Plant compensation to grazing and soil carbon dynamics in a tropical grassland MarkE.Ritchie DepartmentofBiology,SyracuseUniversity,Syracuse,NY,USA ABSTRACT Theeffectsofgrazingonsoilorganiccarbon(SOC)dynamics,particularlyinthe tropics,arestillpoorlyunderstood.Plantcompensationtograzing,wherebyplants maintainleafarea(Cinputcapacity)despiteconsumption(Cremoval)bygrazers, hasbeendemonstratedintropicalgrasslandsbutitsinfluenceonSOCislargelyun- explored.Here,theeffectofgrazingonplantleafareaindex(LAI)wasmeasuredin afieldexperimentinSerengetiNationalPark,Tanzania.LAIchangedlittleforgraz- ingintensitiesupto70%.TheresponsecurveofLAIversusgrazingintensitywas usedinamassbalancemodel,calledSNAP,ofSOCdynamicsbasedonprevious datafromtheSerengeti.ThemodelpredictedSOCtoincreaseatintermediategraz- ingintensity,butthentodeclinerapidlyatthehighestgrazingintensities.TheSNAP modelpredictionswerecomparedwithobservedSOCstocksinthe24grazedplots ofa10-yeargrazingexclosureexperimentateightsitesacrosstheparkthatvaried inmeanannualrainfall,soiltexture,grazingintensityandplantligninandcellulose. ThemodelpredictedcurrentSOCstocksverywell(R2 > 0:75),andsuggeststhat compensatoryplantresponsestograzingareanimportantmeansofhowherbivores mightmaintainorincreaseSOCintropicalgrasslands. Subjects Ecology,EcosystemScience,EnvironmentalSciences,SoilScience Keywords Herbivory,Grazing,Plants,Compensation,Leafarea,Soilcarbon,Modeling,Grasslands, Tropical,Serengeti INTRODUCTION Submitted 20August2013 Accepted 5December2013 Soilorganiccarbon(SOC)ingrasslandsandsavannasrepresentsoneofthelargest Published 28January2014 reservoirsofcarbononearth(Conant etal.,2001;Lal etal.,2007).Duetowidespread Correspondingauthor unsustainablelanduseduetoovergrazing,lossofnativeherbivores,excessivefires,and MarkE.Ritchie,[email protected] thepotentialtoreversetheimpactsoftheseuses,SOCisoneofthemoreimportant Academiceditor potentialsinksofgreenhousegasesintheefforttomitigateclimatechange(Mannetje, JianguoWu 2007;Ellis&Ramankutty,2008;Smithetal.,2008).Amajorquestioniswhat AdditionalInformationand Declarationscanbefoundon managementpracticesingrasslands,intheformofgrazing,fire,fertilization, page22 re-vegetationandrestoration,etc.,canleadtonetsequestrationofcarbon. DOI10.7717/peerj.233 Recentreviewsdemonstratethatherbivorescanhavedramaticallydifferenteffectson soilorganiccarbon(SOC),bothpositiveandnegative,dependingonsoiltype, Copyright 2014Ritchie precipitation,plantspeciescompositionandgrazingintensity(Milchunas&Lauenroth, Distributedunder 1993;Derner&Schuman,2007;Pineiroetal.,2010;McSherry&Ritchie,2013).These Creative-CommonsCC-BY3.0 resultshavebeenobtainedlargelyintemperategrasslandsgrazedbylivestock(McSherry &Ritchie,2013),andrelativelylittleisknownaboutgrazingimpactsonSOCintropical OPENACCESS HowtocitethisarticleRitchie(2014),Plantcompensationtograzingandsoilcarbondynamicsinatropicalgrassland.PeerJ2:e233; DOI10.7717/peerj.233 grasslands.Inthetropicsandincertaintemperateclimates,plantspeciescompositionis dominatedbywarm-season(C4)grassesthat,amongotheradaptationstograzing,invest heavilyinrhizomesandotherstorageorgansthatallowthemtorespondquicklytoboth rainfallanddefoliation(McNaughton,1985;Milchunas&Lauenroth,1993;Dubeux etal., 2007).Compensatoryresponsestograzingcaninvolvesacrificingstemsforleavesand thusmaintainleafareaandpotentialcarboninputsdespitecarbonoff-takebyherbivores (Anderson,Dong&McNaughton,2006;Zheng