Validation of a Smartphone Image-Based Dietary Assessment Method for Pregnant Women

20 Pages·2017·3.53 MB·English

remote sensing
Article
MODIS Time Series to Detect Anthropogenic
Interventions and Degradation Processes in
Tropical Pasture
DanielAlvesAguiar1,2,*,MarcioPupinMello3,SandraFurlanNogueira4,
FabioGuimarãesGonçalves5,MarcosAdami6andBernardoFriedrichTheodorRudorff1
1 AgrosatéliteGeotecnologiaAplicada,Florianópolis88032,Brazil;[e-mail]
2 DivisãodeSensoriamentoRemoto,InstitutoNacionaldePesquisasEspaciais,
SãoJosédosCampos12227,Brazil
3 TheBoeingCompany,BoeingResearch&Technology—Brazil,SãoJosédosCampos12227,Brazil;
[e-mail]
4 BrazilianAgriculturalResearchCorporation(EMBRAPA),MonitoramentoporSatélite,
Campinas70770,Brazil;[e-mail]
5 CanopyRemoteSensingSolutions,Florianópolis88032,Brazil;[e-mail]
6 CentroRegionaldaAmazônia,InstitutoNacionaldePesquisasEspaciais,Bélem66077-830,Brazil;
[e-mail]
* Correspondence:[e-mail];Tel.:[phone]
AcademicEditors:MartinBrandt,TorbernTagesson,StephanieHorion,RasmusFensholt,AlfredoR.Huete,
RichardGloaguenandPrasadS.Thenkabail
Received:1September2016;Accepted:1January2017;Published:14January2017
Abstract: Theunavoidabledietchangeinemergingcountries,projectedforthecomingyears,will
significantlyincreasetheglobalconsumptionofanimalprotein. ItisexpectedthatBrazilianlivestock
production,responsibleforcloseto15%ofglobalproduction,bepreparedtoanswertotheincreasing
demandofbeef. Consequently,theevaluationofpasturequalityatregionalscaleisimportantto
informpublicpoliciestowardsarationallandusestrategydirectedtoimprovelivestockproductivity
in the country. Our hypothesis is that MODIS images can be used to evaluate the processes of
degradation, restoration and renovation of tropical pastures. To test this hypothesis, two field
campaigns were performed covering a route of approximately 40,000 km through nine Brazilian
states.Tocharacterizethesampledpastures,biophysicalparametersweremeasuredandobservations
aboutthepastures,theadoptedmanagementandthelandscapewerecollected.Eachsampledpasture
wasevaluatedusingatimeseriesofMODISEVI2imagesfrom2000–2012,accordingtoanewprotocol
basedonsevenphenologicalmetrics,14Booleancriteriaandtwonumericalcriteria. Thetheoretical
basisofthisprotocolwasderivedfrominterviewswithproducersandlivestockexpertsduringathird
fieldcampaign. TheanalysisoftheMODISEVI2timeseriesprovidedvaluablehistoricalinformation
onthetypeofinterventionandonthebiologicaldegradationprocessofthesampledpastures. Ofthe
782pasturessampled,26.6%experiencedsometypeofintervention,19.1%wereunderbiological
degradation, and54.3%presentedneitherinterventionnortrendofbiomassdecreaseduringthe
periodanalyzed.
Keywords: EVI2timeseries;biophysicalparameters;phenologicalmetrics;biologicaldegradation;
pasturereformation;pasturerenewal;pasturerecovery
1. Introduction
Theincreasingdemandforfoodasaresultofpopulationgrowthandtheglobalurbanization
process[1],thegrowingdemandforrenewableenergysources[2]andtheimprovedmanagementof
RemoteSens.2017,9,73;doi:10.3390/rs9010073 www.mdpi.com/journal/remotesensing
RemoteSens.2017,9,73 2of20
naturalresourceshavebeencallingtheattentionofgovernments,thescientificcommunityandsociety
onissuessuchasagricultureandlivestock,aimingatabetteruseoftheterritory[3,4].
Beefsupplierswillberequiredtoreacttofulfillapredictedincreaseof44%intheglobaldemand
for beef in 2030 as a consequence of income increase in developing countries [5]. Brazil plays an
importantroleasthesecondlargestbeefproducerandthenumberonebeefexporterintheworld.
Nevertheless, the increase of beef supply must be associated exclusively to the improvement in
livestockproductivity,giventhegoalssetbygovernmentsofdevelopingcountriestoreduceratesof
deforestationandgreenhousegasemissions[6,7],andlandcompetitionforproductionofgrainand
biofuels[8–11].
The most important premise for the increase in livestock productivity is the improvement of
pasturesintermsofquality,andthedegradationprocessisthemaincauseofqualitylossintropical
countries,wheremostoftheherdisonpastureyearround. Thisisachronicproblem,beingalimiting
factortotheincreaseinlivestockproductivity. Dias-Filho[12],focusingontropicalpasture,classified
the process of pasture degradation into two types, i.e., biological and agricultural. The biological
degradationisthelossofthepastureabilityinsustainingitsproductivityduetosoilconditions,while
theagriculturaldegradationrepresentstheinabilityofthepasturetobeeconomicallyfeasibledueto
competitionwithinvasiveplants.
Remotesensing(RS)imagesacquiredbyEarthobservationsatellitesmakeitpossibletoevaluate
the changes in land-use/land-cover over the past four decades, on scales never explored before
(e.g.,[13–16]). Thesynopticandrepetitivecoverageprovidedbytheseimages,alongwithecological
indicatorsandvegetationbiophysicalparameters,allowstheapplicationofintegrateddataanalysis
andthedevelopmentofpredictivemodelsusefulfordecisionmakers[17,18].
However,therearenostudiesthatdistinguish,onaregionalscaleandbasedonRSimages,the
consequencesoftheadoptedmanagementpracticesintropicalpastures(includinggrazing), from
thosearisingfromvariationsintrinsictotheecosystem[19].
