A model inter-comparison study to examine limiting factors in …papers.agrivy.com/webfiles/papers/2016-BIOGEOSCIENCES... · 2019-03-14 · Africa (Koch et al., 1995). Soils range

Biogeosciences 13 3245ndash3265 2016wwwbiogeosciencesnet1332452016doi105194bg-13-3245-2016copy Author(s) 2016 CC Attribution 30 License

A model inter-comparison study to examine limiting factors inmodelling Australian tropical savannasRhys Whitley1 Jason Beringer2 Lindsay B Hutley3 Gab Abramowitz4 Martin G De Kauwe1 Remko Duursma5Bradley Evans6 Vanessa Haverd7 Longhui Li8 Youngryel Ryu9 Benjamin Smith10 Ying-Ping Wang11Mathew Williams12 and Qiang Yu7

1Department of Biological Sciences Macquarie University North Ryde NSW 2109 Australia2School of Earth and Environment University of Western Australia Crawley WA 6009 Australia3School of Environment Charles Darwin University Casuarina NT 0810 Australia4Climate Change Research Centre University of New South Wales Kensington NSW 2033 Australia5Hawkesbury Institute for the Environment University of Western Sydney Penrith New South Wales 2751 Australia6Faculty of Agriculture and Environment University of Sydney Eveleigh NSW 2015 Australia7CSIRO Ocean and Atmosphere Canberra 2601 Australia8School of Life Sciences University of Technology Sydney Ultimo NSW 2007 Australia9Department of Landscape Architecture and Rural Systems Engineering Seoul National University Seoul South Korea10Department of Physical Geography and Ecosystem Science Lund University Lund Sweden11CSIRO Ocean and Atmosphere Aspendale Victoria 3195 Australia12School of GeoSciences University of Edinburgh Edinburgh UK

Correspondence to Rhys Whitley (rhyswhitleymqeduau)

Received 17 November 2015 ndash Published in Biogeosciences Discuss 2 December 2015Revised 16 May 2016 ndash Accepted 17 May 2016 ndash Published 3 June 2016

Abstract The savanna ecosystem is one of the most dom-inant and complex terrestrial biomes deriving from a dis-tinct vegetative surface comprised of co-dominant tree andgrass populations While these two vegetation types co-existfunctionally demographically they are not static but are dy-namically changing in response to environmental forces suchas annual fire events and rainfall variability Modelling sa-vanna environments with the current generation of terrestrialbiosphere models (TBMs) has presented many problemsparticularly describing fire frequency and intensity phenol-ogy leaf biochemistry of C3 and C4 photosynthesis vege-tation and root-water uptake In order to better understandwhy TBMs perform so poorly in savannas we conducteda model inter-comparison of six TBMs and assessed theirperformance at simulating latent energy (LE) and gross pri-mary productivity (GPP) for five savanna sites along a rain-fall gradient in northern Australia Performance in predictingLE and GPP was measured using an empirical benchmarkingsystem which ranks models by their ability to utilise meteo-rological driving information to predict the fluxes On aver-

age the TBMs performed as well as a multi-linear regressionof the fluxes against solar radiation temperature and vapourpressure deficit but were outperformed by a more compli-cated nonlinear response model that also included the leafarea index (LAI) This identified that the TBMs are not fullyutilising their input information effectively in determiningsavanna LE and GPP and highlights that savanna dynam-ics cannot be calibrated into models and that there are prob-lems in underlying model processes We identified key weak-nesses in a modelrsquos ability to simulate savanna fluxes andtheir seasonal variation related to the representation of veg-etation by the models and root-water uptake We underlinethese weaknesses in terms of three critical areas for devel-opment First prescribed tree-rooting depths must be deepenough enabling the extraction of deep soil-water stores tomaintain photosynthesis and transpiration during the dry sea-son Second models must treat grasses as a co-dominantinterface for water and carbon exchange rather than a sec-ondary one to trees Third models need a dynamic repre-sentation of LAI that encompasses the dynamic phenology

Published by Copernicus Publications on behalf of the European Geosciences Union

3246 R Whitley et al Limiting factors in modelling Australian tropical savannas

of savanna vegetation and its response to rainfall interannualvariability We believe that this study is the first to assesshow well TBMs simulate savanna ecosystems and that theseresults will be used to improve the representation of savannasecosystems in future global climate model studies

1 Introduction

Savanna ecosystems are a diverse and important biome thatplay a significant role in global land-surface processes (vander Werf et al 2008) Globally they occupy regions aroundthe wetndashdry tropical to sub-tropical equatorial zone cover-ing approximately 15 to 20 of the terrestrial surface andcontributesim 30 to global net primary production (Grace etal 2006 Lehmann et al 2014) Savannas are water-limitedecosystems where rainfall is often seasonal or monsoonaland have a spatial extent that can cover an area with annualrainfall in the range of 500 to 2000 mm (Bond 2008 Kan-niah et al 2010 Sankaran et al 2005) The variability inthe amount and timing of annual rainfall coupled with lo-cal topo-edaphic properties and the frequency and intensityof seasonal fires strongly influence the structure and func-tion of savanna vegetation (Beringer et al 2007 Kanniahet al 2010 Ma et al 2013 Sankaran et al 2005) Sa-vannas are characterised by a multi-layer stratum of vege-tation where an open and discontinuous canopy overstoreyis seasonally dominated by understorey grasses (Scholes andArcher 1997) These tree and grass layers are distinctly andfunctionally different fixing carbon using different photo-synthetic pathways C3 and C4 photosynthesis respectively(Bond 2008 Scholes and Archer 1997 R J Williams etal 1996) The canopy overstorey can be either evergreenor deciduous (depending on the evolutionary history) whilethe grass understorey is annual active only in the wet sea-son and senescing at the end of this period (R J Williamset al 1996) Consequently water carbon and nutrient cy-cling in savannas is largely determined from the balance andco-existence of these two life forms (Lehmann et al 2009Sankaran et al 2005)

Given the complex nature of savannas modelling the landsurface exchange and vegetation dynamics for this biome ischallenging for terrestrial biosphere models (TBMs) Herewe define TBMs to broadly encompass stand land surfaceand dynamic global vegetation models (Pitman 2003) Mostland surface schemes that feed into larger earth system mod-els use simplistic representations of vegetation and these willhave difficulty describing the complex structure of savannaecosystems Such issues may be simplistic assumptions in re-lation to rooting depth and inadequate responses to drought(De Kauwe et al 2015 Li et al 2012) ignoring the mul-tilayered nature of savannas and the differing structural (in-cluding radiation) functional (including different plant func-tional types) and phenological differences (Whitley et al

2011) and in some cases neglecting the C4 photosyntheticpathway entirely (Parton et al 1983 Schymanski et al2007) It is therefore critical that TBMs meet the challengesthat savanna dynamics present if water and carbon exchangeare to be correctly simulated in response to global change

Despite these issues there have been significant advancesin modelling savanna dynamics in recent years and thesehave been focused on integrating important features specificto savanna ecosystems namely frequent fire and treendashgrasscompetitive interactions which are processes that shape sa-vanna structure and function (Haverd et al 2016 Higginsand Scheiter 2012 Scheiter and Higgins 2007 Scheiter etal 2014 Simioni et al 2003) Nevertheless little work hasbeen undertaken to critically evaluate the performance andprocesses of TBMs when used to capture water and carboncycling in savannas most notably in West Africa (Simioniet al 2000) and Australia (Schymanski et al 2007 20082009 Whitley et al 2011) Many global ecosystem mod-els moreover use broad plant functional types (PFTs) withsingle parameter values to describe whole biomes (Pitman2003) making them unable to represent changing vegeta-tion structure (tree grass ratio) in the continuum of grass-land to woodland savanna Approaches have been developedthat can account for savanna dynamics such as using mixedtiles whereby trees and grasses are simulated as separate sur-faces that are then aggregated together (Kowalczyk et al2006) However this approach fails to capture the compe-tition between trees and grasses for light water and nutrientresources

In this study we take six TBMs of distinctly different con-ceptual frameworks and assess their ability to simulate sa-vanna water and carbon exchange along the North AustralianTropical Transect (NATT) which is defined by a strong rain-fall gradient Australian tropical savannas can be consideredlargely intact compared to South American and African sa-vannas and provide a ldquoliving laboratoryrdquo to understand thelinks between vegetation structure and function and how itresponds to environmental change (Hutley et al 2011) Wechallenge the models by evaluating them along the rainfallgradient which extends over a broad biogeographical ex-tent and strong interannual variability in climate (Koch et al1995) The aim of this study is to highlight critical processesthat may be missing in current TBMs and are required to ade-quately simulate savanna ecosystems Specifically we exam-ine whether a TBMrsquos structural framework such as the rep-resentation of the understorey grasses (C4 photosynthesis)tree rooting depth and description of phenology (prescribedvs dynamic) can adequately replicate observed carbon andwater fluxes To achieve this we measure the performance ofeach TBM by comparing its predictions to a set of empiricalbenchmarks that describe a priori expected levels of modelperformance We identify regions of low performance amongsites and seasons to diagnose under what climate conditionsreduced model performance occurs We then infer what pro-cesses (present or missing) may be the cause for reduced per-

Biogeosciences 13 3245ndash3265 2016 wwwbiogeosciencesnet1332452016

R Whitley et al Limiting factors in modelling Australian tropical savannas 3247

formance when applied to savanna ecosystems Our intentionis that these results can be used to flag high priorities for fu-ture development by the terrestrial biosphere modelling com-munity

2 Methodology

21 Observational data

The NATT is a sub-continental rainfall gradient in the wetndashdry tropical climate zone of northern Australia covering adistance of approximately 1000 km over a latitudinal rangeof minus12 to minus23 S and a decline in mean annual precipitation(MAP) from 1700 to 300 mm (Hutley et al 2011) It is one ofthree savanna transects established in the mid-1990s form-ing part of the International Geosphere Biosphere Program(IGBP) along with the SAvannas in the Long Term (SALT)transect in West Africa and the Kalahari Transect in southernAfrica (Koch et al 1995) Soils range from sand-dominatedred kandosols to black cracking clay soils that are more ex-tensive in the southern end of the NATT that are limiting towoody plant growth (Hutley et al 2011 R J Williams et al1996) Kandosols are ancient and weathered such that theyhave been leached of nutrients by the large monsoonal rain-fall (McKenzie et al 2004) Close to the northern coastlinevegetation is comprised primarily of evergreen Eucalyptusand Corymbia tree species that overly an understorey of C4Sorghum and Heteropogon spp grasses Inland tree biomassleaf area index (LAI) and cover tends to decline and byminus18 S savanna vegetation transitions to less dense Acaciawoodlands shrublands and grasslands that are dominated byAstrebla grass species (Hutley et al 2011) Fires occur reg-ularly in these environments increasing in frequency withhigher rainfall (MAP gt 1000 mm) and are fuelled by the ac-cumulation of understorey C4 grasses that cure in the dryseason (Beringer et al 2015 Russell-Smith and Edwards2006)