etal.,2010;Ziter&MacDougall,2013). Secondly,C4grassesgenerallycontainhigherlevelsofligninandcellulose(Bartonetal., 1976),whicharegenerallyrecalcitranttodecomposition,andproduceextensive abovegroundlitterthatfrequentlyburnsduringextendeddryseasons(Trollope,1982; McNaughton,Stronach&Georgiadis,1998;Sankaran,Ratnam&Hanan,2008).Thirdly, benigntemperaturesallowfortheprevalenceofmacro-decomposers,suchastermites anddungbeetles,torapidlyincorporatesenescedplantmaterialandherbivoredunginto soil(Hanski&Cambefort,1991;Freymannetal.,2008;Dungait etal.,2009;Risch, Anderson&Schutz,2012).Thesefeaturesmaydramaticallyalterherbivoreimpactson SOCintropicalgrasslandsascomparedtotemperategrasslandswithequivalent precipitationandsoiltype(McSherry&Ritchie,2013). Ofthesethreeimportantfeatures,theconsequenceofplantcompensationforsoil carbonisprobablytheleastwellunderstood.Leafareaindex(leafarea/areasampled),or LAI,hasbeenshowntodecreaseinresponsetograzinginsomecontextsbutnotothers (Anderson,Dong&McNaughton,2006;Zhaoetal.,2009;Cruz etal.,2010;Zheng etal., 2010;Ziter&MacDougall,2013).StudiesinSerengetiNationalPark(McNaughton,1985; McNaughton,Milchunas&Frank,1996)andwestAfrica(deMazancourt,Loreau& Abbadie,1999)haveshownbiomasscompensationtograzing,butnotcompensationfor lostleafareaspecifically.Furthermore,thequantitativemannerinwhichLAIchangesas aresponsetovaryinggrazingintensityhasnotbeenfullyexplored.Amajorquestionis whetherLAIcanbemaintainedorevenincreasewithincreasinggrazingintensityandat whatlevelofgrazingintensityitdeclines. TheresponseofLAItoherbivorymaybeakeydriverofsoilcarbonresponseto grazing.Leafareaishighlycorrelatedwithgrossphotosynthesisandnetcarbon(C) assimilation(Craineetal.,2001;Ellsworthetal.,2004;Reich,Wright&Lusk,2007). MaintenanceofleafareafollowingdefoliationpotentiallyallowsCinputsintoimportant carbonpools,suchasbiomassandSOC,tobelittleaffectedbyherbivory,whilelossof leafareacandramaticallyreduceCinputs.Becausemostcurrentsoilcarbondynamic modelslackanexplicitaccountingofherbivoreimpactsonleafarea(Paustian,Parton& Persson,1992;Schimel etal.,1994;Partonetal.,1995),therelativeimportanceofleaf area-relatedCinputsrelativetoClossesfromfire,decomposition,andherbivore consumptionhavenotbeenexplored. Thegoalofthispaperistoshowhowleafareaindex(LAI),orleafarea/areasampled, andplantallocationtostemandleafchangeswithgrazingintensityinatropicalsavanna andthentoexplorehowsuchpossiblecompensatoryresponsesmightaffectsoilcarbon dynamics.Hence,thispaper(1)presentstheresultsofa2-yeargrazingexclosure Ritchie(2014),PeerJ,10.7717/peerj.233 2/27 experimentinSerengetiNationalParktomeasureeffectsofvaryinggrazingintensityon plantLAIandproportionofbiomassallocatedtostemsvs.leaves,(2)developsamodelof soilcarbondynamicsbasedonextensivepastresearchintheSerengetiasaframeworkfor evaluatingtheimpactofLAIresponsetograzing,and(3)comparespredictionsofthis modelwithobservedsoilcarbonstocksacrosstheeightsitesofa9-yeargrazingexclosure experimentinwhichkeymodelinputparameterswereindependentlymeasured. Themodelofcarbondynamics,calledSNAP(referringtoitsorigininSerengeti NationalPark)accountsexplicitlyforfixedcarbon,asdeterminedbyLAI,soiltexture andrainfall,anditsallocationtorootsversusshoots,anditsfate:decomposition, consumptionbyherbivores,combustioninfireorresidenceinalong-termrecalcitrant SOCpool.Themodelhasfiveinputvariables:meanannualrainfall,grazingintensity,fire