestablishinsuitablelandscapeniches,withthegreatestexpansiononbasaltlandscapes.The
exactamountofexpansionisunpredictablebecauseoftheinfluenceoffireactivityandfiremanagement

|
11
ONDEI Et al.
5 | CONCLUSION
Wehaveshownthatthedensityofmonsoonrainforestinthenorth-
western Australian savanna is aected by moisture availability,
substrate, and re. The eect of these drivers appears to involve
complicatedfeedbacksandinteracon,suchasthecombinedeects
ofpotenalreproteconandincreasedproducvityintopographi-
callycomplexterrains.Weacknowledgethatourcorrelaveanalysis
cannotseparatecause andeect,ortestthere-drivenASSmodel
in explaining the distribuon of rainforests. To do this demands
analysis of vegetaon boundary dynamics coupled with contrasts
of substrate type and re history. This can be achieved through
carefully designed regional-scale analysis of rainforest boundaries
trends,suchas eldsurveysandhistoricalsequences ofaerialpho-
tography(Banfai&Bowman,2006;Butleretal.,2014;MacDermo,
Fensham,Hua,&Bowman,2016)andisthesubjectofasubsequent
paper.Despiteits limitaons, ourapproachis an important stepin
understandingthe eect ofclimatechange and anthropogenicdis-
turbancesonnaturallyfragmentedrainforestselsewhereinthetrop-
icalsavannabiome.
ACKNOWLEDGMENTS
This study was supported by the Wunambal Gaambera Aboriginal
Corporaon and Bush Heritage Australia as part of their “Healthy
Country”landmanagement plan. We thank the UunguuRangersand
TradionalOwnersforthesupportprovidedintheeld.Wewouldalso
liketothanktheanonymousreviewersfortheirvaluablecommentsand
suggesonswhichgreatlyhelpedimprovingthequalityofthepaper.
CONFLICT OF INTEREST
Nonedeclared.
REFERENCES
Ash,J.(1988).Thelocaonandstabilityofrainforestboundariesinnorth-
easternQueensland,Australia.Journal of Biogeography,15(4),619–630.
Banfai,D. S.,& Bowman,D.M.J.S. (2006).Fortyyears oflowland mon-
soonrainforestexpansioninKakaduNaonalPark,NorthernAustralia.
Biological Conservaon,131(4),553–565.
Banfai,D.S.,&Bowman,D.M.J.S.(2007).Driversofrain-forestboundary
dynamicsin KakaduNaonalPark,northernAustralia:Aeldassess-
ment.Journal of Tropical Ecology,23(1),73–86.
Bartoń,K.(2009).MuMIn: Mul-Model Inference.Rpackageversion0.12.2.
Retrievedfrom:hp://CRAN.R-project.org/package=MuMIn.
Beard,J. S. (1976). The monsoon forest ofthe Admirality Gulf, Western
Australia.Vegetao,31,177–192.
Beard,J. S. (Ed.) (1979). Vegetaon survey of Western Australia: Kimberley
1:1 000 000 vegetaon series sheet 1. Perth,Australia: University of
WesternAustraliaPress.
Bond,W.J.(2010).Donutrient-poorsoilsinhibitdevelopmentofforests?A
nutrientstockanalysis.Plant and Soil,334(1),47–60.
Bond,W.,Midgley,G., Woodward,F.,Homan,M.,&Cowling,R.(2003).
Whatcontrols SouthAfricanvegetaon-climate orre? South African
Journal of Botany,69(1),79–91.
Bond,W.J.,Woodward,F.I.,&Midgley,G.F.(2005).Theglobaldistribuon
ofecosystemsinaworldwithoutre.New Phytologist,165(2),525–537.
Bowman,D.(1992).Monsoonforestsinnorth-westernAustralia.II.Forest-
savannatransions.Australian Journal of Botany,40(1),89–102.
Bowman,D.M.J.S.(2000).Australian rainforests: Islands of green in a land of
re.Cambridge,UK:CambridgeUniversityPress.