The five flux tower sites along the NATT used in thisstudy are outlined in Table 1 which describes stand soiland vegetation characteristics as well as a summary of lo-cal meteorology (Hutley et al 2011) These sites representa sampling of savanna environments covering a wide rangeof MAP and a much smaller range of mean annual temper-ature (Fig 1) At each site an eddy covariance system wasused to measure the ecosystemndashatmosphere exchange of ra-diation heat water and CO2 Quality assurance and controland corrections on the fluxes were carried out on the 30 mindata set using the OzFlux QCQA protocol (v285) devel-oped by the OzFlux community under creative commons li-censing (wwwozfluxorgau see Eamus et al 2013) Miss-ing or rejected data were gap-filled using the DINGO (Dy-namic INtegrated Gap filling and partitioning for Ozflux)system (see Moore et al 2016) Gross primary productiv-ity (GPP) was not observed but determined from the differ-

ence between measured net ecosystem exchange (NEE) andmodelled ecosystem respiration (Re) Values of Re were de-termined by assuming nocturnal NEE equals Re under theconditions for sufficient turbulent transport Values that meetthese requirements are then used to make daytime predic-tions of Re using an artificial neural network (ANN) withsoil moisture and temperature air temperature and the nor-malised difference vegetation index used as predictors Ad-ditionally the effect of fire on the water and carbon fluxes arequantified and incorporated into the data sets accounting forthe nonlinear response in productivity (becoming a carbonsource) during the post-fire recovery period (Beringer et al2007) Because the TBMs used here do not attempt to sim-ulate stochastic fire events (and other disturbance regimes)these post-fire recovery periods were removed when deter-mining the benchmarks and model performance as describedbelow

Finally we use the definitions for water and carbon ex-change as outlined by Chapin et al (2006) whereby the sub-daily rate of GPP is expressed in micromol mminus2 sminus1 and uses anegative sign (minus) to denote the removal of CO2 from the at-mosphere Similarly latent energy (LE) is expressed in termsof energy as W mminus2 and uses a positive sign to denote the ad-dition of H2O to the atmosphere

22 Terrestrial biosphere models

The six TBMs used in this study cover a wide spectrum ofcharacteristics of operation scale and function and includedifferences in operational time step (30 min vs daily) scopeof simulated processes (soil hydrology static or dynamicvegetation multi-layer or big leaf description of the canopy)and intended operational use (coupled to earth system mod-els offline prediction driven by remote sensing products)These characteristics along with what we define as a modelldquofunctional classrdquo are given in Table 2 and are defined asfollows Stand models (SMs) give detailed multi-layer de-scriptions of canopy and soil processes for a particular pointoperating at a sub-daily time step (SoilndashPlantndashAtmospheremodel (SPA) and MAESPA) Land-surface models operate atthe same temporal resolution as SMs but they adopt a sim-pler representation of canopy processes allowing them to beapplied spatially (Community Atmosphere Biosphere LandExchange model (CABLE) and BIOS2 a modified version ofCABLE) Dynamic global vegetation models (DGVMs) sim-ulate water and carbon much like the other models but theysimulate dynamic rather than static vegetation that changesin response to climate and disturbance (LundndashPotsdamndashJenaGeneral Ecosystem Simulator LPJGUESS) Lastly remotesensing models are driven by remotely sensed atmosphericproducts and infer water stress of vegetation through changesin fractional cover rather than detailed soil hydrological pro-cesses (Breathing Earth System Simulator BESS) Some ofthe TBMs share similar structural frameworks in parts forexample both SPA and MAESPA use similar belowground

wwwbiogeosciencesnet1332452016 Biogeosciences 13 3245ndash3265 2016

128deg E 130deg E 132deg E 134deg E 136deg E 138deg E

NATT sites

Adelaide RiverDaly UnclearedDry RiverHoward SpringsSturt Plains

20 22 24 26 28 30

Mean annual temperature (degC)

0 500 1000 1500 2000

Total annual precipitation (mm)

Figure 1 The Northern Territory of Australia and the North Australian Tropical Transect (NATT) showing (a) the flux site locations withan accompanying 30-year (1970 to 2000) expression of the average meteorological conditions for (b) mean annual temperature and (c) totalannual precipitation derived from ANUCLIM v61 climate surfaces (Hutchinson and Xu 2010)

soil hydrology and root-water uptake schemes while BIOS2is a fine spatial resolution (005) offline modelling environ-ment for Australia in which predictions of CABLE (with al-ternate parameterisations of drought response and soil hy-drology) are constrained by multiple observation types (seeHaverd et al 2013) Although these similarities reduce thenumber of truly functionally independent models used in theexperiment the presence of such overlap can be useful inidentifying whether particular frameworks are the cause formodel success or failure

23 Experimental protocol

All TBMs were parameterised for each of the five savannasites using standardised information on vegetation and soilprofile characteristics (Table 1) For TBMs that requiredthem parameter values pertaining to leaf biochemistry suchas maximum Rubisco activity (Vcmax) and leaf nitrogen con-tent per leaf area (Narea) were assigned from Cernusak etal (2011) who undertook a physiological measurement cam-paign during the SPECIAL program (Beringer et al 2011)Parameters relating to soil sand and clay content were takenfrom the Australian Soil Classification (Isbell 2002) whileroot profile information was sourced from Chen et al (2003)and Eamus et al (2002) Each TBM was set up to describe aC3 evergreen overstorey with an underlying C4 grass under-storey and conforms well with the characteristics of savannasin northern Australia (Bowman and Prior 2005) All TBMs(excluding LPJGUESS) prescribed LAI as an input in orderto characterise the phenology of vegetation at each site In

these cases LAI was determined from MODIS-derived ap-proximations that were well matched to ground-based es-timations of LAI at the SPECIAL sites (Sea et al 2011)The fraction of C3 to C4 vegetation was handled differentlyby each model and was determined for each as follows ForMAESPA and SPA the models allowed for time-varying treeand grass fractions to be assigned as direct inputs and thesetime-varying fractions were determined using the methodof Donohue et al (2009) BIOS2 similarly used the samemethod to extract time-varying fractions while CABLE useda static fraction that did not change The BESS model de-rived the C3 C4 fraction from the C3 and C4 distributionmap of Still et al (2003) while for LPJGUESS this fractionis a prognostic determination resulting from the competitionbetween trees and grasses (see Smith et al 2001) Modelsimulations were driven using observations of solar radia-tion air temperature relative humidity (or vapour pressuredeficit VPD) rainfall atmospheric CO2 concentration andLAI (if prescribed) and they included a spin-up period of 5years to allow internal states such as the soil-water balanceand soil temperature to reach equilibrium The exception tothe above was the BIOS2 model which was run using grid-ded meteorological inputs and had its model parameters op-timised through a modelndashdata fusion process (see Haverd etal 2013)

Simulations for each savanna site covered a period of2 to 10 years depending on the availability of data fromeach flux site (Table 1) and results were standardised tothe ALMA (Assistance for Land-surface Modelling Activ-ities) convention Model predictions of LE and GPP were

Table 1 Summarised data set information for each of the five savanna sites used in this study This includes site descriptions pertaining to localmeteorology vegetation and belowground soil characteristics Where data were not available the abbreviation na is used Definitions forthe species genus mentioned in the table are as follows Eucalyptus (Eu) Erythrophleum (Er) Terminalia (Te) Corymbia (Co) Planchonia(Pl) Buchanania (Bu) Themda (Th) Hetropogan (He) and Chrysopogon (Ch) Eddy covariance data sets relating to each of the five siteshere can be download from wwwozfluxorgau

Howard Springsa Adelaide Riverb Daly Unclearedc Dry Riverd Sturt Plainse

Years (inclusive) 2001ndash2012 2007ndash2009 2008ndash2012 2008ndash2012 2008ndash2012

Co-ordinates 1229prime3912primeprime S 1304prime3684primeprime S 1409prime3312primeprime S 1515prime3162primeprime S 1709prime0276primeprime S13109prime09primeprime E 13107prime0408primeprime E 13123prime1716primeprime E 13222prime1404primeprime E 13321prime0114primeprime E

Elevation (m) 64 90 110 175 250

Meteorologyf

Annual rainfall (mm) 1714 1460 1170 850 535Minmax daily temperature (C) 220330 218353 208350 200348 190342Minmax absolute humidity (g mminus3) 110185 89177 86151 78123 6190Minmax soil moisture (m3 mminus3) 00601 009014 003006 003005 00401Soil temperature (C) 327 357 328 na 302Solar radiation (W mminus2) 2565 2581 2706 2665 2697Bowen ratio 17 31 32 46 158

Vegetationf

Overstorey species Eu Miniata Eu tectifica Te grandiflora Eu tetrodonta naEu tetrodonta Pl careya Eu tetrodonta Co terminalisEr chlorostachys Co latifolia Co latifolia Eu dichromophloia

Understorey species Sorghum spp Sorghum spp Sorghum spp Sorghum intransHe triticeus Ch fallax He triticeus Th Tiandra Astrabla spp

Ch fallaxBasal area (m2 haminus1) 97 51 83 54 naCanopy height (m) 189 125 164 123 02LAI (m2 mminus2) 104plusmn 007 068plusmn 007 080plusmn 012 058plusmn 011 039plusmn 011Total leaf nitrogen (g mminus3) 142plusmn 020 127plusmn 018 135plusmn 019 197plusmn 015 237plusmn 017

Type Red kandosol Yellow hydrosol Red kandosol Red kandosol Grey vertosol

A horizon Texture Sandy loam Sandy loam Loam Clay Loam

Clay PSD () 15 20 20 50 20Sand PSD () 60 50 40 25 40Thickness (m) 030 030 020 015 020Bulk density (Mg mminus3) 129 160 139 120 139Hydraulic conductivity (mm hminus1) 9 7 9 3 9Field capacity (mm mminus1) 156 132 147 140 147

B horizon Texture Clay loam Clay Clay loam Clay Clay loam

Hdl references are given by order of column Jason Beringer (2013) a hdl 10210010014228 b hdl 10210010014239 c hdl 10210010014229 d hdl 10210010014234 e hdl 10210010014230Site meteorology is given as 30-year averages with values taken from f Hutley et al (2011) Soil descriptions are taken from the Digital Atlas of Australian Soils (wwwasriscsiroau) g Isbell (2002)

evaluated against local observations at each site from theeddy covariance data sets and benchmarked following themethodology proposed by the Protocol for the Analysisof Land-surface models (PALS) and the PALS Land SUr-face Model Benchmarking Evaluation PRoject (PLUMBER)(Abramowitz 2012 Best et al 2015) as described below