frequency,abovegroundproportionofcelluloseClignin,andsoiltexture(percentsand), basedonthesuggestedmostimportantfatesofstablecarbonintheSerengetiecosystem (McNaughton,1985;McNaughton,Banyikwa&McNaughton,1998;McNaughton,Stronach &Georgiadis,1998;Holdoetal.,2007,Holdoetal.,2009).Themodelanditsfunctional relationshipsarepopulatedwithparametervaluesavailablefromasubstantialliterature ongrazing,plants,soils,andmicrobesfromSerengetiNationalParkplussomeadditional newdataongrazingintensity,plantligninandcelluloseandsoilmoisturegatheredfrom along-term(9year)grazingexperimentacrosseightsitesinSerengetiNationalParkthat varyingrazingintensity,rainfall,soiltexture,firehistory,andplantspeciescomposition. Toevaluatetheroleofplantcompensation,asensitivityanalysisofmodelparameters, includinggrazingintensity,intheSNAPmodelisconducted.Themodelalsoisusedto makedeductivepredictionsofcurrentsoilcarbonstocksundertheassumptionofamass balancesteady-stateresultingfromthepersistenceofmeanrainfall,plantligninand cellulose,soiltexture,firehistory,andgrazingintensityconditionsoveranextended periodofprioryears.Thesepredictionsarethencomparedwithobservedsoilcarbon measuredingrazedplotsofthe9-yearexclosureexperimentusinglinearregression,with testsforoverallmodelfit(R2)andtestsfordifferencesinslopefromavalueof1andan interceptvalueofzero.Thisanalysisprovidesanexplicitandquantitativetestofthe hypothesisthatplantcompensatoryresponsesareimportantintheinfluenceof herbivoresonsoilcarbon. Thisapproachappliesadeductive(Belovsky,1984;Belovsky,1994;Overmars,deGroot &Huigen,2007)ratherthaninductive(Burnham&Anderson,2004)evaluationofthe impactofpossibleplantcompensation.Theuseofadeductivemodelexplicitlycombines thelogically-derivedhypothesizedmechanismsbywhichdifferentfactorsmightdrive soilcarbondynamicsbasedonpriorknowledgeoftheSerengeti.Asensitivityanalysisof themodelassessestherelativeimportanceofagivenvariableorparameterrelativeto others.Theadvantageofadeductiveapproachisthatitavoidstheuncertaintyin inferencederivedfromassociatingsoilcarbondynamicswithmultiple,possibly auto-correlated,independentvariablesinaninductiveapproach(Burnham&Anderson, 2004).Further,adeductivemodelbuildsinnon-linearityofrelationships,whichislargely avoidedininductiveapproachesbutmaybelikelytooccurintherelationshipbetween Ritchie(2014),PeerJ,10.7717/peerj.233 3/27 LAIandgrazing(McNaughton,1985;deMazancourt,Loreau&Abbadie,1999;Anderson, Dong&McNaughton,2006). MATERIALS AND METHODS Study area MostofthedatapresentedinthisstudywerecollectedinSerengetiNationalPark(SNP), Tanzania.SNPcontainsnearly3millionmigratorywildebeest(Connochaetestaurinus), zebra(Equusburchelli),andThomson’sgazelles(Gazellathomsoni),plusmultiplespecies ofresidentherbivores>10kginsize,alldistributedheterogeneously(Andersonetal., 2010)over25,000km2(ameandensityof120/km2)(McNaughton,1985;Sinclair etal., 2007).Theseherbivoresconsumeamajorfraction(mean61.8%(cid:6)12.2SEM)of abovegroundprimaryproduction(McNaughton,1985).VegetationisdominatedbyC 4 grasses,featuringThemedatriandrainmorelightlygrazedareasandDigitaria macroblepharaandPennisetummenzianuminmoreheavilygrazedareas.TheSerengeti regionfeaturesapronouncedrainfallgradientdrivenmixedsomewhatindependently withasoilgradientofsilty,organicmatter-richsoilsinthesoutheasttoclaysoilsinthe westandsandy,lownutrientsoilsinthenorth(Andersonetal.,2007;Sinclair etal.,2007; Andersonetal.,2010).Theclustereddistributionofresidentherbivores(Andersonetal., 2010)andvariablemovementsofmigratoryspeciesresultsinhighlyvariablehistorical