Bowman,D.M.J.S., Brown,G. K.,Braby,M. F.,Brown,J.R., Cook,L. G.,
Crisp, M. D., … Ladiges, P.Y. (2010). Biogeographyof the Australian
monsoontropics.Journal of Biogeography,37(2),201–216.
Bowman,D.,&Fensham,R.(1991).Responseofamonsoonforest-savanna
boundary to re protecon, Weipa, northern Australia. Australian
Journal of Ecology,16(1),111–118.
Bowman,D. M.J.S., Perry,G.L. W.,& Marston, J.B. (2015). Feedbacks
andlandscape-levelvegetaondynamics.Trends in Ecology & Evoluon,
30(5),255–260.
Bowman, D. M.J. S., Walsh, A., & Milne, D.J. (2001). Forest expansion
andgrasslandcontraconwithinaEucalyptussavannamatrixbetween
1941and 1994atLitcheldNaonalPark intheAustralian monsoon
tropics.Global Ecology and Biogeography,10(5),535–548.
Bowman,D.M.J.S.,Wilson,B.A.,&McDonough,L.(1991).Monsoonforests
in NorthwesternAustralia I. Vegetaonclassicaon and the environ-
mentalcontroloftreespecies.Journal of Biogeography,18(6),679–686.
Burnham,K. P.,&Anderson,D. R.(2002).Model selecon and mulmodel
inference: A praccal informaon-theorec approach. New York, NY:
Springer.
Burnham, K. P., & Anderson, D. R. (2004). Mulmodelinference under-
standing AIC and BIC in model selecon. Sociological Methods &
Research,33(2),261–304.
Butler,D. W.,Fensham, R.J., Murphy, B.P.,Haberle,S. G., Bury,S.J., &
Bowman, D.M. J. S. (2014). Aborigine-managed forest, savanna and
grassland: Biome switching in montane eastern Australia. Journal of
Biogeography,41(8),1492–1505.
Clayton-Greene, K., & Beard, J. (1985). The re factor in vine thicket
and woodland vegetaon of the Admiralty Gulf region, north-west
Kimberley, WesternAustralia. Proceedings of the Ecological Society of
Australia,13,225–230.
Congalton,R. G. (1991).A reviewof assessingthe accuracy ofclassica-
ons of remotelysensed data. Remote Sensing of Environment,37(1),
35–46.
Dahlin,K.M.,Asner,G.P.,&Field,C.B.(2014).Linkingvegetaonpaerns
to environmental gradients and human impacts in a mediterranean-
typeislandecosystem.Landscape Ecology,29(9),1571–1585.
Dantas,V. D.L., Batalha, M. A., & Pausas, J. G. (2013). Fire drivesfunc-
onal thresholds on the savanna–forest transion. Ecology, 94(11),
2454–2463.
Dantas,V.D.L.,Hirota,M.,Oliveira,R.S.,&Pausas,J.G.(2016).Disturbance
maintainsalternavebiomestates.Ecology Leers,19(1),12–19.
Daws, M., Mullins, C., Burslem, D. R. P., Paton, S., & Dalling,J. (2002).
Topographicposionaectsthewaterregimeinasemideciduoustrop-
icalforestinPanamá.Plant and Soil,238(1),79–89.
Dunn,P.K. (2014). Tweedie: tweedie exponenal family models.R package
version 2.2.1. Retrieved from: hps://cran.r-project.org/web/pack-
ages/tweedie/index.html
Eldridge,M.D.,Poer,S.,&Cooper,S.J.(2011).Biogeographicbarriersin
north-westernAustralia: Anoverviewand standardisaonof nomen-
clature.Australian Journal of Zoology,59(4),270.
Fensham,R.J.(1995). Floriscsand environmentalrelaonsofinland dry
rainforest in North Queensland, Australia. Journal of Biogeography,
22(6),1047–1063.
Fensham, R. J., & Butler, D. W. (2004). Spaal paern of dry rainfor-
est colonizing unburnt Eucalyptus savanna. Austral Ecology, 29(2),
121–128.
