Mexico Deforestation Vulnerability Analysis and Capacity Building

Full text

i


 



Mexico Deforestation Vulnerability Analysis and
Capacity Building. Final Project Report

Environmental Defense Fund (Consortium Lead),
Conservation International, and Center for Global
Development
September 11, 2014


ALIANZAMÉXICOPARALAREDUCCIÓNDE
EMISIONESPORDEFORESTACIÓNYDEGRADACIÓN

PolíticapúblicaDesarrollodecapacidadesArquitecturafinancieraMRVComunicación

ii





www.alianza‐mredd.org.mx

ThisreportwasmadepossiblebythegeneroussupportoftheAmericanpeoplethroughtheUnited
StatesAgencyforInternationalDevelopment(USAID)underthetermsofitsCooperative
AgreementNumberAID‐523‐A‐11‐00001(M‐REDDProgram)implementedbyprimerecipientThe
NatureConservancyandpartners(RainforestAlliance,WoodshallResearchCenterandEspacios
NaturalesyDesarrolloSustentable).Thecontentsandopinionsexpressedhereinarethe
responsibilityoftheM‐REDDPROGRAManddonotnecessarilyreflecttheviewsofUSAID.Views
expressedinthispaperdonotnecessarilyreflectthoseoftheanyoftheauthors’institutions.Any
remainingerrorsaretheauthors’soleresponsibility.

TheauthorsarealsothankfulforvaluableguidancefromBronsonGriscomandPeterEllisofTNC,
helpfulcommentsfromYvesPaiz,JuanFranciscoTorresOrigel,andJoseCantoVergaraofTNC,
earlyadviceandsupportfortheprojectfromLeticiaGutierrezLoranidiofTNC,aswellasvaluable
inputfromparticipantsintheworkshopon“ModelingDeforestationintheYucatanandBeyond”
convenedbyTNCandMREDDfromApril30toMay2,2014inMerida,Mexico.WealsothankJose
CarlosFernandezofINECCforinspirationandinvaluablesupportfromthefirstinceptionofthe
nationalanalysis.Inaddition,theauthorsaregratefulforimportantmodelingcontributionsfrom
RuohongCaiofEDF,GISsupportfromJeremeyProvilleofEDF,andeditorialassistancefromPasha
FeinbergofEDF.Theauthorsgratefullyacknowledgekeydataonmaximumcroprevenuesfrom
JensEngelmannofCGDaswellasdataoncommunalpropertiesfromLeonardGoffandAllen
BlackmanatRFF.

iii




Acknowledgments
Authors
RubenLubowski,MaxWright,KalifiFerretti‐Gallon,A.JavierMirandaArana,MarcSteininger,and
JonahBusch
Contactdetails:
Projectlead RubenLubowski;rlubowski@edf.org
Literaturereviewandmeta‐analysis KalifiFerrettiGalon(lead);kalifi.fg@gmail.com
JonahBusch;jbusch@cgdev.org
Nationalmodeling RubenLubowski(lead)
A.JavierMirandaArana(consultant);javmi@yahoo.com
Localmodeling MaxWright(lead);twright@conservation.org
MarcSteininger;msteininger@conservation.org

OrganizationCore Staff
EnvironmentalDefenseFund RubenLubowski(ChiefNaturalResourceEconomist)
A.JavierMirandaArana(Consultant)
ConservationInternational MarcSteininger(ScientificDirector)
MaxWright(RemoteSensingandGeospatialAnalyst)
CenterforGlobalDevelopment JonahBusch(ResearchFellow)KalifiFerretti‐Gallon
(ResearchAssistant)



iv



Contents
1.ExecutiveSummary1
1.1.KeyFindings1
2.Introduction5
2.1.GlobalGreenhouseGasesandMexico’sForests5
2.2.ReportOutline5
3.LiteratureReviewofDriversofDeforestationinMexico11
3.1.Introduction11
3.2.Overviewofdeforestation11
3.2.1.DeforestationinMexico 11
3.2.2.DeforestationintheYucatán 12
3.3.Overviewoflandtenure,ruralagriculturalsupport,andpaymentsfor
ecosystemsservicesinMexico12
3.3.1.LandTenure 12
3.3.1.1.Privatelands12
3.3.1.2.PublicLands12
3.3.1.3.CommunalLands12
3.3.2.RuralAgriculturalSupport 14
3.3.3.PaymentsforEcosystemsServices 15
3.4.Databaseregressionresults16
3.4.1.AMeta‐analysisofDriversofDeforestationinMexico:Methods 16
3.4.2.ResultsforMexicoandSEsub‐regions 17
4.Analysisofdeforestationatnationallevel/OSIRIS21
4.1.Introduction21
4.2.EmpiricalModel21

v

4.2.1.EconometricSpecification 21
4.2.1.1.Relationshipofdeforestationtoavailableforestareawithina900mgridcell22
4.2.1.2.Observedandunobservedcomponentsofnetreturnsfromlandconversion24
4.3.HistoricalSimulations26
4.3.1.SimulationScenario 26
4.3.2.SimulationResults 28
4.3.2.1.ChangesinAgriculturalReturns28
4.4.Futureprojections32
4.4.1.1.“Business‐as‐usual”projection33
4.4.1.2.CarbonIncentiveProjections37
5.LocalModelingofDeforestation46
5.1.Introduction46
5.1.1.Overallapproach 46
5.1.2.Definitionofextents 46
5.2.DataandMethods46
5.2.1.Deforestationdata 46
5.2.2.Otherdata 48
5.2.3.Spatialmodeling 50
5.3.Results51
5.3.1.Deforestationsince2000 51
5.3.2.Modeleddeforestationbeyond2012 53
5.4.Predictingdeforestationinthefuture:56
5.5.Conclusions66
6.Conclusion69
6.1.Summaryofreportfindingsanddirectionsforfutureresearch69
6.1.1.Literaturereviewandmeta‐analysis 69
6.1.2.Nationalmodeling 70
6.1.3.LocalModeling 71

vi

7.Workscited73

Listoftables,figuresandmaps
Tables
Table3.4.1DriversofdeforestationinMexico,bydrivercategory17
Table4.2.1.Principalexplanatoryvariablesusedinnationalregressions(900mcell)25
Table4.3.1.Simulationscenariosoverhistoricalperiodindataset,2000‐201226
Table4.3.2.NationalSimulationResults29
Table4.3.3.RegionalSimulationResultsforSensitivitytoAgriculturalReturns31
Table4.4.1.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,byAATR
referenceregionsandlandownershipcategory35
Table4.4.2.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,bynational
regionsandAATRReferenceRegions36
Table4.4.3.FuturePredictions,2014‐2024,Business‐as‐Usualand$10/tonCO2PolicyCase,
forAATRandnon‐AATRregions40
Table5.2.1.Driverindependentvariablesusedforspatialmodelsatthelocallevel.49
Table5.3.1.Summaryofforestcoverin2000anddeforestationfrom2000to2012among
AATRs.52
Table5.3.2.Relativeimportanceofthedifferentdrivervariablesformodelsrunineachof
thelocalstudyareas.SeeTable5.3.1forthelistofvariables.54
Table5.4.1.Predicteddeforestationfrom2012‐202258

Figures
Figure2.2.1.ProjectFlowchart9
Figure3.4.1DriversofDeforestationinMexico:ResultsofMeta‐Analysis18
Figure3.4.2.DriversofDeforestationintheYucatánPeninsulaasComparedtotheRestof
Mexico:ResultsofMeta‐Analysis20
Figure4.4.1.EstimatedcostcurvesforCO2emissionsreductionsfromabove‐groundforest
carbonlossesinMexico,byregion45
Figure4.4.2.Estimatedcostcurvesforreducingemissionsfromabove‐groundforestcarbon
lossesinMexico,byAATRandnon‐AATRregions.45
Figure5.4.1.Predicteddeforestation2012‐2022,OaxacaIstmo.AATRsitehighlightedin
yellowthatching.59

vii

Figure5.4.2.Predicteddeforestation2012‐2022,OaxacaMixteca.AATRsitehighlightedin
yellowthatching.60
Figure5.4.3.Predicteddeforestation2012‐2022,OaxacaSierraNorte.AATRsite
highlightedinyellowthatching.60
Figure5.4.4.Predicteddeforestation2012‐2022,SierraChiapas.AATRsitehighlightedin
yellowthatching.62
Figure5.4.5.Predicteddeforestation2012‐2022,CutzmalaValleBravo.AATRsite
highlightedinyellowthatching.63
Figure5.4.6.Predicteddeforestation,SierraPUCC.AATRhighlightinyellowthatching64
Figure5.4.7.Predicteddeforestation,SierraRaramuri.AATRsitehighlightinyellowthatch
65

Maps
Map4.4.1.Projected“BusinessasUsual”(BAU)ForestLoss2014‐202441
Map4.4.2.ProjectedAvoidedForestLoss2014‐2024,with$10/tonCO2incentive42
Map4.4.3.ProjectedRemainingForestLosswith$10/tonCO2incentive,2014‐202443
Map4.4.4.ProjectedAvoidedEmissions2014‐2024,with$10/tonCO2incentive44















ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

1

1. ExecutiveSummary
In2010,Mexicoranked8thamongcountrieswiththelargestareaofprimaryforest(FAO
2010).Mexico’sforests,coveringaboutathirdofthenation,provideanumberofservicesincluding
carbonsinks,highlevelsofendemismandspeciesrichness,andsubsistenceresourcesforlocal
population.TheseservicesarebeingerodedasMexicocontinuestoexperienceforestcoverloss.
Mexicohaslostabouthalfitsforestareasince1950.From2005‐2010,thecountrymaintainedan
averagedeforestationrateof0.24%accordingtoFAO,reducingitscapacityforcarbon
sequestrationandincreasinglandconversionrelatedemissions.Landuse,land‐usechangeand
forestrywasrecentlyestimatedtoemitabout10%ofMexico’stotalGHGemissions.
TheMexico‐REDDAlliance(MREDD)issupportingMexico’seffortstoreduceitsemissions
fromdeforestationandforestdegradationandtoenhanceforestcarbonstocksarecurrently
supportedby.TheprogramhasidentifiedEarlyActionAreas(ÁreasdeAcciónTempranaorAATR),
orhighrisk–highrewardareaslocatedinMexicanstatesthatarerecognizedashavinghigh
biodiversity,culturaldiversityaswellashighratesofdeforestation.Researchrelatedtoforest
coverlossinMexicohassofarfocusedondriversofdeforestation,includingtheimpactofland
ownershiptypesuniquetothecountry(communityforestry,protectedareas,andprivatelands).
Missingfromthesizeableliteraturearetwotopicsofparticularimportancefortheidentificationof
vulnerableregionsandthedesignofconservationstrategiesunderMREDD:thefirstisananalysis
ofdriversofdeforestationthatisspecifictotheAATRs.Thesecondisananalysisoftheeffectof
geographiccharacteristicsorpolicymeasuresthatisdisaggregatedbylandownershiptype.
Tohelpaddressthesegaps,weconductaseriesofanalysesthatcombinebothnationaland
localscalemodelingtoaidtheMREDDAlliancepartnersinassessingthevulnerabilityofMexico’s
foreststodeforestation.TheseanalysesfocusonthevulnerabilityofforestedlandswithinMexico’s
AATRs,accountingforMexico’suniqueforestmanagementdynamicsthroughdisaggregatingthe
resultsbylandownershiptypes.Theseanalysesareultimatelymeanttoinformnationaland
subnationalpolicy,pavingthewayforincentivebasedprograms,andultimatelyreduced
deforestationvulnerabilityinMexico.Ourmethodologyincludesthreedifferentand
complementaryapproaches:(i)reviewingtheexistingliterature,(ii)anationaleconometric
analysisandassociatedscenariosimulationmodeling,and(iii)local‐levelspatialspatialmodeling
foreachAATR.Keyfindingsfromeachofthesethreepartsofthereportaresummarizedbelow.
1.1.KeyFindings
LiteratureReviewofDriversofDeforestationinMexico.
 Whiledeforestationhasdecreasedoverthepastdecades,forestlosscontinuesatabout
0.24%peryear,accoridingtoUN‐FAO,generatingabout6%ofthecountry’stotal
greenhousegasemissionsin2010.Deforestationismostlyoccurringinthemoredensely
forestedareasofsouth‐easternMexicandlargelyattributedintheliteraturetocropand
cattledevelopment.