grazingintensityacrossthelandscape.Strongpreferencesbyherbivoresforplantsfound atthetopratherthanbottomofslopesplusatendencytoavoidtheedgesofwoodlands resultsinvariablegrazingintensityoverdistancesof<50m(Gwynne&Bell,1968; Anderson,Dong&McNaughton,2006;Andersonetal.,2007). Design PlantcompensationtograzingwasassessedneartheSerengetiWildlifeResearchCentre ◦ ′ ◦ ′ nearSeronera(34 50 Eand2 25 S)insidethePark.TheareaisopenAcaciatortilis savannawoodland,withdominantgrassesThemedatriandraandDigitaria macroblephara,andhostsresidentherdsofimpala(Aepycerosmelampus),buffalo (Synceruscaffer),andThomson’sgazelles(Gazellathomsoni),amongothers,andis routinelyvisitedbytheannualwildebeestandzebramigration.Plantresponsestograzing weremeasuredinsideandoutsideofgrazingexclosuresestablished2yearspreviouslyina 10haopengrasslandwithlessthan5%treecoverneartheSerengetiWildlifeResearch Centre(2-yearexclosureexperiment).Theproportionofbiomassasstemversusleafwas measuredforallplantsclippedfromsixrandomlyselected15(cid:2)15cmquadrats,three insideandthreeoutsideeachoftwelveexclosuresofvarioussize(10(cid:21)200m2)erected between2009and2010toprotectexperimentalgardensornutrientaddition experiments.LAI,proportionofstem,andabovegroundbiomassweremeasuredinthe unused‘‘bufferzone’’justinsideeachfenceandtheninquadrats4mperpendicularto eachfencepairedwitheachquadratinsidethefence.Theobjectivewastoencounter differentlocalizedgrazingintensitiesthatresultedfromsamplingbeneaththornyshrubs orotherplantprotectionsversuspatcheswithhighlypreferredforagespecies. Ritchie(2014),PeerJ,10.7717/peerj.233 4/27 Togenerateadditionalinputdataforamodelofsoilcarbondynamics,alonger-term(9 year)grazingexclosureexperimentestablishedbySamMcNaughtonin1999(Anderson, Ritchie&McNaughton,2007)wasusedtoprovidedataonsoilmoistureasafunctionof rainfall,andindependentmeasuresofgrazingintensity,plantligninandcellulose,and SOCtotesttheSOCdynamicmodel.Alineofsixplots(4(cid:2)4m),spaced10mapartina linetransectwereestablishedinOctober1999ateachof8siteslocated10ormorekm apart,ingrasslandareasvisitedalmostentirelybygrazing,ratherthanbrowsing,ungulate species(Anderson,Ritchie&McNaughton,2007).Siteswerechosentobewithin1kmof large,permanentconcentrationsofgrazingherbivores(Andersonetal.,2010)andvaried considerablyinannualrainfall,soiltype,andfirefrequency.Threerandomlyselected plotswithinthetransectateachsitewerefencedwith2mhigh,8cmmeshwire.These fenceseffectivelyexcludedallgrazingmammals>10kg,sinceanimalspreferredtogo aroundratherthanjumpfences.Theremainingthreeplotsateachsitewereunfenced controls.Woodyvegetationatourexperimentalsiteswastypicallysparse(lessthan10 woodyplants/ha)butherbaceouslegumes,suchasIndigoferavolkensii,wereubiquitous andreasonablyabundant(>2%cover)atallourstudysites.Oureightsitesincludedtwo onthetreelessshortgrassplainsofsoutheasternSNP,andtheothersixsiteswere representativeofwoodysavannasbroadlydistributedacrossSNP(Ruess&Seagle,1994). Measurements Theeffectofgrazingonplant(grass)allocationtoleafversusstemandonleafareaindex, LAI(cm2leaf=cm2groundarea)wasassessed.Clippedplantsforallplantspeciesina quadratweretakenbacktothelaband,whilefresh,laidoutandtracedongraphpaper, ◦ andthenleaveswereclippedfromstemsandstored,driedat45 C,andweighed separately(aswithfieldbiomass).Theproportionofleafforeachquadratwascalculated, alongwithLAI(totalleafarea(cm2)dividedbyquadratarea(cm2)). Anotherkeyvariable,GI(grazingintensity)varieslocallyandwasestimatedinboth