ContentslistsavailableatScienceDirect
Agricultural
Water
Management
j o ur na l h o me pa g e : w w w . e l s e v i e r . c o m / l o c a t e / a g w a t
Performance
assessment
of
the
FAO
AquaCrop
model
for
soil
water,
soil
evaporation,
biomass
and
yield
of
soybeans
in
North
China
Plain
P.
Paredes
a,1,
Z.
Wei
b,c,1,
Y.
Liu
b,c,
D.
Xu
b,c,
Y.
Xin
b,
B.
Zhang
b,
L.S.
Pereira
a,∗aCEER-BiosystemsEngineering,InstituteofAgronomy,UniversityofLisbon,TapadadaAjuda,1349-017,Lisbon,Portugal
bStateKeyLaboratoryofSimulationandRegulationofWaterCycleinRiverBasin,ChinaInstituteofWaterResourcesandHydropowerResearch,China cNationalCenterofEfficientIrrigationEngineeringandTechnologyResearch,Beijing,100048,China
a
r
t
i
c
l
e
i
n
f
o
Articlehistory: Received25July2014 Accepted10December2014 Availableonline22January2015 Keywords:
Cropcoefficientscurve Dualcropcoefficientapproach Partitioningcropevapotranspiration Planttranspiration
Soilwaterbalance SIMDualKcmodel
a
b
s
t
r
a
c
t
FouryearsofsoybeanexperimentaldataobservedatDaxing,NorthChinaPlain,wereusedtoassessthe
abilityoftheAquaCropmodeltopredictsoybeanfinalbiomassandyield.Themodelwasparameterized
andcalibratedusingfielddataonleafareaindex(LAI),availablesoilwater,soilevaporation,biomassand
finalyielddata.Themodelwasassessedusingcalibratedanddefaultparameters.DataonLAIwereused
toderivethefractionofgroundcoverandtocalibratethegreencanopycover(CC)curve.Anaccurate
cal-ibrationoftheCCcurvewasperformed,withlowrootmeansquareerrors(RMSE<7.3%).Resultsrelative
tosoilwaterbalancesimulationsshowahighvariabilityofthepredictions,thusabiasoftheestimation,
withR2ranging0.22–0.86andlowNash-SutcliffeefficiencyEF,rangingbetween−0.47and0.82.The
esti-mationerrorswererelativelyhigh,withRMSEnotexceeding22.9mm.AquaCropwascomparedwith
thesoilwaterbalancemodelSIMDualKc,thathasshownbetterperformancewithR2≥0.83,EFgenerally
greaterthan0.75andRMSEsmallerthan12.5mm.Thesoilevaporation(Es)simulationswerecompared
withtheobservationsperformedusingmicrolysimeters;resultsforAquacrophaveshownacleartrend
forunder-estimationofEs,with“goodness-of-fit”resultsworsethanforSIMDualKc(Weietal.,2015).
Ingeneral,AquaCrophasshownseriouslimitationstoestimatecroptranspirationorsoilevaporation,
whichislikelyduetoabandoningtheFAOdualKcapproach.However,themodelperformedwellrelative
tobiomassandyieldpredictions,withayieldRMSEof302kgha−1.Overall,resultsshowtheadequacy
ofAquaCropforestimatingsoybeanbiomassandyieldwhenthemodelisappropriatelyparameterized.
However,AquaCropisnotappropriatetosupportirrigationscheduling.
©2014ElsevierB.V.Allrightsreserved.
Abbreviations:ARE,Averagerelativeerror(%);ASW,Availablesoilwater(mm);B,Abovegrounddrybiomass(tha−1);b,Regressioncoefficient(non-dimensional);BWP*,
Biomass(water)productivityadjustedforEToandCO2(gm−2);CC,Greencanopycover(%);CC*,Actualcropcanopycoveradjustedformicro-advectiveeffects(%);CCo,
Canopycoverat90%ofemergence(cm2perplant);CC
x,Maximumgreencanopycover(%);CDC,Canopydeclinecoefficient(%GDD−1or%day−1);CGC,Canopygrowth
coefficient(%GDD−1or%day−1);CGDD,Cumulativegrowingdegreedays(oC);CN,Curvenumber(non-dimensional);CR,Capillaryrisefromshallowwatertable(mm);
DP,Deeppercolation(mm);EF,Modellingefficiency(non-dimensional);Es,Soilevaporation(mm);ET,cropevapotranspiration(mm);ETc,Potential(non-stressed)crop
evapotranspiration(mm);ETc act,Adjustedoractualcropevapotranspiration(mm);ETd,Cropevapotranspirationdeficit(mm);ETo,Referenceevapotranspiration(mm);fc,
Fractionofsoilcoverbyvegetation(non-dimensional);few,Fractionofsoilwettedandexposedtosolarradiation(non-dimensional);fK,Declinefactor(non-dimensional);
GDD,Growingdegreedays(oC);HI
o,Referenceharvestindex(%);Kc,Cropcoefficient(non-dimensional);Kcmax,Maximumvalueofcropcoefficient(followingrainor
irrigation)(non-dimensional);Kcb,Basalcropcoefficient(non-dimensional);Kcbact,Actualoradjustedbasalcropcoefficient(non-dimensional);Kcbend,Basalcropcoefficient
atendofthelateseasongrowthstage(non-dimensional);Kcbini,Basalcropcoefficientduringtheinitialgrowthstage(non-dimensional);Kcbmid,Basalcropcoefficient
duringthemid-seasongrowthstage(non-dimensional);Kcmax,Maximumvalueofthecropcoefficient(Kc)followingrainoranirrigationevent(non-dimensional);Kc,Tr,
Croptranspirationcoefficient(non-dimensional);Kc,Tract,Actualcroptranspirationcoefficient(non-dimensional);Kc,Trx,Maximumstandardcroptranspirationcoefficient
(non-dimensional);Kex,Soilevaporationcoefficientforfullywetandnon-shadedsoilsurface(non-dimensional);Kd,Densitycoefficient(non-dimensional);Kr,Evaporation
reductioncoefficient(non-dimensional);Ks,Waterstresscoefficient(non-dimensional);Ksat,Saturatedhydraulicconductivity(cmd−1);Ky,Yieldresponsefactor
(non-dimensional);LAI,Leafareaindex(cm2cm−2);R2,Determinationcoefficient(non-dimensional);RAW,Readilyavailablesoilwater(mm);REW,Readilyevaporablesoilwater
(mm);RMSE,Rootmeansquareerror(sameunitsasobservations);RO,Runoff(mm);RYL,Relativeyieldlosses(%);Ta,Actualtranspiration(mm);TAW,Total(plant)available
soilwater(mm);Tc,Croptranspiration(mm);Td,transpirationdeficit(mm);TEW,Totalevaporablewater(mm);Wrel,Relativesoilwatercontent(%);Ya,Actualyield(tha−1);
Ze,Evaporablelayerthickness(m);Zr,Rootdepth(m);FC,Volumetricwatercontentatfieldcapacity(m3m−3);sat,Volumetricwatercontentatsaturation(m3m−3);WP,
Volumetricwatercontentatwiltingpoint(m3m−3).
∗ Correspondingauthor.Tel.:+351213653480;fax:351213653287.
E-mailaddresses:lspereira@isa.ulisboa.pt,luis.santospereira@gmail.com(L.S.Pereira).
1 Theseauthorscontributedequallytothepresentstudy.
http://dx.doi.org/10.1016/j.agwat.2014.12.007
1. Introduction
SoybeansareamainsummercropinNorthChinaPlain,where
theyare cropped during the rainyseason, thus only requiring
supplementalirrigationtofulfilthecropwaterrequirement.The
predictionofthesoybeanyieldandoftheyieldresponsetowater
ismandatoryfordevelopingstrategiesforirrigationmanagement
and tosupport related farmers’ decision-making under limited
wateravailabilityconditions.
Toassesstheimpactsofdifferentirrigationscheduling
strate-giesonyield,variousmodellingapproachesmaybeusedsuchas
couplinga soilwater balancemodelwithwateryieldfunctions
describingtherelationshipsbetweencropevapotranspirationor
croptranspirationwith yield.A successful approach appliedto
soybeansisdescribedbyWeietal.(2015)thatadoptedthe
SIM-DualKcsoilwaterbalancemodel(Rosaetal.,2012)coupledwith
theStewart’swater-yieldmodel(Stewartetal.,1977).The
SIM-DualKcmodelappliestheFAOdualcropcoefficientapproachfor
computingandpartitioningthedailycropevapotranspiration(ETc,
mm)intocroptranspiration(Tc,mm)andsoilevaporation(Es,mm).
SIMDualKcperformsadailysoilwaterbalancefortheentireroot
zoneandadailywaterbalanceofthesoilevaporationlayer
adopt-ingthetwostagesRitchie’sevaporationapproach(Ritchie,1972;
Allenetal.,1998;AllenandPereira,2009).Deeppercolation(DP,
mm)andcapillaryrise(CR,mm)arecomputedusingtheparametric
equationsproposedbyLiuetal.(2006),andrunoff(RO,mm)is
esti-matedusingthecurvenumber(CN)approach(Allenetal.,2007).
SIMDualKcallowscomputingthedailyETcandTcdeficits(ETdand
Td)definedrespectively asthedifferencebetweenthestandard
ETc andtheactualET(ETc act)andasthedifferencebetweenTc
andactualtranspiration(Ta).Thosevaluesmaybeusedwiththe
Stewart’swater-yieldmodel(Stewartetal.,1977),thusadoptinga
simple,linearcrop-waterproductionfunctionthatrelatesseasonal
ETdorTdwiththerelativeyieldloss(RYL)throughan
appropri-atewater-yieldfactor,Ky,asreportedbyParedesetal.(2014a)for
maize.