Fensham,R.J., Fairfax,R.J., &Archer,S. R. (2005).Rainfall, landuse and
woody vegetaon cover change in semi-arid Australian savanna.
Journal of Ecology,93(3),596–606.

12
|
ONDEI Et al.
Goetze,D.,Hörsch,B.,&Porembski,S.(2006).Dynamicsofforest–savanna
mosaics in north-eastern Ivory Coast from1954 to 2002. Journal of
Biogeography,33(4),653–664.
Hanan,N.P.,Tredennick,A.T.,Prihodko,L.,Bucini, G.,& Dohn,J.(2014).
Analysis of stable statesin global savannas: Is the CART pulling the
horse?Global Ecology and Biogeography,23(3),259–263.
Hanan,N.P.,Tredennick,A.T.,Prihodko,L.,Bucini, G.,& Dohn,J.(2015).
Analysisofstablestatesinglobalsavannas—AresponsetoStaverand
Hansen.Global Ecology and Biogeography,24(8),988–989.
Hijmans,R.J., Cameron,S. E.,Parra,J. L.,Jones,P.G., &Jarvis,A.(2005).
Veryhighresoluoninterpolatedclimatesurfacesforgloballandareas.
Internaonal Journal of Climatology,25(15),1965–1978.
Hirota,M.,Holmgren,M.,vanNes,E.H.,&Scheer,M.(2011).Globalre-
silience oftropical forest and savanna to crical transions. Science,
334(6053),232–235.
Homann,W.A.,Geiger,E.L.,Gotsch,S.G.,Rossao,D.R.,Silva,L.C.R.,
Lau, O. L., … Lloret, F. (2012). Ecologicalthresholds at the savanna-
forestboundary:Howplanttraits,resourcesandregovernthedistri-
buonoftropicalbiomes.Ecology Leers,15(7),759–768.
Homann,W.A.,Jaconis,S.Y.,McKinley,K.L.,Geiger,E.L.,Gotsch,S.G.,&
Franco,A.C.(2012).Fuelsormicroclimate?Understandingthedrivers
ofrefeedbacksat savanna–forestboundaries.Austral Ecology,37(6),
634–643.
Ibanez,T.,Borgniet,L.,Mangeas,M.,Gaucherel,C.,Geraux,H.,&Hely,C.
(2013).RainforestandsavannalandscapedynamicsinNewCaledonia:
Towards a mosaic of stable rainforest and savanna states? Austral
Ecology,38(1),33–45.
Jenness, J. (2006). Topographic Posion Index (tpi_jen.avx) extension for
ArcView 3.x. v. 1.3a. Jenness Enterprises.Retrieved from: hp://www.
jennessent.com/arcview/tpi.htm.
Kenneally,K. F.,Keighery,G. J.,&Hyland, B. P.M. (1991). Floriscs 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,Australia:SurreyBeayandSons.
Kimber, P. C., Forster, J. E., & Behn, G. A. (1991). Mapping Western
Australianrainforests—Anoverview.InN.L.McKenzie,R.B.Johnston
& P. G. Kendrick (Eds.), Kimberley ranforests of Australia. Chipping
Norton,NSW,Australia:SurreyBeay&Sons.
Lawes, M. J.,Midgley, J. J., & Clarke,P.J. (2013). Costs and benets of
relavebarkthicknessinrelaontoredamage:Asavanna/forestcon-
trast.Journal of Ecology,101(2),517–524.
Leal, I. R., da Silva,J., Cardoso, M., Tabarelli, M., & Lacher,T. E. (2005).
Changing the course of biodiversityconservaon in the Caanga of
northeasternBrazil.Conservaon Biology,19(3),701–706.
Lehmann,C.E.R.,Anderson,T.M.,Sankaran,M.,Higgins,S.I.,Archibald,S.,
Homann,W.A.,…Bond,W.J.(2014).Savannavegetaon-re-climate
relaonshipsdieramongconnents.Science,343(6170),548–552.