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

2

 Landtenure(communitylandmanagement,includingejidos),ruralagriculturalsupport,
andpaymentsforecosystemsservicesaremajorfocusesoftheliterature.Conclusionson
theroleofthemajorlandtenuretypeinMexico,communitylandmanagement,aremixed.
Studiesarealsoindisagreementontheroleofsuchruralagriculturalsupportprogramsas
PROCAMPO.
 Moststudiesagreethatpaymentsforecosystemsservicesdecreasedeforestationrisk,with
somecaveatsrelatedtoregionaldifferencesandstartingdeforestationrisk.
 Theserelationshipsweremirroredinthemeta‐analysis:regressionresultsweremixedfor
ejidosandruralincomesupport,whileresultsforPEStendedtobeassociatedwith
decreaseddeforestation.Furthermore,resultsfromthemeta‐analysisrevealedother
variableswithconsistentrelationshipstodeforestationinMexico.Thevariablesmost
associatedwithreduceddeforestationinMexicowereassociatedwithprotectionmeasures
(asproxiedbyprotectedareasandPES),reducedaccessibility(elevation),reducedresource
competition(propertysize)andcommunityforestry.
 Thevariablesmostassociatedwithincreaseddeforestationwererelatedtoareaswhere
economicreturnstoagriculturearehigher(proximitytoagricultureandagriculture
returns),biophysicalconditionsforconversionarefavorable(soilsuitability),and
competitionforresourcesarehigh(population).
 MostoftheserelationshipswererobustwhenresultsweredisaggregatedtotheYucatán
Peninsula.Notablyhowever,atthenationallevel,povertyappearstobelinkedtoincreases
indeforestation,whileintheYucatánPeninsulapovertyisassociatedwithdecreased
deforestation.Conversely,indigenouspopulationisassociatedwithdecreased
deforestation.
NationalAnalysisofDeforestationinMexico.
 Thenationalanalysisrevealsthatacriticaldriverofdeforestationhasbeentheanticipated
economicreturnsfromlandconversion,specificallyfromagricultureasproxiedbycrop
productioninourstudy.Keyfactorsmodulatingdeforestationvulnerabilityincludeland
ownershiptypeandinitialforestareawithinagridcell.
 Weestimatetheresponsivenessofgrossdeforestationtochangesinneteconomic
incentivesforlandconversion.A1%decreaseinpotentialagriculturalreturnsover2000‐
2012wouldhavedecreasedcumulativegrossdeforestationnationallyoverthisperiodby
anestimated0.24%.Conversely,a1%increasewouldhaveboostedgrossdeforestationby
anestimated0.26%.Similarly,a10%decreaseinpotentialagriculturalreturnsover2000‐
2012wouldhavedecreasedcumulativegrossdeforestationnationallyoverthisperiodby
anestimated2%.Conversely,a10%increasewouldhaveraisedgrossdeforestationbyan
estimated3.3%.
 Apreliminaryexaminationsuggeststhatdecreasingpotentialcropreturns(orincreasing
benefitstolowemissionsactivitiesthatavoiddeforestation)bytheamountofPROCAMPO
subsidiesonejidosandagrariancommunitylandswouldhavedecreaseddeforestationby
about5%over2000‐2012.

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

3

 Basedontheeconomicprofitabilityofagricultureandstartingforestcoverin2012,the
modelpredictsanoverall27%“business‐as‐usual”increaseinannualdeforestationin
Mexicooverthenexttenyears,relativeto2000‐2012.
 Ontheonehand,thereisrelativelyhighsensitivitytoagriculturalreturnsandhigh
estimatedfuturevulnerabilitytodeforestationamongforestremantsinareaswith
relativelysparserforestcover,includingintheNorthwestandBajioandNortheastregions.
Comunidadesandprotectedareaswerethelandtypesprojectedtohavethebiggest
proportionalincreaseinforestlossesoverthenext10yearsandarealsoestimatedtohave
thegreatestpercentdeclinesinresponsetoapotentialcarbonincentive.
 Themostsensitiveareas,however,arenotthatimportantinabsoluteterms.Thegreatest
amountofdeforestationisprojectedtooccurintheSouthandYucatanPeninsularegion,as
wellaswithinejidosandprivatelandtypes.Theseareas,particularlytheYucatan
Peninsula,holdthelion’sshareofestimatedpotentialforreducingdeforestationand
emissions.
 ThesevenAATRsarenotallconcentratedintheareaswiththehighestprojectedfuture
deforestation,andsomeofsitesarelocatedinareaswithlowhistoricalratesofforestloss,
comparedtothenationalaverage.Nevertheless,overallasagroup,AATRsandtheir
surroundingregionshavehigherprojecteddeforestationincreasesthanotherforested
areasnationally,aswellasregionally,aswellasthemajorityofthepotentialtocost‐
effectivelyavoiddeforestation.
 Weuseourstatisticalparameterstoestimatenationalandregionalcarbonemissionscost
curves,basedonahypotheticalcarbonincentivefocusingonlyonabove‐groundforest
carbon.Wefindthatthereisrisingpotentialnationallytoreduceemissionsatcostsranging
from$5to$100/tonCO2,atwhichpointabout90%oftheemissionsareavoided.About
halfoftheestimatedreductionsavailableatpricesof$10/tonCO2orbelowandmorethan
twothirdsoftheestimatedreductionsavailableatpricesof$20/tonCO2orbelow.The
nationalandregionalcostcurvesarerisingatanincreasingrate,indicatingthatitcosts
moreandmoretoavoiddeforestationonlandswithgreateragriculturalpotentials.
LocalModelingofDeforestationinMexico.
 Judgingfromtheprojecteddeforestationscenarios,thegreatestbenefitsfrom
implementingREDD+oranotherincentivebasedconservationactivitywouldbefeltin
AATRsitesthatareprimarilyunfragmentedforest,meaningthattheystillcontainlarge
areasofundisturbedcoreforest,andareexperiencingfrontierexpansion,usuallystemming
frompopulationcentersoraccesspoints.SitessuchasSierraPUCCCheneandOaxaca
IstmodisplaythesecharacteristicsascomparedtositeslikeOaxacaMixtecaorSierra
Raramuriwhicharehighlyfragmentedandexperiencelowerratesofdeforestation.
 Variablesrelatedwithaccessibilityandmarketsweremostinfluentialintheless
fragmentedreferenceregions,whilevariablesrelatedtobiophysicalsuitabilityweremost
influentialinthefragmentedsites.Thevariable,distancetomegacites,wasimportantin

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

4

thetworegionsthatcontainedthem(SierraRararmurinearCuliacunandCutzemalaValle
BravonearToluca).
 Theremaybemultiplepatternsofforestchangepresentinthereferenceregions;lossof
primaryforest,lossofsecondaryforest,fallowrotationsandagro‐forestry.Modelscouldbe
strengthenedbyaddressingtheseseparatelyorfocusingonaparticularpattern.
 Theinterpretationofthelocalmodelsshouldincludeboththesoftandhardpredictions
underthevariousscenariosaswellasthegeneralpatternofthesofttransitionsurface.
 Futureworkcouldincludeamorethoroughexaminationoftheeffectsofland‐usepractices
withincomunidadesandejidos,asthesedesignationshadsomeinfluenceoverthemodels,
howevertheresultsweremixed.




ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

5

2. Introduction
2.1.GlobalGreenhouseGasesandMexico’sForests
Greenhousegasemissionsfromagriculture,forestryandotherland‐useactivitiesaccount
foranestimated24%ofglobalemissions,secondonlytoemissionsproducedbyfossilfuel
combustion(IPCC5thAssessmentReport,2014).In1990,officialestimatesarethatdeforestation,
forestdegradation,andotherland‐usechangesinMexicoproducedover100MtCO2eofemissions
peryear,accountingfor18.2%ofnationalemissions.Morerecentlyin2010,forestsandland‐use
changesproducedcloseto47MtCO2eorabout6.3%oftotalemissions(SEMARNAT/INECC,
2012).Mexicoiscurrentlyundertakingeffortstoreduceitsemissionsfromdeforestationand
forestdegradationandtoincreasesequestrationbyenhancingforestcarbonstocks(REDD+),
supportedbytheMexico‐REDD(MREDD)Allianceprogram.1Crucialtothesuccessofanti‐
deforestationpoliciesisanunderstandingofhowspatialvariationingeographiccharacteristics,
landownership,economicprofitability,andpolicymeasuresaffectMexico’svulnerabilitytoforest
coverloss.Itisalsoimportanttounderstandhowpotentialchangesinthesefactorsovertime
mightaffectdeforestationinthefuture.
Mexicohaslostroughlyhalfitsforestareasince1950.From2005to2010,thecountrylost
155,000hectaresofforestcover,anaveragedeforestationrateof0.24%(FAO,2010).Forest
conservationinMexicoprovidesbiodiversityco‐benefitsbeyondclimate,asthecountryboasts
bothhighlevelsofendemismandspeciesrichness(Barsimantov&Kendall,2012).While
deforestationrateshavedecreasedandreforestationeffortsareevident(FAO,2010),widespread
deforestationcontinuestothreatencommunitiesandecosystemsthatdependonforests.Thereisa
needtobetterunderstanddeforestationinMexicotoidentifyvulnerabilitiesandinformpolicies
thataimtoreduceforestloss.
2.2.ReportOutline
Missingfromthesizeableliteratureonland‐usechangeinMexicoaretwotopicsof
particularimportancefortheidentificationofvulnerableregionsandthedesignofandlow‐
emissionsdevelopmentstrategiesunderMREDD:thefirstisananalysisofdriversofdeforestation
thatisspecifictotheREDD+earlyactionareas(AATRs)undertheMREDDprogram.2Thesecondis
ananalysisoftheeffectofgeographiccharacteristicsorpolicymeasuresthatisdisaggregatedby
landownershiptype(e.g.ejido,protectedarea,privatelands).

1TheAllianzaMREDD+isapartnershipofTheNatureConservancy,RainforestAlliance,theWoodsHole
ResearchCenter,Mexico’sgovernment,andcivilsocietytohelplaythebasisforeffortstoreduceemissions
fromdeforestation,forestdegradation,andotherforestryactivities(i.e.REDD+).(see:www.alianza‐
mredd.org)
2ÁreasdeAcciónTemprana(AATR)areREDD+EarlyActionareaslocatedinMexicanstateswithhigh
biodiversity,culturaldiversityandhighratesofdeforestation,butalsogreatREDD+potential.Lessons
learnedinthesesubnationaltargetareascouldhelpscaleupbestpractices.