AmorecomplexmodellingapproachisusedintheFAOcrop
yield model AquaCrop (Raes et al., 2012; Steduto et al., 2012),
thatsimulatescropbiomassandyieldinresponsetowater and
otherabioticstresses(temperature,fertilization,salinityandCO2).
AquaCropusesanempiricalapproachtoestimateTcandEs
depend-inguponthecanopycovercurve,whichisdifferentfromtheFAO
dualKcapproachdescribedinFAO56(Allenetal.,1998).Itisbased
uponaKccurvethatdoesnotrelatewiththecommonFAOKccurve
adoptedinFAO24andFAO56(DoorenbosandPruitt,1977;Allen
etal.,1998).AquaCropperformsadailysoilwaterbalanceand
esti-matesROalsousingtheCNmethod.DifferentlyfromSIMDualKc,
itusesasemi-empiricDPestimationprocedurethatrequiresthe
knowledgeofthesaturatedhydraulicconductivity,Ksat,
through-outthesoilprofile(Raesetal.,2006).
TheAquaCropmodelwasalreadyappliedtoseveralannualfield
cropsbutonlyafewapplicationsanalysedthemodelbehaviour
relative tothesoil water and evapotranspiration, e.g.,Farahani
etal.(2009)forcotton,Katerjietal.(2013)fortomatoandParedes etal.(2014b)formaize.However,assessmentsofETpartitioning
inAquaCroparelimited(Pereiraetal.,2015b)andthereareno
assessmentsforthemodel’sabilitytopredictsoilevaporation.
Consideringtheabovediscussionsonthepossible
appropriate-nessandlimitationsoftheAquaCropmodel,aswellastheprevious
resultsobtainedwhenusingtheSIMDualKcandStewart’s
mod-ellingapproachesdescribedbyWeietal.(2015),theobjectivesof
thepresentstudyare:(1)assessingtheperformanceofAquaCrop
forsoybean yieldestimation whenusing calibratedand default
parameters;(2)totesttheAquaCropabilitytopartitioncropET
incomparisonwiththeFAOdualKc approach;(3)toassessthe
abilityofAquaCroptosimulatesoilevaporationcomparingwith
observationsperformedwithmicrolysimetersalongfoursoybean
seasons;and(4)tocomparetheAquaCropapproachesusedforsoil
waterbalanceandsoilevaporationestimationwiththeonesused
bythesoilwaterbalanceSIMDualKc.ThesamedatausedbyWei
etal.(2015)areusedinthisstudy.
2. Materialandmethods
2.1. Experimentalsitecharacterisationandobservations
Soybean(GlycinemaxL.)experimentswereperformedatthe
Irrigation Experiment Station of the China Institute of Water
Resources and Hydropower Research (IWHR)locatedat Daxing
(39◦37N, 116◦26E, and 40.1m altitude). The soybean variety
ZhonghuangNo. 13wassownusingconventional tillagewitha
plantdensityof15plantsm−2 andaninter-rowspacingof0.4m.
This variety is a high-yielding semi-determinate cultivar that
belongstothematuritygroupIIandtakesanaverageof96days
toreachfullmaturity(Wangetal.,2013).Theexperimentswere
performedfrom2008to2011,withsowingbymid-Juneand
har-vestingbyearlyOctober.FurtherinformationisprovidedbyWei
etal.(2015).
Theclimateintheexperimentalsiteissub-humidofmonsoon
type,withcoldanddrywinterandhotandhumidsummer,which
isclassifiedasDwaaccordingtotheKöppenclassification(Kottek
etal.,2006).Climaticdatausedinthestudywerecollectedfroman
automaticmeteorologicalstationinstalledinsidetheexperimental
station.Dailydatausedincludedprecipitation,maximumand
min-imumairtemperature,relativehumidity,globalandnetradiation
andwindspeedat2mheight.Theclimaticdatasetswerechecked
forqualityassessmentasrecommendedbyAllenetal.(1998).The
referenceET(ETo)wascomputedwiththeFAOPenman–Monteith
method(Allenetal.,1998).Table1presentstheclimatic
charac-terizationofthefourcropseasonsfortheperiod2008to2011.
Detailedinformationonweatherdatarelativetothefouryearsof
observationsisprovidedbyWeietal.(2015).Climaticdata(Table1)
suggestthatdifferencesamongyearsweresmallexceptrelativeto
precipitation,whichwashigherin2009and2011.
Thesoils in theexperimentalfield aresilty soils formedby
depositsoftheloessformations.By2007,fourundisturbedsoil
samplesof250cm3foreachsoillayertoadepthof1mwere
col-lectedinvariousplotstodeterminethesoilwaterretentioncurve
andthehydraulicconductivitycurveinlaboratory.Theku-pf
appa-ratus(Umwelt-Geräte-Technik,Müncheberg,Germany)wasused.
Averagedvalues ofbasicsoilhydraulicpropertiesarepresented
inTable2.TheKsat valuesareintherangeofthoseproposedby
Rawlsetal.(1998)andRaesetal.(2012)forsiltloamsoils,
how-evertheyarehigherthanthoseformerlyobservedintheregion
(Pereiraetal.,2003).Capillaryrisefromthegroundwaterwasnot
consideredbecausetheaveragegroundwatertablewasdeep,near
18m,inallfouryearsofobservations.
Theirrigationschedulesweresetusingtwosoilwater
thresh-olds,of75%and60%ofFC,respectivelytreatmentT1andT2.Thus,
irrigationwasperformedwheneverthesoilwatercontentreached
thosethresholds.Treatmentswereperformedwiththree
replica-tionsinplotsof30m2 each.Sincethecropdevelopsduringthe
monsoonrainyseason,lowerirrigationthresholdscouldnotbe
selected.In2009and2011,duetoabundantrainfall,no
distinc-tioncouldbemadebetweentreatmentsthus,resultinginatotalof
sixdatasets.
Fieldobservationsincluded:
(a)Thedatesofeachcropgrowthstage(Table3).
(b)The leaf area index (LAI, cm2cm−2), that was measured
Table1
Averagemonthlyweatherdatarelativetothesoybeanseason,years2008–2011.
Year Jun Jul Aug Sep Oct
Max.airtemperature,◦C 2008 27.7 31.5 30.3 25.9 20.2
2009 31.5 31.4 29.8 25.7 21.2
2010 29.3 32.7 30.7 25.5 18.5
2011 32.0 31.6 30.4 25.2 19.5
Min.airtemperature,◦C 2008 17.4 21.7 20.3 14.0 6.0
2009 17.5 21.1 20.2 14.2 6.3
2010 18.2 23.2 20.1 14.0 6.1
2011 18.8 22.1 20.6 12.3 6.0
Min.RelativeHumidity,% 2008 63 64 66 64 54
2009 44 64 69 65 47 2010 59 63 63 62 57 2011 48 66 70 57 60 Solarradiation,MJm−2d−1 2008 21.4 21.5 19.8 15.9 11.4 2009 22.5 21.5 19.4 14.9 12.1 2010 22.3 19.4 19.0 15.2 11.2 2011 22.7 20.9 20.1 16.0 11.4 ETo,mm 2008 118 132 113 72 42 2009 147 140 113 70 52 2010 132 135 115 72 43 2011 146 137 119 81 40 Precipitation,mm 2008 103 66 64 123 42 2009 81 206 104 21 6 2010 87 42 141 69 32 2011 72 210 72 33 10 Table2
AveragesoilhydraulicpropertiesofDaxingexperimentalfields.
Layer Depth(m) sat(cm3cm−3) FC(cm3cm−3) WP(cm3cm−3) Ksat(cmd−1)
1 0.00–0.10 0.46 0.32 0.09 97.1
2 0.10–0.20 0.46 0.34 0.13 91.5
3 0.20–0.40 0.47 0.35 0.10 98.5
4 0.40–0.60 0.45 0.33 0.11 88.0
5 0.60–1.00 0.44 0.31 0.16 37.1
FC,WPandsatrepresentthesoilwatercontentatfieldcapacity,wiltingpointandsaturationrespectively;Ksatisthesaturatedhydraulicconductivity.
(AccuPAR LP-80,Decagon Devices) and a leaf area scanner
(F915900model,Cannon).
(c)Therootdepth(Zr,m),observedinrandomlydistributedplants;
byfullsoilcover,Zrreached1.0mdepth,henceinagreement
withdatareportedbyYan(2007).
(d)Thesoilwatercontent,thatwasmonitoredwithapreviously
calibrated TDRsystem(TRIME®-T3/IPH,IMKO GmbH).
Mea-surementswereperformedateach0.10muntilthemaximum
depthof1.0m,withthreereplications,everyfivedays.When
precipitation or anirrigationevent occurred, thesoil water
content wasmeasuredin the following day.For simulation
purposes,thesoilwatercontentwasconvertedintoavailable
soilwater,whichisthedepthofwaterstoredinthesoilabove
thewiltingpoint(ASW,mm).
(e)Soilevaporation,thatwasmeasuredusingtwomicrolysimeters
ineachplot;themicrolysimetersweremadeofPVCwithan
internaldiameterof0.10mandadepthof0.17m.Weighing
wasperformedeverydayaroundsunset,whenenergyavailable
forevaporationandtranspirationwasreduced.