Lloyd,J.,Bird,M.I.,Vellen,L.,Miranda,A.C.,Veenendaal,E.M.,Djagbletey,
G.,…Farquhar,G.D.(2008).Contribuonsofwoodyand herbaceous
vegetaontotropicalsavannaecosystemproducvity:Aquasi-global
esmate.Tree Physiology,28(3),451–468.
Lyons, M., Keighery, G., Gibson, L., & Handasyde, T. (2014). Flora and
vegetaon communies of selected islands othe Kimberley coast
of Western Australia. Records of the Western Australian Museum,
Supplement,81,205–243.
MacDermo,H.J.,Fensham,R.J.,Hua, Q.,&Bowman,D.M.J.S.(2016).
Vegetaon,re and soil feedbacks of dynamic boundaries between
rainforest, savanna and grassland. Austral Ecology, doi:10.1111/
aec.12415.
Mangglamarra, G., Burbidge, A. A., & Fuller, P. J. (1991). Wunambal
wordsforrainforestandotherKimberleyplants and animals.InN. L.
McKenzie,R. B.Johnston &P.G.Kendrick (Eds.),Kimberley rainforest
Australia(pp.413–421).Canberra,Australia:SurreyBeayandSons.
McKenzie, N. L., Belbin, L., Keighery, G. J., & Kenneally, K. F. (1991).
Kimberley rainforest communies: Paerns of species composion
and Holocene biogeography.In N. L. McKenzie, R. B. Johnston& P.
G. Kendrick (Eds.), Kimberley rainforests of Australia (pp. 423–452).
ChippingNorton,NSW,Australia:SurreyBeay&Sons.
Murphy,B.P.,&Bowman,D.M.J.S.(2012).Whatcontrolsthedistribuon
oftropicalforestandsavanna?Ecology Leers,15(7),748–758.
Murphy,B.P.,Paron,P.,Prior,L.D.,Boggs,G.S.,Franklin,D.C.,&Bowman,
D.M.J.S.(2010).Usinggeneralizedautoregressiveerrormodelstoun-
derstandre-vegetaon-soilfeedbacksin amulga-spinifexlandscape
mosaic.Journal of Biogeography,37(11),2169–2182.
O’Connor, S. (1995). Carpenter’s Gap rockshelter 1: 40,000 years of
Aboriginaloccupaon inthe NapierRanges,Kimberley,WA.Australia
Archaeology,40,58–59.
Oliveira-Filho,A.T.,& Raer,J.A. (2002).Vegetaonphysiognomies and
woody ora ofthe cerrado biome. In P. S. Oliveira,& R. J. Marquis
(Eds.),The Cerrados of Brazil(pp.91–120). New York,USA:Columbia
UniversityPress.
Ondei, S., Prior,L. D., Vigilante, T.,& Bowman, D. M. J. S. (2015). Post-
reresproungstrategiesofrainforestandsavannasaplingsalongthe
rainforest–savannaboundaryintheAustralianmonsoon tropics.Plant
Ecology,217(6),711–724.
Paiva,A.,Silva,L.,&Haridasan,M.(2015).Producvity-eciencytradeos
in tropical gallery forest-savanna transions: Linking plant and soil
processesthrough lierinput and composion.Plant Ecology, 216(6),
775–787.
Pausas, J. G. (2015). Bark thickness and re regime.Funconal Ecology,
29(3),315–327.
Price, O. F. (2006). Movements offrugivorous birds among fragmented
rainforestsintheNorthernTerritory,Australia.Wildlife Research,33(6),
521–528.
RCoreTeam.(2013). R: A language and environment for stascal comput-
ing[Online].Vienna,Austria: R Foundaonfor StascalCompung.
Retrievedfromhp://www.R-project.org/
RibeiroJr, P.J., & Diggle, P.J. (2001). geoR: A package forgeostascal
analysis.R News,1(2),15–18.