24 Empirical benchmarking

The paradigm for model assessment outlined by PALS(Abramowitz 2012) suggests that model assessment is moremeaningful when a priori expectations of performance in anygiven metric can be defined Such benchmarks can be createdusing simple empirical models built on statistical relation-ships between the fluxes and drivers and establish the de-

Table 2 Summary table of the ecosystem models used in the experiment highlighting differences and similarities in model structure andshared processes Information is broken down into how each model describes aboveground canopy and belowground soil processes

Model name SPA MAESPA CABLE BIOS2 BESS LPJGUESS

Model definition SoilndashPlantndashAtmosphere model MAESTRA-SPA Community AtmosphereBiosphereLand-surface ExchangeModel

Modified CABLE (CABLE + SLI+ CASA-CNP)

Breathing Earth System Simu-lator

LundndashPotsdamndashJena GeneralEcosystem Simulator

Version 10 10 20 20 10 21Reference M Williams et al (1996) Duursma and Medlyn (2012) Kowalyzck et al (2006)

Wang et al (2011)Haverd et al (2013) Ryu et al (2011 2012) Smith et al (2001)

Temporal resolution 30 min 30 min 30 min Daily (30 min time steps are gener-ated from daily time series)

Snapshot with MODIS over-pass then up-scaled to a dailyand 8-day time series

Spatial resolution Point Point 005 (5 km) 005 (5 km) 005 (5 km) Patch (c 01 ha)Functional class Stand model Individual plant or stand model Land-surface model Land-surface model Remote sensing model Dynamic global vegetation model

Canopy description

C3 assimilation Farquhar et al (1980) Farquhar et al (1980) Farquhar et al (1980) Farquhar et al (1980) Farquhar et al (1980) Collatz et al (1991)C4 assimilation Collatz et al (1992) Collatz et al (1992) Collatz et al (1992) Collatz et al (1992) Collatz et al (1992) Collatz et al (1992)Stomatal conductance M Williams et al (1996) Medlyn et al (2011) Leuning (1995) Leuning (1995) Ball et al (1987) Haxeltine and Prentice (1996)Transpiration Penman-Monteith calculated at leaf

scale accounting for gb and limita-tion of soil-water supply via 9l

Penman-Monteithcalculated at the leaf scale

Penman-Monteith Penman-Monteith Penman-Monteith Haxeltine and Prentice (1996)

Boundary layer resistance f (wind speed leaf width air tem-perature)

f (wind speed leaf width air temperatureand atmospheric pressure)

f (wind speed leaf width airtemperature

f (wind speed leaf width air tem-perature

Not modelled Huntingford and Monteith (1998)

Aerodynamic resistance f (wind speed canopy height) Not calculated unless transpiration is cal-culated at the canopy scale in which casegb above is not calculated

f (wind speed canopy height) f (wind speed canopy height) f (wind speed canopy height) Huntingford and Monteith (1998)

Leaf area index Prescribed (MODIS) Prescribed (MODIS) Prescribed (MODIS) Prescribed (MODIS) Prescribed (MODIS) Prognostic(C allocation)

Canopy structure Canopy + understoreydivided into 10 layers

Individual plant crownsspatially explicit locations and uniformunderstorey

Two (treegrass) big leaf (sun-litshaded)

5-year agesize cohorts for treessingle-layer grass understorey

C3 C4 fraction Dynamic ratio variable with timecompete for water and light

Dynamic ratio variable with time com-pete for water and light

Simulated as independent lay-ers

Dynamic ratio variable with timecompete for water not light

Still et al (2003)Ratio changes 7030 to 1090down transect

Prognostic determined as the out-come of the competition with trees

Canopy interception Yes Yes Yes Yes No YesSimulates growth No No No No No Yes

Soil profile description

Soil profile structure Profile divided into n

layers (prescribed ndash 20 in this case)Profile divided into n

layers (prescribed ndash 20 in this case)Profile divided into 6 layers Profile divided into 12 layers (ad-

justable)Not modelled Two layers (0ndash05 05ndash2 m) with

10 cm evaporation sub-layerSoil hydraulic properties Function of sand and clay particle

size distributionsFunction of sand and clay particle sizedistributions

Prescribed Australian Soils Resource Informa-tion System (ASRIS)

Not modelled Sitch et al (2003)

Soil depth 65 m 50 m 45 m 100 m Not modelled 2 mRoot depth 65 m 50 m 45 m 05 m (grasses) 50 m (trees) Not modelled 2 mRoot distribution Prescribed exponential decay as a

function of surface biomass and thetotal root biomass of the column

Prescribed exponential decay as a func-tion of surface biomass and the total rootbiomass of the column

Prescribed exponential decay Prescribed exponential decay Not modelled PFT-specific trees have deeperroots on average

Soil-water stress modifier Et via gs is increased to meet atmo-spheric demand while 9l remainsabove a critical threshold

Maximum transpiration rate calculatedfrom hydraulic conductance (soil-to-leaf)sets limit on actual transpirationor uses the Tuzet et al (2003) model ofstomatal conductance

Supplydemand gs scaled by a soil moisture limita-tion function related to extractablewater accessible by roots

Assumes LAI and seasonalvariation of Vcmax reflect soil-water stress

Supplydemand

Hydraulic pathway resistance Rsoil+Rplant Rsoil+Rplant Not modelled Not modelled Not modelled Not explicit min(supply demand)determines sap flow

gree to which models utilise the information available in theirdriving data about the fluxes they aim to predict Addition-ally these empirical models are simple in the sense that theyare purely instantaneous response to time-varying meteoro-logical forcing and contain no internal states or expression ofecophysiological processes This is in comparison to TBMsthat are complex having some 20+ soil and vegetation pa-rameters internal states partitioning of light as well as soiland vegetation carbon and nitrogen pools (Abramowitz etal 2008)

We created a set of three empirical models of increasingcomplexity following the procedure of Abramowitz (2012)which we compared with the TBMs The first benchmark(emp1) is simply a linear relationship between a turbulentflux (LE or GPP) and downward short-wave radiation (Rs)The second benchmark (emp2) is slightly more complex andis a multi-linear regression between a flux and Rs air temper-ature (Ta) and VPD Finally the third benchmark (emp3) isthe most complex and is a nonlinear regression of the fluxesagainst Rs Ta VPD and LAI determined from an ANNThis benchmark is constructed using a self-organising lin-ear output map that clusters the four covariates into 102 dis-tinct nodes and performs a multi-linear regression between

the fluxes and the four covariates at each node resulting ina nonlinear (piece-wise linear) response to the meteorolog-ical forcing data (Abramowitz et al 2008 Hsu 2002) Ina departure from Abramowitz (2012) we include LAI as anadditional covariate as the seasonal variance of savanna wa-ter and carbon exchange is strongly coupled to the phenol-ogy of the grasses and to the deciduous and semi-deciduouswoody species (Moore et al 2016) The seasonal behaviourof the empirical benchmark drivers along the NATT can bereferred to in the Supplement Empirical benchmarks are cre-ated for each of the five flux sites using non gap-filled dataand are parameterised out of sample such that they use datafrom all sites except the one in question For example theHoward Springs empirical benchmark models would use in-formation from Adelaide River Daly Uncleared Dry Riverand Sturt Plains to establish their parameter values but wouldexclude Howard Springs itself Constructing the benchmarkout of sample results in what is effectively a generalised re-sponse to an independent data set Once the empirical modelswere calibrated for each site benchmarks were then createdfor both fluxes using the same meteorological forcing usedto run the TBMs

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

Figure 2 Time series of daily mean latent heat (LE) flux and gross primary productivity (GPP) depicting an average year for each of thefive savanna sites using a smoothed 7-day moving average The sites are ordered from wettest to driest (a) Howard Springs (b) AdelaideRiver (c) Daly River (d) Dry River and (e) Sturt Plains The joined black dots are the tower flux time series while the grey lines are theperformance benchmarks (emp1 emp2 emp3) Predictions of LE and GPP for each of the six terrestrial biosphere models are given by aspectrum of colours described in the legend

Finally we assess ecosystem model performance in termsof a ranking system following the PLUMBER methodologyof Best et al (2015) The performance of each individualecosystem model in predicting both LE and GPP at each sitewas determined using four statistical metrics that describethe mean and variability of a model compared to the observa-tions These metrics included the correlation coefficient (r)standard deviation normalised mean error and mean bias er-ror (see Table A1 in Appendix A) Similarly the same met-rics were determined for each of the three benchmarks ateach savanna site Each TBM was then ranked against the

benchmarks (independently of the other models) for each ofthe metrics listed above where the ranking is between 1 and4 (1 model + 3 benchmarks) and the best performing modelfor a given metric is ranked as 1 An average ranking is thendetermined across all metrics for each TBM and all bench-marks to give a final ranking of performance for each savannasite The ranks denote the number of metrics being met by themodels and are not a measure of the smallest absolute errorIn determining the average ranks the metrics were evaluatedat the daily timescale as this was the lowest temporal res-olution common amongst the six TBMs Additionally days

where either driver or flux had been gap-filled were removedHere we use the term performance to relate to how well theTBMs compare to the benchmarks as expressed by the ranks

3 Results

31 Model predictions

Figure 2 shows the daily time course of LE and GPP fromthe flux tower models and benchmarks at each of the fivesavanna sites Models benchmarks and observations are rep-resented as a smoothed time series (7-day running mean)and have been aggregated into an ensemble year to ex-press the typical seasonality of savanna water and carbonexchange Visually the TBMs showed varying levels of per-formance across the rainfall gradient None of the modelsshowed a clear consistency in simulating either flux and eachresponded differently to the meteorological drivers acrosssites Additionally some of the models such as CABLE andLPJGUESS showed difficulty in simulating the seasonalityof the fluxes across the transect particularly GPP Differ-ences among model-simulated LE and GPP were larger in thewet season than the dry season However modelled LE andGPP appeared to co-vary quite strongly overall both fluxeswere underestimated across sites by most models Simu-lations by SPA and MAESPA were the exception to thisbroadly capturing tower GPP despite consistently underes-timating LE across sites