Inthissense,timeseriesoftwo-bandEnhancedVegetationIndex(EVI2;[20])derivedfromthe
ModerateResolutionImagingSpectroradiometer(MODIS)instrumentmaybeusedtounderstand
phenologicalpatternsandsupporttheassessmentofanthropogenicinterventionsanddegradation
processontropicalpastures. Startingfromthispremise,theobjectiveofthisstudywastoproposeand
testanewprotocolforidentifyinganthropogenicinterference(i.e.,pasturemanagement)andpasture
degradationprocessesusingphenologicalmetricsextractedfromtimeseriesofMODIS/EVI2.
TheoreticalBasisforTropicalPasturesAssessmentUsingEVI2TimeSeries
Climatic and edaphic factors, anthropogenic interventions, cattle grazing and the occurrence
of fires define the spectral dynamics of pasture lands in time. Thus, models that are based on the
assumptionthatthevegetationbehaviorislinearandgradualarenotvalidtodescribethebehaviorof
tropicalpastureundergrazing[17,21,22].
Thepastureresponsetothecomplexinteractionofthesefactorscanbegradual,linear,non-linear
orabrupt[17],withindicatorsthatcanbeobservedbysatellitesensors,allowingtheassessmentof
pasturebehaviorinrelationtochangesintheircomposition. Metricsfromtimeseriesofvegetation
indices,associatedwiththeempiricalknowledgeoftheaforementionedfactors,allowthemonitoring
ofvegetationinsearchforindicatorsofchange[13,23–25]. Unlikeagriculturalcrops—withitswell
definedplantingandharvestingschedules—andforests—wheretheseasonalvariabilityinbiomassis,
ingeneral,notrelevant—thereisnotawelldefinedscheduleforpastureandtheannualvariabilityin
biomasscanbehigh,dependingonthegrasstype,soilandclimateconditions[17,26–28].
A tropical pasture under biological degradation [12], combined with the lack of a proper
management,coverandsoilnutrientloss,causestheemergenceofpatchesofexposedsoil[27,29,30].
Time series of vegetation indices extracted from satellite images are sensitive to changes of this
kind[22]. Ontheotherhand,pasturesunderagriculturaldegradation—wheretherearechangesinthe
RemoteSens.2017,9,73 3of20
biologicalcompositionduetosecondarysuccession—orexperiencingundergrazing,showanincrease
inbiomass[12],aprocesswhichisalsodetectedintimeseriesofvegetationindices[31].
Infact,thetypicalpasturebehaviorovertimeisduetotheseasonalvariationinbiomassrelated
to water availability [27,32]. This variation is expressed in time series of vegetation indices as a
seasonal behavior that characterizes, in each crop year, dry (with low values for the index) and
wet (with high values) periods [24,27]. On the other hand, significant changes in pasture and
grazingmanagement(e.g.,increasedstockingrate,changefromcontinuoustorotationalgrazing,or
anthropogenicinterventionsforpastureimprovement)arecharacterizedbymeaningfulchangesinthe
typicaltemporalbehaviorofthevegetationindex[17].
ForMacedoetal.[30],therecoveryofadegradedpastureischaracterizedbythereestablishment
ofthepastureproduction,withtheplantingofthesamespeciesorcultivar.Alternatively,therenewalof
adegradedpastureischaracterizedbytheintroductionofanewpasturespecies[30]. Thereformation
isrelatedtocorrectionsand/orrepairsmadeaftertheestablishmentofthepasture. Boththerecovery
andtherenewalcanbeperformeddirectlyandindirectly,i.e.,withorwithouttheuseofsummercrops
orannualpasture.
Inthisstudy,therecoveryandrenewalprocessesweregroupedtogetherduetothedifficultyof
distinguishingthembecauseofthescaleoftheanalysis. Bothprocessesarecharacterizedbysoiltilling,
whichsignificantlychangestheminimumvaluesofEVI2inthebeginningoftheseprocesses,allowing
theiridentificationinEVI2timeseries.
Thus,thehypothesesfortheestablishmentoftheassessmentprotocolbasedonEVI2timeseriesof
tropicalpastureare:(i)anthropogenicinterventionstypicallyassociatedwithpasturereformationcause
anincreaseinthepasturebiomassandarenotprecededbysoilexposure[30,33];(ii)anthropogenic
interventionstypicallyassociatedwithrenewal/recoveryprocessescauseanincreaseinthepasture
biomass and are preceded by tillage [30,33]; (iii) biological degradation of pastures result in the
decreaseofbiomassovertime[12];and(iv)vegetationindicesextractedfromtimeseriesofsatellite
imagesaresensitivetothechangesdescribedabove.
A pasture that has gone through some sort of intervention in a given period could also be
subjecttodegradationaftertheintervention,resultinginlossofbiomass. Factorssuchasinadequate
management,problemsduringplanting,insufficientfertilization,andadoptionofstockingratesabove
thepasturecarryingcapacitycanacceleratethedegradationprocess. However,itwasassumed,forthe
purposeofthisstudy,thatonlypasturesthathavenotgonethroughanysortofhumanintervention,
and showed a tendency of decrease in biomass during the analyzed period, could be classified as
degradedpasture.
2. MaterialsandMethods
2.1. StudyAreaandFieldCampaigns
Twofieldcampaignswereconductedfordatacollectionandcharacterizationofpasturecover
conditions,bothinthecontextof“RallydaPecuária”[34]. Theroutesofthecampaignsweredefined
based on the distribution of cattle production, according to the number of heads per census tract
(Figure1;[35]),takingintoaccountroadconditionsandlogisticalconstraints. Theaccessrestrictions
observedduringthefirstcampaign,andthechangesobservedinthefieldrelativetothecensusdata,
weretakenintoaccountwhendefiningtherouteofthesecondcampaign. Thefirstcampaignwas
heldbetween25Septemberand12November2011,whilethesecondcampaignwasheldbetween
22Augustand1October2012,bothcampaignsincludingtheSouth,Southeast,MidwestandNorth
regionsofBrazil(Figure1). Thetotaldistancecoveredinthesecampaignswas40,000kmandthe
numberofsamplescollectedineachwas369and413,respectively.