(f) Thefinal crop biomassand yield, that were determinedby
harvestingplantsamplesineachplot,withatotalof3
sam-plespertreatment.Thesampleswereplacedinrefrigerated
Table3
Soybeancropgrowthstagesdatesandcumulativegrowingdegreedays(CGDD)forallexperimentalyears.
Cropgrowthstages Soybean
2008 2009 2010 2011
Initial Dates 24-06to13-07 14-06to09-07 25-06to18-07 22-06to07-07
CGDD(◦C) 372 463 468 302
Cropdevelopment Dates 14-07to07-08 10-07to31-07 19-07to20-08 08-07to07-08
CGDD(◦C) 910 948 1163 969
Mid-season Dates 08-08to16-09 01-08to09-09 21-08to19-09 08-08to13-09
CGDD(◦C) 1642 1690 1713 1676
LateSeason Dates 17-09to09-10 10-09to02-10 20-09to08-10 14-09to02-10
containersuntiltheywereweighedin thelab toobtainthe
freshweightand,later,ovendriedat65±5◦Ctoobtainthedry
weight.
Furtherdescriptionof theexperiments andfurtherobserved
data,namelyrelativetorootdepthsandirrigationschedules,are
presentedbyWeietal.(2015).
2.2. Modellingapproaches.ETpartitioningandyieldprediction
TheAquacropmodel(Stedutoetal., 2012;Raesetal., 2012)
isbasicallyacropyieldmodelthatcomputesbiomassandyield
consideringtheactualtranspiration(Ta,mm).Itseparately
com-putesTaandEsusingadailytimestep.ThedailyactualcropET
(ETcact,mmday−1)isobtainedasthesumofTaandEs,whichare
respectivelycomputedas
Ta=KsCC∗Kc,TrxETo (1)
and
Es=Kr(1−CC∗)KexETo (2)
where, ETo is reference evapotranspiration (mmday−1), Kc,Tr x
is the maximum standard crop transpiration coefficient
(non-dimensional),ormaximumbasalcropcoefficientwhenCC=100%;
CC*istheactualgreencropcanopycover(%)adjustedfor
micro-advectiveeffects;Ks(0–1)isthewaterstresscoefficient;Kex is
thesoilevaporationcoefficientforfullywetandnon-shadedsoil
surface(non-dimensional);andKr(0–1)istheevaporation
reduc-tioncoefficient(Raesetal.,2012).Thus,thecomputationofboth
ETcomponentsismainlytiedtothesimulatedcropcanopycover
(internallyadjusted for micro-advective effects),CC*.Thisis an
approachverydifferentfromFAO56(Allenetal.,1998),thusfrom
theapproachusedwithSIMDualKc(Rosaetal.,2012)despitethis
modelestimatesKcbwithadensitycoefficientKdthatisafunction
ofthefractionofgroundcoveredorshadedbythecrop(fc,
non-dimensional)andofthecropheight(h,m)asdescribedbyAllen
andPereira(2009).
ThesoilevaporationcoefficientiscomputedinAquaCropusing
adeepmodificationoftheRitchie’stwostagesapproach,including
askinlayermodification(Raesetal.,2012).StageIisdetermined
bytheavailableenergyatthesoilsurface,thusnotlimitedbythe
evaporablewateravailableinthesurfacesoillayer;Esisthenat
itspotentialrateandit is assumedthat waterevaporates from
athinsoilsurfacelayerwith0.04mdepththatdirectlycontacts
withtheatmosphere(Raesetal.,2012).Whenwaterisevaporated
fromthisthinsurfacelayer,anupwardsfluxfromthesoillayer
underneathoccursandevaporationisinstageII.Atthisstage,
evap-orationislimitedbythesoilwateravailabilityandthesoilhydraulic
propertiesdeterminingthetransferofwaterfromtheunderneath
soillayertotheevaporativesurfacelayer.Thesoilwatercontent
intheunderneathsoillayerprogressivelydecreasesresultingin
a decreaseof theevaporation ratewithtime. AquaCrop uses a
mechanisticapproachtodescribethefallingevaporationrate
dur-ingStageIIwhichmakesKrcomputedasafunctionoftimeand,
amongotherfactors,oftheamountofwaterextractedbythecrop
rootsfromtheuppersoillayerandofthedeclineoftheunsaturated
hydraulicconductivitywiththedecreaseofthesoilwatercontent.
Anexponentialequationisusedtorelatetheevaporation
reduc-tioncoefficientKrwiththerelativewatercontentofthesurface
soillayer,whichdependsonadeclinefactor(fK)andoftherelative
watercontentofthesoillayer(Wrel)throughwhichwatermoves
totheevaporativesurfacelayer.ThedefaultvaluefK=4isprovided
inthemodelmanual,aswellasofthethicknessofthesurfacelayer
(0.15mincludingtheskinlayer).However,accordingtoRaesetal.
(2012),whenWreldropsbelowathresholdof0.4,thethicknessis
expandedtoadepthsetbydefaultas0.3m.Theapproachis
rea-sonablebuthasnotbeenprovedinpracticeormentionedinany
modelusers’publication.Thereisnoreferencetotheneedfor
cal-ibrationoftherequiredparametersorrelativetotheircalibration
bytheusers.
SIMDualKc,contrastingly,computesKewiththeRitchie’smodel
andbyperformingadailysoilwaterbalanceoftheevaporablelayer
(Allenetal.,2005;Rosaetal.,2012).InSIMDualKc,theevaporable
layerischaracterizedbyitsthickness(Ze,m),thatisassumedto
varyfrom0.10to0.15m,thetotalevaporablewater(TEW,mm),
whichisthemaximumdepthofwaterthatcanbeevaporatedfrom
thatsoillayerafterithasbeenfullywetted,andthereadily
evap-orablewater (REW,mm),which isthedepthofwater thatcan
beevaporatedwithoutwateravailabilityrestrictions,i.e.,during
stageIofsoildrying.Furthermore,theevaporationcoefficientKeis
maximumwhenthesoiliswetbutislimitedbytheenergy
avail-ableatthesoilsurface,thusitsvaluecannotexceedthedifference
Kc max−Kcb,betweenthemaximum dailyKc andthebasal crop
coefficientatthesame day.As thetopsoil dries andless water
isavailablefor evaporation,areduction inEs occurs in
propor-tiontotheamountofwaterremaininginthesurfacesoillayer,
whichisusedtodefinetheevaporationreductioncoefficient(Kr,
non-dimensional)asdescribedbyAllenetal.(1998,2005).Due
toadoptingdifferentapproaches,Krisobviouslydifferentinboth
models.
AccordingtoRaesetal.(2012),themodelestimatestheCC
evo-lutionalongtimeinthreedistinctphases:(a)thefirstphasebegins
atthecropemergenceandendswhenhalfofthemaximumCC
(CCx)isattained;(b)thesecondphasefollowsthepreviousone
untilCCxisreached;and(c)thethirdphasebeginswhen
senes-cencestartsandendsatharvest.Anexponentialfunctionoftime
isusedinthe1stand2ndphases,beginningwiththecanopycover
when90%emergencehasoccurred(CCo)andusingagrowthrate
definedbythecanopygrowthcoefficient(CGC).Thethirdandlast
phasereferstothedecline ofgreencanopycover anditsshape
isdefinedbythecanopy declinecoefficient(CDC).Therefore,to
parameterizetheCCcurvesitisnecessarytousethefourreferred
parametersCCo,CCx,CGCandCDC.Incontrast,insteadof
simu-latingCC,SIMDualKcadoptstheobservedfcfractionofsoilcover
byvegetationasdefinedFAO56(Allenetal.,1998).Inadditionto
thereferredadjustmentofKcbvaluesthroughadensitycoefficient,
fc isalsousedtolimitKe(Ke≤fewKc max)wherethefractionof
soilwettedandexposedtosolarradiation(few,non-dimensional),
fromwheresoilevaporationoriginates,isdependingonfc(Allen
etal., 1998,2005).The approachesused inAquaCrop and with
thedualKc methodologyusedinSIMDualKcaretherefore
differ-ent.
AquaCropestimatestheabovegrounddrybiomass(B,tha−1)by
multiplyingthewatertranspiredbythecropalongtheseason(Ta)
bytheadjustedbiomass(water)productivity(BWP*,gm−2).BWP*
representstheabovegroundbiomassproducedperunitofland
areaconsideringboththecumulativetranspiration,after
adjust-mentforatmosphericCO2concentrationandETo(Raesetal.,2012).
Theactualyield(Ya,tha−1)ispredictedbythemodelusingthe
ref-erenceharvestindex(HIo,%)andB.TheHIoisadjustedwhenwater
stressoccursbyafactorintegratingfivewaterstressfactors
rela-tivetotheinhibitionofleafgrowth,inhibitionofstomata,reduction
ingreencanopydurationduetosenescence,reductioninbiomass
duetopre-anthesisstress,andpollinationfailure(Raesetal.,2012).
Relativetoyieldestimation, differencesbetweenAquaCrop and
theStewart’smodelareenormousbecausethelatterusesjustan
empiricalrelationbetweentherelativeyieldlossandtheseason
ETdeficitwhileAquaCropusesadeterministicapproachasbriefly
describedabove.
FurtherdescriptionsoftheAquaCropmodelandauxiliary
2.3. Modelparameterization,calibrationandvalidation
TheAquaCropmodelusesalargenumber ofparameters
rel-ativetocropcharacteristicsand effectsofcropmanagement on
thecropgrowth.Someofthoseparameters,namedconservative
parameters,areexpectedtochangelittlewithtime,with
manage-mentconditionorwiththelocation.Theyareidentified,described
andtabledbyRaesetal.(2012).Thesetabledvalueswereusedto
startthesimulationsandwereadjustedusingfieldobservations.