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

6

Toaddressthesegaps,weconductaseriesofanalysesthatcombinebothnationalandlocal‐
scalemodelingtosupporttheMREDDAlliancepartnersinassessingthevulnerabilityofforested
landstodeforestationinMexico,focusingonthevulnerabilityofforestedlandswithinMexico’s
AATRs,withtheresultsdisaggregatedbylandownershiptypes.Thegoalistoinformnationaland
subnationalpolicy,pavingthewayforincentivebasedprograms,andultimatelyreduced
deforestationvulnerabilityinMexico.Ouranalysisonlyconsideredforestlosses,ratherthangains,
duetodatalimitations.WhileincreasingforestgainscouldbeanimportantpieceofREDD+
programs,afocusonavoidingdeforestationshouldcapturethelargestnear‐termopportunitiesfor
reducingnetemissionsfromforests.
Thisprojectgeneratedseveralanalyticresultsaswellasdataproducts,including:
‐ Avulnerabilitydataset:aspatiallyexplicitrasterdatasetinwhicheachcellhasavalue
indicatingtherelativeriskoffuturedeforestation,bothatthenationalandregionalscale.
‐ Afuturedeforestationprojection:aspatialdatasetprojectinglocationsoffuture
deforestationasafunctionofthevulnerabilitydatasetandpredictedratesoffuture
deforestation.
‐ Adatabaseofallvariablesinthemodelinganalyses.
‐ AdatabaseofeconometricstudiesofthedriversofdeforestationinMexico(andother
countries.

Thisreportdescribesourvulnerabilityanalysisandkeyfindings,alongwiththemethods
usedtogeneratethe“soft”and“hard”deforestationprojections‐‐thevulnerabilitymapand
deforestationprojections,respectively.Ourmethodologyincludesthreedifferentand
complementaryapproaches:(i)reviewingtheexistingliterature,(ii)conductinganational
econometricanalysisandbuildinganassociatedpolicysimulationmodel,and(iii)conductinglocal‐
levelspatialanalyses.TheflowdiagraminFigure2.2.1illustratestheroleofthedifferent
projectcomponentsandassociatedinputsandoutputs.
Wepresentanddiscussthemainresultsfromeachofthesethreeunderlyinganalyses.The
modelingincludedtestingthepredictivepowerofaseriesofindividual“driver”datasets,which
mayormaynotactuallycausedeforestation,butarepotentiallycorrelatedwithit.Wediscussthe
relativepredictivepowerofthedifferentdriverdatasets,withspecialfocusontheircorrelation
withthespatialdistributionofhistoricandprojecteddeforestationwitheachoftheseven
identifiedAATRs.Wealsoseektounderstandhowdeforestationmightchangecausallyinthe
futurewithchangesintheeconomicincentivesgoverningforestcoverloss.
Thefirstapproachisaliteraturereviewandmeta‐analysisofexistingstudiesof
deforestationandland‐usechangeinMexicoandelsewheregloballytoidentifytrends,
contradictions,andtoprovidecontextonland‐usedecision‐makinginMexico,aswellasinother
countries(Ferretti‐Gallon&Busch,2014).Thisreviewuncoversgapsintheliterature,informsthe
selectionofdrivervariablesforthenationalandlocalmodelingdescribedbelow,andprovides
contextforevaluatingthemodelingresults.

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

7

Thesecondapproachmodelstheimpactofdifferentdriversofland‐usechangeatthe
nationalscale,tocomplementandprovideinputstothelocalanalysesconductedusingtheIDRISI‐
SelvaLandChangeModeler(LCM).ThenationalanalysisforMexicoadaptstheapproachofthe
OpenSourceImpactsofREDD+Incentives(OSIRIS)model,whichwasdevelopedforanalyzingthe
impactofalternativeREDD+policiesinBolivia,Madagascar,PeruandIndonesia(Busch,etal.,
2012).3OurnationalanalysisforMexicofocusesonidentifyingtheimpactofonevariablethatis
arguablyofcausalimportancefordeforestation:theneteconomicreturnsperhectarefrom
convertinglandfromforesttonon‐forestlanduses.Usingthislargergeographicscaleisespecially
importanttocapturebroadervariationineconomicvariablesinordertoexplicitlymeasuretherole
ofchangingeconomicreturnsfromcompetinglanduses.Inparticular,wemodeldeforestationin
relationtovariationinestimatedgrossagriculturalrevenuesandproxiesforfixedandvariable
costsusingobservablesitecharacteristics.Theestimatedresponsivenesstotheeconomic
profitabilityofagriculturallanduseprovidesthebasisforsimulatingdeforestationunder
alternativescenarioswithdifferenteconomicincentivesforforestprotection,includingtheeffectof
potentialREDD+policies.
Thenationalsimulationyieldsanestimateddeforestationvulnerabilitymapatthenational
scaleata900mresolution.Weusethenationaleconometricmodeltoconductaseriesof
simulationsthatyieldregionalpredictionsofdeforestationunderabusiness‐as‐usual(BAU)
referencescenarioaswellasasetofhypotheticalpolicycases.Theseregionalpredictionsprovide
aninputtothelocalscaleanalysestomakepredictionsonfuturedynamicsofforestcoveratseven
AATRs.
ThethirdapproachusesLCMinordertodrawonitspredictivespatialmodelingcapacityto
morefinelydisaggregatetheregionalresultsacrossthelandscapeinthelocalstudyareas.Foreach
ofthesevenAATRs,theLCMmodelsexaminetherelationshipbetweenpotentialdrivervariables
andobservedpatternsofdeforestation.Thesemodelsgeneratea“soft”vulnerabilitymapaswellas
a“hard”predictionofdeforestationunderaseriesofhistoricalandalternativescenarios,informed
bythemoreaggregatepredictionsofthenationallevelmodel.
TheempiricalanalysesinthisstudyuseanewglobaldatasetfromtheUniversityof
Maryland,basedonLandsatsatelliteinformation,justreleasedinJanuaryofthisyear(Hansen,et
al.,2013).Toourknowledge,thisstudyisthefirsteconometricstudytoexploittherichspatial
detailandmultipletimeperiodsfromthesenewdata.Assuch,resultsfromouranalysisand
approachforMexicocouldprovideinsightsforanalyzingdeforestationinothercountriesand
regionsaswell.

3TheOpenSourceImpactsofREDD+Incentives(OSIRIS)modelisasuiteoffree,transparent,open‐source,
spreadsheet‐baseddecisionsupporttools.OSIRISgoesbeyondpredictionsofthespatialdistributionandrate
offuturedeforestationtoestimateandmaptheclimate,forestandrevenuebenefitsofalternativepolicy
decisionsforREDD+.See:http://sp10.conservation.org/osiris/Pages/overview.aspx


ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

8

Thisreportisdividedinto6sections.Section3describestheliteraturereviewofdriversof
deforestationinMexico.Section4discussesthenational‐scaleeconometricanalysis.Section5
presentsthelocalmodelingfortheAATRs.Section6concludes.

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

9

Figure2.2.1.ProjectFlowchart

ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report

10




11

3. LiteratureReviewofDriversofDeforestationinMexico
3.1.Introduction
Wecompiledadatabaseofeconometricstudiesofdeforestation,including117
studiesglobally,ofwhich23studiesfocusonMexico.AppendixTableA‐1providesan
annotatedbibliographyoftheMexicostudies.Fromouranalysis,drivervariables
associatedwithlowerratesofdeforestationinMexicoincludedprotectedareas,community
forestry,andpaymentsforecosystemsservices.Drivervariablesassociatedwithhigher
ratesofdeforestationinMexicoincludeagriculturalactivity,population,soilsuitabilityand
proximitytourbanarea.Theseassociationsbetweendifferent“drivers”anddeforestation
donotnecessarilyindicatecausalrelationships.CausalstudiesofprotectedareasinMexico
havefoundtheseterritoriestobelinkedwithdecreaseddeforestation.Causalstudiesof
ejidoshavenotbeenperformed,suggestingtheneedforfurtherstudy.
3.2.Overviewofdeforestation
3.2.1.DeforestationinMexico
AllknowncategoriesofMexicanforestcover(tropicaldry,tropicalwet,and
montaneforests)havebeensubjecttodeforestation(Vaca,etal.,2012).Deforestationis
occurringmostlyinSouthernMexico,withthehighestratesoccurringinthestatesof
CampecheandQuintanaRoo.Whilerecentstudiesobserveapatternofnetdeforestationin
Mexico(Vaca,etal.,2012),recentlythenation’stotalannualdeforestationhasdecreased.
Between1990and2000,Mexicolost354,000ha/year;from2000to2005,thearea
deforestedannuallyhaddecreasedto235,000;and,from2005‐2010,Mexico’sforestloss
furtherdeclinedto155,000haperyear(FAO,2010).
ReforestationhasoccurredinsomeregionsofMexico(about178,000ha/yearfrom
1990‐2010)(FAO,2010).Thistrendhasbeenattributedtoplantedforestswithproduction
astheirprimaryfunction(FAO,2010).Reforestationthroughtreeplantationsisaresultof
increaseddemandforoilpalm,eucalyptus,andcitrusproducts.Regenerationofforest
coverisalsoseenasaresultofpassivetransition,wherefarmersabandonlandandmigrate
toareaswithbetterpaidfarmjobs.Itcanalsobearesultofactivetransition,inwhichthe
growingscarcityofforestproductsencouragegovernmentsandlandownerstoplanttrees,
i.e.sustainablecommunityforestmanagement(Vaca,etal.,2012).Thereislittleevidenceof
naturalforestregeneration.
AlthoughthedeforestationrateinMexicohasdeclined,widespreadforestcoverloss
persists.Mostdeforestationprocessesareattributedtoagriculture(mainlycoffee,maize,
beans,andsugarcane)andcattledevelopment.Otherhistoricdriversofdeforestationhave
includedhumansettlement,monocultureforestry(inSouthernMexico)andnatural
phenomena(e.g.,hurricanesandfiresintheYucatán)(Vaca,etal.,2012).Population
growth,poverty,andphysiogeographicvariablesareclaimedtobesignificantdriversof
forestlossinMexico(Barsimantov&Kendall,2012).However,literatureonthesubject
rendersconflictingconclusionsontheeffectsondeforestationofotherdrivervariables,