RemoteSens.2017,9,73 4of20
Remote Sens. 2017, 9, 73 4 of 20

Figure 1. Location of the pastures sampled during the 2011 and the 2012 campaigns, and total number
Figure1.Locationofthepasturessampledduringthe2011andthe2012campaigns,andtotalnumber
of cattle heads per census tract [35].
ofcattleheadspercensustract[35].
2.2. In Situ Pasture Data Collection
2.2. InSituPastureDataCollection
The characterization of the pasture cover (sampling) was performed every 20 or 40 km,
Tdheepecnhdairnagc toenr itzhaet ipoantho fletnhgethp.a Fsotur reeaccho vpearst(usraem sapmlinpgle), wwae smpeearsfuorremd ebdioepvheyrsyic2a0l poarr4a0mkemter,sd oefp tehne ding
onthevepgaetthatiloenn gttoh .dFetoerrmeainceh pthaes tpuarsetusraem sptalned,w, teoomke paasunoreradmbiico apnhdy sviecratlicpaal rpahmoetotegrraspohfst, hceolvleecgteedta tion
to detgeeromgrianpehtichael cpoaosrtduirneatsetsa, nadnd, tooboskerpveadn ochraarmacitcerainstdicsv oefr ttihcea lppashtuorteo ganradp thhse, lcaondllseccatpeed, ignecolugdrianpgh ical
coordpinasattuerse, asnpdecioebs,s etorvpeodgrcahpahrya, cstoeirli stetixctsuroef, tahnedp oatshteurrse. aAnlld dtahtea lwanerdes craecpoer,diendc lound innugmpbaesrteudr efieslpde cies,
sheets, which were associated to the photographs taken and the EVI2 time series.
topography,soiltexture,andothers. Alldatawererecordedonnumberedfieldsheets,whichwere
Biophysical parameters considered for determination of the pasture stand were: (i) number of
associatedtothephotographstakenandtheEVI2timeseries.
invasive plants, stools, and termite mounds (i.e., degradation and management indicators) found
Biophysicalparametersconsideredfordeterminationofthepasturestandwere: (i)numberof
within three circles with a radius of 3 m randomly placed within each sampled pasture; (ii) number
invasive plants, stools, and termite mounds (i.e., degradation and management indicators) found
of plants and dry weight of the pasture harvested from a 1 m2 quadrat placed within the first circle,
withinfotllhorweiencgi rthclee mswethitohdoaloragdy ioufs [3o6f];3 amndr (aiini)d hoemiglhyt, pmlaeacseudrewdi atht itnhreeea clohcsaatimonps lwedithpians etaucrhe ;sa(imi)pnleudm ber
ofplapnatsstuarned (sdeery Awppeeigndhitxo Af—thFeigpuarset uAr1e fohra irlvluessttreadtiofnro).m a1m2quadratplacedwithinthefirstcircle,
followingTthhies msete tohf opdaroalmogeyterosf w[3a6s] c;oannsdid(eiriei)dh feoirg thhet, dmeteearmsuirneadtioant othf rtehee ploacsatutiroen sstawndit bhyin a esapcehciaslaismt, pled
pastuwreh(os eeevAalpuapteendd tihxeA s—tatFuisg uorfe aAll1 pfaosrtuilrleuss tvriastiitoedn,) .classifying them as degraded, intermediate, or
Tahppisrospertiaotef. pTahrea rmeseutletsr sofw thaiss ceovnalsuidateiorend wfeorre tchoemdpeatreerdm toin tahtei orensuolftst hoebtpaiansetdu rreemstoatneldy.b yaspecialist,
whoevaluatedthestatusofallpasturesvisited,classifyingthemasdegraded,intermediate,orappropriate.
Theresultsofthisevaluationwerecomparedtotheresultsobtainedremotely.
Thefollowingruleswereadoptedforselectionofsamplingpointswithinthepastures: (i)the
samples were located at least 150 m away from the edges of the pastures to compensate for GPS
RemoteSens.2017,9,73 5of20
Remote Sens. 2017, 9, 73 5 of 20
positioningerrorsandedgeeffectsthatcouldpotentiallyinfluencethebiophysicalparameters;and
(ii)thesTithese cfhololosewninwge rruelreesp wreesreen atadtoivpeteodf ftohre sceolnecdtiitoino nosf osfatmhpelpinags tpuoreinstssa wmipthleind ,thi.ee .,ptahsetuhreetse: r(oi)g ethnee ity
samples were located at least 150 m away from the edges of the pastures to compensate for GPS
ofthepasturewastakenintoaccountafteravisualinspectioninthefield.
positioning errors and edge effects that could potentially influence the biophysical parameters; and
2.3.(iRi)e mthoet elsyitSese ncsehdosDeant awere representative of the conditions of the pastures sampled, i.e., the
heterogeneity of the pasture was taken into account after a visual inspection in the field.