BecauseCCdeterminesthepartitionofETintocroptranspiration
andsoilevaporation(Eqs.(1)and(2)),theparameterizationfirstly
focusedontheCCcurve,i.e.,theparametersCCo,CCx,CGCandCDC.
TheCCmeasuredvalueswerederivedfromtheobservedleafarea
index(LAI)andcropheightusingtheapproachproposedbyAllen
andPereira(2009),previouslyusedwithSIMDualKc(Weietal., 2015).CCowasobtainedfromplantdensityafteremergence,CCx
wasthemaximumCCobservedfornowaterstressconditions,and
atrialanderrorprocedurewasusedforCGCandCDC.Thetrial
anderrorprocedurewasinitiatedusingparametervaluestabled
byRaesetal.(2012)andendedwhenthesimulatedCCcurvefitted
theobservedCCvaluesasdescribedbyParedesetal.(2014b)ina
previousapplicationtomaize.Subsequently,atrialanderror
pro-cedurefocusedonadjustingtheKc,Tr xbycomparingsimulatedand
observedfielddataofavailablesoilwater.Inthisapplication,the
initialvalueforKc,Tr xwas1.10asproposedbyRaesetal.(2012).
ThemodelwascalibratedwiththedataofTreatment1in2008and
wastestedwiththeremainingdatacollectedinthecropseasons
of2008to2011.
Tosimulatesoilevaporation,avalueforthereadilyevaporable
water(REW,mm),wasobtainedfromsoiltexturalandhydraulic
propertiesofthesoilasdefinedbyAllenetal.(1998).Toestimate
runoff,thecurvenumber(CN)methodisusedbythemodel;theCN
valueswereobtainedfromtabledvaluesproposedbyAllenetal.
(2007)forsoilswithmediumtexturewhosepreviouscropwasa
cereal.REWandCNwerenotobtainedthroughthetrialanderror
procedurebecausetheywereavailablefromtheprevious
calibra-tionofthemodelSIMDualKc(Weietal.,2015).
The reference harvest index HIo was obtained from yield
data observations performed in all seasons, without
occur-ring water stress; it resulted the average value HIo=0.38.
The retained HIo value is within the range of HIo values,
0.30–0.43, reported by Donatelli et al. (1997) and is only
slightly lower thanvalues reported byCui and Yu(2005) with
HIo=0.41.
TheBWP*wasobtainedusingatrialanderrorprocedureaimed
atminimizingdifferencesbetweenpredictedandobservedabove
grounddrybiomass.BWP* wassetat17gm−2,whichisslightly
higherthan thetabledvalueproposedbyRaes etal.(2012)for
soybean[15gm−2],andtherangeofvaluesproposedbySteduto
etal.(2012)[12–16gm−2];howeveritiswithintherangeofvalues
proposedbythesameauthorsforC3plants.
The“goodness-of-fit”relativetoboththecalibrationand
vali-dationprocesseswasassessedusingasetofindicatorsdescribed
previously(Weietal.,2015).Alinearregressionforcedthroughthe
originwasperformedtocompareobservedand simulated
(pre-dicted)values,OiandPi,whosemeansarerespectivelyOandP;
theregressioncoefficient(b)andthedeterminationcoefficient(R2)
werethenassumedasmainstatisticalindicators.Toanalysethe
residualestimationerrorsasetofindicatorswerealsoused,mainly
therootmeansquareerror(RMSE)toexpressthevarianceoferrors
(Bowermannetal.,2005),andtheaveragerelativeerror(ARE)to
indicatetheaveragesizeoftheestimatederrors.Theseindicators
Table4
ConservativeandcalibratedcropparametersofAquaCropmodel.
Description Unitsorsymbolmeaning Value
Conservativeparameters Default*
Basetemperature ◦C 5
Cut-offtemperature ◦C 30
Canopycoverat90% emergence(CCo)
cm2perplant 5.00
Soilwaterdepletionthreshold forcanopyexpansion
Upperthreshold 0.15
Soilwaterdepletionthreshold forcanopyexpansion
Lowerthreshold 0.65
Shapefactorforwaterstress coefficientforcanopy expansion
Curveshapemoderately convexcurve
3.0
Soilwaterdepletionthreshold forstomatalcontrol
FractionofTAWatwhich stomatastarttoclose
0.50 Shapefactorforwaterstress
coefficientforstomatal control
Highlyconvexcurve 3.0
Soilwaterdepletionthreshold forfailureofpollination
FractionofTAWatwhich pollinationstartstofail
0.85
Calibratedparameters Default* Calibrated
Cropcoefficientfor transpirationatCC=100%
Basalcropcoefficient(Kc,Trx) 1.10 1.12
BWP* Biomass(water)productivity
adjustedforEToandCO2
(gm−2)
15 17
HIo Referenceharvestindex(%) 0.40 0.38
Canopycovercurveparameters Default* 2008 2009 2010 2011
Maximumgreencanopycover, CCx,
% 99 98 98 98 98
Canopygrowthcoefficient,CGC %GDD−1 0.45 0.71 0.62 0.72 0.74
Canopydeclinecoefficient,CDC %GDD−1 1.50 1.04 0.870 1.04 1.50
*DefaultparametersaretabledbyRaesetal.(2012).
arecomputedfromthepairsofobservedandpredictedvaluesOi andPi(i=1,2,...,n)as RMSE=
⎡
⎢
⎢
⎢
⎢
⎣
n i=1 (Pi−Oi)2 n⎤
⎥
⎥
⎥
⎥
⎦
0.5 (3) and ARE=100n n i=1Oi−Pi Oi
(4)
Theseindicatorswerecalculatedateachiterationofthetrialand
errorproceduretosupportfindingofthecalibratedparametersthat
leadtotheminimizationoftheestimationerrors.
In addition, an indicator of the quality of modelling was
used,theNashandSutcliffe(1970)modellingefficiency(EF,
non-dimensional),thatisanormalizedstatisticwhichcorrespondsto
theratiobetweentheresidualvarianceandtheobservations
vari-ance EF=1.0− n
i=1 (Oi−Pi)2 n i=1 Oi−O 2 (5)ThetargetvalueforEFis1.0,whileanullornegativevalue
indi-catesthatthemeanofobservationsisasgoodorabetterpredictor
thanthemodel.
3. Resultsanddiscussion
3.1. AquaCropmodelparameterization,calibrationand
validation
Aspreviouslyreferred,modelcalibrationwasperformedusing
dataof2008-T1byminimizingthedifferencesbetweenobserved
andsimulated canopy cover(CC, %), availablesoil water (ASW,
mm),soilevaporation (Es,mm),biomass (B,kgha−1)and yield
(Ya,kgha−1).Duetotheinterdependenceofthecalibratedmodel
parameters,themainfocuswastheaccuracyofBandYa
predic-tions.TheparameterizationoftheCCcurvewasfirstperformed.
Table4presentsthedefaultvalues(Raesetal.,2012)ofthemain
modelparametersusedtoinitiatethemodelapplication,aswellas
thecalibrated valuesofthepreviouslyreferredparametersCCo,
CCx,CGC,CDC,Kc,Tr x,BWP*andHIo.
AspreviouslydiscussedconsideringEqs.(1)and(2),an
accu-rateparameterizationoftheCCcurveisofmajorimportancefor
appropriatecomputationofTaandEsand,subsequently,
estimat-ingbiomassandyield.SpecificCGCandCDCwereobtainedforeach
year(Table4).Differencesamongyearsrelatewiththeobserved
differencesinCGDDrequiredbythecroptoreachthecrop
devel-opmentandmid-season stages(Table3).Selectedresultsofthe
fittedandthedefaultCCcurvesarepresentedinFig.1.Relativeto
thecalibratedCCcurves,resultsshowthatthemodeltendsto
over-estimateCCduringtheinitialandcropdevelopmentstagesandto
slightlyunder-estimateCCduringthemid-seasonin2008and2009
(e.g.Fig.1a)buttoover-estimatein2010and2011(e.g.Fig.1b).This
mayrelatewiththefactthatAquaCropinternallyadjustedthe
cali-bratedCCx(Table4)to97.9%forbothtreatmentsin2008and2011,
to90.8%in2009,to93.9%and95.2%respectivelyforT1andT2in
2010.Raesetal.(2012)explainsinternalmodeladjustmentsdue
tostress,butstresswasnotobservedorcomputedbythemodel
inthepresentstudy.Ifdefaultparameters(CCx,CGCandCDCin
Table4)areusedthereisacleartrendforunder-estimationofthe
observedCCvaluesmainlyduringtheinitial,cropdevelopmentand
midstageperiods.
The“goodness-of-fit”indicatorsrelativetotheCCcurveswhen
usingdefaultandcalibratedparametersarepresentedinTable5.
AfteraccuratecalibrationoftheCCcurveparameters(CCo,CCx,
CGCand CDC), resultsdo not show anyclear trendto over or
under-estimation,withbrangingfrom0.96to1.04(Table5).The
determinationcoefficientsarehigh(R2>0.91),indicatingthatthe
CCmodelisabletoexplainthevarianceofobservedCCvalues.EF
valuesarealsohigh(≥0.91),showingthatthevarianceofresiduals
wasmuchsmallerthanthemeasureddatavariance.Consequently,
theestimationerrorsarelow,withRMSErangingfrom4.3%to7.4%
(ofcanopycover)andAREnotexceeding10.2%.Ifdefault
param-eters(CCx,CGCandCDCinTable4)areusedthereisacleartrend
forunder-estimationof theobserved CCvalues, withbranging
from0.73to0.86.Estimationerrorsarehigh,withAREexceeding
26.7%,thusdefinitelynonnegligible.Themodelefficiencyvalues
arelow,includinganegativevalue,−0.71,in2011,thuswhenb
isthesmallestand errorsarethehighest.Thelessgood results
obtainedwhenusingdefaultparametersrelatewiththenatural
variabilityofcropgrowth(Table3),whichrelateswiththe
vari-abilityofclimaticconditions(Table1)despitetheseweregenerally
smallexceptforrainfall.