12

includinglandownership,subsidyprograms,roaddensityandpercapitaincome
(Barsimantov&Kendall,2012).
3.2.2.DeforestationintheYucatán
InMexico,mostoftheGulfCoastlowlandshavealreadybeendeforested,and
significantlandclearanceoccurredintheinteriorLacandonforestsofChiapas(TurnerII,et
al.,2001).TheforestsofsouthernCampecheandQuintanaRoohavebeenconsideredthe
lastfrontierinthe“westtoeastmovementoftropicallowlanddevelopment”inMexico
(TurnerII,etal.,2001).TheSouthernYucatánhasbeenidentifiedasadeforestationhot
spot(Rueda,2010).Itisconsideredtobeoneoftheworld’simportantforestedregions,
characterizedbytheCalakmulBisophereReserveandtheMesoamericanBiological
Corridor(Busch&Geoghegan,2010).Itisthereforecrucialtounderstanddriversofland‐
useandland‐coverchangeintheregion. 
3.3.Overviewoflandtenure,ruralagriculturalsupport,andpaymentsfor
ecosystemsservicesinMexico
3.3.1.LandTenure
Mexicohasalonghistoryofpolicyreformsfocusedonpropertyrightsandtherole
oflandtenureonlandcoverchange(Bonilla‐Moheno,etal.,2013).Therearethreetypesof
landmanagementinMexico:Private,public(protectedareas,publicenterprises,etc.),and
communal(comunidadesagrariasandejidos).
3.3.1.1.Privatelands
Asof2011,privatelandsthatareownedand/ormanagedbycompanies,
sharecroppers,andlandlessruralpopulationrepresent37%oftheMexicanagrarian
landscape.Theseprivatelands,however,onlyencompass26%ofthecountry’sforests
(Corbera,etal.,2010).
3.3.1.2.PublicLands 
Publiclands,inturn,belongtofederalorregionalpublicagencies,aswellasto
publicenterprises.Theselandsrepresentjustover8%oftheagrarianlandscapeandcover
only4%offorestedareas,primarilyincludingprotectedareasandbodiesofwater
(Corbera,etal.,2010).
3.3.1.3.CommunalLands
Landsundercommonmanagementisthemostcommontypeofmanagement,
representing52%oftheMexicanagrarianlandscapeand70%oftheforests(Corbera,etal.,
2010).Therearetwomaintypesoftenurearrangements:comunidadesagrarias(agrarian
communities)andejidos.Comunidadesagrariasrefertorepatriatedindigenouslandsand
ejidosarelandsgrantedbythepostrevolutiongovernment(Barsimantov&Kendall,2012).
Botharecommunallyownedlands.NucleosAgrariosisageneraltermforejidosand
comunidadesagrariasinMexico.CarrilloandMota‐Villanueva(2006)explainthatthis
generalizationisbasedonsharedcharacteristicslikelegalstatusandlandownershipgiven
byPresidentialActorbytheHighAgrarianCourtofJustice.


13

A.HistoryofCommunalLands
A.aComunidadesagrarias
TheSpanishCrowngrantedtheselandrightstogroupsconsideredoriginalsettlers.
Thecommunitiesthatdeveloped,therefore,consistofpeoplewhohavehistorically
inhabitedaregionandsharelanguage,traditionsandgoverninginstitutions.Landholder
typesinthisformofmanagementconsistofagrariancommunitiesandindividualrights
holders(comuneros).Forestregulationisgovernedbyacommunalassemblymadeupofall
comuneros(someofwhommaybewomen).Acouncilofauthoritiesisrenewed
periodically,normallyeverythreeyears(Corbera,2010).
A.bEjidos
Ejidos,ontheotherhand,areamorespecificformoflandmanagementthan
comunidadesagrarias.Theywereestablishedwhenagroupoffamiliesclaimsrightsovera
territory,andtheparceloflandgrantedtothesegroupsremainsundercommunal
ownership.Anyrentalorlandsalesareprohibited.Landcanonlybegivenbyoneejido
landholder(ejidatario)toasingledescendant.Forestandlandforpasture(forfuelwood
collection,timberharvestingandgrazing)areusuallymanagedincommon.Forestfor
timberharvesting,inparticular,isorganizedthroughcommunitymembersandgroups,or
throughexternalconcessions.Ejidotimberconcessionsareorganizedthroughextraction
quotasandcorrespondingbenefitsaredefinedanddistributedthroughtheejidoassembly
and/orthecouncilofauthorities.
Bothcomunidadesagrariasandejidoshavemembers(avecindados)whohavebeen
givenaparceltofarmandanothertoliveon,butwhodonothaverightstobenefitsfrom
theforest.Itisestimatedthatthereareover30,000agrariancommunitiesandejidosinthe
country,occupyingover50%ofthetotalnationalterritory(PROCEDE,2010).Community
landmanagementinMexicoisoftenclaimedtohavepositiveenvironmentaland
socioeconomicoutcomes(Barsimantov&Kendall,2012).
B.Historyofcommunallandmanagement
Mexico’scurrentsystemoflandmanagementdevelopedfrompost‐revolution
governmentlandmanagementreform.AftertheMexicanRevolutioninthe1910s,Article
27ofthe1917Constitutiondeclaredthatalllandsandwatersoriginallybelongedtothe
nationandthatthenationwouldgrantprivatepropertyrightsundercertainconditions
(CamaradeDiputados,2008).Article27limitedthesizeofprivateproperties,parceled
largeprivatelandholdingsand,mostimportantly,grantedrightstoruralcommunitiesand
groupsoffamiliestoownlandtomeettheirbasicneedsortorestorecustomaryrightsheld
beforethe1800s(Corbera,etal.,2010).Theshareofcommunallandincreasedupuntilthe
early1980s.Intheearly1990s,Article27wasreformed,legalizingtheformationofjoint
venturesbetweencommunallandholdersandprivatecapital.Thisallowedcommunityland
managementmembersandejidomemberstobecomeprivateowners,andtorentandsell
landtothirdparties.Forests,however,couldnotbesubdividedandsold,excludingthem
fromprivatization(Corbera,etal.,2010).


14

C.Impactofcommunityforestryondeforestation
Amajorityofpublishedacademicstudieshaveconcludedthatcommunityforestry
doesnotinfluencedeforestation.Forinstance,Perez‐Verdinconcludedthatdeforestationis
drivenbyresource‐specificcharacteristics,suchaslocationandsoilproductivity,andnotby
ejidos’attributes(Perez‐Verdin,etal.,2009).However,a2012studyreviewedevidence
relatedtocommunityforestmanagementandforestcover,findingthatcommonproperty
andcommunityforestryaresignificantlyrelatedtoreducedratesofdeforestationand
increasedratesofforestrecoveryofconiferousforestsinMexico(Barsimantov&Kendall,
2012).Theirresultssuggestthatcommonpropertycanleadtogreaterforestconservation
whenthereisaneconomicallyvaluableassettoprotect(coniferousforests)andwhenthere
aremanagementplansinplacetoformalizetheextractionprocessandrevenue
distribution.Anotherstudyconfirmedthatcommunitylandmanagementpracticeshave
resultedinthemaintenanceofforestedlandscapeinsomeareasofMexico(Bray,etal.,
2004).Butotherstudiesconcludedthatcommunitymanagementhasmixed,ifnota
negativeeffectonforestcover(Vance&Iovanna,2006)(Alix‐Garcia,2007).Astudyin
2010demonstratedthatthecharacteristicsoftheejido,ratherthanthepresenceorabsence
anejidalsystem,determinetheimpactondeforestation:populationdensity,agricultural
productionandintensificationwithinejidosaffecteddeforestationrates(Rueda,2010).
VanceandGeoghegan(2002)observedincreasingdeforestationasejidodemographics
change,withageandpopulationdensitybeingsignificantlypositivelyrelatedto
deforestation.Geogheganetal.(2004)supportsthisconclusionandfurtherpositedthat
deforestationprimarilyfollowsagriculturalexpansionbytheejidosector,thepredominate
formoflandtenureinthesouthernYucatán.
3.3.2.RuralAgriculturalSupport
TheroleofgovernmentagriculturalsubsidiesondeforestationinMexicoismixed.
In1999,astudywasdonecontrastingtheeffectswhichtheBancodeDesarrolloRuralor
RuralDevelopmentBank(BANRURAL)creditandtechnicalassistancehaveon
deforestation.Itwasinitiallythoughtthatthistypeofaidwouldincreaseagricultural
intensification,therebyrelievingpressureonnearbyforestsforfutureconversion.The
studyrevealedthat“governmentsubsidizedcreditfailedtospuraprocessofagricultural
intensificationthatcouldhavesubstitutedforcuttingdownforests”(Deininger&Minten,
1999).Thesameauthorsproducedanotherstudyafewyearslaterthatdeterminedthat
BANRURALis,infact,associatedwithsignificantlyhigherlevelsofdeforestation,andthat
thesecreditsubsidies“seemtohaveencouragedthecuttingdownofforests”(Deininger&
Minten,2002).
Asecondstudythatsameyearconfirmedthatanotherruralsubsidyprogram,
ProgramadeApoyosDirectosalCampoorFamersDirectSupportProgram(PROCAMPO),is
alsoassociatedwithhigherlevelsofdeforestation(Vance&Geoghegan,2002).PROCAMPO
isaMexicanruralsupportprogramcreatedtoalleviatethefinancialimpactoftheNAFTA
onagriculturalworkersin1994(Klepeis&Vance,2003).Theprogramwasalso
implementedwiththeintentionofdecreasingenvironmentaldegradationthroughthe
promotionofmoreefficientlanduse,usingfundstointensifyproductionanddecrease


15

pressureonremainingforests(Klepeis&Vance,2003).Theresultingincreasein
deforestationputstheprogramatoddswithitsintent.VanceandGeoghegan(2002)
suggestpoorintegrationoflandownersintomarketsthatwouldotherwiseencourageland‐
intensivechemicalinputsasareasonforincreasedagriculturalexpansionand,
consequently,decreasedforestcover.Thesamestudyalsosuggeststhatthespecificterms
oftheprogram,whichstipulatetheareaandlocationsupportedbyPROCAMPObe
maintainedundercontinuousproduction,discouragesaforest/fallowagriculturalmethod
thatmaintainsthefertilityofsoilsused.LaterstudiesofPROCAMPOreportedmixed
results(Geoghegan,etal.,2004)orinsignificantrelationships(Chowdhury,2006).
Alternatively,anotherstudyfoundthateachhectareregisteredinPROCAMPOactually
decreasedthehazardofdeforestationby2.21%(Vance&Iovanna,2006).
AthirdcreditprogramthatmayaffectdeforestationistheProgramaNacionalde
SolidaridadorMexico’sNationalSolidarityProgram(PRONASOL).Themostrecentstudy
onPRONASOLandforestcoverchangedeterminedthattheprogram’ssubsidiesinnorthern
municipalitiesarecausingaconsiderableincreaseinforestloss,whilesubsidiesinthe
southandeastarenot(Jaimes,2010).TheeffectofMexico’sruralagriculturalsupport
programsondeforestationrequiresfurtherstudyofthetypesofruralagriculturalsubsidies
andwhereandtowhatextenttheyarerelatedtodeforestation.
3.3.3.PaymentsforEcosystemsServices
Mexicohasalreadydesignedandimplementedapaymentsforecosystemsservices
(PES)program,apaymentsforhydrologicalservicesprogram(PSAH),whichisdesignedto
incentivizetheincreasedproductionofhydrologicalservicesthroughforestconservation
(Alix‐Garcia,etal.,2012).ThroughPSAH,theMexicanfederalgovernmentpays
participatingforestownersforthebenefitsofwatershedprotectionandaquiferrechargein
areaswherecommercialforestryisnotcurrentlycompetitive(Munoz‐Pina,2008).Most
studieshavefoundthatthisapplicationofPESinMexicoreducesdeforestationtosome
extent.Anumberofstudiesonprotectedforestsrevealthatacombinationoflegalforest
protectionandfinancialincentiveshashelpedreducedeforestationinMexico(Honey‐
Roses,etal.,2011).In2011,astudyfoundthatacombinationoflegalprotectionandPES
hashelpedprotectforesthabitatforthemonarchbutterflyinMexico.Thestudyestimated
thatwithoutthejointconservationinitiative,lossesofforestwouldhavebeen3%and11%
higherinareaswithjustaloggingbanorwithdensecanopy,respectively(Honey‐Roses,et
al.,2011).In2012,inanotherstudyanalyzingPSAH,resultssuggestedPESinMexico
reduceddeforestationthatwouldhaveoccurredunderBAUscenarios,butresultresults
wereuneven.Itwasfurtherrevealedthattheprogramseemedtobemoreeffectivein
generatingavoideddeforestationwherepovertyislowerandinthesouthernandnorth‐
easternstatesofMexico(Alix‐Garcia,etal.,2012).A2008studyrevealedthatwhilePSAH
isassociatedwithreduceddeforestation,theprogram’spaymentshavebeeninareaswith
lowdeforestationrisk,suggestingthattheselectioncriteriabemodifiedtobettertarget
higherriskareas(Munoz‐Pina,2008).Thereisroomforfurtherstudyonsocio‐economics
oftheareaunderPSAHaswellasotherpotentialPESprogramdesigns.