MODIS16-daycompositeEVI2timeseries[20]andmonthly25kmTropicalRainfallMeasuring
Mis2s.3io. nRe(mToRteMlyM Se)nrsaedin Dfaaltla estimates from 2000 to 2012 were obtained for all points sampled in the
fieldcampaigns[37]. ToreducenoiseintheMODISEVI2timeseries,wefirstremoveddateswhere
MODIS 16-day composite EVI2 time series [20] and monthly 25 km Tropical Rainfall Measuring
thereflectancewasgreaterthan10%inthebluebandandwheretheviewzenithanglewasgreater
Mission (TRMM) rainfall estimates from 2000 to 2012 were obtained for all points sampled in the field
than 32.5◦, following methods of [38]. The time series were filtered using the wavelet transform,
campaigns [37]. To reduce noise in the MODIS EVI2 time series, we first removed dates where the
following[39].
reflectance was greater than 10% in the blue band and where the view zenith angle was greater than
Pixelsrepresentingthevisitedpasturewerechosenwiththeaidofhighspatialresolutionimages
32.5°, following methods of [38]. The time series were filtered using the wavelet transform, following [39].
availablPeiaxteltsh reevpirretsueanltginlogb theeo vfiGsioteodgl peaMstauprse, wanedrei mchaogseesn owfimthe tdhieu amids opfa htiigahl rsepsaotliuatl iroensosleuntisoonr sim(TaMgesa nd
ETMav+a)ilfarbolme a2t 0t0h0e tvoir2tu01a2l ,galocbqeu iorfe dGoiongtleh eMdarpys, (aAnpdr iimltaogOesc toofb mere)daiunmd wspeattpiaelr rieosdoslu(Mtioany steonSsoerpst e(TmMb er).
Thiasnidn fEoTrmMa+t)i ofrnowma 2s0i0n0t etgor a2t0e1d2,i nactoquairwede bi-nb athseed dtroyo (lAapdraipl tteod Ofrcotombe[r4)0 a].nPda wsteutr epsersimodasl le(Mrtahya ntot he
pixeSlepsitzeem(b2e5r0).m Th×is2 5in0fomrm),aotrionno twreaps riensteengtreadtebdy ianntoy a“ pwuereb”-bpaisxeedl (tio.eo.l, aadpaixpeteldto tfarollmy c[o4n0t]a. iPnaesdtuwreitsh in
thespmaasltluerr et)h,awn ethree epxixcelul sdiezde (f2r5o0m mt h×e 25a0n amly),s oisr. not represented by any “pure” pixel (i.e., a pixel totally
contained within the pasture), were excluded from the analysis.
2.3.1. ProtocolforPastureAssessmentUsingVegetationIndexTimeSeries
2.3.1. Protocol for Pasture Assessment Using Vegetation Index Time Series
Forthedevelopmentofthisprotocol,sevenfarmswerevisitedinthestatesofSãoPaulo,Mato
GrossoFdoor tShuel ,deavnedloPpamreánitn ofO thctiso bperorto2c0o1l2, ,seinvetnh fearcmons twexerteo vfistihteedG ine othdee gstraatdese opf rSoãjoe cPtacuoloo,r Mdiantaot ed
byGErMosBsRo AdoP ASuSl,a atenldli tPeaMrá oinn iOtocrtionbger[ 4210]1.2,D inu rthineg cothnetesxet voifs tihtse, Gexepoderetgsrawdee rpercoojencst uclotoerdditnoateesdta bbyli sh
EMBRAPA Satellite Monitoring [41]. During these visits, experts were consulted to establish the
the criteria that relate the EVI2 time series to the interventions (renovation/recovery, reform, and
criteria that relate the EVI2 time series to the interventions (renovation/recovery, reform, and
degradationprocess)thatoccurredfrom2000to2012. Fromtheserelationships,phenologicalmetrics
degradation process) that occurred from 2000 to 2012. From these relationships, phenological metrics
wereestablished,asdescribedinthefollowingsection.
were established, as described in the following section.
2.3.2. PhenologicalMetrics
2.3.2. Phenological Metrics
Exceptforthevegetativevigor,allphenologicalmetricsextractedfromthetimeserieswerebased
Except for the vegetative vigor, all phenological metrics extracted from the time series were
ontheidentificationofmaximum(max)andminimum(min)valuesofEVI2duringagivencropyear
based on the identification of maximum (max) and minimum (min) values of EVI2 during a given
(Figure 2). For the identification of these values, it was established a time window in which an R
crop year (Figure 2). For the identification of these values, it was established a time window in which
language[42]algorithmidentifiedtheminimumEVI2valueandthedateonwhichitoccurred(dmin).
an R language [42] algorithm identified the minimum EVI2 value and the date on which it occurred
For the first crop year, the algorithm searched the lowest observed value in the eight months that
(dmin). For the first crop year, the algorithm searched the lowest observed value in the eight months
suctcheaetd seudccteheededda tteheo fdtahtee ofifr tshteo fbirssetr ovbasteiorvnaotifotnh oef stehrei esse,riwesh, ewrehaesrefaosr faolrl atlhl ethfeo lfloolwlowinigngc rcorpopy yeeaarsrs,,t he
algothreit halmgohriatdhmas hiatsds atas rittisn sgtaprtoiinngt pthoeindtm thine iddmeinnt iifideendtiifnietdh einp trheev iporuesvicoruosp cyroeapr ypelaurs p1l2usm 1o2n mthosn.tBhas.s ed
onBthaastedd aotne, tihtaste adracthee, dit fsoerarthcheemd info(ra tnhde tmhiend (manind) twheit hdimnina)t iwmitehwini nad toimweo wfeinigdhotwm oofn etihgsh,tc monosnidthesr,i ng
foucromnsoidnethrisnbg effoourre manodntfhosu bremfooren athnsd affotuerr mthoendthast ea.fter the date.

Figure 2. Average EVI2 time series of a visited pasture, highlighting the Local Minimum Limit and
Figure2.AverageEVI2timeseriesofavisitedpasture,highlightingtheLocalMinimumLimitandthe
the duration of the dry period (ddp), the intensity of the dry period (idp), and the vegetative vigor (vv)
durationofthedryperiod(ddp),theintensityofthedryperiod(idp),andthevegetativevigor(vv)for
for the 2002/2003 crop year.
the2002/2003cropyear.