Resultsdiscussedaboveindicatethattheuseofdefaultvalues
mayleadtolargeinaccuraciesinthecomputationsthatusethe
CCcurves.Therefore,thereisaneedforanappropriaterevision
a)
b)
0 10 20 30 40 50 60 70 80 90 100 24 /0 6 01 /0 7 08 /0 7 15 /0 7 22 /0 7 29 /0 7 05 /0 8 12 /0 8 19 /0 8 26 /0 8 02 /0 9 09 /0 9 16 /0 9 23 /0 9 30 /0 9 07 /1 0 Ca no py co ve r(%) 0 10 20 30 40 50 60 70 80 90 100 25 /0 6 02 /0 7 09 /0 7 16 /0 7 23 /0 7 30 /0 7 06 /0 8 13 /0 8 20 /0 8 27 /0 8 03 /0 9 10 /0 9 17 /0 9 24 /0 9 01 /1 0 08 /1 0 Ca no py co ve r( % )Fig.1.Bestandworsesimulatedcanopycover(CC)curveswhenusingdefaultparameters(—)andaftercalibration( )comparedwithobservedvalues( ):(a)T1in2008, and(b)T2in2010.
Table5
“Goodness-of-fit”indicatorsrelativetothecanopycovercurveusingdefaultandcalibratedparameters.
Yearandtreatment Numberofobservations b R2 RMSE(%) ARE(%) EF
Usingdefaultparameters 2008,T1 13 0.81 0.93 20.0 31.5 0.49
2008,T2 13 0.80 0.93 20.3 31.8 0.48 2009 14 0.85 0.92 18.4 26.7 0.38 2010,T1 12 0.76 0.81 25.2 36.1 0.23 2010,T2 11 0.86 0.91 16.4 26.7 0.72 2011 11 0.73 0.74 30.1 37.6 −0.71 Usingcalibrated parameters 2008,T1 13 1.01 0.94 7.4 10.2 0.93 2008,T2 13 1.01 0.94 7.2 9.9 0.94 2009 14 0.97 0.95 5.7 9.0 0.94 2010,T1 12 0.96 0.99 5.3 9.4 0.97 2010,T2 11 1.04 0.99 4.3 7.2 0.99 2011 11 1.00 0.99 5.4 7.9 0.95
oftheCCdefaultvaluesbythemodeldevelopers.Meanwhile,it
isadvisablethatatleastacoupleofCCcurvesshouldbelocally
calibratedandrelatedresultsusedasdefaultparameters.
3.2. Simulationoftheavailablesoilwater
Selectedresultsrelativetocomparingtheobservedand
sim-ulatedASWafterappropriatemodelcalibrationarepresentedin
Fig.2.Alsoincludedthevaluessimulatedwiththedefault
param-eterslistedinTable4.ASWwereobservedandsimulatedforthe
maximumrootdepth.ThetargetupperlimitofASWisthethetotal
availablesoilwater(TAW,mm)thatcorrespondstotheASWstored
atfieldcapacityin therootzone,andthetargetlowerlimitfor
ASWwithoutwaterstressisthereadilyavailablewater,RAW=p
TAW,wherepisthedepletionfractionfornostress(Allenetal.,
1998).Inthisapplication,pwaspreviouslycalibratedwith
SIMD-ualKc(Weietal.,2015).ResultsinFig.2showthatASWgenerally
variedbetweenRAWandTAW,thusevidencingthatonlynegligible
waterstressmayhaveoccurred.Thisisduetothefactthatirrigation
treatmentsT1andT2weredesignedfordepletionfractionssmaller
thanp.
Fig.2showsthatthemodeltendstounder-estimateASWduring
mostofthemid-seasonandendseasonandtoover-estimateASW
duringcropdevelopmentandpartofthemid-season.This
behav-iorisparticularlyevidentfor2010and2011(Fig.2candd).Results
indicateabiasintheestimationoftheASWalongtheseason.Bias
inthesoilwatersimulationbyAquaCropwasidentifiedin
stud-iesrelativetobarley(Pereiraetal.,2015b),cotton(Farahanietal.,
2009)andmaize(Paredesetal.,2014b).TheanalysisbyPereira etal.(2015b)suggeststhatthereferredbiasedestimationofASW
islikelyduetoAquaCropabandoningtheFAOKc andKcbcurve,
changingtheprocedurestocomputeKe,andthusabandoningthe
FAOdualKcapproachtomakeTaandEstodependtoomuchfrom
theCCcurve.ThisisanalyzedinSection3.3throughcomparing
theKcbandKecurvesrelativetoAquaCropandtoSIMDualKcwhen
usingtheFAOdualKcapproach.Resultsbecameworsewhenusing
defaultparametersin2008,butnotthosefor2009,2010and2011.
OnecannotfindareasonforthatwhenknowingthattheCCcurves
simulatedwithdefaultparametersshowmuchlargererrorsthan
those usingcalibrated values(Table5).This isprobably dueto
internaladjustmentsofthemodelnotopentotheusers.
Resultsofthe“goodness-of-fit”indicatorsrelativetotheASW
simulationswhenusingdefaultandcalibratedparametersare
pre-sentedinTable6togetherwithresultsobtainedbyWeietal.(2015)
whenusingSIMDualKc.AfterAquaCropcalibration,the
determina-tioncoefficientsvaryinawiderange,indicatingthatthevariability
oftheASWobservationsisnotwellcapturedbythemodel.Errors
aregenerallyrelativelysmallbutRMSEmayreach24mmandARE
canattain 14.1%.Theregression coefficients arearound1.0 but
thesebvaluesmasktheover-estimationofASWinthefirsthalf
of theseasonand theunder-estimationinthesecondhalf.The
worseresultsarefor2011whenEFtakesanegativevalue,thus
indicatingthattheresidualsvariancewaslargerthanthemeasured
datavariance;onthecontrary,theotherdatasetshadpositiveEF
indicating thattheresiduals variance waslowerthan the
mea-sureddatavariance. Differently,the“goodness-of-fit”indicators
relativetoSIMDualKcareoverall betterthanthosereportedfor
AquaCrop,particularlywithquitegoodresultsfor2010and2011
whenAquaCropresultsareworse(Fig.2andTable6).Thus,thepoor
simulationresultsrelativetothesedatasetsarenotduetosome
peculiardatacombinationbuttosomeinsufficiencyinmodelling,
likelyduetoabandoningtheFAOwellprovenapproachesandto
internalmodeladjustmentsthatchangetheparametervalues.
Whenusingdefaultparameters,someindicatorsareworseand
otherarebetterthanusingcalibratedparameters(Table6).This
factistiedwithproblemsofcalibrationbecause,asdiscussedby
Pereiraetal.(2015b),thecalibratedparameterssuchasKc,Tr x,or
theCCxparameters,areinternallychangedbythemodel.Thisfact
identifiesadifficultyintheuseofthemodelbecausetheuserhas
nocontrolontheparameterizationand/orcalibrationprocesses.
Thus,overall,theindicatorsof“goodness-of-fit”oftheAquaCrop
modelusingcalibratedordefaultparametersfailedthelimitsfor
R2andEF,0.80and0.70respectively,proposedbyMaetal.(2011)
foragriculturalmodels.
3.3. ETpartitioningandevaporationandtranspirationcrop
coefficients
Fig. 3 presents examples of the daily variation of the crop
coefficients under standard conditions (Kcb and Kc,Tr), adjusted
towater stress(Kcb act and Kc,Tr act)as wellas theevaporation
coefficient(Ke)relativetothesamedatasetsofFig.2estimated
withAquaCropandSIMDualKc.ThepicturesrelativetoAquaCrop
includethecropcoefficientandevaporationcurvescomputedwith
calibratedanddefaultparametersaimingatexplainingthe
simu-lationresultsinFig.2.Rainfallandirrigationarerepresentedinthe
picturesrelativetotheKcbandKecurvescomputedwith
SIMDu-alKc.
Fig.3clearly showsthedifferencesin Kcb curves by
SIMDu-alKc,followingthelinearshapedcropcoefficientscurveassumed
inFAO56(Allenetal.,1998);incontrastwiththosederivedwith
AquaCrop, where theKc,Tr curves consistof curvilinear shaped
curvesproportionaltotheCCcurve(e.g.,Fig.1).Thus,Fig.3clearly
showthatAquaCropnearlyabandonedtheFAO“Kc–ETo”approach
despite,asdiscussedbyPereiraetal.(2015a),itssimplicity,
accu-racyandcommonapplication.
In AquaCrop, the crop coefficients curves are built using a
dailymodifiedKc,TrthatdependsupontheCCadjustedfor
micro-advectionandoftheKc,Tr x,thatistheKc,TrvaluewhenCC=100%.