the2-yearand9-yearexperimentsasthefractionaldifferenceinstandingaboveground biomass(AGB):GI D 1(cid:0)AGB =ABG (Anderson,Ritchie&McNaughton,2007).GIis g ug anindirectmeasureofthefractionofannualproductionconsumedbygrazers (McNaughton,1985;McNaughton,Milchunas&Frank,1996).AGBwasmeasuredby clippingandweighingallabovegroundplantmaterial,excludinglitter,afterdryingat ◦ 45 Cforthreedaystoconstantmass.AGBwasmeasuredinthe2-yearexclosure experimentinthepairedquadratsinsideandoutsideexclosuresusedtomeasureLAI. ThequadratinsidetheexclosurewasassignedGI D 0,andthepairedquadratoutsidethe exclosurewasassignedGIbasedonthedifferenceinAGBbetweenthepairedquadrats. Inthe9-yearexclosureexperiment,biomasswasclippedfromfour15(cid:2)15cmquadrats withineachgrazedandexclosureplotinlateMayorJuneattheendofthewetseasonin 2006,2009,and2010.Biomassmeasuredineachofthethreegrazedplotsateachsitewas comparedtobiomassineachoftheirnearestneighborexclosureplots(Anderson,Ritchie &McNaughton,2007),yieldingthreeestimatesofgrazingintensityateachsite. Tomeasureligninpluscellulosefraction(LIGCELL),driedplantmaterialfromthe entireclippedquadratwasgroundthrougha0.9mm(40mesh)screeninaWileymill Ritchie(2014),PeerJ,10.7717/peerj.233 5/27 andthensubjectedtoasulfuricacidhydrolysismethod(Sluiter etal.,2010),i.e., sequentialdigestioninfirstneutraldetergentmanufacturedbyAnkomTechnologyCorp., Macedon,NewYork,USA(sodiumlaurylsulfate,50.02%,EDTAdisodium,dehydrate, 31.03%.,sodiumborate,decahydrate11.35%,andsodiumphosphate,dibasic,anhydrous, 7.6%),followedbyAnkommanufacturedaciddetergent(20gcetyltrimethylammonium bromide(CTAB)dilutedin1L1.00NH SO ),andthenin72%sulfuricacid,followed 2 4 ◦ finallybyashinginamufflefurnaceat500 Cfor24h.Ligninandcellulosefractionswere inferredbythemasslostbetweentheremainderfollowingneutraldetergentdigestion andtheremainderafterashing,expressedasaproportionoftheoriginalsamplemass. Annualrainfalloverthe9yearsofthestudywasobtainedfromraingaugedataat23 sitescollectedbytheSerengetiEcologicalMonitoringProgram(Andersonetal.,2007). FirefrequencywasdeterminedbyinterpretingMODISsatellitevisibleandnearinfra-red imagery(Dempewolf etal.,2007)forJuly(cid:21)Octoberfor2000(cid:21)2008togeneratemapsof fireoccurrenceforeachyearwitharesolutionof250m.GrazingexperimentsiteGPS ⃝ locationswereoverlaidwithfiremapsforeachyearinaGIS(ArcGIS9.1byESRIR)to determinethenumberoftimesasiteburnedovertheperiod2000(cid:21)2008. Effectsofprecipitationonsoilmoistureandnumberofdaysinwhichmicrobeswere likelytobeactive(WETDAYS)weredeterminedfromfieldmeasurementsofsoil moistureinthelong-termexperiment(unpublisheddatagenerouslyprovidedbySam McNaughton).Fivecmdiametersoilcoresweretakentoadepthof10cmeachmonth fromDecember1999toJune2002(atotalof31days)ateachofthreegrazedplotsatthe eightsitesofthelong-termSerengetigrazingexperiment.Moisturecontentwas determinedgravimetricallyandmonthlyrainfallwasmeasuredfromraingauges mountedon1mtallfencepostsateachsite.Arecentreview(Manzoni,Schimel& Porporato,2012)suggeststhat,insemi-aridsoils,microbialactivityceaseswhenwater potentialdropsbelow(cid:0)14cmH O.Waterpotentialwasconvertedintosoilmoisture, 2 calibratedforthetextureofSerengetisoils(Table1)(Cosby etal.,1984)andsuggested thatManzoni,Schimel&Porporato(2012)thresholdcorrespondedonaverageto10% gravimetricwatercontent.Thenumberofdaysateachoftheeightsitesthatsoilmoisture was10%orhigherwasdeterminedandthenusedtobuildaregressiontoestimatethe