Figure 3 shows the probability density functions (PDFs)for the wet (NovemberndashApril) and dry season (MayndashOctober) fluxes at each site Tower and model PDFs weredetermined by binning each flux into the respective seasonsand using kernel density estimation (Bashtannyk and Hynd-man 2001) to determine smoothed distributions The shapeand mean position of the distributions indicate the ability ofthe models to capture the extremes (day-to-day variability)and the seasonality of the fluxes respectively highlightingpossible predictive biases (ie the over- or underestimation ofthe tower fluxes) Across the NATT the PDFs for the towerfluxes tended to shift to low values and became narroweras annual rainfall declined and this was most prominent inthe dry season A change in the spread and mean positionof the flux tower PDFs demonstrate the strong seasonalityof water and carbon exchange at all sites The PDFs of themodel simulations did not replicate this trend having highdensities and being mostly stationary across sites Regard-ing savanna water use the distributions of the BIOS2 andSPA models were similar to those of the flux towers TheBESS model also showed a similar distribution of LE de-spite the fact that it did not simulate soil-water extractionThe LPJGUESS model which had the shallowest simulatedtree rooting depth displayed PDFs of high density that werebiased towards low LE (20ndash40 W mminus2) across all sites andseasons The MAESPA model showed a similar behaviour

despite this model having a much deeper simulated rootingdepth and a root-water extraction scheme that is equivalent tothe SPA model The distributions for the CABLE and BIOS2models were largely disparate despite these models beingfunctionally equivalent Notably CABLE wet season LE wasmore broadly distributed (5ndash200 W mminus2) than the flux tow-ers and other models at all sites while dry season LE wasnarrower In relation to savanna carbon uptake all modelsshowed wet and dry season PDFs of high density that be-came more closely aligned with the flux tower distributionsas the sites became drier The behaviour of the modelled GPPdistributions were otherwise similar to those of the modelledLE distributions The differences among TBM and flux towerPDFs indicated possible issues in simulated processes thatare active during the wet season

The benchmarks set low to high levels of expectedTBM performance across the NATT Additionally they alsodemonstrated the level of model complexity that is requiredto simulate water and carbon exchange at these sites Thesimplest of the benchmarks represented as a linear regres-sion of the fluxes against Rs (emp1) which was capable ofpredicting the magnitude and daily time course of the towerfluxes (data not shown) but there was not enough informa-tion in Rs to capture the seasonality or the distribution of thefluxes expressed by the tower data The intermediate bench-mark that included additional meteorological information onTa and VPD (emp2) demonstrated an improved capability incapturing the flux distributions but could not replicate thefull seasonality of the fluxes across the NATT It was onlyby including additional phenological information (LAI) to-gether with site meteorology (Rs Ta and VPD) that the sea-sonality and distribution of the fluxes could be captured asdemonstrated by the most complex benchmark (emp3) Thisindicated that in order for the TBMs to achieve the best pos-sible performance at simulating water and carbon exchangealong the NATT the correct implementation and utilisationof phenological information by the models was required AllTBMs used in this study utilised this breadth of informationbut only some of the models were capable of meeting the ex-pected level of performance set by the emp3 benchmark andthen only for specific sites and seasons

32 Residual analysis

An analysis of the model residuals was conducted to showhow model structure affects the prediction of savanna fluxesacross the rainfall gradient To do this we examined the stan-dardised model residuals from each TBM determined by ex-pressing the residual error in terms of its standard deviationFigure 4 shows the residual time series for model-predictedLE and GPP at each savanna site and provides an effec-tive way of examining how a model responds to progressivechanges in the environment through the expression of modelbias and error (Medlyn et al 2005)

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

GPP (micromol mminus2 sminus1 )

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

Figure 3 Probability densities (expressed in scientific notation) of daily mean latent heat (LE) flux and gross primary productivity (GPP) ateach of the five savanna sites where the distributions for each flux are partitioned into wet and dry seasons The order of the sites are fromwettest to driest (a) Howard Springs (b) Adelaide River (c) Daly River (d) Dry River and (e) Sturt Plains The grey region is the towerflux while the dotted lines are the empirical benchmarks Predicted LE and GPP probability densities from each of the six process-basedmodels are given by a spectrum of colours described in the legend

The model residuals demonstrated that there was signifi-cant bias and heteroscedasticity in predicted LE and GPP inalmost all cases The residual time series showed that modelerror was largest in the wet season but declined with the tran-sition into the dry season Additionally the models underes-timated LE and GPP more significantly during the wet sea-

son A possible explanation for this behaviour is that dur-ing the wet season multiple land-surface components thesoil surface the understorey grasses and the tree canopy(ie three sources for potential error) contribute to the bulkfluxes while during the dry season only the tree canopy con-tributes (ie one source for potential error) It is likely that

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

Howard Springs Adelaide River Daly Uncleared Dry River Sturt Plains

Figure 4 Standardised model residuals for latent energy (LE) and gross primary productivity (GPP) expressed in units of standard deviations(SD) [(modelled flux ndash observed flux)SD(observed flux)] Residuals are presented for each model (a) CABLE (b) BIOS2 (c) LPJGUESS(d) MAESPA (d) BESS and (e) SPA where each flux site is represented by a bluendashgreenndashyellow gradient For both fluxes the residuals areplotted against time (ensemble average year) and against the flux prediction (bias)

the reduction in residual error between wet and dry seasonswas a result of the declining influence of the grasses andthe soil surface to ecosystem land-surface exchange duringthe latter period (via senescence and low surface soil mois-ture respectively) The bias towards the underestimation ofwet season fluxes was more pronounced at the mesic sites(Howard Springs Adelaide River) despite some models sim-ulating relatively deep root profiles (eg BIOS2 MAESPA)

Differences in how the TBMs simulated root-water extrac-tion also had no effect on reducing this bias (eg MAESPASPA) Given that soil-water was not a limiting factor at themesic sites during this period deep root profiles offered lim-ited advantage towards model performance Nonetheless thesimulated tree root zone appeared to be an important fac-tor for all sites during the dry season with shallow rootdepths (LPJGUESS 2 m) andor inadequate root-water up-

10BIOS2

(a) Annual performanceLPJGUESS

10MAESPA

10BESS

10BIOS2

(b) Wet season performanceLPJGUESS

10MAESPA

10BESS

10BIOS2

(c) Dry season performanceLPJGUESS

10MAESPA

10BESS

emp3 emp2 emp1 model

Figure 5 Average rank plot showing the performance of the terrestrial biosphere models for all sites across the North Australian TropicalTransect (NATT) ordered in terms of annual rainfall as follows Howard Springs (HowSpr) Adelaide River (AdrRiv) Daly Uncleared(DalUnc) Dry River (DryRiv) and Sturt Plains (StuPla) Models are individually ranked against the benchmarks on the order of 1 to 4(1 model + 3 benchmarks) and express the amount of metrics the models are meeting listed in Table S1 The rankings are determinedindividually for latent energy (LE) and gross primary productivity (GPP) The coloured lines represent each of the six models in the studywhile the grey lines represent the empirical benchmarks The average ranking for each model was determined for (a) a complete year (b) thewet season and (c) the dry season

take schemes (CABLE concentrated in the upper soil pro-file) the likely cause for underestimation during this periodHowever as the sites became drier (eg Sturt Plains) a shal-low root profile was suitable to give flux estimates of a rea-sonably low error Despite model error reducing with the in-crease in ecosystem water limitation that occurs in both space(down the NATT) and time (wet to dry season) there arestill patterns of model bias that may be unrelated to simu-lated soil-water dynamics This is particularly obvious duringthe wet-to-dry transition periods (eg BIOS2 SPA) when theC4 grass understorey senesces indicating possible problemswith the how the models translate information on phenology

33 Model performance

Figure 5 shows a comparison of individual TBM perfor-mance ordered by site from wettest (Howard Springs) todriest (Sturt Plains) and in terms of their annual wet anddry season predictions for each flux Despite differences inmodel complexity (Table 1) the TBMs showed a similar per-formance across sites and seasons For almost all sites theTBMs outperformed the emp1 benchmark for annual fluxpredictions (Fig 5a) However there were some exceptionsto this and good performance in one flux did not necessar-ily result in good performance in the other For exampleMAESPA was unable to beat the emp1 benchmark for LEat sites where MAP gt 1000 mm but performed better thanthe emp2 benchmark for GPP In general there was a slightpattern of increased model performance as annual rainfall de-clined though with a degree of site-to-site variability in therankings for some of the TBMs

In order to examine how seasonal changes affect modelperformance we additionally determined the metrics andrankings for the wet and dry season periods (Fig 5bndashc) Sea-sonal differences were immediately obvious Model perfor-mance for wet season LE and GPP was low to moderateand the majority of the TBMs showed a performance thatranged between the emp1 and emp2 benchmarks In contrastthere were noticeable improvements to dry season modelperformance amongst the TBMs For dry season LE halfthe models (BIOS2 BESS and SPA) were able to consis-tently outperform the emp2 benchmark and come close tomeeting the same number of metrics as the emp3 bench-mark particularly at the drier sites In comparison predicteddry season GPP saw a larger enhancement in model perfor-mance with TBMs more frequently outperforming the emp2benchmark and even some outperforming the emp3 bench-mark (LPJGUESS BESS and SPA at the Daly Unclearedsite) The exception to all this was the CABLE model whichshowed surprisingly little loss or gain in performance despitethe season The results give an indication that as a whole in-put information was better utilised by each TBM at drier sitesand in the dry season suggesting that there are problems inwet season processes

4 Discussion

The NATT which covers a marked rainfall gradient presentsa natural ldquoliving laboratoryrdquo with which a modelrsquos abilityto simulate fluxes in savanna ecosystems may be assessedOur results have highlighted that there is a clear failure ofthe models to adequately perform at predicting wet seasondynamics as compared to the dry season and suggests thatmodelled processes relating to the C4 grass understorey areinsufficient This highlights a key weakness of this group ofTBMs which likely extends to other models outside of thisstudy The inability of these TBMs to capture wet season dy-namics is highlighted by the benchmarking where the per-formance for many of the models was at best equivalent tothat of a multi-linear regression against Rs Ta and VPD(emp2) and in some cases no better than a linear regres-sion against Rs (emp1) Given that this subset of TBMs aresophisticated process-based models that represent our bestunderstanding of land surfacendashatmospheric exchange pro-cesses we would expect them to perform as well as a neu-ral network prediction (emp3) Consequently there is an ev-ident underutilisation of the driving information (ie a fail-ure to describe the underlying relationships in the data) im-peding the performance of these models when predicting sa-vanna fluxes However there were instances where some ofthe TBMs were able to reach similar levels of performancewith the emp3 benchmark which strongly suggests that eachof these models is capable of replicating savanna dynamicsunder certain conditions (eg during the dry season)

Our results suggest that errors among models are likely tobe systematic rather than related to calibration of existingparameters For example BIOS2 had previously optimisedmodel parameters for Australian vegetation (see Haverd etal 2013) but was still unable to out-perform the emp3benchmark in most cases although it performed better thanan un-calibrated CABLE to which it is functionally simi-lar Similarly MAESPA and SPA which used considerablesite characteristic information to parameterise their simula-tions did not significantly outperform un-calibrated models(eg CABLE) Additionally despite these models using thesame leaf root and soil parameterisations both SPA andMAESPA displayed markedly different performances in pre-dicting LE Consequently improving how models representkey processes that drive savanna dynamics is critical to im-proving model performance across this ecosystem