)
b
)
a
c)
d)
e)
f)
g)
h)
0 50 100 150 200 250 24 /0 6 01 /0 7 08 /0 7 15 /0 7 22 /0 7 29 /0 7 05 /0 8 12 /0 8 19 /0 8 26 /0 8 02 /0 9 09 /0 9 16 /0 9 23 /0 9 30 /0 9 07 /1 0 AS W( m m ) TAW RAW 0 50 100 150 200 250 24 /0 6 01 /0 7 08 /0 7 15 /0 7 22 /0 7 29 /0 7 05 /0 8 12 /0 8 19 /0 8 26 /0 8 02 /0 9 09 /0 9 16 /0 9 23 /0 9 30 /0 9 07 /1 0 AS W (mm ) TAW RAW 0 50 100 150 200 250 14 /0 6 21 /0 6 28 /0 6 05 /0 7 12 /0 7 19 /0 7 26 /0 7 02 /0 8 09 /0 8 16 /0 8 23 /0 8 30 /0 8 06 /0 9 13 /0 9 20 /0 9 27 /0 9 AS W (m m ) TAW RAW 0 50 100 150 200 250 14 /0 6 21 /0 6 28 /0 6 05 /0 7 12 /0 7 19 /0 7 26 /0 7 02 /0 8 09 /0 8 16 /0 8 23 /0 8 30 /0 8 06 /0 9 13 /0 9 20 /0 9 27 /0 9 AS W( m m ) TAW RAW 0 50 100 150 200 250 25 /06 02 /07 09 /07 16 /07 23 /07 30 /07 06 /08 13 /08 20 /08 27 /08 03 /09 10 /09 17 /09 24 /09 01 /10 08 /10 ASW (mm) TAW RAW 0 50 100 150 200 250 25/ 06 02/ 07 09/ 07 16/ 07 23/ 07 30/ 07 06/ 08 13/ 08 20/ 08 27/ 08 03/ 09 10/ 09 17/ 09 24/ 09 01/ 10 08/ 10 AS W (mm ) TAW RAW 0 50 100 150 200 250 22 /0 6 29 /0 6 06 /0 7 13 /0 7 20 /0 7 27 /0 7 03 /0 8 10 /0 8 17 /0 8 24 /0 8 31 /0 8 07 /0 9 14 /0 9 21 /0 9 28 /0 9 AS W (mm ) TAW RAW 0 50 100 150 200 250 22 /0 6 29 /0 6 06 /0 7 13 /0 7 20 /0 7 27 /0 7 03 /0 8 10 /0 8 17 /0 8 24 /0 8 31 /0 8 07 /0 9 14 /0 9 21 /0 9 28 /0 9 AS W (mm) TAW RAWFig.2.Observed( )andsimulated( )dailyavailablesoilwater(ASW)withAquaCrop,ontheleft,comparedwithSIMDualKc,ontheright(datafromWeietal.,2015): (a,b)T1in2008,calibration;(c,d)2009,(e,f)T2in2010,and(g,h)2011(errorbarsrefertothestandarddeviationofASWobservations).Alsoincludedthesimulationwith AquaCropwhenusingdefaultparameters(—)
Fig.3. Seasonalvariationofthesoilevaporationcoefficient(Ke)andcropcoefficients(KcborKc,Tr)relativeto:(a,b)T1in2008,calibration;(c,d)2009,(e,f)T2in2010,and
Table6
“Goodness-of-fit”indicatorsrelativetosimulationsofASW(mm)withAquaCropwhenusingdefaultandcalibratedparametersandtoSIMDuaKcmodel(Weietal.,2015). Yearandtreatment Numberofpairs b R2 RMSE(mm) ARE(%) EF
AquaCrop Usingdefaultparameters 2008,T1 18 1.10 0.69 20.4 12.0 0.28
2008,T2 18 1.16 0.35 26.8 17.5 −1.11 2009 21 0.99 0.83 11.4 5.9 0.79 2010,T1 33 0.99 0.92 6.4 3.7 0.90 2010,T2 33 1.06 0.81 11.6 7.0 0.56 2011 32 1.06 0.70 19.9 10.2 −0.12 Usingcalibrated parameters 2008,T1 18 1.01 0.85 11.4 6.0 0.78 2008,T2 18 0.99 0.85 10.6 6.8 0.67 2009 21 0.98 0.86 10.7 6.1 0.82 2010,T1 33 0.94 0.82 12.3 6.5 0.64 2010,T2 33 0.95 0.22 15.6 10.0 0.19 2011 32 1.01 0.58 22.9 14.1 −0.47
SIMDualKc(Weietal., 2015) 2008,T1 18 0.99 0.93 10.7 6.0 0.80 2008,T2 18 0.98 0.84 12.5 8.4 0.54 2009 21 0.97 0.95 7.8 3.5 0.90 2010,T1 33 1.00 0.84 9.2 5.5 0.79 2010,T2 33 1.02 0.83 8.6 5.1 0.76 2011 32 1.03 0.86 9.3 4.8 0.76
aspreviouslydiscussed,themodelinternallyadjustsCCxwithout
apparentreason.InSIMDualKc,differently,theKcbcurvesarebuilt
takingintoconsiderationtheKcbvaluesfortheinitial(Kcb ini),mid
(Kcbmid)andend(Kcbend)stages,withKcbmidandKcbendadjusted
forclimate(Allenetal.,1998,2005)and forcropdensityusing
theobservedfractionofgroundcoverandplantheight(Allenand
Pereira,2009;Rosaetal.,2012).Thereferreddifferentapproaches
arethecauseforthedifferentcropcoefficientcurvesrepresented
inFig.3.
ThecalibratedKc,Tr xwas1.12(Table4),thatishigherthanKcb
mid=1.05obtainedbyWeietal.(2015)whencalibrating
SIMDu-alKcforthesameobservationdatasets.Thedifferenceisexplained
bythefactthatKcbmidreferstotheentiremidseasonwhileKc,Tr x
isthemaximumvalueofKc,TrforCC=100%.InFig.3Kcbmid<1.05,
becauseitisadjustedtoclimateandtocropdensity.AquaCropdid
notcomputeanystressresultingthatKc,TrandKc,Tr actcurvesare
coincident.Differently,SIMDualKcadjustedKcbforanyoccasion
whenastressoccurred,inthepresentapplicationforonlyshort
periods(Fig.3b,dandf).ResultsshowthatKc,Trduringthe
mid-seasonwashigherthanKcbduetothehigherKc,Tr xadopted.The
calibrationcouldhaveselectedasmallerKc,Tr xbutthiswouldlead
topoorresultsforbiomassandyieldpredictionssincethe
vari-ousmodelparametersareinterlinked.Inaddition,asdiscussedby
Pereiraetal.(2015b)theprocedurerelativetotheinternal adjust-mentofKc,Tr xforeverycrophavingCC<100%,shouldberevised
sinceitmakesthiscalibrationadifficulttaskwithinsufficient
con-trolbytheuser.
ThesoilevaporationcoefficientKe(Fig.3)ishighestduringthe
initialstage,continuestobehighbutdecreasesduringthecrop
developmentstage,and becomes small during themidseason,
whenthesoilis wellcovered bythecrop, increasingagainbut
fewduringthelateseason.Thisbehavioriscommontoboth
mod-elsandallcropseasons.Numerouspeaksareshownasresponses
tosoilwettingsbyprecipitation.However,therearedifferences
betweentheKevaluescomputedbybothmodels.Duringthe
ini-tialand cropdevelopmentstages,KefromAquaCrop arehigher
thanKefromSIMDualKc(Fig.3).Bythemid-andlate-season,when
themaximumfcorCCisattained,AquaCropcomputesgenerally
lowerKevaluesandlessKepeaksthanSIMDualKc,thusshowing
apoorreactiontotheprecipitationorirrigationeventsoccurring
then.ThereferreddifferencesinKeresultsalongthecropseasonare
verylikelyduetothedifferencesinmodellingapproachesused,as
discussedinSection2.2,includingdifferencesincalibration
param-eters,moreexigentincaseofSIMDualKc,aspreviouslydiscussed
forabarleyapplication(Pereiraetal.,2015b).
WhenusingdefaultparameterstheKc,TrandKevalueschange
relativetothosecomputedwhencalibratedvalueswereused.This
isexpectedsincebaseparametersusedincomputationsare
differ-ent;theKc,Trcurveisshifteddownandtotheright(Fig.3a,c,eand
g)similarlytotheshiftoftheCCcurve(Fig.1).TheKepeaksare
aboutthesameduringtheinitialstagebutareincreasedwhenthe
Kc,Trcurvesobtainedwithdefaultvaluesarebelowthosecomputed
withcalibratedparameters.However,thisbehaviordoesnotjustify
theupwardordownwardshiftsoftheASWsimulationwithdefault
parametersinFig.2a,c,e,andgorin“goodness-of-fit”indicators
inTable6.
Fig.3aandbshowsthatKc,Trvaluesareverydifferentfromthose
oftheKcbvaluesresultinginTa=Kc,TrETobeingmuchsmallerthan
Tc=KcbEToduringtheinitialandfirstpartofthecropdevelopment
stages,whenKefromAquaCropwerelargerthanforSIMDualKc.It
resulted(Ke+Kc,Tr)tobelikelysmallerthan(Kcb+Ke)duringthis
period,thusindicatinglesswaterusebythen,whichmayexplain
thecorrespondingover-estimationof ASW(Fig.2).For thelast
partofthecropdevelopmentandthemid-seasonKc,Tris larger
thanKcbwhileKeisreduced.Thus(Ke+Kc,Tr)becomeslargerthan
(Kcb+Ke)and theover-estimationof ASWturnsinto an
under-estimation. During thelate-season, (Ke+Kc,Tr) keep larger than
(Kcb+Ke).Therefore,thereisatrendforETcact=(Ke+Kc,Tr)ETotobe
initiallyunder-estimated,thusleadingtoasmallsoilwater
deple-tionand,therefore,toanover-estimationofASWduringthefirst
halfofthecropseason.ETc actbecomesprogressivelyhigherand
likelyover-estimatedinthesecondhalfofthecropseasonwhen,
duetoincreasedwaterdepletion,ASWbecomesunder-estimated.