RemoteSens.2017,9,73 6of20
Afterfindingthedatescorrespondingtotheminimum(min)vegetationindicesforallcropyears,
theidentificationofthemaximumvalues(max),andtheirdatesofoccurrence(dmax),wasperformedby
searchingtheperiodscomprisedbetweenthedatesassociatedwithtwoconsecutiveminima. Themin
andmaxvaluesforeachcropyearofthetimeseriesandtheirrespectivedateswereusedtocalculate
otherphenologicalmetricsdescribedinTable1.
Table1.Descriptionofthesevenphenologicalmetrics.
Metric Description
max MaximumobservedvalueofEVI2(itscorrespondingdateisdmax—JulianDay)
min MinimumobservedvalueofEVI2(itscorrespondingdateisdmin—JulianDay)
amp Amplitude:max–min
gur Green-uprate((max−min)/(dmax−dmin))
ddp Durationofthedryperiod(numberof16-dayEVI2composites)
idp Intensityofthedryperiod
vv Vegetativevigor
Forthecalculationofthedurationandtheintensityofthedryperiod(ddpandidp),wecalculateda
LocalMinimumLimit(LML)basedontheobservationsoftheanalyzedcropyearandtheobservations
ofthetwopreviouscropyears,whichdefinedtheso-called“dryperiod”. TheLMLwascalculatedas:
a−b
LML = min + (1)
year
4
wheremin istheminobservedforthecropyearunderanalysis;aisthesmallestmaxvalueobserved
year
forthelastthreecropyears(max ,max andmax );andbisthesmallestminvalueobserved
year year-1 year-2
forthelastthreecropyears(min ,min andmin ).
year year-1 year-2
TheLMLisrepresentedbyahorizontallineinthegraphofthetimeseries(Figure2). AllEVI2
values that were lower than the LML were considered to represent a dry period. The number of
compositions of a crop year in which the EVI2 values were lower than the LML defines the ddp.
TheareabetweenthetimeseriescurveandtheLMLdefinestheidp(Figure2).
Thevegetativevigor(vv)metricwascalculatedasthearea,ineachcropyear,betweenthetime
series curve and the mean EVI2 calculated considering all the values of the time series (Figure 2).
Theamplitude(amp)wascalculatedasthedifferencebetweenmaxandmin,andthegreen-uprate(gur)
wascalculatedas:
max−min
(2)
dmax−dmin
BooleanCriteriaandNumericComparisons
Theuseofphenologicalmetricsallowscomparisonoftwoormorecropyears. Tounderstandthe
processesthatoccurovertime(suchaspasturedegradation),orevenprocessesthatarecharacterized
by detectable changes in a specific time period (for example, human intervention in pasture), it is
commontocomparephenologicalmetricsofagivencropyearwiththosefromcropyearsthatprecede
and/orsucceedit[26,43]. Thisstudywasbasedoncomparisonsbetweenthecropyearandthetwo
previousyearsusing14Booleancriteria(i.e.,trueorfalse). Additionally,14numericalcomparisons
wereperformed(Table2).
RemoteSens.2017,9,73 7of20
Table2.Descriptionofthe14Boolean(bc)andnumerical(nc)criteriausedforcomparisonsbetweenthe
phenologicalmetricsofacropyear(year)withthosefromthetwopreviousyears(year-1andyear-2).
bc BooleanCriterion(bc) nc NumericCriterion(nc)
bc1 ismaxyeargreaterthanmaxyear-1? nc1 (maxyear−maxyear-1)/maxyear-1
bc2 ismaxyeargreaterthanmaxyear-2? nc2 (maxyear−maxyear-2)/maxyear-2
bc3 isminyearlesserthanminyear-1? nc3 (minyear−minyear-1)/minyear-1
bc4 isminyearlesserthanminyear-2? nc4 (minyear−minyear-2)/minyear-2
bc5 isampyeargreaterthanampyear-1? nc5 (ampyear−ampyear-1)/ampyear-1
bc6 isampyeargreaterthanampyear-2? nc6 (ampyear−ampyear-2)/ampyear-2
bc7 isguryeargreaterthanguryear-1? nc7 (bgiyear−bgiyear-1)/iveyear-1
bc8 isguryeargreaterthanguryear-2? nc8 (iveyear−iveyear-2)/iveyear-2
bc9 isddpyeargreaterthanddpyear-1? nc9 ddsyear−dpsyear-1
bc10 isddpyeargreaterthandppyear-2? nc10 ddsyear−dpsyear-2
bc11 isidpyeargreaterthanidpyear-1? nc11 idpyear−ipsyear-1
bc12 isidpyeargreaterthanidpyear-2? nc12 idpyear−ipsyear-2
bc13 isvvyeargreaterthanvvyear-1? nc13 (ivvyear−ivvyear-1)/ivvyear-1
bc14 isvvyeargreaterthanvvyear-2? nc14 (ivvyear−ivvyear-2)/ivvyear-2
2.3.3. IdentificationofAnthropogenicInterventioninthePasture
It is possible to define, with the help of a livestock specialist, procedures for identifying
interventionprocesses. Pasturereformation,forinstance,ischaracterizedbyafastandsteepincrease
invegetativevigor,whichisreflectedasanincreaseinthevegetationindex,especiallyinthemaximum
observedvaluesofEVI2(max)andinthegreen-uprate(gur). Thegurreflectsabruptchangesinthe
vegetation[44]and,inthecaseofpasture,isassociatedwithfertilization,controlofinvasiveplants
andpests,amongotherfactors.