Russell-Smith,J.(1991). Classicaon,speciesrichness, andenvironmen-
tal relaons ofmonsoon rain forest in northernAustralia. Journal of
Vegetaon Science,2(2),259–278.
Russell-Smith, J., & Bowman, D. M. J. S. (1992). Conservaon of mon-
soonrainforest isolatesin the NorthernTerritory,Australia.Biological
Conservaon,59(1),51–63.
Russell-Smith,J.,Stanton, P.J.,Edwards,A.C., &Whitehead,P.J.(2004).
Rain forest invasion ofeucalypt-dominated woodland savanna, Iron
Range,north-easternAustralia:II. Ratesof landscapechange.Journal
of Biogeography,31(8),1305–1316.
Sánchez-Azofeifa,G.A.,Kalacska,M.,Quesada,M.,Calvo-Alvarado,J.C.,
Nassar,J.M.,&Rodríguez,J.P.(2005).Needforintegratedresearchfor
asustainablefuturein tropicaldryforests.Conservaon Biology,19(2),
285–286.
Schmidt,S., &Stewart,G.R.(2003).δ15Nvalues oftropicalsavannaand
monsoon forestspecies reect root specialisaons and soil nitrogen
status.Oecologia,134(4),569–577.
Sco,K., Seereld, S.A., Douglas,M.M., Parr,C. L., Schatz,J. O.N.,&
Andersen,A.N.(2012).Doeslong-termreexclusioninanAustralian
tropicalsavannaresultinabiomeshi?Atestusingthereintroducon
ofre.Austral Ecology,37(6),693–711.
Silva, L. R., Homann, W., Rossao, D., Haridasan, M., Franco, A., &
Horwath,W.(2013).Cansavannasbecomeforests?Acoupledanalysis
ofnutrient stocks and rethresholdsin central Brazil.Plant and Soil,
373(1–2),829–842.
Silva, L. C. R., Sternberg, L., Haridasan, M., Homann,W.A., Miralles-
Wilhelm,F., &Franco,A. C.(2008).Expansionof gallery forestsinto
centralBraziliansavannas.Global Change Biology,14(9),2108–2118.
Smyth, G., Hu, Y., Dunn, P., Phipson, B., & Chen, Y. (2015). Statmod:
Stascal modelling. R package version 1.4.21. Retrievedfrom: hp://
cran.r-project.org/package=statmod.

|
13
ONDEI Et al.
Speck, N., Bradley, J., Lazarides, M., Twidale, C., Slatyer, R., Stewart, G.,
& Paerson, R. (2010).The lands and pastoral resources of the North
Kimberleyarea,W.A.No.4.CSIRO Land Research Surveys,2010(1),1–116.
Staver,A.C.,Archibald, S.,&Levin,S. (2011a).Treecoverin sub-Saharan
Africa:Rainfallandreconstrainforestandsavannaasalternavesta-
blestates.Ecology,92(5),1063–1072.
Staver,A.C.,Archibald,S.,&Levin,S.A.(2011b).Theglobalextentandde-
terminantsofsavanna andforestas alternavebiomestates.Science,
334(6053),230–232.
Staver,A. C., & Hansen, M. C. (2015).Analysis of stable states in global
savannas:IstheCARTpulling thehorse?—Acomment.Global Ecology
and Biogeography,24(8),985–987.
Thornthwaite,C.W.(1948).Anapproachtowardaraonalclassicaonof
climate.Geographical Review,38(1),55–94.
Trauernicht,C.,Brook, B.W.,Murphy,B.P.,Williamson,G.J.,&Bowman,
D.M. J.S.(2015). Local andglobal pyrogeographic evidencethat in-
digenousremanagementcreatespyrodiversity.Ecology and Evoluon,
5(9),1908–1918.
Trauernicht,C.,Murphy,B.P.,Portner,T.E.,&Bowman,D.M.J.S.(2012).
Treecover–reinteraconspromotethepersistenceofare-sensive
conifer in a highly ammable savanna. Journal of Ecology, 100(4),
958–968.