numberofdaysperyearwithactivesoilmicrobesasafunctionofannualrainfall. Asatestofthenewsoilcarbonmodel,SOCwasmeasuredatthesametimeasbiomass ineachplotfrom8.3cmdiameter(cid:2)40cmdeepcoresfromeachofthethreebiomass quadratsperplot.Eachcorewasspreadonaportabletabletopandcrushedwitharolling pintobreakupsoilaggregates.Allpebbles,rocks,andvisiblerootswereremoved,and thensoilwassievedthrough0.7mmmesh.Coresofknownvolumewerepooled,stored ◦ insealedplasticbags,andthendriedfor6daysat45 Candthenweighedtodetermine bulkdensity(gsoil/cm3).BothSOCandnitrogencontentsweredeterminedwiththe Walkley-BlackmethodattheSoilsAnalysisLaboratoryatSokoineUniversityof AgricultureinMorogoro,Tanzania.Reportederrorwas(cid:6)0.004g/gC.Soilcomposition (sand,siltandclayfractions)wasdeterminedbyamicropipettemethod.Soilcarbon densitywasmeasuredbymultiplyingSOCcontentbybulkdensity(g/m2)to40cmdepth Ritchie(2014),PeerJ,10.7717/peerj.233 6/27 Table1VariablesandkeyparametersintheSNAPmodel,theirunits,andsourcesofdata. Parameter Units Type Source ANPPmax gm(cid:0)2yr(cid:0)1 Literature Maximumabovegroundproduction(McNaughton,1985) ANPPest gm(cid:0)2yr(cid:0)1 Model Grazer-modifiedANPP BNPPest gm(cid:0)2yr(cid:0)1 Literature Belowground production (McNaughton, Banyikwa & Mc- Naughton,1998)(Fig.3B) DDSOC gCm(cid:0)2yr(cid:0)1 Model Dung-derivedSOC FIRE #/yr Measured Numberoffiresperyear,monitored(Table4) GI proportion Measured 1-(grazedbiomass/ungrazedbiomass)(Table4) LAI - Measured LeafAreaIndex(dimensionless);Thisstudy(Fig.2B) LIGCELL proportion Measured CelluloseClignin(%)abovegroundbiomass,(Table4) MRESP gCm(cid:0)2d(cid:0)1 Literature (Ruess&Seagle,1994)(Fig.3D) PL proportion Measured Proportionleaf;Thisstudy(Fig.2A) PDSOC gCm(cid:0)2yr(cid:0)1 Model Plant-derivedSOC RAIN mm/yr Interpolated Meanannualrainfall(Andersonetal.,2010)(Table4) SAND% percent Measured Fromastandardsoilanalysis,measured(Table4) WETDAYS dyr(cid:0)1 Measured #dayswithsoil>10%water;Thisstudy(Fig.3C) WHC proportion Literature Waterholdingcapacity(Ruess&Seagle,1994)(Fig.3A) Table2ParametersforMonteCarlosimulations.Parametersandtheirerrorsusedinpredictingsoilcarbon stocksinSerengetigrasslandsandintheMonteCarlosimulations. Parameter Input Coefficient B0 SE B1 SE B2 SE PL GI 0.597 0.061 0.24 0.023 LAI GI 1.15 0.027 (cid:0)0.015 0.0011 4.6 0.2 ANPPmax RAIN (cid:0)27.5 17.6 0.84 0.07 WHC SAND% 1.04 0.05 (cid:0)0.0070 0.0008 BNPPest RAIN 958.8 165.0 (cid:0)0.82 0.27 WETDAYS RAIN (cid:0)0.025 0.033 0.00043 0.00006 MRESP SOC (cid:0)0.58 0.45 0.00044 0.00007 CARBON N/A 0.45 0.02 ofthepooledcoresineachplot.A40cmdepthwasused,asthistypicallycorrespondsto thedepthof>90%ofroots(McNaughton,Stronach&Georgiadis,1998),andmultiple sitesfeaturedahardpanordenseclaylayerat40(cid:21)50cmthatpreventedreasonable samplingtogreaterdepths. Literature data and re-analyses Literaturedata,mostlyfrompriorstudiesinSerengeti,werere-analyzedtospecifically estimatesoilwaterholdingcapacity(WHC),abovegroundproduction,belowground production,andsoilmicrobialrespirationofcarbon.Forsourcesandunitsofliterature data,seeTable1.TherelationshipsusedarereportedintheResultsandDiscussion (Table2). Ritchie(2014),PeerJ,10.7717/peerj.233 7/27 Figure1 Hypotheticalmajorfatesofcarbonintropicalgrassland,asthebasisforapracticalsoil carbondynamicmodel.Netfixedcarbonbecomesresidentsoilorganiccarbon(SOC)throughtwoma- jorpathways:plant-derivedSOCnotconsumedbymicrobes(insoilorthegutsofmacro-decomposers