There is certainly enough information in the time-varyingmodel inputs to be able to adequately simulate wet and dryseason dynamics as is evidenced by the benchmarks Wetherefore consider the implications of our results presentpossible reasons below for why this group of TBMs is failingto capture water and carbon exchange along the NATT andmake suggestions as to how this could be improved

255075

CABLE LPJGUESS MAESPA SPA

255075

Jun Jul

255075

Jun Jul

Vegetation transpiration Soil evaporation

Jun Jul

rt Pla

Figure 6 Average year outputs of vegetation transpiration (grass+ trees) and soil evaporation as well as their percentage contributions tototal latent energy (LE) for each of the six terrestrial biosphere models at each of the five savanna sites

41 Water access and tree rooting depth

During the late dry season surface soil moisture in the sandysoils declines to less than 3 volumetric water content withan equivalent matric potential of 3 to 4 MPa (Prior et al1997) During this seasonal phase the grass understorey be-comes inactive and LE can be considered as equivalent totree transpiration such that it is the only active componentduring this period (OrsquoGrady et al 1999) Using this equiv-alence one can infer the relative effect that rooting depthhas on LE during this period Previous studies have shownthat for these savanna sites along the NATT tree transpira-tion is maintained throughout the dry season by deep rootsystems that access deep soil-water stores which in turn arerecharged over the wet season (Eamus et al 2000 Hutley

et al 2001 Kelley et al 2007 OrsquoGrady et al 1999) Inorder for models to perform well they will need to set ade-quate rooting depths and distributions along with root-wateruptake process to enable a model response to such seasonalvariation Examining performance across the models we caninfer this to be a key deficiency As expected TBMs thatprescribed shallow rooting depths (eg LPJGUESS) did notsimulate this process well and underestimated dry season LEat three of the five savanna sites by up to 30 to 40 Thetwo sites at Adelaide River and Sturt Plains were an excep-tion to this with the TBMs displaying a low residual errorwhich is likely to be a consequence of heavier textured soilsand trees at these sites having shallow root profiles At Ade-laide River shallow root profiles are a consequence of shal-low heavier textured soils however dry season transpiration

is sustained due to the presence of saturated yellow hydrosolsoils Sturt Plains is a grassland (the end member of the sa-vanna continuum) where C4 grasses dominate and no treesare present such that transpiration is close to zero in the dryseason The few small shrubs that are established have shal-low root profiles that have adapted to isolated rainfall eventsdriven by convective storms (Eamus et al 2001 Hutley etal 2001 2011) Consequently the TBMs would be expectedto perform better at these sites as water and carbon exchangewill be modulated by the soil-water status of the sub-surfacesoil layers For the other sites models which assumed a rootdepth gt 5 m (BIOS2 SPA and MAESPA) showed the mostconsistent performance in predicting dry season LE and wesuggest that for seasonally water-limited ecosystems suchas savanna deeper soil-water access is critical Our resultshighlight the need for data with which to derive more mech-anistic approaches to setting rooting depth such as that ofSchymanski et al (2009)

Interestingly a low residual error for LE in the dry seasondid not translate as good performance in the overall modelranking This suggests that other processes along the soilndashvegetationndashatmosphere continuum need to be considered toimprove simulated woody transpiration Such processes mayinclude root-water uptake (distribution of roots and how wa-ter is extracted) and the effect of water stress and increasedatmospheric demand at the leaf level (adjustment of stomatalconductance due to changes in leaf water potential) More de-tailed model experiments that examine how each TBM sim-ulates these processes would help identify how they can beimproved

An exception to the above is the BESS model which for-goes simulating belowground processes of soil hydrologyand root-water uptake entirely Rather this model assumesthat the effects of soil-moisture stress on water and carbonexchange is expressed through changes in LAI (and by exten-sion Vcmax) which acts as a proxy for changes in soil mois-ture content (Ryu et al 2011) The fact that BESS performedmoderately well along the NATT coupled with the fact thattree transpiration continues through the dry season suggeststhat there may be enough active green material for remotesensing proxies of water stress to generally work rather wellfor savanna ecosystems It is notable that BESS overesti-mated both GPP and ET in dry season at the driest site SturtPlains (Fig 2e) implying that greenness detected by satelliteremote sensing might not capture carbon and water dynamicswell in such a dry site

42 Savanna wet season dynamics

The relative performance of the TBMs at predicting LE wasmuch poorer in the wet season compared to the dry seasonThe reason for this difference is that wet season LE is the sumof woody and herbaceous transpiration (Eveg) as well as soiland wet-surface evaporation (Esoil) in contrast dry seasonLE is predominantly woody transpiration as described previ-

ously During the wet season up to 75 of total LE arisesfrom understorey herbaceous transpiration and soil evapora-tion (Eamus et al 2001 Hutley et al 2000 Moore et al2016) and of this fraction the C4 grasses contribute a sig-nificant daily amount (Hutley et al 2000) In the absenceof observations of understory LE it can be difficult to de-termine whether grass transpiration is being simulated cor-rectly However separating out the components of wet sea-son LE into soil and vegetation can help identify which ofthese components are causes for error

Separating the outputs of simulated Eveg and Esoil fromeach TBM (excluding BESS which did not determine theseas outputs during the study) shows that simulated wet sea-son Eveg was particularly low for a lot of the models despitehigh LAI and non-limiting soil-water conditions (Fig 6) Aprevious study at Howard Springs by Hutley et al (2000)observed that during the wet season the grass understoreycould transpire sim 28 mm dminus1 while the tree canopy tran-spired only 09 mm dminus1 (Eveg = 37 mm dminus1) Of the sixTBMs at Howard Springs only CABLE and SPA were ableto predict an Eveg close to this level while the other modelspredicted values closer to tree transpiration (ie an underes-timate) This pattern is similar for other NATT sites wherepredicted wet season Eveg remained low and was dominatedby Esoil at the southern end of the NATT An underestima-tion of wet season LE could be due to underestimated Esoilin some of the models Conversely CABLE and BIOS2 pre-dicted a higher Esoil than the other models and this couldbe a reason for their higher LE performance during the wetseason Although Esoil has been reported to reach as highas 28 mm dminus1 at Howard Springs (Hutley et al 2000) pre-dicted Esoil by these models may still be overestimated giventhat vegetation cover during this period is at a seasonal peak(limiting energy available at the soil surface) and transpira-tion is only limited by available energy not water (Hutley etal 2000 Ma et al 2013 Schymanski et al 2009 Whit-ley et al 2011) Given the limited data for Esoil along theNATT it is difficult to determine how large Esoil should beHowever the ratios displayed by the TBMs appear to be rea-sonable with vegetation acting as the predominant pathwayfor surface water flux

Grass transpiration is thus clearly being underrepresentedby most of the TBMs and reasons for this could be due tomultiple factors The evolution of C4 grasses to fix carbonunder low light low CO2 concentrations and high temper-atures has resulted in a gas-exchange process that is highlywater-use efficient (von Caemmerer and Furbank 1999)Consequently this life form is abundant in tropical water-limited ecosystems where it can contribute to more than50 of total LAI (20 to 25) particularly at high rainfallsites (Sea et al 2011) The annual strategy of the C4 grassesat these sites is to indiscriminately expend all available re-sources to maximise productivity during the monsoon pe-riod for growth and to increase leaf area This therefore al-lows grass transpiration to exceed tree transpiration during

the peak wet season as evergreen trees will be more conserva-tive in their water use allowing them to remain active in thedry season (Eamus et al 2001 Hutley et al 2000 Scholesand Archer 1997) Following this logic our results suggestthat the TBMs are either (i) incorrectly ascribing leaf areato the understorey (ie the C4 fractional cover is too low)(ii) incorrectly describing the C4 leafndashgas exchange physi-ology (iii) incorrectly describing the understory micro cli-matic environment (Rs Ta VPD) or (iv) a combination ofthese causes Furthermore it should be noted that the TBMsused in this study are not truly modelling grasses but approx-imating them Grasses are effectively simulated as ldquostem-lessrdquo trees and the distinction between the two life formsis reliant on different parameter sets (eg Vcmax height) andslight modifications of the same process (eg rate of assim-ilation respiration) While our results and the tower data donot allow us to directly determine how C4 grasses may bemisrepresented in these TBMs they clearly indicate that fu-ture development and evaluation should be focused on theseissues Eddy covariance studies of understorey savanna veg-etation as conducted by Moore et al (2016) will be critical tothis process

43 Savanna phenology

The results from this study have shown that to simulate sa-vanna fluxes TBMs must be able to simulate the dynamics ofsavanna phenology expressed by LAI This was highlightedby the empirical benchmarks where the results showed thatwhile Rs Ta and VPD were important drivers LAI was re-quired to capture the seasonality and magnitude of the fluxesto achieve good performance LAI integrates the observedstructural changes of the savanna as annual rainfall declineswith reduced woody stem density driving water and carbonexchange as a result (Kanniah et al 2010 Ma et al 2013Sea et al 2011) When LAI is prescribed in a model it isimportant that leaf area is partitioned correctly between thetrees and grass layers to describe their respective phenologyThis partitioning is important as the C4 grass understoreyexplains most of the seasonal variation in LAI and is a con-sequence of an annual phenology that exhibits rapid growthat the onset of the wet season and senescence at the onsetof the dry (R J Williams et al 1996) By contrast the ev-ergreen eucalypt canopy shows modest reductions in canopyleaf area during the dry season especially as mean annualrainfall declines (Bowman and Prior 2005 Kelley et al2007) The strong seasonal dynamics of the grasses result inlarge changes in LAI with levels varying between 07 and 25at high rainfall sites (Sea et al 2011) The phenological strat-egy of the C4 grasses also changes with rainfall interannualvariability with the onset of the greening period becomingprogressively delayed as sites become drier to become even-tually rain-pulse driven as the monsoonal influence weakens(Ma et al 2013)