Tun,C.E.G.,White,L.J.T.,&Mackanga-Missandzou,A.(1997).Theuse
by rain forestmammals of natural forest fragments in an equatorial
Africansavanna.Conservaon Biology,11(5),1190–1203.
Veenendaal, E., Torello-Raventos, M., Feldpausch, T., Domingues, T.,
Gerard,F.,Schrodt,F.,…Ford,A.(2015).Structural,physiognomicand
above-groundbiomassvariaoninsavanna–foresttransionzoneson
three connents–howdierent are co-occurring savanna and forest
formaons?Biogeosciences,12(10),2927–2951.
Vigilante,T.,Bowman, D. M.J. S., Fisher,R.,Russel-Smith, J.,& Yates,C.
(2004). Contemporary landscape burning paerns in the far North
Kimberleyregionofnorth-westAustralia:Humaninuencesandenvi-
ronmentaldeterminants.Journal of Biogeography,31,1317–1333.
Vigilante,T.,Murphy, B.P.,& Bowman, D.M. J.S.(2009). Aboriginalre
useinAustraliantropicalsavanna:Ecologicaleectsandmanagement
lessons. In M.A. Cochrane (Ed.), Tropical re ecology: Climate change,
land use and ecosystem dynamics(pp.143–167).Heidelberg,Germany:
Springer-Praxis.
Vigilante,T.,Toohey,J.,Gorring,A.,Blundell,V.,Saunders,T.,Mangolamara,
S.,…Morgan,K.(2013).Islandcountry:Aboriginalconnecons,values
and knowledge of the Western Australian Kimberley islands in the
contextofanislandbiologicalsurvey.Records of the Western Australian
Museum, Supplement,81,145–181.
Vourlis,G.,deAlmeidaLobo,F.,PintoJr,O.,Zappia,A.,Dalmagro,H., de
Arruda,P.,&de SouzaNogueira,J.(2015).Variaonsin aboveground
vegetaon structure along a nutrient availability gradient in the
Brazilianpantanal.Plant and Soil,389(1–2),307–321.
Warman,L., Bradford,M. G., &Moles,A.T.(2013).A broadapproach to
abrupt boundaries: Looking beyond the boundary at soil aributes
withinandacrosstropicalvegetaontypes.PLoS One,8(4),e60789.
Warman,L., & Moles,A.T.(2009).Alternave stable statesinAustralia’s
WetTropics:Atheorecalframeworkfortheelddataandaeld-case
forthetheory.Landscape Ecology,24(1),1–13.
Webb,L.J. (1968).Environmentalrelaonshipsof thestructural typesof
Australianrainforestvegetaon.Ecology,49(2),296–311.
Webb,L.,& Tracey,J.(1981).Australian rainforests:Paernsand change.
In A. Keast (Ed.), Ecological biogeography of Australia (pp. 605–694).
Boston,MAandLondon,UK:TheHagueandDr.W.JunkbvPublishers.
Xu,C.,Hantson,S.,Holmgren,M.,van Nes,E.H., Staal,A.,&Scheer,M.
(2016).Remotelysensedcanopyheightrevealsthreepantropicaleco-
systemstates.Ecology,97(9),2518–2521.
Zomer,R. J., Trabucco,A., Bossio, D. A., &Verchot,L. V. (2008). Climate
changemigaon:Aspaalanalysisofgloballandsuitabilityforclean
development mechanism aorestaonand reforestaon. Agriculture,
Ecosystems & Environment,126(1),67–80.
SUPPORTING INFORMATION
AddionalSupporngInformaonmaybefoundonlineinthesupport-
inginformaontabforthisarcle.
How to cite this arcle:OndeiS,PriorLD,WilliamsonGJ,
VigilanteT,andBowmanDMJS.Water,land,re,andforest:
Mul-scaledeterminantsofrainforestsintheAustralian
monsoontropics.Ecol Evol.2017;00:1–13.doi:10.1002/
ece3.2734.