liketermites)(heavysolidarrows),anddung-derivedSOC(heavyshort-dashedarrows)notassimilated bygrazersandgutorsoilmicrobes.Allothercarboniscombustedthroughfire(blackarrow),orrespired bygrazersandmicrobesinsoilorthegutsofmacro-decomposers(longdashedarrows). Model development Theflowoffixedcarbonthroughhypotheticaltropicalgrasslandwasdescribedaccording tofourcriteria.First,massbalancewasassumed,sothatallfixedcarbonresidedinpools, suchasbiomassorSOC,orsufferedthefateofbeingcombusted,consumed,orrespired. Second,thisflowwasdescribedintheminimumnumberoffluxesandpoolsnecessaryto accountforthefateofcarbon(Fig.1).Third,themostparsimonioussetofinputvariables thatcoulddescribethechangesinthesepoolswasused.Finally,atimestepofoneyear withadistinctwetanddryseasoneachyearwasassumed,withproductionandgrazing resultinginastandingbiomassatthebeginningofthedryseasoneachyearthat subsequentlyeithercombustediftherewasfireorwasotherwisedecomposed.More complicatedtimedynamicsofdailyorseasonalchangeshavebeenusedrecently(Holdo etal.,2007,2009),butthisintroducespossiblyunnecessarycomplexity.Above-and belowgroundproductionarethereforeseasonalaccumulationsofbiomassasmodifiedby averagegrazingintensityoverthecourseofthewetseason.Biomassnotconsumedby herbivoresmaythenburnatsomefrequencyduringthedryseason.Theseassumptions ledtoarelativelysimplesetofpathwaysforstablecarbon(Fig.1):(1)inputintosoil organicmatterfromdecomposingshootsorroots,(2)combustioninfire,or(3)input intosoilorganicmatterthroughdung.Allnon-ligninorcellulosecompoundswere assumedtoturnoverwithin1(cid:21)2yearsorless(Ruess&Seagle,1994)andwere consequentlyignored. Themodelwasbasedonfourmajorassumptions.First,tropicalgrassespotentially exhibitcompensatoryresponsestodefoliationthatmaintainsimilarleafarea,andthus photosyntheticcapacity,acrossabroadrangeofgrazingintensities(McNaughton,1985). Second,thelargestcarboninputstothesoilorganicmatterpooloccurthrough Ritchie(2014),PeerJ,10.7717/peerj.233 8/27 decompositionofabove-andbelowgroundbiomassandthroughincorporationof herbivoredungintosoil.Third,themajorlossesofSOCderivefromcombustion(fire), herbivorerespirationandsoilmicrobialrespiration.Finally,allplantCinplanttissue otherthanligninandcellulosewasassumedtobeassimilatedandrespiredbyherbivores, microbesinsoilorthegutsofmacro-decomposers.Ligninandcellulosearethemost recalcitrantformsofcarbontodecomposition,andlikelyaccountforthemajority (>90%)oflong-livedSOC(Ganjegunteetal.,2005;Frank,Pontes&McFarlane,2012). TheSOC(g/m2to40cmdepth)measuredin2009inthegrazedplotsinthelong-term grazingexperimentrepresentedSOC underthereasonableassumptionthatconditions eq intheseplotshadbeenapproximatelysimilarintermsofwildlifeuse,plantspecies composition,andclimatesince1979(Sinclair etal.,2007;Holdoetal.,2009).Theaverage ofgrazingintensitiesmeasuredin2006,2009,and2010wasassumedtoreflecthistorical long-termgrazing.Theotherinputvariableseitherwerealreadyalong-termaverage (RAIN,FIRE)orareunchangedbylanduse(SAND%). Statistics TheassociationbetweenLAIandgrazingintensity(GI)wastestedwithlinearregression inSPSSversion19(IBM,Armonk,NewYork,USA).Non-linearrelationshipswerealso explored,andbasedontheappearanceofthedata,theRscript‘‘nls’’wasusedwitha functionB (cid:0)B exp.