16

3.4.Databaseregressionresults 
3.4.1.AMeta‐analysisofDriversofDeforestationinMexico:Methods
Recenttechnologicalandmethodologicaladvancementshaveencouragedthe
proliferationofeconometricstudiesofdeforestationgroundedinremotelysensedevidence
offorestcoverloss.Wehavecompiledacomprehensivedatabaseof117econometric
studiesofdeforestation,including23studiesinMexico,publishedbetween1996and2014.
Tobeincludedinthedatabase,studieshadtomeetfivecriteria:(1)thedependentvariable
mustmeasureforestcoverorforestcoverchange;(2)thedependentvariablemustbe
remotelysensed;(3)thedependentvariablemusthaveresulted,inpart,from
anthropogeniccauses;(4)thearticlemustincludeatableofmultivariateregression
outputs;and,(5)thearticlemusthavebeenpublishedinapeer‐reviewedjournal.The
databaseismeanttobeasinglesourceforalleconometricstudiesofdeforestation,allowing
easyaccessandanalysisofdeforestation.Thisdatabasewascreatedtoprovidean
overviewofcurrentscientificunderstandingofforestcoverloss,toimprovepolicy
implementationaimedatdeforestationmitigation,andtoidentifygapsinscientific
evidencerequiringfurtherresearch.
Fromtheindividualstudieswecategorizeddrivervariables(n=1159)into“meta‐
variables”suchaselevation,proximitytoroad,oragriculturalactivity,ofwhich33were
includedinthestudiesofdeforestationinMexico(Table3.4.1).Asinglemeta‐variableis
thesumofallregressionresultsfromindicatorsmeasuringthesamephenomenon.For
instance,themeta‐variableElevationiscomprisedofvariableslabelled“Elevation,”“Mean
Elevation,”“Altitude”etc.WhileTable3.4.1presentsacomprehensivelistofdriver
variablescollectedinthedatabasefromstudiesinMexico,somevariableshaveyettobe
analyzedduetothecomplexityofinterpretingthevariable(e.g.SoilType).
Foreachmeta‐variable,withineachstudy,wesummedthenumberofregression
outputsormatchingoutputsthatfoundtheassociationbetweenthatmeta‐variableand
deforestationtobenegativeandsignificant,notsignificant,orpositiveandsignificant.
Theseresultswerethenorganizedintoadatabaseuponwhichwebasedouranalysis.We
termedthemeta‐variabletobeconsistentlyassociatedwithlower(orhigher)deforestation
iftheratioofpositiveandsignificantoutputstonegativeandsignificantoutputswas
statisticallysignificantlylessthan(orgreaterthan)1:1inatwo‐tailedt‐testatthe95%
confidencelevel.Wetermedthemeta‐variabletobenotconsistentlyassociatedwithlower
orhigherdeforestationiftheratioofpositiveandsignificantoutputstonegativeand
significantoutputswasnotstatisticallysignificantlydistinguishablefrom1:1.







17

Table3.4.1DriversofdeforestationinMexico,bydrivercategory
BiophysicalBuilt
Infrastructure
Agriculture,Pasture,
andWorkingForests
Demographics,
Poverty,and
Income
Land
Management
Elevation
(n=15)
Slope(n=16)
Wetness(n=8)
ForestArea
(n=3)
SoilSuitability
(n=6)
Proximityto
Clearing(n=9)
Proximityto
Water(n=3)
Proximityto
Road(n=13)
Proximityto
UrbanArea
(n=12)

AgriculturalActivity
(n=9)
ProximitytoAgriculture
(n=8)
AgriculturalPrices(n=4)
EconomicActivity(n=2)
LivestockActivity(n=2)
TimberActivity(n=1)
TimberPrice(n=1)
UseofFuelwood(n=1)
Population(n=10)
Poverty(n=14)
Education(n=8)
Indigenous
Population(n=8)
Age(n=1)
Presenceof
Females:(n=1)
PropertySize(n=7)
RuralIncome
Support(n=8)
Off‐Farm
Employment(n=3)

TenureSecurity
(n=6)
ProtectedAreas
(n=6)
PlotSize(n=4)
LandUse(n=4)
Logging
Activities(n=3)
PES(n=2)
Community
Forestry/Ejidos
(n=15)

Note:“n”indicatesthenumberofstudiesthathaveanalyzedthemeta‐variableinrelationto
deforestationinMexico,outofatotal23studies.Wecategorizedeveryregressionresultreportedin
theincludedstudiesintooneofthreecategories.Regressionresultsshowinganegativeand
significantrelationshipbetweenadrivervariableanddeforestationwerecodedas“‐“;regression
resultsshowingapositiveandsignificantrelationshipbetweenadrivervariableanddeforestation
werecodedas“+“;regressionresultsshowingnosignificantrelationshipbetweenadrivervariable
anddeforestationwerecodedas“n.s.“
3.4.2.ResultsforMexicoandSEsub‐regions
Theresultsforhoweachmeta‐variableisassociatedwithdeforestationacross
statisticalstudiesofdeforestation,areshowninFigures3.4.1and3.4.2attheendofthis
sectionandFigureA‐1intheAppendix.Figure3.4.1presentsthedatabaseresultsforall
studiesfocusedonMexico.InMexico,variablesmostassociatedwithdecreasesin
deforestation,includeprotectedarea,propertysize,elevation,communityforestry,and
paymentsforecosystemsservices(PES).Therearesomepredictableresults:thatprotected
areasandPESareassociatedwithdecreaseddeforestationisnotsurprising.Forestsin
areasofhigherelevationmaywellbemoreremoteandhavemorelimitedaccess.That
increasedpropertysizeisassociatedwithlowerdeforestationcouldreflectthatbigger
propertiesimplyfewerlandusers,andconsequentlyreducedcompetitionforforest
resources.
Variablesassociatedwithincreaseddeforestationincludeproximitytoagriculture,
population,agriculturalactivityandsoilsuitability.Again,theserelationshipsareprobably
notsurprising:deforestationinMexicooccurswhereeconomicreturnstoagricultureare
higher(asproxiedbyproximitytoclearedlandandagriculturalactivity)andwhere
biophysicalconditionsarefavourable(asindicatedbysoilsuitability).Populationisalso
generallyassociatedwithincreaseddeforestation,asitsuggestsincreasedcompetitionfor
forestresources.


18

Figure3.4.1DriversofDeforestationinMexico:ResultsofMeta‐Analysis

Note:ThisgraphpresentsregressionresultsfromstudiesondeforestationinMexico.Resultsare
orderedbyratioofnegativetopositiveassociationwithdeforestation.
Mostvariablesthatarenotconsistentlysignificantareperhapsalsonotsurprising.
Asexpected,resultsforruralincomesupportaremixed.Surprisingly,however,community
forestryismoreconsistentlyassociatedwithlessdeforestation,whereastheeffectofejidos
ondeforestationismixed.Weseparatedvariablesreferringspecificallytoejidosandthose
referringtothebroadertermofcommunityforestry.Thisdiscrepancysuggestsmorestudy
isneededofthedifferencesbetweenvariouscommunitylandtenuresinMexicoandtheir
respectiverelationshipswithdeforestationrates.Alsosurprising,variablesindicating
indigenousterritoryarenotsignificantlyrelatedtodeforestation,eitherpositiveor
negative,inMexico.Inourglobalstudywefoundindigenouslandtenureiscommonly
associatedwithdecreaseddeforestation(Ferretti‐Gallon&Busch,2014).


19

FigureA‐1intheAppendixcomparesresultsontherelationshipsbetweenthe
variablesanddeforestationatthegloballevelandinMexico.Duetospacelimitations,the
figureonlyincludesthe15topvariablesthathavebeenmostincludedinregression
analysesattheMexicolevel.Still,thefiguresuggeststhatvariablesaffectingdeforestation
aregenerallythesameinMexicoasatthegloballevel.Protectedareaextentandelevation
arebothassociatedwithdecreasedratesofdeforestationandarerobustatbothlevelsof
study.Ontheotherhand,globally,communalforestmanagementisassociatedwith
increaseddeforestation,whileattheMexicolevel,thecommunityforestry(includingboth
ejidosandothervariablesrelatedtocommunallandownership)isassociatedwithlower
deforestation.Similarly,ruralincomesupportisassociatedwithincreasesindeforestation
atthegloballevel,buttheresultsaremoremixedattheMexicolevel.Finally,attheglobal
level,povertyisassociatedwithlowerdeforestation,whileinMexicoincreasedpoverty
appearstobeassociatedwithhigherdeforestation.
Figure3.4.2comparesresultsdisaggregatedfromtheMexicoleveltotheYucatán
Peninsula(includingtheYucatán,QuintanaRoo,andCampeche,butexcludingTabasco).
Duetospacelimitations,thegraphagainonlyincludesthe15topvariablesthathavebeen
mostregressedattheYucatánlevel.Variablesassociatedwithlessdeforestation(property
sizeandelevation)andvariablesassociatedwithmoredeforestation(population,proximity
toagricultureandpopulation)arerobustatthislevelofdisaggregation.Notably,poverty
againhasaninconsistentassociationwithdeforestation.Atthenationallevel,poverty
appearslinkedtoincreasesindeforestation,whileintheYucatánPeninsulapovertyis
associatedwithdecreaseddeforestation.Asimilarinconsistencyisnotedwithindigenous
populations.WhileatthenationallevelIndigenousterritoryisassociatedwithdecreased
deforestation,thesamevariableisassociatedwithincreaseddeforestationattheYucatan
Peninsulalevel.TheseinconsistenciesperhapssupportthewidelyheldviewthatMexico’s
landscapeandtherelateddriversofdeforestationvarygreatlybyregion.
Itisimportanttoemphasizethedistinctionbetweencorrelation,orassociation,and
causation.Toprovideonewell‐knownexample,ratesofdeforestationmightbelower
withinprotectedareasbecauseprotectedareasarepreventingdeforestationfrom
occurring(causality).Thisrelationshipmightalsobebecauseareasthathavelowratesof
deforestationforotherreasonssuchasgeographicremotenesshavegreaterintact
biodiversity,whichledtoprotectedareasbeingdesignatedinthoselocations(anexample
ofreversecausality).Disentanglingtheseeffectsrequiresspecializedtechniquessuchas
matchingmethods,whichhavebeenperformedinMexicoforprotectedareasandpayments
forecosystemservices(Honey‐Roses2011),butnotyetforejidos,suggestinganavenuefor
furtheranalysis.