Inthisstudy,pasturereformationwascharacterizedbypixelswheretheBooleancriteriabc1,bc2,
bc5,bc6,bc7,andbc8weretrue,whereasthecriteriabc11andbc12werefalse(seeTable2). Inaddition,
thepixelsshouldhaveamaxvalueinthecurrentcropyearatleast15%higherthanintheprevious
cropyear;i.e.,nc1≥0.15(Table2). Thisthresholdwasestablishedbasedonexamplesofreformed
pastureobservedinthemonitoredfarms(Section2.3.1). Weexpectthattheproposedprotocolwill
detectonlysuccessfulreformations,whereanincreaseofthepasturebiomassisobservedasaresult.
However,wenotethatnotallreformationswillresultinincreasedbiomassduetofactorsthatrange
fromthechoiceofthereformationstrategytotheavailabilityoffinancialresources[30,33]. Insuch
cases,nosignificantchangesareexpectedfortheEVI2values.
Therenewal/restorationprocesseswerecharacterizedbypixelswhereallBooleancriteriawere
true,exceptforbc7andbc8,whichcouldbeeithertrueorfalse. Thiswasbecauseafastincreaseof
the vegetative vigor is a more striking feature in the reformation process than in the renewal and
restorationprocesses,forwhichonemustconsiderthetimeofformationofthenewpasture. Moreover,
therenewal/restorationwerecharacterizedbypixelsshowingavegetativevigor(seeFigure2)inthe
currentcropyearatleastthreetimesgreaterthanthatofthepreviousyear;i.e.,nc13≥2(seeTable2).
This threshold was also established based on data of recovered and renewed pastures in the
monitored farms and expert consultation. Due to the high cost of the operations associated with
renewal/recoveryprocesses[30,33],itwasassumedthattheseinterventionsoccuredonlyinpastures
thathadundergoneseveredegradation. Thesepasturesshowalowamplitudebetweenthemaximum
andminimumvaluesintheEVI2timeseries. ThelowEVI2amplitude,theperiodofexposedsoil
between the removal of the degraded pasture and the planting of the new pasture (reflected as a
declineintheEVI2values),andthemaximumvaluesofEVI2associatedwiththenewpastureinthe
wetperiod,allinfluencedthechoiceofthenc13threshold.
Foreachsample, thealgorithmidentifiedtheoccurrenceofreformationorrenewal/recovery
processes.Forsamplesinwhichnointerventionwasdetectedintheanalyzedcropyears,thealgorithm
searchedfordegradationprocesses. Assumingthatthetypicaldegradationprocessobservedinthe
Remote Sens. 2017, 9, 73 8 of 20
RemoteSens.2017,9,73 8of20
algorithm searched for degradation processes. Assuming that the typical degradation process
ovbissietervdefda rimn sthien vSiãsoitePda ufalormans dinM Saãtoo PGaruolsos oanddo MSualtois Gchroasrsaoct deroi zSeudl bisy crheadruaccttieornizoedf vbeyg eretadtuivcetiovnig oofr
v(ie.eg.e,tbaitoimvea svsi)gaonrd (,ic.eo.n, sbeiqoumenastlsy), aanredd, uccotniosneqinuethnetlyp,a sat urreedcuacrtriyoinn ginca tphaec itpya,sdteugrrea dcaartiroyninwg acsadpeaficniteyd,
dbeagseraddoantiothne wvvasp dheefninoleodg bicaaslemd eotnri cth,ea svdv epshcreinboeldogbieclaolw m.etric, as described below.
22..33..44.. IIddeennttiiffiiccaattiioonn ooff DDeeggrraaddaattiioonn PPrroocceessss iinn PPaassttuurree
TThhee iiddeennttiiffiiccaattiioonn ooff ddeeggrraaddaattiioonn pprroocceessss wwaass ppeerrffoorrmmeedd uussiinngg lliinneeaarr rreeggrreessssiioonn aannaallyyssiiss [[4455,4,466]]. .
FFoorr eevveerryy ppiixxeell iinn wwhhiicchh nnoo iinntteerrvveennttiioonn wwaass iiddeennttiiffiieedd,, tthhee 1122 ccaallccuullaatteedd vvaalluueess ooff vvvv ((22000000––22001122))
wweerree tatakkeenna saosb soebrsveartvioantisonofs aodfe pae nddeepnetnvdaerniat blvea,rwiahbillee, tiwmhei(lye eatirms)ew (aysedaerfis)n ewdaass tdheefiinnedde paens dtehnet
ivnadreiapbelned.eBnot thvavraiarbialeb.l eBsowthe rveanrioarbmleasl iwzeedreb entowrmeeanlizzeerdo baentdweoenne zbeerfoo reanfidt tionnge thbeefroerger efiststiionng ltihnee
r(ee.ggr.e,sFsiigounr eli3n)e. O(en.gc.e, tFhigeuarsesu 3m). pOtinocnes tfhoer raesgsruemsspiotnioannsa floyrs irsewgreersesmioent a[4n7a]l,yasissl owpeertee smtuets i[n4g7]S, tau dselonpt’es
ttewsta usscionngd Sutuctdeednitn’s otr wdears tcoovnedruifcyteidft ihne orredgerer stsoi ovnersilfoyp ief wthaes rseiggrneisfiscioannt salotpthee w1a0s% sliegvneifli,cianndti caat ttihneg
1th0e%o lcecvuerlr,e inncdeicoaftbiniogl othgeic oalccduergrreandcaet ioofn b.iFoilgougrieca3l sdheogwrasdthateioscna. tFteigrpulroet 3o fshthoewnso trhmea sliczaetdtevrpalloute sofo fthvev
nvoerrsmuaslitzimede vfoarlutehse o1f2 vavn vaelyrszuesd tcimroep fyoera trhsef 1o2r oannaeloyfzethde cproixpe ylsemaros nfoitro orende ionf tthhee Gpiexoedlse gmroadnietoprreodje icnt ,
twhiet hGtehoedeesgtriamdaet epdrorjeegcrte, swsiiothn tlhinee e.stimated regression line.