(cid:0)B GI),whereB arefittedcoefficients.LinearregressioninSPSS 0 1 2 i wasusedtobuildregressionrelationshipsfromliteraturedata(Table2)andtocompare observedSOCtopredictedSOC fromtheSNAPmodel.Inthecomparisonofthe eq observedandpredictedSOC,thehypothesesthattheinterceptoftheregressionlinewas equaltozeroandthattheslopewasequaltoonewereevaluatedwithz-tests. TheSNAPsoilcarbondynamicmodelwasconstructedfromanumberofempirical relationshipsthatcontainparameters(intercepts,slopes,etc.)estimatedwitherror.A MonteCarloanalysis(Ogleetal.,2010)wasconductedbysamplingthenormal distributionsinferredfrommeasuredmodelinputparametersandparametersfrom relationshipsusedtoconstructthemodel.Thisprocesswasrepeated100timesto generateameanands.e.m.fortheestimatedSOC foreachsiteandeachpairedgrazed eq andungrazedplotinthe9-yearexclosureexperiment. RESULTS AND DISCUSSION New and prior data analyses Theproportionofbiomassinleavesincreasedsignificantly(P < 0:001)withincreasing grazingintensity(Fig.2A).Inconcordancewiththispattern,LeafAreaIndex(LAI) declinedsignificantlywithincreasinggrazingintensity(GI)accordingtolinearregression (R2 D 0:36,dfD1,34,PD0.01).Howeveranon-linearfunction,LAI D B (cid:0)B exp 0 1 .B GI/fitthedatamuchbetter(R2 D 0:58,dfD1,33,P < 0:001)andvisiblymatched 2 bettertheconsistentlyhighLAIfoundatlowtointermediategrazingintensityandthe dramaticdeclineinLAIaboveagrazingintensityof70%(Fig.2B).Thesetwopatterns supportthehypothesisthatplantsmaintainCinputstotheSerengetiecosystemdespite Ritchie(2014),PeerJ,10.7717/peerj.233 9/27 Figure2 PlantcompensatoryresponsestograzinginSerengeti.Keyrelationshipsforassessingplant compensationtograzing:(A)proportionofleaves(PL)and(B)LeafAreaIndex(LAI,cm2=cm2)of grazedandungrazedplants,eachasafunctionofgrazingintensity(GI). heavyremovalsofbiomassbygrazers.ThemaintenanceofLAIacrosssuchabroadrange ofgrazingintensitiescontrastswiththetypicalpatternofdramaticreductionsinLAIseen mostlyintemperatesystemsunderrelativelymoderategrazingintensities(30%(cid:21)60%) (Zhaoetal.,2009;Cruz etal.,2010;Zheng etal.,2010).Thepatterncorroboratesresults ofenhancedproductivityundergrazing(McNaughton,1985;Milchunas&Lauenroth, 1993;Frank,Kuns&Guido,2002)thatareatleastpartlyderivedfromcompensatory morphologicalchangesbyplantstograzing(Anderson,Dong&McNaughton,2006;Ziter &MacDougall,2013).Despitethesepreviousresults,itisstillnotclearwhytheC grasses 4 typicallydominantintheSerengetiandelsewhereresponddifferentlytomoderately intensegrazingthanC grasses(Derner&Schuman,2007;McSherry&Ritchie,2013). 3 Asexpected,waterholdingcapacityexhibitedastrong,significant(P < 0:001) negativeassociationwithincreasingsandcontent(Ruess&Seagle,1994)(Fig.3A),and rootproductionshowedasignificant(P D 0:0015)declineinassociationwithhigher rainfall(McNaughton,Banyikwa&McNaughton,1998)(Fig.3B).Theproportionofdays soilheldmorethan10%moistureincreasedsignificantly(P D 0:006)withmeanannual rainfallacrosstheeightlong-termgrazingexperimentalsites(Fig.3C).Finally,maximum microbialrespirationrateswerestronglyandsignificantly(P < 0:001)associatedwith existingSOCstocksacrosstheSerengetilandscape(Ruess&Seagle,1994)(Fig.3D). Regressionequationsarereportedinthefiguresandwereusedinthedevelopmentofthe SOCdynamicmodel. SOC dynamic model BasedontheresultsforLAIresponsetograzing(Fig.2B),theSOCmodelforSerengeti NationalPark(hereafterreferredtoastheSNAPmodel)wasconstructedwithfiveinput variables:RAIN,meanannualrainfall(mm/yr);GI,grazingintensity(1-(grazed biomass/ungrazedbiomass))(McNaughton,1985);FIRE,firefrequencyorthenumberof firesrecorded,dividedbythenumberofyearsoverwhichfiresweremonitored; Ritchie(2014),PeerJ,10.7717/peerj.233 10/27