20

Figure3.4.2.DriversofDeforestationintheYucatánPeninsulaasComparedtothe
RestofMexico:ResultsofMeta‐Analysis

Note:ThisgraphdisplaysregressionresultsfromstudiesfocusedontheYucatánPeninsula
(includingCampeche,QuintanaRooandYucatán)ascomparedtoresultsfromthestudiesfocusedon
therestofMexico.Foreachmeta‐variable,twosetsofresultsarereported:thefirstsetrepresents
resultsfortheYucatánPeninsulainlightercolors,whilethesecondsetrepresentsresultsforMexico
indarkercolors.Resultspermeta‐variableareorderedbyratioofaveragenegativetoaverage
positiveassociationwithdeforestation.


21

4. Analysisofdeforestationatnationallevel/OSIRIS
4.1.Introduction
WeconductedaneconometricanalysisofdeforestationinMexicoatthenational
scaleinordertocalibrateasimulationmodeltoexploretheimpactofalternativeeconomic
andpolicyscenarios.Inparticular,weanalyzeddetailedspatially‐explicitdataonannual
forestcoverlossesacrossallofMexicoover2000‐2012.Oureconometricanalysisisbased
ontheideathatlandowners4willchoose,fromasetofpotentiallanduses,theoptionthat
bringsthehighestexpecteddiscountedreturnsThegoalistoexplicitlycapturethe
influenceoftheeconomicnetbenefitsfromconvertinglandfromforesttonon‐forestuses
forthepurposesofcalibratingapolicy‐simulationmodelthatcan,forexample,analyzethe
impactofdifferentREDD+policystructures,orotherpotentialpaymentsforecosystem
services.
Thenationalmodelservesto1)measuretheimpactofdifferenthistoricaldriversof
land‐usechange2)generateaspatialdistributionofprobabilityoffuturedeforestation
underalternativepolicyandmarketscenarios,3)helptoidentifycost‐effectivemitigation
opportunitiesandestimatetheopportunitycostsofabatingcarbonemissionsfrom
deforestation,and4)provideabasisforexaminingpolicydesignelementssoastocreate
economicincentivesfortheimplementationofREDD+inMexico.Inparticular,resultsfrom
aneconometricanalysisservetocalibratethesimulationandestimationonthedistribution
andtotalrateofdeforestationacrossMexicounderasetofeconomicandpolicyscenarios
thataltertheeconomiccalculusforlandconversion,lookingretrospectivelyover2000‐12
aswellasoutintothefutureoverthenext10years.Thenationalmodelpredictssite‐level
deforestationbasedonfittedvaluesfromtheeconometricmodel,estimatedusingobserved
deforestation.Inparticular,wemodeldeforestationinrelationtovariationinestimated
grossagriculturalrevenuesandproxiesforfixedandvariablecostsusingobservablesite
characteristics.Theresultsfromthesimulationprovideregionaldeforestationratesasan
inputtotheLCMmodelingofthesevenAATRs.
4.2.EmpiricalModel
4.2.1.EconometricSpecification
Severalchallengesariseindevelopinganempiricallytractablespecificationto
identifytheroleofeconomicreturnsindrivingdeforestationinMexico.Oureconometric
approachfocusesonaddressingtwomainsetsofissues.Thefirstsetofissuesrelatestothe
structureofourdependentvariable,whichisanaggregationofthenativedataatthe30m
cellresolution.Theaggregationintroducesthechallengeofmodelingarangeofpotential
changesinforestareawithinalargergridcell,wherethepotentialmagnitudeofchangesis

4InMexico,approximately70%offorestlandhasacommunalformofownership(Corbera,etal.,
2010).Therefore,forouranalysisbothprivateindividualsandcommunitiesaretherelevantland
ownersormanagers.


22

linkedtotheamountofforestareawithineachgrid.Thesecondsetofissuesrelatestothe
factthatweonlyhaveimperfectobservationsofeconomicreturnsforourunitsof
observation,asmentionedabove.Afulldiscussionofournationalmodel,econometric
approach,data,andestimationresultsareprovidedinAppendixI.
4.2.1.1.Relationshipofdeforestationtoavailableforestareawithina900mgridcell
Ourunderlyingdatasourcefordeforestation,ourdependentvariableinterest,
providesbinaryinformationonthepresenceornon‐presenceofforestsatthe30mcelllevel
foreachyearbetween2000and2012,providingatotalof11observedannualchanges
(Hansen,etal.,2013).Whileweconductthelocalscaleanalysesatthismostdetailedlevel
ofresolution,ananalysisatthislevelofdetailisnotcomputationallytractableforallof
Mexicoasthiswouldinvolveover1billionpointsperyearoralmost13billiondatapoints
acrossall11observedyearlychangeperiods.Tomakethenationalanalysis
computationallyfeasible,weaggregateour30mx30mcellsintolarger900mx900mcells,
eachofwhichcontain900potentiallyforestedsmallercellsatthe30mresolution.This
procedurereducesthesizeofthedatasettoabout1.39millionobservationsannually,after
eliminatingany900mgridcellsnotcontaininganyofthesmaller30mforestedcellsinthe
year2000.5Atthisscale,ourpreferredspecificationstilltookabout24hourstorunonour
mostpowerfulcomputerwith24GBofRAM.
Ourconstructeddependentvariableisthustheannualchangeinforestcoverfrom
2000through2012oneach900mcellcontainingforests,spanningallthecontinentalland
areaofMexico(i.e.,islandswereexcluded).Ourunitsofanalysisthusmeasure900mx
900mor810,000m2(equivalentto81haor0.81km2).Werestrictattentionto900mcells
thatcontainatleastoneforested30mcell.Thechangewithineachoftheseunitsis
measuredintermsofthenumberofconstituent30mcellsthatareforestedatthestartof
theyearbutthenchangefromforesttonon‐forestcoverovertheyear.Whileweassignthe
sameexplanatoryvariablestoallthesmaller30mcellswithineachofour900munits,we
thusmodelchangesin30mcellincrements.Thesechangesmightrepresentdecisionsby
oneormorelandownerswithineach900mcell.Wedonothavecomparableannualdata
forpossibleforestgainsonthesecells,soonlyconsiderforestlossesinourmodel.6Thus,if
a900mcelllosesallofitsforestcoverinaparticularyear,thatcelldoesnotenterintoour
econometricanalysisinanysubsequentyears.

5Givenavailabledata,weonlyexaminelossesofforestcoverinareasthatwereforestedin2000.
Thusourdeforestationanalysiscannotconsiderdeforestationonareasthatwerenotforestedin
2000butcouldhavesubsequentlygainedandlostforestbetween2000and2012.Thisis
appropriategivenourfocusontheREDD+policyandthegreatercarbonandbiodiversityvalues
associatedwithmorematureforests,ratherthanrecentlyregeneratingforests.
6While(Hansen,etal.,2013)doprovidedataoncumulativeforestgainsfrom2000to2012,an
analysisofthesedatawouldhaverequiredaseparateanalysisandwasbeyondthescopeofthe
currentstudy.


23

Thestructureofourdependentvariableraisesseveralissues.Thefirstissueisthat
ourdatahasa“count”structure,asforestareaandchangesinareaaremeasuredindiscrete
units,rangingfrom0upto900,themaximumnumberof30mcellswithinalarger900m
gridcell.Giventhiscountstructure,oureconometricestimationmethodisaPoissonquasi‐
maximumlikelihoodestimator(QMLE)whichisconsistentwithestimatingacountvariable
generatedbyindependent,binarydecisionsatthe30mcellresolution(Wooldridge,2002).
Forrobustness,wealsoconducttheanalysisusinganegativebinomialmodel,which
modifiesthePoissonregressionmodelwithamultiplicativerandomeffecttorepresent
unobservedheterogeneity.Thisisawaytoaddresspotential“over‐dispersion,”whichisa
commonsituationinanalysesofcountdata,wheretheobservedvarianceofthedependent
variableexceedsthevarianceofthetheoreticalmodel,indicatingthemodelisnotagood
representationoftheunderlyingphenomenon.
Thereisanotherimportantissuetoconsiderwhenestimatingthemagnitudeof
changesinforestareawithinarelativelysmallfixedgeographicboundary:theamountof
deforestationoveragivenperiodiscloselylinkedtotheamountofforestavailabletobe
deforestedwithineachcellatthebeginningoftheperiod.Oneissueisthatthereisasimple
physicalconstraint.Theamountofforestthatcanbelostinanygivenyearislimitedbythe
availabilityofforestwithinthegridcell.Givenourdatasetwithoutforestgains,moreforest
cannotbelostoverayearthanexistsatthestartoftheyear.Rather,whendeforestation
progressesovertime,theavailableforestdeclinesand,insomecases,iscompletely
exhaustedwithina900mgridcell.
Althoughthestartingforestcoversetsaphysicallimitonthepotential
deforestationwithineach900mcell,therearealsoeconomicfactorsatwork.Thedifficulty
ofaccessinganddeforestinga30mforestcellislikelytobegreaterthefartherawaythat
cellisfromnon‐forestareas,includingpreviouslyforestedlandthathasalreadybeen
cleared,givengreatercostsintermsoftraveltimeandefforttotransportpeopleand
machinerythroughforestsascomparedtomoreopenareas.Asaresult,asacellis
progressivelydeforested,moreandmoreofthecell’sforestedareasbecomeaccessibleand
easier(lowercost)tocutdown.Thus,generallyspeaking,thecostsofconvertingahectare
offorestwithina900mcellarelikelytobeinverselyrelatedtothetotalamountofforest
areainthecell.Thisignores,forthetimebeing,thedispositionofthesurroundingcellsas
wellasdifferencesinthespatialconfigurationoftheforestareaatthe30mresolution
withinthe900mcell.
Anothereconomicconsiderationisthefactthatforestlosswithina900mgridcellis
notlikelytobedistributedinacompletelyrandommanner.Peopleshouldhavean
incentivetopreferentiallydeforestthoseareasyieldingahighernetreturn,eitherbecause
ofhighernetrevenuesorbecauseoflowercostsofconversion.Thus,onewouldexpect
peopletotendtofirstcutthoseareasthataremosteasilyaccessibleorbestsuitedfor
agriculture.Asaresult,thefactthatwhileacertainshareoftheforesthasbeencleared,
anothershare(oneminusthedeforestedshare)stillremainsinforestcovermayconvey
certaininformationabouttherelativeprofitabilityofconvertingthoseremainingforests.
Forexample,iffivepercentoftheoriginalforestextent(e.g.45outof900possible30m