With the exception of LPJGUESS all models prescribedLAI as an input driver Prescribing LAI can be problematicdepending on the timescale and how it is partitioned betweentrees and grass layers At large time steps (months) it will failto capture the rapidly changing dynamics of vegetation dur-ing the transition periods and this is particularly true for theonset of the wet season (SeptemberndashNovember) especially atdrier sites that are subject to larger interannual rainfall vari-ability (Hutley et al 2011) Additionally as the sites becomedrier the tree grass ratio will become smaller and this dy-namic can be difficult to predict although methods do exist(see Donohue et al 2009) From the results we infer thatTBMs that prescribe LAI and allow for a dynamic repre-sentation of tree and grass ratios are better able to capturethe changing dynamics of the savanna system This is a pos-sible explanation for the better performance of the BIOS2MAESPA and SPA models in simulating GPP as these mod-els dynamically partition leaf area between trees and grassesat the sub-monthly timescale rather than using a bulk valueHowever models that prescribe LAI have limited capabilityin simulating the land-surface response of savannas to chang-ing climate as tree and grass cover is the outcome of the en-vironmental forcing (particularly rainfall variability and dis-turbance) and not a driver of the system (Sankaran et al2005) DGVMs that consider dynamic vegetation and use aprognostic LAI can simulate the feedback between the cli-mate and the relative cover of trees and grasses which shapesthe savanna continuum This feedback allows the simulatedsavanna structure to potentially shift to alternate states (eggrassland or forest) in response to changes in annual rainfalland fire severity (Scheiter and Higgins 2007 2009) WhileLPJGUESS was the only TBM to use a prognostic LAI in ourstudy it achieved only moderate performance and this maybe due to how carbon is allocated from the pool on an annualtime step such that it is not as dynamic as it could be How-ever its capability to simulate the feedback between climateand LAI is critical for simulating how savanna dynamics maychange from year to year There may also be issues with howphenology is simulated particularly as it is determined fromempirical formulations which are (i) not specifically devel-oped for savanna environments and (ii) calculated before thegrowing season begins Such formulations are therefore notmechanistic and do not respond to actual season dynamics(eg limiting soil water) but they are empirically determined(Richardson et al 2013)

5 Conclusions

This study set out to assess how well a set of functionally dif-ferent state-of-the-art TBMs perform at predicting the bulkexchanges of carbon and water over savanna land surfacesOur model inter-comparison has identified key weaknesses inthe assumptions of biospherendashatmosphere processes whichdo not hold for savanna environments Our benchmarking

has identified low model performance by TBMs is likely aresult of incorrect assumptions related to (i) deep soil-wateraccess (ii) a systematic underestimation of the contributionof the grass understorey in the wet season and (iii) the useof static phenology to represent dynamic vegetation Our re-sults showed that these assumptions as they currently existin TBMs are not wholly supported by ldquoobservationsrdquo of sa-vanna water and carbon exchange and need to be addressedif more reliable projections are to be made on how savannasrespond to environmental change Despite this our bench-marking has shown that all TBMs could potentially operatewell for savanna ecosystems provided that the above issuesare developed We suggest that further work investigate howparticular processes in the models may be affecting overallpredicted water and carbon fluxes and may include testingvariable rooting depths alternate root-water uptake schemesand how these might affect leaf-level outputs (eg stomatalconductance leaf water potential) among TBMs and dif-ferent phenology schemes The issues highlighted here alsohave scope beyond savanna environments and are relevant toother water-limited ecosystems The results from this studyprovide a foundation for improving how savanna ecosystemdynamics are simulated

Data availability

Half-hourly eddy-covariance data sets pertaining to each ofthe savanna sites used in this study are available from httpdataozfluxorgau (Isaac and van Gorsel 2016) Soil de-scriptions for each savanna site are derived from the Digi-tal Atlas of Australian Soils available at wwwasriscsiroau(Isbell 2016)

Appendix A

Table A1 Definition of common metrics used to determine ranks against the empirical benchmarks The terms M and O stand for modeland observations respectively while n denotes the length of the data and i is the datum

Statistical metric Definition

Correlation coefficient (r)n

nsumi=1

(OiMi )minusnsum

nsumi=1

Miradicradicradicradicradicn

nsumi=1

O2i minus

nsumi=1

M2i minus

Standard deviation (SD)

∣∣∣∣∣∣∣∣1minusradic

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

Normalised mean error (NME)

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

∣∣Normalised mean bias (MBE) 1

nsumi=1

(Mi minusOi)

The Supplement related to this article is available onlineat doi105194bg-13-3245-2016-supplement

Acknowledgements This study was conducted as part of theldquoAustralian Savanna Landscapes Past Present and Futurerdquo projectfunded by the Australian Research Council (DP130101566) Thesupport collection and utilisation of data was provided by theOzFlux network (wwwozfluxorgau) and Terrestrial EcosystemResearch Network (TERN wwwternorgau) and funded by theARC (DP0344744 DP0772981 and DP130101566) PALS waspartly funded by the TERN ecosystem Modelling and ScalinginfrAStructure (eMAST) facility under the National CollaborativeResearch Infrastructure Strategy (NCRIS) 2013ndash2014 budgetinitiative of the Australian Government Department of IndustryRhys Whitley was supported through the ARC Discovery Grant(DP130101566) Jason Beringer is funded under an ARC FT(FT110100602) Vanessa Haverdrsquos contribution was supported bythe Australian Climate Change Science Program We acknowledgethe support of the Australian Research Council Centre of Excel-lence for Climate System Science (CE110001028)

Edited by N Kljun

References

Abramowitz G Towards a public standardized diagnostic bench-marking system for land surface models Geosci Model Dev 5819ndash827 doi105194gmd-5-819-2012 2012

Abramowitz G Leuning R Clark M and Pitman A Evaluatingthe Performance of Land Surface Models J Climate 21 5468ndash5481 doi1011752008JCLI23781 2008

Ball J T Woodrow I E and Berry J A A model predictingstomatal conductance and its contribution to the control of photo-synthesis under different environmental conditions in Progressin Photosynthesis Research Martinus-Nijhoff Publishers Dor-drecht the Netherlands 221ndash224 1987

Bashtannyk D M and Hyndman R J Bandwidth selection forkernel conditional density estimation Comput Stat Data An36 279ndash298 doi101016S0167-9473(00)00046-3 2001

Beringer J Hutley L B Tapper N J and Cernusak L ASavanna fires and their impact on net ecosystem productiv-ity in North Australia Glob Change Biol 13 990ndash1004doi101111j1365-2486200701334x 2007

Beringer J Hutley L B Hacker J M Neininger B andPaw U K T Patterns and processes of carbon water andenergy cycles across northern Australian landscapes Frompoint to region Agr Forest Meteorol 151 1409ndash1416doi101016jagrformet201105003 2011

Beringer J Hutley L B Abramson D Arndt S K Briggs PBristow M Canadell J G Cernusak L A Eamus D EvansB J Fest B Goergen K Grover S P Hacker J HaverdV Kanniah K Livesley S J Lynch A Maier S Moore CRaupach M Russell-Smith J Scheiter S Tapper N J andUotila P Fire in Australian Savannas from leaf to landscapeGlob Change Biol 21 62ndash81 doi101111gcb12686 2015

Best M J Abramowitz G Johnson H R Pitman A J Bal-samo G Boone A Cuntz M Decharme B Dirmeyer PA Dong J Ek M Guo Z Haverd V van den Hurk BJ Nearing G S Pak B Peters-Lidard C Santanello Jr JA Stevens L and Vuichard N The plumbing of land surfacemodels benchmarking model performance J Hydrometeorol16 1425ndash1442 2015

Bond W J What Limits Trees in C4 Grasslands andSavannas Annu Rev Ecol Evol Syst 39 641ndash659doi101146annurevecolsys39110707173411 2008

Bowman D M J S and Prior L D Why do evergreen trees dom-inate the Australian seasonal tropics Aust J Bot 53 379ndash399doi101071BT05022 2005

Cernusak L A Hutley L B Beringer J Holtum J AM and Turner B L Photosynthetic physiology of eu-calypts along a sub-continental rainfall gradient in north-ern Australia Agr Forest Meteorol 151 1462ndash1470doi101016jagrformet201101006 2011

Chapin F S Woodwell G M Randerson J T Rastetter E BLovett G M Baldocchi D D Clark D A Harmon M ESchimel D S Valentini R Wirth C Aber J D Cole J JGoulden M L Harden J W Heimann M Howarth R WMatson P A McGuire A D Melillo J M Mooney H ANeff J C Houghton R A Pace M L Ryan M G RunningS W Sala O E Schlesinger W H and Schulze E D Recon-ciling carbon-cycle concepts terminology and methods Ecosys-tems 9 1041ndash1050 doi101007s10021-005-0105-7 2006

Chen X Hutley L B and Eamus D Carbon balance of atropical savanna of northern Australia Oecologia 137 405ndash16doi101007s00442-003-1358-5 2003

Collatz G J Ball J T Grivet C and Berry J A Physiologi-cal and environmental regulation of stomatal conductance pho-tosynthesis and transpiration a model that includes a laminarboundary layer Agr Forest Meteorol 54 107ndash136 1991

Collatz G J Ribas-Carbo M and Berry J A Coupledphotosynthesis-stomatal conductance model for leaves of C4plants Funct Plant Biol 19 519ndash538 1992

De Kauwe M G Zhou S-X Medlyn B E Pitman A J WangY-P Duursma R A and Prentice I C Do land surface modelsneed to include differential plant species responses to droughtExamining model predictions across a mesic-xeric gradientin Europe Biogeosciences 12 7503ndash7518 doi105194bg-12-7503-2015 2015

Donohue R J Mc Vicar T R and Roderick M L Climate-related trends in Australian vegetation cover as inferred fromsatellite observations 1981ndash2006 Glob Change Biol 15 1025ndash1039 doi101111j1365-2486200801746x 2009

Duursma R A and Medlyn B E MAESPA a model to studyinteractions between water limitation environmental drivers andvegetation function at tree and stand levels with an example ap-plication to [CO2]times drought interactions Geosci Model Dev5 919ndash940 doi105194gmd-5-919-2012 2012

Eamus D OrsquoGrady A P O and Hutley L B Dry season condi-tions determine wet season water use in the wet-tropical savannasof northern Australia Tree Physiol 20 1219ndash1226 2000

Eamus D Hutley L B and OrsquoGrady A P O Daily and seasonalpatterns of carbon and water fluxes above a north Australian sa-vanna Tree Physiol 21 977ndash88 2001

Eamus D Chen X Kelley G and Hutley L B Root biomassand root fractal analyses of an open Eucalyptus forest in a sa-vanna of north Australia Aust J Bot 50 31ndash41 2002

Eamus D Cleverly J Boulain N Grant N Faux R andVillalobos-Vega R Carbon and water fluxes in an arid-zoneAcacia savanna woodland An analyses of seasonal patterns andresponses to rainfall events Agr Forest Meteorol 182ndash183225ndash238 doi101016jagrformet201304020 2013

Farquhar G D von Caemmerer S and Berry J A A Biochem-ical Model of Photosynthetic CO2 Assimilation in Leaves of C3species Planta 149 78ndash90 1980

Grace J Jose J S Meir P Miranda H S and Montes R AProductivity and carbon fluxes of tropical savannas J Biogeogr33 387ndash400 doi101111j1365-2699200501448x 2006

Haverd V Raupach M R Briggs P R Canadell J G IsaacP Pickett-Heaps C Roxburgh S H van Gorsel E ViscarraRossel R A and Wang Z Multiple observation types reduceuncertainty in Australiarsquos terrestrial carbon and water cyclesBiogeosciences 10 2011ndash2040 doi105194bg-10-2011-20132013