Figure 3. Scatterplot of the normalized values of vv versus time for the 12 crop years of a sample pixel,
Figure3.Scatterplotofthenormalizedvaluesofvvversustimeforthe12cropyearsofasamplepixel,
with the estimated regression line.
withtheestimatedregressionline.
3. Results
3. Results
3.1. Regional Analysis
3.1. RegionalAnalysis
Figure 4 presents the average EVI2 time series for the 267, 389, and 106 samples collected in the
Figure4presentstheaverageEVI2timeseriesforthe267,389,and106samplescollectedinthe
Amazon, Cerrado, and Atlantic Forest biomes, respectively. On average, the pastures sampled in the
Amazon,Cerrado,andAtlanticForestbiomes,respectively. Onaverage,thepasturessampledinthe
Amazon presented maximum values of EVI2 higher than those sampled in other biomes. It is possible
AmazonpresentedmaximumvaluesofEVI2higherthanthosesampledinotherbiomes. Itispossible
that the difference in maximum values of EVI2 among the three biomes is related not only to soil and
that the difference in maximum values of EVI2 among the three biomes is related not only to soil
climate characteristics, but also to differences in stocking rates, which are generally higher in the
andclimatecharacteristics,butalsotodifferencesinstockingrates,whicharegenerallyhigherinthe
Atlantic Forest and the Cerrado [48], directly impacting the availability of forage.
AtlanticForestandtheCerrado[48],directlyimpactingtheavailabilityofforage.
In addition to differences in precipitation regimes, the differences in the minimum values of the
In addition to differences in precipitation regimes, the differences in the minimum values of
EVI2 curves among biomes in Figure 4 can be partially explained by the predominant type of
the EVI2 curves among biomes in Figure 4 can be partially explained by the predominant type of
degradation process in each biome. Invasive plants tend to remain green during the dry season due
degradationprocessineachbiome. Invasiveplantstendtoremaingreenduringthedryseasondue
to their greater ability to absorb water from the soil in dry conditions. As a result, pastures infested
totheirgreaterabilitytoabsorbwaterfromthesoilindryconditions. Asaresult,pasturesinfested
with invasive plants tend to maintain EVI2 values higher than pastures without invasive plants [12].
withinvasiveplantstendtomaintainEVI2valueshigherthanpastureswithoutinvasiveplants[12].
In the Amazon, the main cause of pasture degradation is the change in the biological composition
IntheAmazon,themaincauseofpasturedegradationisthechangeinthebiologicalcomposition
due to secondary forest succession [21,49]. Pastures that have management problems during the
due to secondary forest succession [21,49]. Pastures that have management problems during the
establishment stage, or those attacked by insect pests, are more susceptible to this type of
establishmentstage,orthoseattackedbyinsectpests,aremoresusceptibletothistypeofdegradation.
degradation.
RemoteSens.2017,9,73 9of20
Remote Sens. 2017, 9, 73 9 of 20

Figure 4. Mean values of EVI2 (2000–2012) for the pastures sampled in field campaigns in the
Figure4.MeanvaluesofEVI2(2000–2012)forthepasturessampledinfieldcampaignsintheAmazon,
Amazon, Cerrado and Atlantic Forest biomes (267, 389 and 106 samples, respectively).
CerradoandAtlanticForestbiomes(267,389and106samples,respectively).
Figure 4 can also be used to assess changes in the precipitation regime during the analyzed
Figure4canalsobeusedtoassesschangesintheprecipitationregimeduringtheanalyzedperiod
period and the impacts of these changes on the pasture. However, the influence of rainfall on
andtheimpactsofthesechangesonthepasture. However, theinfluenceofrainfallonvegetation
vegetation indices derived from satellite images varies according to the type of vegetation and soil
indicesderivedfromsatelliteimagesvariesaccordingtothetypeofvegetationandsoilproperties[50].
properties [50]. In the case of pasture, there is yet another important variable that must be taken into
Inthecaseofpasture,thereisyetanotherimportantvariablethatmustbetakenintoaccount: grazing.
account: grazing. The distinction of the impact of each of these variables on the pasture can only be
The distinction of the impact of each of these variables on the pasture can only be performed in
performed in controlled environments, where the history of grazing and precipitation is known
controlledenvironments,wherethehistoryofgrazingandprecipitationisknown[51,52].
[51,52].
3.2. DegradationandIntensificationTrendsintheSampledPastureLands
3.2. Degradation and Intensification Trends in the Sampled Pasture Lands
Theresultsoftheapplicationoftheprotocoltothepasturessampledinthefieldcampaignsare
The results of the application of the protocol to the pastures sampled in the field campaigns are
presentedinTable3. Only14.1%ofthepastureswerereformedbetween2003and2011,whichisless
presented in Table 3. Only 14.1% of the pastures were reformed between 2003 and 2011, which is less
thantherequiredforrehabilitatingdegradedpasturelandinBrazilby2020(15Mha),asestablishedby
than the required for rehabilitating degraded pastureland in Brazil by 2020 (15 Mha), as established
theSectorialPlanforMitigationandAdaptationtoClimateChangefortheConsolidationofaLow-Carbon
by the Sectorial Plan for Mitigation and Adaptation to Climate Change for the Consolidation of a Low-Carbon
EconomyinAgriculture(ABCprogramme;[53]). Ofthesampledpastures,19.0%presentedasignificant
Economy in Agriculture (ABC programme; [53]). Of the sampled pastures, 19.0% presented a
declineinbiomassasmeasuredbytheregressionanalysisdescribedinSection2.3.4.
significant decline in biomass as measured by the regression analysis described in Section 2.3.4.
Table 3. Results of the application of the protocol for pasture assessment using vegetation index
Table 3. Results of the application of the protocol for pasture assessment using vegetation index time
timeseries.
series.