ThisbehaviorisapparentforallcasesinFig.2butover-and
under-estimationsvaryinintensity,whichisnotexplainedwithresults
inFig.3.Moreover,thebehaviorofthemodelwhenusingdefault
parameters,which produceKc,Trcurves withmuchsmaller
val-ues,isnotexplainable.Itmaydependuponinternaladjustments
unknownfortheuser.In ourconditionofmodelusers,we just
considernecessaryadeepchangeinproceduresandabetter
con-troloftheparameterizationbytheusers.Otherwise,thesoilwater
computationsshouldbeusedonlyforBWP*calibrationaimingat
biomassandyieldpredictions.
3.4. Soilevaporationtesting
TheAquaCropmodelwastestedforsoilevaporationcomputed
withEq.(2) comparedwithmicrolysimeter observationsofsoil
evaporation(Fig.4).Es simulationsshow thatthe model tends
a)
b)
c)
d)
e)
0 1 2 3 4 5 6 7 24 /0 6 01 /0 7 08 /0 7 15 /0 7 22 /0 7 29 /0 7 05 /0 8 12 /0 8 19 /0 8 26 /0 8 02 /0 9 09 /0 9 16 /0 9 23 /0 9 30 /0 9 07 /1 0 Soi lev apor a on (mm) 0 1 2 3 4 5 6 7 24 /0 6 01 /0 7 08 /0 7 15 /0 7 22 /0 7 29 /0 7 05 /0 8 12 /0 8 19 /0 8 26 /0 8 02 /0 9 09 /0 9 16 /0 9 23 /0 9 30 /0 9 07 /1 0 Soi lev ap or a on (m m ) 0 1 2 3 4 5 6 7 14 /0 6 21 /0 6 28 /0 6 05 /0 7 12 /0 7 19 /0 7 26 /0 7 02 /0 8 09 /0 8 16 /0 8 23 /0 8 30 /0 8 06 /0 9 13 /0 9 20 /0 9 27 /0 9 Soil ev ap or a on (mm ) 0 1 2 3 4 5 6 7 25 /0 6 02 /0 7 09 /0 7 16 /0 7 23 /0 7 30 /0 7 06 /0 8 13 /0 8 20 /0 8 27 /0 8 03 /0 9 10 /0 9 17 /0 9 24 /0 9 01 /1 0 08 /1 0 Soil ev ap or ao n (m m) 0 1 2 3 4 5 6 7 22/06 29/06 06/07 13/07 20/07 27/07 03/08 10/08 17/08 24/08 31/08 07/09 14/09 21/09 28/09 So il eva po ra on (mm )Fig.4. Dailysoilevaporation(Es)dynamicsalongsoybeansseasonswhenusingtheAquaCropafteradequatecalibration( )andSIMDualKc(----)comparedwith
microlysimetersobservations()for:(a)T1in2008,calibration,(b)T2in2008;(c)2009,(d)T1in2010,and(e)2011(SIMDualKcdatafromWeietal.,2015)
during the mid- and late-season. This type of behaviour was
expectedbecausemicrolysimeterevaporationishigherthansoil
evaporationsincethelatterisaffectedbywateruptakebyroots
activeintheevaporativesoillayerwhilethoserootsdonotexistin
themicrolysimeters.AsimilarbehaviourwasobservedbyWeietal.
(2015)whenusingSIMDualKcbutunder-estimationsweresmaller
(Fig. 4).Relative toothercrops, under-estimationsof lysimeter
observationswerereportedbyKlockeetal.(1990,1996),Jaraetal.
(1998)and,formaizeinthesameexperimentalarea,byZhaoetal. (2013).Inadditiontounder-estimations,alargevariabilityofEs
measuredvaluesiscommonlyreported.
Thelargerunder-estimationsby AquaCroprelative to
SIMD-ualKc maybe explained by different approaches relative to Es
computations,mainlyreferringtothereductioncoefficientKr
(pre-viouslydiscussed in Section 2.2)since theKe values computed
bybothmodelsdonotshowlargedifferencesasanalysedbefore
(Section3.3).ThelimitationsofEsestimationbyAquaCropwere
discussedbyPereiraetal.(2015b)forabarleyapplicationunder
contrastingwetnessanddrynessconditions.
Table7presentstheresultsofthe“goodness-of-fit”indicators
oftheEsestimationswhenusingbothmodels.Resultsshowthat,
aftercalibration,AquaCroppresentsacleartendencyfor
under-estimationofEs,withlowervaluesofbfor2010and2011,which
aretheseasonswhenmostobservationswereperformedduring
midandlate seasons(Fig.4dand e).Theseseasonswerethose
whereworseresultswereobtainedforsimulatingASWasreported
inSection3.2.ThismayindicatethatwhenEsislargely
underesti-mated,particularlyinthoseseasons,thesimulationofASWisalso
under-estimated(seeFig.2eandg).Intheseyears,R2 variedin
awiderange,thusreflectingabiasintheestimation,particularly
duringthesecondhalfofthecropseason.Consequently,errorsare
high,bothRMSEandARE;however,RMSEaresmallerin2010and
2011relativeto2008and2009,whichlikelydependonthesize
Table7
Indicatorsof“goodness-of-fit”relativetosimulatingsoilevaporationwithAquaCropandSIMDualKcmodels. Yearand
treatment
Numberof observations
b R2 RMSE(mmd−1) ARE(%) EF
AquaCrop Usingdefaultparameters 2008,T1 65 0.96 0.72 0.66 59.7 0.61
2008,T2 54 1.08 0.84 0.48 68.2 0.72 2009 82 1.04 0.72 0.76 78.8 0.65 2010,T1 47 1.16 0.71 0.55 57.1 0.18 2011 42 1.42 0.77 0.69 68.2 -0.17 Usingcalibrated parameters 2008,T1 65 0.88 0.73 0.68 71.9 0.59 2008,T2 54 0.98 0.81 0.52 72.7 0.68 2009 82 0.89 0.76 0.66 56.0 0.74 2010,T1 47 0.84 0.65 0.50 66.3 0.31 2011 42 0.66 0.73 0.45 80.4 0.50
SIMDualKc(Weietal., 2015) 2008,T1 65 0.97 0.85 0.52 57.7 0.76 2008,T2 54 0.99 0.89 0.48 60.1 0.73 2009 82 0.96 0.79 0.65 56.4 0.74 2010,T1 47 0.90 0.86 0.38 60.5 0.62 2011 42 0.95 0.84 0.24 39.5 0.85
largelyrefertotheperiodswhenthecanopycoverwaslowandEs
waslarge(Fig.4a,bandc).Asexpected,theEFvaluesaresmallerfor
2010and2011becausethereferredunder-estimationofEsleads
tolargerresidual’svariance.
Lessexpectedaretheresultswhendefaultvalueswereused
withAquaCrop.For2008and2009resultsonR2,RMSE,AREand
EFaresimilartothoserelativetousingcalibratedparametersbutb
valuesindicateatendencyforover-estimationofEs.Incontrast,for
2010and2011,largerRMSEandsmallerAREvalueswereobtained,
aswellasanimportantover-estimationwithb=1.41for2011.Low
EFvaluesresulted,witha negativevaluefor 2011.Thecontrast
betweenresultsfor2011whenusingcalibratedordefaultvaluesis
enormousintermsofbvalues,RMSEandEFvalues.Thisexplains
whysimulationsofASWin2011werealsocontrasting,withworse
resultswhenusingcalibratedparameters(Table6).Thisbehaviour
isthereforeassociatingtheASWsimulationwiththeEsestimation,
thusindicatingthat betterASWsimulation resultsareobtained
whenEsvaluesarehighduringthemid-andlate-seasonsincea
worseASWunder-estimationoccursduringthesecropstages.This
identifiesaprobleminthecomputationofEs,asalreadydiscussed
byPereiraetal.(2015b),thatmayrelatetothealreadydiscussed
mechanisticapproachusedforcomputingthereductioncoefficient
Kr(Section2.2).Thus,itisadvisablethatmodeldevelopersrevise
theEsestimationprocedures.
WhenusingtheSIMDualKcmodel,resultspresented byWei
etal. (2015)show aslight tendencyfor under-estimation,
par-ticularly during 2010 (Table 7). As previously pointed out,
under-estimationislikelyduethefactthatmostofEsobservations
wereperformedduringmidandlateseasonswhenmicrolysimeters
areexpectedtoevaporatemorethanthesurroundingsoilwhere,in
additiontoevaporation,waterisalsoextractedbycroproots.The
RMSEandAREvaluesaresmallerthanthosebyAquaCrop(Table7)
whileR2andEFvaluesareconsistentlyhigher.Thesegoodresults
evidencethegoodnessoftheFAOdualKcapproachrelativetothe
Esestimation.
Table8presentstheresultsofcomparingAquaCropand
SIM-DualKcsimulated Es and Es/ETc act for the various cropstages
andallseasonsandtreatments.Duringtheinitialstage,theratios
Es/ETcactcomputedwithAquaCropareconsistently higherthan
thoseobtainedwithSIMDualKc,howeverwithsmalldifferencesin
termsoftheEsamount(Fig.3),whichisevensmallerincaseof
AquaCropforthe2008-T1and2009experiments.Thisbehaviour
resultsfromanunderestimationofthetranspirationduringthat
periodbecauseTaiscomputedfromCC(Eq.(1)),whichisextremely
smallduringtheinitialcropstage.Itresults,therefore,averylarge
proportionofEsrelativetoETc.act,rangingfrom91%to97%while
thatpercentage variesfrom72% to85%withSIMDualKc. These
resultsindicatethatTaareunder-estimatedbyAquaCropduring
theinitialcropstage.