Haverd V Smith B Raupach M Briggs P Nieradzik LBeringer J Hutley L Trudinger C M and Cleverly J Cou-pling carbon allocation with leaf and root phenology predictstree-grass partitioning along a savanna rainfall gradient Biogeo-sciences 13 761ndash779 doi105194bg-13-761-2016 2016

Haxeltine A and Prentice I C A general model for the light-use efficiency of primary production Funct Ecol 10 551ndash561doi1023072390165 1996

Higgins S I and Scheiter S Atmospheric CO2 forces abrupt veg-etation shifts locally but not globally Nature 488 209ndash212doi101038nature11238 2012

Hsu K Self-organizing linear output map (SOLO) An artifi-cial neural network suitable for hydrologic modeling and analy-sis Water Resour Res 38 1ndash17 doi1010292001WR0007952002

Huntingford C and Monteith J L The behaviour of a mixed-layer model of the convective boundary layer coupled to a bigleaf model of surface energy partitioning Bound-Lay Meteo-rol 88 87ndash101 doi101023A1001110819090 1998

Hutchinson M F and Xu T ANUCLIM v61 Fenner School ofEnvironment and Society Australian National University Can-berra ACT 2010

Hutley L B OrsquoGrady A P O and Eamus D Evapotranspirationfrom Eucalypt open-forest savanna of Northern Australia FunctEcol 14 183ndash194 2000

Hutley L B Grady A P O and Eamus D Monsoonal influenceson evapotranspiration of savanna vegetation of northern Aus-tralia Oecologia 126 434ndash443 doi101007s0044200005392001

Hutley L B Beringer J Isaac P R Hacker J M and Cer-nusak L A A sub-continental scale living laboratory Spa-tial patterns of savanna vegetation over a rainfall gradient innorthern Australia Agr Forest Meteorol 151 1417ndash1428doi101016jagrformet201103002 2011

Isaac P and van Gorsel E OzFlux Australian and New ZealandFlux Research and Monitoring available at httpdataozfluxorgau last access 1 June 2016

Isbell R The Australian Soil Classification Revised Edn CSIROPublishing Collingwood Victoria 2002

Isbell R F Digital Atlas of Australian Soils available at httpwwwasriscsiroaudownloadsAtlassoilAtlas2Mzip last ac-cess 1 June 2016

Kanniah K D Beringer J and Hutley L B The compara-tive role of key environmental factors in determining savannaproductivity and carbon fluxes A review with special refer-ence to northern Australia Prog Phys Geogr 34 459ndash490doi1011770309133310364933 2010

Kelley G OrsquoGrady A P Hutley L B and Eamus D A compar-ison of tree water use in two contiguous vegetation communitiesof the seasonally dry tropics of northern Australia The impor-tance of site water budget to tree hydraulics Aust J Bot 55700ndash708 doi101071BT07021 2007

Koch G W Vitousek P M Steffen W L and Walker B HTerrestrial transects for global change research Vegetation 12153ndash65 doi101007BF00044672 1995

Kowalczyk E A Wang Y P and Law R M The CSIRO Atmo-sphere Biosphere Land Exchange (CABLE) model for use in cli-mate models and as an offline model Aspendale Victoria 2006

Lehmann C E R Prior L D and Bowman D M J S Decadaldynamics of tree cover in an Australian tropical savanna AustralEcology 34 601ndash612 doi101111j1442-9993200901964x2009

Lehmann C E R Anderson T M Sankaran M Higgins SI Archibald S Hoffmann W A Hanan N P WilliamsR J Fensham R J Felfili J Hutley L B Ratnam JSan Jose J Montes R Franklin D Russell-Smith J RyanC M Durigan G Hiernaux P Haidar R Bowman DM J S and Bond W J Savanna vegetation-fire-climate re-lationships differ among continents Science 343 548ndash552doi101126science1247355 2014

Leuning R A critical appraisal of a combined stomatal-photosynthesis model for C3 plants Plant Cell Environ 18339ndash355 1995

Li L Wang Y P Yu Q Pak B Eamus D Yan J Van GorselE and Baker I T Improving the responses of the Australiancommunity land surface model (CABLE) to seasonal drought JGeophys Res-Biogeo 117 1ndash16 doi1010292012JG0020382012

Ma X Huete A Yu Q Coupe N R Davies K Broich MRatana P Beringer J Hutley L B Cleverly J BoulainN and Eamus D Spatial patterns and temporal dynam-ics in savanna vegetation phenology across the North Aus-tralian Tropical Transect Remote Sens Environ 139 97ndash115doi101016jrse201307030 2013

McKenzie N N Jacquier D Isbell R and Brown K Australiansoils and landscapes an illustrated compendium CSIRO Pub-lishing Collingwood Victoria 2004

Medlyn B E Robinson A P Clement R and McMurtrie RE On the validation of models of forest CO2 exchange usingeddy covariance data some perils and pitfalls Tree Physiol 25839ndash857 2005

Medlyn B E Duursma R A Eamus D Ellsworth D S Pren-tice I C Barton C V M Crous K Y De Angelis P Free-man M and Wingate L Reconciling the optimal and empiri-cal approaches to modelling stomatal conductance Glob ChangeBiol 17 2134ndash2144 doi101111j1365-2486201002375x2011

Moore C E Beringer J Evans B Hutley L B McHugh Iand Tapper N J The contribution of trees and grasses to pro-ductivity of an Australian tropical savanna Biogeosciences 132387ndash2403 doi105194bg-13-2387-2016 2016

OrsquoGrady A P O Eamus D and Hutley L B Transpiration in-creases during the dry season patterns of tree water use in euca-lypt open-forests of northern Australia Tree Physiol 19 591ndash597 1999

Parton W J Anderson D W Cole C V and Stewart J W BSimulation of soil organic matter formation and mineralizationin semiarid agroecosystems in Nutrient Cycling In AgriculturalEcosystems 23 533ndash550 1983

Pitman A J The evolution of and revolution in land surfaceschemes designed for climate models Int J Climatol 23 479ndash510 doi101002joc893 2003

Prior L D Eamus D and Duff G A Seasonal Trends in Car-bon Assimilation Stomatal Conductance Pre-dawn Leaf WaterPotential and Growth in Terminalia ferdinandiana a DeciduousTree of Northern Australian Savannas Aust J Bot 45 53ndash691997

Richardson A D Keenan T F Migliavacca M Ryu YSonnentag O and Toomey M Climate change phenol-ogy and phenological control of vegetation feedbacks tothe climate system Agr Forest Meteorol 169 156ndash173doi101016jagrformet201209012 2013

Russell-Smith J and Edwards A C Seasonality and fire severityin savanna landscapes of monsoonal northern Australia Int JWildland Fire 15 541ndash550 doi101071WF05111 2006

Ryu Y Baldocchi D D Kobayashi H van Ingen C Li JBlack T A Beringer J van Gorsel E Knohl A LawB E and Roupsard O Integration of MODIS land andatmosphere products with a coupled-process model to esti-mate gross primary productivity and evapotranspiration from1 km to global scales Global Biogeochem Cy 25 GB4017doi1010292011GB004053 2011

Ryu Y Baldocchi D D Black T A Detto M Law BE Leuning R Miyata A Reichstein M Vargas R Am-mann C Beringer J Flanagan L B Gu L HutleyL B Kim J McCaughey H Moors E J Rambal Sand Vesala T On the temporal upscaling of evapotranspira-tion from instantaneous remote sensing measurements to 8-day mean daily-sums Agr Forest Meteorol 152 212ndash222doi101016jagrformet201109010 2012

Sankaran M Hanan N P Scholes R J Ratnam J AugustineD J Cade B S Gignoux J Higgins S I Le Roux X Lud-wig F Ardo J Banyikwa F Bronn A Bucini G Caylor KK Coughenour M B Diouf A Ekaya W Feral C J Febru-ary E C Frost P G H Hiernaux P Hrabar H Metzger KL Prins H H T Ringrose S Sea W Tews J Worden Jand Zambatis N Determinants of woody cover in African sa-vannas Nature 438 846ndash849 doi101038nature04070 2005

Scheiter S and Higgins S I Partitioning of root and shoot com-petition and the stability of savannas Am Nat 170 587ndash601doi101086521317 2007

Scheiter S and Higgins S I Impacts of climate change on the veg-etation of Africa an adaptive dynamic vegetation modelling ap-proach Glob Change Biol 15 2224ndash2246 doi101111j1365-2486200801838x 2009

Scheiter S Higgins S I Beringer J and Hutley L B Climatechange and long-term fire management impacts on Australian sa-vannas New Phytol 205 1211ndash1226 doi101111nph131302015

Scholes R J and Archer S R Tree-grass interac-tions in savannas Annu Rev Ecol Syst 28 517ndash544doi101146annurevecolsys281517 1997

Schymanski S J Roderick M L Sivapalan M Hutley L Band Beringer J A test of the optimality approach to mod-elling canopy properties and CO2 uptake by natural vegeta-tion Plant Cell Environ 30 1586ndash98 doi101111j1365-3040200701728x 2007

Schymanski S J Sivapalan M Roderick M L Beringer J andHutley L B An optimality-based model of the coupled soilmoisture and root dynamics Hydrol Earth Syst Sci 12 913ndash932 doi105194hess-12-913-2008 2008

Schymanski S J Sivapalan M Roderick M L Hutley L Band Beringer J An optimality-based model of the dynamicfeedbacks between natural vegetation and the water balanceWater Resour Res 45 W01412 doi1010292008WR0068412009

Sea W B Choler P Beringer J Weinmann R A Hutley LB and Leuning R Documenting improvement in leaf areaindex estimates from MODIS using hemispherical photos forAustralian savannas Agr Forest Meteorol 151 1453ndash1461doi101016jagrformet201012006 2011

Simioni G Roux X Le Gignoux J and Sinoquet H Treegrassa 3-D process-based model for simulating plant interactions intreendashgrass ecosystems Ecol Modell 131 47ndash63 2000

Simioni G Gignoux J and Le Roux X Tree layer spatial struc-ture can affect savanna production and water budget Resultsof a 3-D model Ecology 84 1879ndash1894 doi1018900012-9658(2003)084[1879TLSSCA]20CO2 2003

Sitch S Smith B Prentice I C Arneth A Bondeau ACramer W Kaplan J Levis S Lucht W Sykes M Thon-icke K and Venevski S Evaluation of ecosystem dynamicsplant geography and terrestrial carbon cycling in the LPJ dy-namic vegetation model Glob Change Biol 9 161ndash185 2003