24

cells)remainsstanding,whiletheotherninety‐fivepercenthasbeencutdown,thismay
indicatethatthelastfivepercentisrelativelydifficultorotherwiseunprofitabletoconvert.
Thismayalsoprovidesomeinformationregardingthelikelydegradationandpotential
timbervalueoftheremainingforestcover.
Wetakethesedynamicsintoaccountinourmodelbydirectlycontrollingforthe
startingforestareaineach900mgridcell.Inparticular,westratifythesampleinto20
startingforestareacategories,withthebinschosentocontainroughlysimilarnumbersof
900mgridcells(giventhattheseobservationsareourunitofanalysis).Thisincludesabin
forcellswith100%forestcover(themaximum900countofforested30mcells).Wethen
includedummyvariablesforeachofthesestartingforestareacategoriesaswellas
additionalmultiplicativetermsthatcapturetheinteractionsbetweenthisinitialsetof
dummyvariableandeachofourkeyexplanatoryvariablesintheregression.Thisallowsus
toestimatehoweachofthesedifferentvariablesaffectthelikelihoodandscaleof
deforestationwithinagridcell,dependingonthestartingareaoftheforest.Inthisway,we
cancaptureboththephysicalconstraintsimposedbythedifferentavailablequantitiesof
forestaswellasthedifferenteconomicdynamicsofforestclearingatdifferentstagesof
deforestationwithina900mcell.
Untilnow,thediscussionhasfocusedonhowdeforestationwithina900mcell
dependsontheextentofforestclearancewithinthegridcellitself.Thesurroundingarea
outsidethecellshouldmatterbothintermsofmakingthecellmoreorlessaccessibleand
thusincreasingordecreasingthecostsofconversion,asdiscussedearlier.Wecontrolfor
thesurroundinglandscapebycalculatingameasureoftheaveragedistanceofagridcellto
allofthenon‐forest30mcellsinthesurroundingarea,withina2.5kmradius.Weusea
“kerneldensity”tointerpolatetheinfluenceofthenon‐forestareaoverspace,assuming
decreasing“gravity”oftheseareasasdistanceincreases,uptothechosen2.5kmradius,at
whichpointtheinfluenceofnon‐forestareaisconsideredzero.
4.2.1.2.Observedandunobservedcomponentsofnetreturnsfromlandconversion
Theprincipalchallengeindevelopingamodelforempiricalestimationisthatwe
onlyhavepartialinformationonthepotentialnetreturnsthatlandownerscouldobtain
fromthemostprofitablenon‐forestlanduse.Weproxyforsomedifferencesinthecostsof
conversionandheterogeneousqualityofagriculturallandwithinagridcellbyaccounting
forthestartingforestareaonitsownaswellasininteractionwithourkeyexplanatory
variables.Ourmainexplanatoryvariableofinterestisanestimateofthepotential
economicreturnsperhectarefromcropproduction,whichweconsiderasaproxyforthe
potentialreturnsfromconvertingland.Wedonothavedataonthecostsofproducing
cropsintermsoflabor,fertilizer,chemicals,andanyotherinputsnordowehavedataon
thecostsoftransportinganyproductstothemarket.Wealsodonothavedataonthe(one‐
time)costsofconversion(aswellasanypotentialone‐timebenefitsofconversionsuch
salesoftimber).Bothfixedandvariablecostsaswellasrevenueswilldeterminethe
economicrationaleforconvertingforests.


25

Toaccountforthesedifferentcosts,ourapproachistointroduceadditionalcontrol
variablesatthelevelofthe900mcellthatweexpectwillbecorrelatedwithproductionand
conversioncosts.Alltime‐varyingexplanatoryvariablesarelaggedoneyearsoasnottobe
contemporaneouswiththedependentvariable.Thestartingforestareacategories,
describedabove,provideoneproxyforpotentialconversioncostsaswellaspotential
differencesinagriculturalreturnswithinthegridcell.Aswiththestartingforest
categories,eachoftheothercontrolvariablesinourmodelisincludedindependentlyandin
interactionwithourmeasureofpotentialrevenuesforeachgridcell.Whenthesevariables
areincludedindependently,theestimatedparametersontheseadditionalvariableswill
adjusttheinterceptinthemodel,capturingpotentialone‐timeconversionorotherfixed
costs(orbenefits).Whenthevariablesareincludedininteractionswiththeagricultural
revenues,theestimatedeconometricparameterswillscaletheresponsetotheestimated
economicreturnsbasedontheproxiesforadditionalcostfactors.
OurprincipalvariablesarelistedinTable4.2.1.Whilethesevariableshelptoadjust
thefixedcostsandtoscaletheeffectsoftheagriculturalreturns,theremaystillbe
significantunobservedfactorsaffectingeconomicprofitabilityoflandconversion.Asa
result,giventhespecificinterestoftheMREDDprogramintheYucatánandSouthern
regions,wealsointroducedregionaldummyvariables,singlyandmultiplicatively(i.e.in
interaction)withagriculturalreturns,toaccountforotherfactors,suchasgovernment
policies,thatmayaffectagriculturalprofitabilityatthebroadregionallevel.
Table4.2.1.Principalexplanatoryvariablesusedinnationalregressions(900mcell)
VariableUnitsVariationoverSpaceVariationover
Time
PotentialCropRevenue MXN$/ha Yes Yes
Starting
f
orestareacategory 0/1 Yes Yes
Non‐forestinfluence km2Yes Yes
Urbaninfluence km2Yes No
Protectedareaexten
t
 m2Yes Yes
Ejidoareaexten
t
 m2Yes No
Comunidade
s
areaexten
t
 m2Yes No
Slope %Yes No
Spatialtrendsurface Lat/long Yes No
WecompileddetailedinformationonPROCAMPOpayments.However,wedidnot
directlyincludePROCAMPOpaymentsinoureconometricmodelbecausereceiptof
paymentsfromPROCAMPO(andothergovernmentprograms)isnotrandom.These
paymentsareafixedamountperhectarebasedonthesizeoffarms,andpaymentsare
concentratedinejidosandagrariancommunityareas.Asaresult,theconstanttermsinour
modelandthevariableonejidosandagrariancommunitylandswithinagridcellsmay
alreadycapturetheroleofthegovernmentpayments.Includingtheseexplicitlyislikelyto


26

capturecharacteristicsofthelandowners(notablyfarmsize)ratherthantheimpactofthe
paymentsthemselves.EconometricallyidentifyingtheroleofPROCAMPOandother
governmentpaymentswouldrequireadistinctempiricalstrategy,exploitingchangesinthe
programcriteria,andwasbeyondthescopeofthisstudy.Nevertheless,weareableto
simulatethepotentialroleofeliminatingagriculturalsubsidiesfromPROCAMPO,building
ontheideathattheroleoftheseprogramsisalreadycapturedinourestimatedparameters.
4.3.HistoricalSimulations
4.3.1.SimulationScenario
Weuseourestimatedmodelparameterstoconductaseriesofsimulationsto
explorealternativescenarios,lookingbackretrospectivelyoverthe2000‐2012period.In
thenextSection4.4,weconsiderforwardlookingscenariosover2014‐2024.Webegin
withanalysesthatarewithinthesampleperiodtobeasconsistentaspossiblewiththedata
usedtoestimatethemodel.Thegoalofthesescenariosistounderstandtherelativeeffect
ofdifferentvariables,aswellastoexploresomealternativepolicyscenarios.Wethen
conductaforward‐lookingsimulationtopredictdeforestationinthefutureinthenext
section(Section3.6).Weconductsixsimulationsoverourhistoricalperiodofanalysis,as
summarizedintable4.3.1.below.
Table4.3.1.Simulationscenariosoverhistoricalperiodindataset,2000‐2012
ScenarionameDescription
1)Factualsimulation Allvariablesheldathistoricallevelsfrom2000to2012.
2)99%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenues fromconvertingforestlands
reducedby1%relativetohistoricallevelsinallyears.
3)101%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenuesfromconvertingforestlands
increasedby1%relativetohistoricallevelsinallyears.
4)90%PotentialAgricultural
ReturnsonForestLands
Potnatialagriculturalrevenuesfromconvertingforestlands
reducedby10%relativetohistoricallevelsinallyears.
5)110%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenues fromconvertingforestlands
increasedby10%relativetohistoricallevelsinallyears.
6)NoPROCAMPOpaymentson
forestedlandsinejidosoragrarian
communities.
Potentialagriculturalreturnsfromconvertingforestlands
withinagrariancommunityandejidosreducedbyvalueof
PROCAMPOpaymentsperprogramhectareinmunicipality
*Thesimulationsregardingchangestoagriculturalreturnsareaimedatrevealingtheestimated
sensitivityofdeforestationtochangesinthenetbenefitsfromconvertingforeststoagriculturaluses.
First,weestablishabaselineforcomparingoursimulationresultsbyconductinga
“factual”simulationusingtheactualhistoricalvaluesofallthevariablesusedinthe
estimation.Thenextfoursimulationsexaminetheimpactofourprimaryvariableof
interest,theestimatedagriculturalreturns.Thisvariableisourbestguessofthepotential
netbenefitsofconvertingforestlandstoanon‐forestuse.Theestimatedsensitivitytothis
variablewillbeusedinourmodelingtoexaminethepossibleimpactsofalternativepolicies
thatcouldchangethenetbenefitsfromconvertingforestedland.Suchchangesinthenet


27

benefitscouldcomethroughchangesintheprofitabilityofthenon‐forestuse(e.g.because
ofchangesingovernmentagriculturalsubsidiesonconvertedforestlands),orthrough
changesintherelativevalueofmaintaininglandinforestcover(e.g.becauseofdifferent
otentialincentivesforforestprotection).
Scenarios2and3explorethesensitivityofdeforestationtoourpotential
agriculturalreturnsvariableby,respectively,decreasingandincreasingestimatedpotential
agriculturalreturnsby1%relativetotheirfactualvalues.Thisprovidesanestimated
elasticityforchangesindeforestationwithrespecttochangingeconomicincentives,as
capturedbyourmodel.Thesesimulationsaregenerallymoreindicativeofthemodel
findingsforsmallerchangesinthevariablesthatarewithintherangeofthedatausedinthe
analysis.Nevertheless,inordertoseehowtheseresultsmightscalewithlargerchangesin
returns,scenarios3and4repeattheexercisewithasomewhatlargerchangeinreturns,
decreasingandincreasingestimatedagriculturalreturnsby10%relativetotheirfactual
values.
Thefourthscenariousestheestimatedparametersonagriculturalreturnsto
simulatechangesintheeconomicincentivesforconvertinglands.Scenario4isa
preliminaryexplorationofthepotentialinfluencehistoricalimpactsofthePROCAMPO
agriculturalsupportprogramundertheassumptionsthatfarmersweighingthepotential
benefitsofconvertinglandtocroplandrespondtoexpectedPROCAMPOpaymentsfromthe
governmentinthesamewayastheyrespondtoexpectedcroprevenuesreceivedfromthe
market.Inreality,farmersmayrespondtothesepotentialincomestreamsindifferent
potentialwaysgivendifferentperceptionsovertheirrelativeuncertaintyandfuture
evolution,forexample.Nevertheless,wemaintainthisassumptionasafirst
approximation.WhilewedidnotexplicitlyincludePROCAMPOpaymentsinthemodel,the
estimatedparametersimplicitlyreflecttheeffectsofthesepayments.Thus,reducing
potentialagriculturalreturnsbytheamountofthesepaymentswillreflecttheeffectof
reducingtheexpectedbenefitsfromcropproduction,takingintoaccountallofthepolicies
inplacefrom2000‐2012.
Aquestionisbywhatamounttoreducepotentialagriculturalreturnsgiventhatnot
allcroplandareaswereeligibletoreceivePROCAMPOpayments.Approximately80%of
currentlyplantedacresoverbothgrowingseasonsreceivedPROCAMPOsupportlastyear.
Between2000and2012,thesepaymentswentlargelytolandsinejidooragrarian
communitydesignations.FromouranalysisofthePROCAMPOdatafrom1999to2011,
about85%oflandsreceivingpaymentsnationallywereclearlyidentifiedasbeingwithin
ejidosoragrariancommunities,whileabout8%wereclearlyidentifiableasprivately
owned.Therelevantissue,however,isnotwhatshareofcurrentcroplandiseligiblefor
PROCAMPOpaymentsbutwhatshareofforestareasthatmightbeconvertedtocropswas
eligibletoreceivepaymentsinthepastandwouldbeeligibletoreceivepaymentsinthe
future.Theshareoflandseligibleforpaymentscouldbesignificantlyhigherinforested
areasifpotentialfarmsizesaresmallerthaninotherareas,whichmightespeciallybecase
onejdooragrariancommunitylands.Givenlackofadditionalinformation,asapreliminary
exploration,ourscenario4assumesthatPROCAMPOpaymentsonlywenttolandsinejidos