Duringthecropdevelopmentstage,theEsvaluessimulatedwith
AquaCropweresmallerthanthoseobtainedwithSIMDualKcinthe
lastpartofthiscropstage(Fig.3),withtheratiosEs/ETc.act
consis-tentlysmaller(Table8).Thismaybeduetoanoverestimationof
TaduringthatsameperiodasdiscussedinSection3.3.During
mid-seasonthesametrendisobservedbutwithaverylowEsandEs/ETc
actvaluescomputedwithAquaCrop,alsoconsistentlysmallerthan
forSIMDualKc(Table8andFig.4).Thesedifferencesarecoherent
relativetoresultsinFig.3,previouslydiscussed.Thetendencyof
AquaCropforunderestimatingEswasmaintainedduringthe
late-season(Fig.4),exceptforthecropseasonof2009,whichisalso
likelyrelatedtothesameinsufficienciesindescribingtheKc,Trcurve
discussedin Section3.3.Consequently,for allcropseasons,the
totalevaporationdepthsandtheratiosEs/ETc actareconsistently
smallerwhencomputedwithAquaCrop(Table8).Thedifference
washigherfortheT2-2010dataset,withAquaCropcomputing
almosthalfoftheEscomputedbySIMDualKc.Therefore,itislikely
thatEscomputationswhenthecanopycoveriscompleteorisnear
completionaretoomuchdependentonCCandlittleonthesoil
evaporativecharacteristics,i.e.,therespectiveparameterization,
referredbefore,isinsufficientrelativetothatproposedbyAllen
etal.(1998,2005)andadoptedinSIMDualKc(Rosaetal.,2012).
ConclusionsreportedbyFarahanietal.(2009),Katerjietal.(2013),
andParedesetal.(2014b),aswellastheanalysisbyPereiraetal. (2015b),showthattheproceduresforestimationofEsinAquaCrop
arequestionableandshouldbeimproved.
3.5. Soybeanbiomassandyieldpredictions
Theobservedfinalharvestedbiomass(B)andyield(Ya)usedto
testtheAquaCroppredictionsarepresentedinTable9.Thehighest
yieldwasobtainedwiththetreatmentT1in2010(4230kgha−1)
whilethelowestrefersto2011(3222kgha−1).Differencesinyield
maybepartially explainedbydifferencesinclimaticconditions
alongthecropseasons.Sinceradiationandtemperatureaverages
alongthosereferredcropseasonsareverysimilar(Table1),
differ-encesinYamayrelatetothediurnalandnocturnaltemperatures
asdiscussedbyTao etal.(2008)relative totheHebeiprovince
orasreportedbyPutehetal.(2013)forMalaysia,namely
refer-ringtohightemperaturestressduringthereproductivegrowth
stage.Differencesincropmanagementmayalsohavecontributed
todifferencesinyield.
AquaCroppredictionsofboth biomassandyield whenusing
onlydefaultparametersshowacleartrendforunder-estimation,
Table8
Soilevaporationandevaporationratio(Es/ETcact)simulatedwhenusingAquaCrop(Aqua)andSIMDualKc(SIM)modelsaftercalibration,foreachdevelopmentstageandfor
theentireseason.
Year/treatment Cropstage
Initial Cropdevelopment Mid-season Lateseason Entireseason
Aqua SIM* Aqua SIM* Aqua SIM* Aqua SIM* Aqua SIM*
2008,T1 Es(mm) 62 66 33 36 1 7 5 6 101 115 Es/ETcact(%) 97 85 36 40 1 6 15 19 31 36 2008,T2 Es(mm) 61 57 31 30 1 7 5 7 98 101 Es/ETcact(%) 97 83 35 34 1 5 14 20 30 32 2009 Es(mm) 72 73 33 38 7 10 6 1 118 122 Es/ETcact(%) 91 80 37 43 5 7 14 4 33 35 2010,T1 Es(mm) 45 44 30 32 2 2 3 7 80 85 Es/ETcact(%) 92 72 24 28 3 3 9 24 27 30 2010,T2 Es(mm) 46 46 14 23 1 3 3 7 64 79 Es/ETcact(%) 91 72 12 22 1 4 9 24 21 28 2011 Es(mm) 47 43 51 52 1 9 2 2 101 106 Es/ETc act(%) 97 80 40 43 1 7 5 7 29 32
*AdaptedfromWeietal.(2015).
Table9
Deviationsbetweenpredictedandobservedsoybeanfinalbiomassandyield(kgha−1),whenusingdefaultandcalibratedparameters,comparedwithStewart’sandSIMDualKc combination.
Year Observed* (kgha−1)
AquaCrop Stewart’smodelwithT
datafromSIMDualKc
Defaultparameters Calibrated
Predicted Deviation Predicted Deviation Predicted Deviation
(kgha−1) (kgha−1) % (kgha−1) (kgha−1) % (kgha−1) (kgha−1) %
Dryfinalabove groundbiomass,B 2008–T1 9631(±690) 6532 −3099 32.2 9742 111 1.2 2008–T2 8948(±906) 6509 −2439 27.3 9722 774 8.6 2009 9118(±651) 6387 −2731 30.0 9425 307 3.4 2010–T1 11840(±617) 6530 −5310 44.8 9731 −2109 17.8 2010–T2 10965(±598) 6949 −4016 36.6 10143 −822 7.5 2011 10757(±510) 6536 −4221 39.2 9830 −927 8.6
Dryfinalyield,Y 2008–T1 3778(±272) 2613 −1165 30.8 3703 −75 2.0 4046 267 7.1
2008–T2 3549(±358) 2607 −942 26.5 3698 149 4.2 4009 461 13.0
2009 3454(±246) 2192 −1262 36.5 3522 68 2.0 3689 234 6.8
2010–T1 4230(±222) 2556 −1674 39.6 3731 −499 11.8 4443 214 5.1
2010–T2 3578(±196) 2819 −759 21.2 3884 306 8.6 4260 682 19.1
2011 3222(±151) 2550 −672 20.9 3637 415 12.9 3374 152 4.7
*Driedat65±5◦C;Standarddeviationbetweenbrackets.
biomass and yield, respectively (Table 9). In contrast, when
themodelwasappropriatelycalibrated thedeviations between
observedandpredictedbiomasswereverysmall,generallysmaller
than9%.Relativetothefinal yield,deviations werebelow8.6%
exceptfortwocases,whentheywereof11.8%and12.9%.Results
maybeconsideredexcellentfor 2008and 2009and reasonably
goodfor2010and2011.Itmaybenoticedthattheseresultsrelate
tothequalityofsoilwatersimulations,notgoodfor2010and2011
(Fig.2andTable6).TheworseBand Yaresultsmayhavebeen
influencedbythedifficultiesinachievingagoodpartitionofETc
act,withpoorestimation ofEs and Ta,thelattercontributingto
theestimationofBandYa.Nevertheless,ifgoodBandYa
estima-tionscouldbeachievedthisisduetothesatisfactorycalibrationof
BWP*,whichhighlyinfluencesbiomassandyieldestimation.One
mayhypothesizethatifabetterestimationofEsandTacouldbe
achieved,betterestimationsofthefinalbiomassandyieldwould
havebeenattained.
ThebiomassRMSEis1056kgha−1,whichrepresentlessthan
12% of the observed final biomass while for the final yield it
was quite low, 302kgha−1, thus representing 8% of the
aver-agedobservedyield.Resultsfallwithinthecategoryof“verygood
modellingresults”asproposedbyHansonetal.(1999)sincethe
deviationswerelowerthan15%oftheobservedvaluesexceptfor
onepredictionofB.DespitethelowaccuracyoftheASW
simu-lations,indicativeofpoorEsandTaestimation,yieldpredictions
could bevery good due to an appropriate BWP* calibration. If
defaultparametersareusedthenBandYaestimationshavemuch
largererrors,probablynotacceptableformostapplications.
Similar yield predictions, with RMSE=381kgha−1, were
obtainedbyWeietal.(2015)withthesamedatasetsbutusingthe
Stewarts’modelcombinedwiththeSIMDualKcmodel,however
withalargerover-prediction.UsingtheCROPGRO-soybeanmodel,
Calvi ˜no et al. (2003) reportedRMSE=512kgha−1, i.e.,
approxi-mately18% of theaverageobserved Ya,while Liu et al.(2013)
foundRMSErepresenting15to22%oftheaverageYainNortheast
China.Stöckleetal.(2003)appliedtheCropSystmodelandreported
RMSEcorrespondingto14%oftheaverageYa.Mohantyetal.(2012)
whenusingtheAPSIMmodelreportedanunder-estimationofonly
100kgha−1,i.e.,lessthan6%ofYa.Overall,resultsofthepresent
studyfallwithinthereportedsoybeanapplications,thus
indicat-ingthatAquaCropmaybefurtherusedforyieldpredictionswhen
appropriatelycalibrated.
4. Conclusions
TheAquaCropmodelwasparameterizedusingfouryearsoffield
observationsofasoybeancropintheNorthChinaPlain.Themodel
wastestedusingdefaultandcalibratedparametersandcompared
withthesoilwaterbalancemodelSIMDualKc(Weietal.,2015),that