Smith B Prentice I C and Sykes M T Representation ofvegetation dynamics in the modelling of terrestrial ecosystemsComparing two contrasting approaches within European climatespace Global Ecol Biogeogr 10 621ndash637 doi101046j1466-822X2001t01-1-00256x 2001

Still C J Berry J A Collatz G J and DeFries RS Global distribution of C3 and C4 vegetation Carboncycle implications Global Biogeochem Cy 17 6-1ndash6-14doi1010292001GB001807 2003

Tuzet A Perrier A and Leuning R A coupled model of stom-atal conductance photosynthesis Plant Cell Environ 26 1097ndash1117 2003

van der Werf G R Randerson J T Giglio L Gobron N andDolman A J Climate controls on the variability of fires inthe tropics and subtropics Global Biogeochem Cy 22 1ndash13doi1010292007GB003122 2008

von Caemmerer S and Furbank R T Modeling C4 Photosynthe-sis in C4 Plant Biology edited by Sage R F and Monson RK 173ndash211 Academic Press Toronto 1999

Wang Y P Kowalczyk E Leuning R Abramowitz G Rau-pach M R Pak B Van Gorsel E and Luhar A Diag-

nosing errors in a land surface model (CABLE) in the timeand frequency domains J Geophys Res-Biogeo 116 1ndash18doi1010292010JG001385 2011

Whitley R J Macinnis-Ng C M O Hutley L B Beringer JZeppel M Williams M Taylor D and Eamus D Is pro-ductivity of mesic savannas light limited or water limited Re-sults of a simulation study Glob Change Biol 17 3130ndash3149doi101111j1365-2486201102425x 2011

Williams M Rastetter E B Fernandes D N Goulden M LWofsy S C Shaver G R Melillo J M Munger J WFan S-M and Nadelhoffer K J Modelling the soil-plant-atmosphere continuum in a Quercus Acer stand at Harvard For-est the regulation of stomatal conductance by light nitrogen andsoilplant hydraulic properties Plant Cell Environ 19 911ndash9271996

Williams R J Duff G A Bowman D M J S and Cook G DVariation in the composition and structure of tropical savannas asa function of rainfall and soil texture along a large-scale climaticgradient in the Northern Territory Australia J Biogeogr 23747ndash756 doi101111j1365-26991996tb00036x 1996

Abstract

Introduction

Methodology

Observational data

Terrestrial biosphere models

Experimental protocol

Empirical benchmarking

Results

Model predictions

Residual analysis

Model performance

Discussion

Water access and tree rooting depth

Savanna wet season dynamics

Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

of savanna vegetation and its response to rainfall interannualvariability We believe that this study is the first to assesshow well TBMs simulate savanna ecosystems and that theseresults will be used to improve the representation of savannasecosystems in future global climate model studies

1 Introduction

Savanna ecosystems are a diverse and important biome thatplay a significant role in global land-surface processes (vander Werf et al 2008) Globally they occupy regions aroundthe wetndashdry tropical to sub-tropical equatorial zone cover-ing approximately 15 to 20 of the terrestrial surface andcontributesim 30 to global net primary production (Grace etal 2006 Lehmann et al 2014) Savannas are water-limitedecosystems where rainfall is often seasonal or monsoonaland have a spatial extent that can cover an area with annualrainfall in the range of 500 to 2000 mm (Bond 2008 Kan-niah et al 2010 Sankaran et al 2005) The variability inthe amount and timing of annual rainfall coupled with lo-cal topo-edaphic properties and the frequency and intensityof seasonal fires strongly influence the structure and func-tion of savanna vegetation (Beringer et al 2007 Kanniahet al 2010 Ma et al 2013 Sankaran et al 2005) Sa-vannas are characterised by a multi-layer stratum of vege-tation where an open and discontinuous canopy overstoreyis seasonally dominated by understorey grasses (Scholes andArcher 1997) These tree and grass layers are distinctly andfunctionally different fixing carbon using different photo-synthetic pathways C3 and C4 photosynthesis respectively(Bond 2008 Scholes and Archer 1997 R J Williams etal 1996) The canopy overstorey can be either evergreenor deciduous (depending on the evolutionary history) whilethe grass understorey is annual active only in the wet sea-son and senescing at the end of this period (R J Williamset al 1996) Consequently water carbon and nutrient cy-cling in savannas is largely determined from the balance andco-existence of these two life forms (Lehmann et al 2009Sankaran et al 2005)

Given the complex nature of savannas modelling the landsurface exchange and vegetation dynamics for this biome ischallenging for terrestrial biosphere models (TBMs) Herewe define TBMs to broadly encompass stand land surfaceand dynamic global vegetation models (Pitman 2003) Mostland surface schemes that feed into larger earth system mod-els use simplistic representations of vegetation and these willhave difficulty describing the complex structure of savannaecosystems Such issues may be simplistic assumptions in re-lation to rooting depth and inadequate responses to drought(De Kauwe et al 2015 Li et al 2012) ignoring the mul-tilayered nature of savannas and the differing structural (in-cluding radiation) functional (including different plant func-tional types) and phenological differences (Whitley et al

2011) and in some cases neglecting the C4 photosyntheticpathway entirely (Parton et al 1983 Schymanski et al2007) It is therefore critical that TBMs meet the challengesthat savanna dynamics present if water and carbon exchangeare to be correctly simulated in response to global change

Despite these issues there have been significant advancesin modelling savanna dynamics in recent years and thesehave been focused on integrating important features specificto savanna ecosystems namely frequent fire and treendashgrasscompetitive interactions which are processes that shape sa-vanna structure and function (Haverd et al 2016 Higginsand Scheiter 2012 Scheiter and Higgins 2007 Scheiter etal 2014 Simioni et al 2003) Nevertheless little work hasbeen undertaken to critically evaluate the performance andprocesses of TBMs when used to capture water and carboncycling in savannas most notably in West Africa (Simioniet al 2000) and Australia (Schymanski et al 2007 20082009 Whitley et al 2011) Many global ecosystem mod-els moreover use broad plant functional types (PFTs) withsingle parameter values to describe whole biomes (Pitman2003) making them unable to represent changing vegeta-tion structure (tree grass ratio) in the continuum of grass-land to woodland savanna Approaches have been developedthat can account for savanna dynamics such as using mixedtiles whereby trees and grasses are simulated as separate sur-faces that are then aggregated together (Kowalczyk et al2006) However this approach fails to capture the compe-tition between trees and grasses for light water and nutrientresources

In this study we take six TBMs of distinctly different con-ceptual frameworks and assess their ability to simulate sa-vanna water and carbon exchange along the North AustralianTropical Transect (NATT) which is defined by a strong rain-fall gradient Australian tropical savannas can be consideredlargely intact compared to South American and African sa-vannas and provide a ldquoliving laboratoryrdquo to understand thelinks between vegetation structure and function and how itresponds to environmental change (Hutley et al 2011) Wechallenge the models by evaluating them along the rainfallgradient which extends over a broad biogeographical ex-tent and strong interannual variability in climate (Koch et al1995) The aim of this study is to highlight critical processesthat may be missing in current TBMs and are required to ade-quately simulate savanna ecosystems Specifically we exam-ine whether a TBMrsquos structural framework such as the rep-resentation of the understorey grasses (C4 photosynthesis)tree rooting depth and description of phenology (prescribedvs dynamic) can adequately replicate observed carbon andwater fluxes To achieve this we measure the performance ofeach TBM by comparing its predictions to a set of empiricalbenchmarks that describe a priori expected levels of modelperformance We identify regions of low performance amongsites and seasons to diagnose under what climate conditionsreduced model performance occurs We then infer what pro-cesses (present or missing) may be the cause for reduced per-

2 Methodology

NATT sites

20 22 24 26 28 30

0 500 1000 1500 2000

Elevation (m) 64 90 110 175 250

Meteorologyf

Vegetationf

Canopy description

Supplydemand

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

2 Methodology

NATT sites

20 22 24 26 28 30

0 500 1000 1500 2000

Elevation (m) 64 90 110 175 250

Meteorologyf

Vegetationf

Canopy description

Supplydemand

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

NATT sites

20 22 24 26 28 30

0 500 1000 1500 2000

Elevation (m) 64 90 110 175 250

Meteorologyf

Vegetationf

Canopy description

Supplydemand

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Elevation (m) 64 90 110 175 250

Meteorologyf

Vegetationf

Canopy description

Supplydemand

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Canopy description

Supplydemand

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

020406080

100120140160180

minus2)

(a) Howard Springs

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(b) Adelaide River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(c) Daly Uncleared

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

20406080

100120140160180

minus2)

(d) Dry River

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

020406080

100120140160180

minus2)

(e) Sturt Plains

JanFeb

MarApr

May Jun JulAug

Sep Oct NovDec

GPP (microm

minus2 sminus

Tower obs

LPJGUESS

MAESPA

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

3 Results

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

0 40 80 120 16000

1e 1Wet season

LE distributions

0 40 80 1200

1e 2Dry season

12 9 6 3 001234567

1e 1Wet season

GPP distributions

6 5 4 3 2 1 0 100

Dry season

10(a) H

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0000510152025303540

6 5 4 3 2 1 0 100

10(b) A

0 40 80 120 1600

0 40 80 12000

101e 1

12 9 6 3 00005101520253035

6 5 4 3 2 1 0 100

10(c) Daly

0 40 80 120 16000

0 40 80 1200

12 9 6 3 0012345678

6 5 4 3 2 1 0 100

0 40 80 120 160

LE (W m2 )

0 40 80 12000

12 9 6 3 0

Tower obs

LPJGUESS

MAESPA

6 5 4 3 2 1 0 100

turt P

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

(σstand)

(a) CABLE

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(b) BIOS2

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(c) LPJGUESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(d) MAESPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(e) BESS

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

(f) SPA

Mar Apr

May Ju

Aug Sep

OctNov Dec

(σstand)

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

10BIOS2

10MAESPA

10BESS

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

4 Discussion

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

255075

Jun Jul

255075

Jun Jul

rt Pla

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

5 Conclusions

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Data availability

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Appendix A

nsumi=1

(OiMi )minusnsum

nsumi=1

O2i minus

nsumi=1

M2i minus

1nminus1

nsumi=1

(MiminusM

)2radic

1nminus1

nsumi=1

(OiminusO

)2∣∣∣∣∣∣∣∣

nsumi=1|MiminusOi |

nsumi=1

∣∣OminusOi

nsumi=1

(Mi minusOi)

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Edited by N Kljun

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Abstract

Introduction

Methodology

Observational data




Results

Model predictions

Residual analysis

Model performance

Discussion



Savanna phenology

Conclusions

Appendix A

Acknowledgements

References

Documents

A model inter-comparison study to examine limiting factors in …papers.agrivy.com/webfiles/papers/2016-BIOGEOSCIENCES... · 2019-03-14 · Africa (Koch et al., 1995). Soils range