2011 2012 Total
PastureStatus 2011 2012 Total
Pasture Status
SamSpalemsples % % SSamamplpelses %% SaSmamppleless % %
RefoRremfoatrimonation 52 52 14.114.1 5252 1122..66 101044 13.133 .3
Renewal/Recovery 50 13.6 49 11.9 99 12.7
Renewal/Recovery 50 13.6 49 11.9 99 12.7
ReformationandRenewal/Recovery 3 0.8 2 0.5 5 0.6
Reformation and Renewal/Recovery 3 0.8 2 0.5 5 0.6
InDegradation 58 15.7 91 22.0 149 19.1
WithouInt IDnteegrvreandtaiotinon 206 58 55.185.7 21991 5232..00 144295 19.514 .3
WithTooutat lIntervention 369206 55.8 412319 53.0 427852 54.3
Total 369 413 782
The results of the intersection between the assessment of the pastures in the field and the
The results of the intersection between the assessment of the pastures in the field and the
classification obtained with the EVI2 time series are presented in Table 4. We found that, of the
classification obtained with the EVI2 time series are presented in Table 4. We found that, of the 778
778samplesevaluatedbyboththeprotocolandthefieldassessment,34(4.4%)displayedasignificant
samples evaluated by both the protocol and the field assessment, 34 (4.4%) displayed a significant
reductioninbiomassasshownintheEVI2timeseries,andwereclassifiedasdegradedwhenthefield
reduction in biomass as shown in the EVI2 time series, and were classified as degraded when the field
surveywasperformed. Thepasturesclassifiedasintermediateordegradedinthefieldrequiresomekind
survey was performed. The pastures classified as intermediate or degraded in the field require some
ofintervention,withouttillingofthesoiland/orchangingtheforage. Wefound30samples(3.8%)in
kind of intervention, without tilling of the soil and/or changing the forage. We found 30 samples
thiscategory,and80(10.3%)sampleswithsuitablestand,butwithadecreasingtrendinbiomass.
(3.8%) in this category, and 80 (10.3%) samples with suitable stand, but with a decreasing trend in
biomass.
Among the 207 pasture samples found to have gone thought some kind of intervention, only 38
were classified as degraded in 2011 or 2012. However, if we consider the interventions carried out in
2009, 2010, and 2011, only one sample was classified as degraded in the field, and this sample was
classified by the protocol as pasture reformed in 2011, after the field visit. From the pastures showing
RemoteSens.2017,9,73 10of20
Table4.Resultsoftheintersectionbetweenthefieldassessmentandthepastureclassificationobtained
withtheEVI2timeseries.
Stand(FieldAssessment)
Intervention/Degradation
Appropriate Intermediate Degraded
(EVI2TimeSeries)
Samples % Samples % Samples %
Reformation 63 14.4 20 11.8 20 11.8
Renewal/Recovery 54 12.3 27 15.9 18 10.7
ReformationandRenewal/Recovery 4 0.9 1 0.6 0 0.0
InDegradation 80 18.2 30 17.6 34 20.1
WithoutIntervention 238 54.2 92 54.1 97 57.4
Total 439 170 169
Amongthe207pasturesamplesfoundtohavegonethoughtsomekindofintervention, only
38 were classified as degraded in 2011 or 2012. However, if we consider the interventions carried
outin2009,2010,and2011,onlyonesamplewasclassifiedasdegradedinthefield,andthissample
was classified by the protocol as pasture reformed in 2011, after the field visit. From the pastures
showing no degradation trend (427), which have not gone through interventions in the analyzed
period,330(42.4%)wereclassifiedasappropriateorintermediate(Table4).
The results of the application of the protocol are presented in Table 5 and Figure 5 by biome.
ThesamplescollectedinthePantanalbiomehadthegreatestpercentageofpastureunderdegradation
(40%). However,thenumberofsampledpasturesinthisbiomewasrelativelysmall(20samples)when
comparedtootherbiomes. Amongthebiomeswiththegreatestnumberofsamples, theAmazon
hadthehighestpercentageofpasturesamplesindegradationprocess,followedbytheCerrado,and
theAtlanticForest. PasturereformationwasamorecommonpracticeinsamplesofAtlanticForest,
whilerenewals/recoveriesweremorecommonintheAmazonpastures, whenthePantanalisnot
considered(Table5). Thesedifferencescanbeexplained,inpart,bytheleveloftechnologyadoptedin
eachbiome[48]. However,factorssuchascattleprice,marketconditions,anddistancetothemain
consumermarketsmayinfluencetheproducer’sdecisiononwhethertointervene[54].
Table5.Pasturestatusforthesamplescollectedinthisstudy.Percentagesareexpressedasafraction
ofthetotalnumberofsamplesperbiome.
Amazon Cerrado AtlanticForest Pantanal
PastureStatus
Samples % Samples % Samples % Samples %
Reformation 31 11.6 53 13.6 18 17.0 2 10.0
Renewal/Recovery 38 14.2 46 11.8 12 11.3 3 15.0
ReformationandRenewal/Recovery 1 0.4 2 0.5 2 1.9 0 0.0
InDegradation 60 22.5 69 17.7 7 6.6 8 40.0
WithoutIntervention 137 51.3 219 56.3 67 63.2 7 35.0
Total 267 389 106 20
InMidwestBrazil,wheretheCerradobiomeispredominant,themostdirectcauseofpasture
degradationisthesystematicuseofstocknumbersthatexceedthepasturecarryingcapacity,causing
defoliation and loss of soil nutrients, aggravated by the lack of a proper management for pasture
recovery[29,54,55]. Overgrazingreducesthevegetationcoverandincreasesthedecompositionrates
of organic matter and erosion, which are features of biological degradation [12]. For this kind of
degradation, the most adopted strategy is the recovery of the soil fertility by manuring, usually
performedduringtheprocessofreformation.