28

andagrariancommunities,andthatallnewcroplandacresinthesedesignationswere
entitledtofulllevelofpayments.Inparticular,wereducethepotentialagriculturalreturn
onejidoandagrariancommunitylandsbytheaveragePROCAMPOpaymentreceivedonthe
PROCAMPOprogramhectaresinthemunicipalityintheprioryear.7Whileitisnotthecase
thatnoforestedlandsoutsideejidosandcomunidadeswouldhavebeeneligibletoreceive
payments,itisalsolikelynotthecasethatalllandswithintheselandstypeswouldhave
receivedpayments.Wesimulateascenariowherenolandsoutsideofejidosand
comunidadesreceivedpaymentsinordertobeconservativeinnotoverstatingtheimpacts
oftheprogram.
RemovingthefullamountofPROCAMPOpaymentsperhectarerepresentsabouta
35%reductionintheestimatedpotentialagriculturalrevenuesonforestedlandsoverthe
historicalperiodforthemediangridcellintheejidosandagrariancommunities.This
scenariowilllikelyunderestimatetheeffectofPROCAMPOoutsideofcommunallandareas,
asweareassumingzeroeffectatfirstapproximation,butwilllikelysomewhat
overestimatetheprogram’seffectswithinthecommunalareasbyassumingallnew
croplandsinthosedesignationsareeligibletoreceivePROCAMPOprogrampayments,
despitethelimitsonpaymentsaccordingtothesizeoffields.
Thesesimulationsexploretheeffectsofchangingjustonevariableinthemodel,
holdingallothersconstant.Inreality,allothervariableswouldnothavebeenconstant,
mostspecificallythestartingforestarea.Forexample,ifdeforestationin2000islower
(higher)duetolower(higher)agriculturalreturns,thenstartingforestareawouldhave
beenhigher(lower)inthesubsequentyear.Wedonottakethisintoaccountinour
historicalsimulationssincethegoalisjusttoexaminethesensitivitytotheonevariable.
Nevertheless,forthepurposesofthefuturepredictions,describedinthenextsection,we
updatethestartingforestareaineachyeartoreflectthedeforestationinthepreviousyear.
4.3.2.SimulationResults
4.3.2.1.ChangesinAgriculturalReturns
ResultsfromthesimulationatthenationallevelaresummarizedinTable4.3.2
below.Wepresentresultsfromourpreferredmodel(the“negativebinomial”),butinclude
resultsfromouralternativemodel(the“poisson”withoutfixedeffects)intheAppendix.8

7Whena900mcellwasonlypartiallyincommunallandownership,weestimatedaweighted
averageofthePROCAMPOpaymentassumingtheejidoandagrariancommunityportionswere
eligibleforthefullpayment,whiletheremainderwasnot.
8Forthepurposesofevaluatingchangesinresponsetoparticularvariables,wepreferthenegative
binomialspecificationasPearsontestindicatesthedataarenotagoodfittothepoissonmodel,even
thoughthelatterhasabetterfittothehistoricaldata.Wereportresultswithbothmodelsfor
comparison.Weonlyreportresultsforthepoissonmodelwithoutfixedeffectsasweareunableto
conductsimulationswiththe“fixedeffects”modelgiventhatwewereonlyabletoestimate
“conditional”fixedeffectsmodel,whichdoesnotactuallyestimatethefixedeffectsforeachofthe
900mcells.Estimatesoftheseeffectsarenecessarytomakeabsolutepredictionsofthedependent



29

Ouralternativemodel(the“poisson”modelwithoutfixedeffects)replicatestheobserved
quantityofdeforestationpreciselyatthenationalaswellasregionallevels.Ourpreferred
modelhasasomewhatlessprecisefit,overestimatingnationaldeforestationoverthe2000‐
2012periodbyabout120thousandhectaresor6.8%,withapredictedtotalforestlossof
1.88millionhectaresversusanobservedlossof1.76million.9Althoughthismodel
providesasomewhatlessprecisefittothedatainabsoluteterms,wefocusonresultsfrom
thismodelasitisourpreferredspecificationforestimatingrelativechangesinforestlossin
responsetochangesinparticularvariables.
Table4.3.2.NationalSimulationResults

 Totalforestloss,
2000‐12
(Ha)
Differencefrom
factualsimulation
(Ha)
Differencefrom
factualsimulation
(%)
Observed(within
sample)*
1,762,854 ‐120,624 ‐6.4%
1)Factualsimulation  1,883,478 0 0.0%
2)99%agricultural
returns
1,878,961 ‐4,517 ‐0.24%
3)101%agricultural
returns
1,888,360 4,882 0.26%
4)90%agricultural
returns
1,845,771 ‐37,707 ‐2.0%
5)110%agricultural
returns
1,946,100 62,623 3.3%
6)NoPROCAMPO
payments
1,789,400 94,078 ‐5.0%
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualdeforestationwas1,997,765haor13%higher,aswecould
notusealltheobservationsduetomissingdataforsomeofthevariables.Note:2000‐12forestlossis
throughtheendof2011butdoesnotincludedeforestationoccurringin2012.Resultsinthistable
arefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfromthe
alternative“poisson”model(withoutfixedeffects)inAppendixTableA‐10.
Attheregionallevel,thepreferredmodelcapturesthegeneraldistributionofforest
loss,byregion,aswellasareaswithinandoutsidetheAATRreferenceregions.A
comparisonoftheobservedversusmodeledforestloss(the“factualsimulation”)for
differentregionsandlandtypesisshownintables4.4.1and4.4.2.Themodelvariesinits
precisionbyregion,underestimatingdeforestationbyalmost10%intheYucatánPeninsula
(region6),byabout4%intheSouthandWest(regions5and3),by7‐8%intheNorthwest

variable.Estimatingactualfixedeffectsprovedcomputationallyimpossibleevenwithdistrict‐level
fixedeffects.
9Forthepurposesofcomparingtotheestimatesfromourmodels,the“observed”forestlossfigure
representstheobserveddeforestationfor900mcellswithinthesampleusedforourestimation.
Actualdeforestationwas1,997,765haor13%higher,aswecouldnotusealltheobservationsdueto
missingdataforsomeofthevariables.


30

(region1)andBajioandNortheast(region2),andbyjust1%intheCenterandEast(region
4).Suchvariationsarenotsurprisinggiventhatwearepredictingregionalandsub‐
regionalforestlossesbasedonanempiricalestimationofdeforestationresponsesacross
thewholecountry,withonlyafewregion‐specificdummiestocaptureregion‐specific
particularities.
Theresultsexaminingthesensitivityofdeforestationtothepotentialnetbenefits
fromconvertingforeststocroplanduseconfirmthatgreaterexpectedpotentialagricultural
returnswereassociatedwithincreasesinannualdeforestation,asexpectedbytheory.The
simulationsbasedonourpreferredmodelindicatethata1%decreaseinpotential
agriculturalreturnsover2000‐2012wouldhavedecreasedcumulativedeforestation
nationallyoverthisperiodby0.24%.Conversely,a1%increasewouldhaveboosted
deforestationby0.26%.Thesimulationsfromthealternativemodelsuggestaverysimilar
deforestationresponse,withdeforestationdecreasing0.26%fora1%fallinagricultural
returns,andincreasing0.27%fora1%increaseinreturns(seeAppendixtableA‐10).
Resultsforthe10%changesinreturnsareroughlyproportional,butshowamore
asymmetricresponse,withforestlossesdecreasing2.0%forat10%decreasein
agriculturalreturnsandincreasingby3.3%fora10%increase.
Ourfinalsimulationsuggeststhatdecreasingcropreturnsbytheamountof
PROCAMPOsubsidiesonejidosandagrariancommunitylandswouldhavedecreased
deforestationbyabout5%.Giventhatthisrepresentsarounda35%decreaseinreturns,
thisisabitlessthanproportionaltoourfindingthata10%decreasewouldhavereduced
deforestationbyabout2%.Mostoftheestimatedreductionsfromeliminatingthe
PROCAMPOpaymentsoncommunallandcategoriesoccurintheYucatánPeninsulaand
Southregions.About46%ofthereductionsoccurintheYucatánPeninsulaandabout
25%intheSouth.
Thefindingthatdeforestationincreasesmorethanitdecreasesforanequivalent
percentincreaseanddecreaseinagriculturalreturns,respectively,isperhapssurprisingif
oneimaginesthatprogressivelymoreandmoremarginalagriculturallandisentering
production,makingitmoreandmoredifficultforlandtocomein.Inpart,thisresult
reflectsthefactthatoureconometricmodelsarenon‐linearcountdatamodels,wherethe
coefficientsarecontributionstoaratesuchthattheydonothaveasimplelinear
interpretationintermsofabsoluteimpacts.
Thesensitivitytomarginalchangesinagriculturalreturnsvariesbyregion,as
showninTable4.3.3.ThemostsensitiveregionsaretheNorthwestandBajioand
Northeast,withtheleastsensitiveregionsbeingtheCenterandEastandtheYucatan
Peninsula.Whiletheformerregionsareestimatedtorespondabout0.5‐0.6%and0.8‐
1.0%,respectively,forevery1%changeinagriculturalreturns,thelatterregionisonly
estimatedtorespondabout0.08%.Inpartthisreflectsthenatureofoursimulations,which
consideredpercentageratherthanabsolutechanges.Asaresult,areaswithlargerabsolute
levelsofreturns,experiencelargerchangesinabsolutereturns,forthesamepercentage
change.Thelargerresponseinregions1and2reflectsthefactthattheseregionshave
higherpotentialagriculturalreturnsandthuslargerabsoluteincreasesanddecreasesin


31

deforestationunderthesescenarios(whichsimulatedpercentage,ratherthanabsolute
changes)and,consequently,havemorenon‐linearchangesinthedeforestationratefora
givenpercentageincreaseinnetreturns.
Theseregionalresultsshouldnotbetakentooliterallygiventhatthemodelismost
appropriatetoreflectnational‐averageresponses.However,themodelisalsopickingup
somedifferencesintheresponsivenesstodeforestationassociatedwithforestcategories.
Thelargerpercentresponseforanincreaseinreturnsinregions1and2alsoreflectsthe
factthatregionscontainmoresmallareasofforest.Breakingoutthesimulationresultsby
startingforestcategorywithineachregionindicatesthattheresponsivenessto1%changes
inagriculturalreturnsgenerallyincreasesasforestcoverdeclines.Thismightindicate
loweraccesscoststothesegridcells,makingthemmoresensitivetochangesingross
revenues.However,insomeregions,notablytheSouth,West,andYucatanPeninsula,there
isaU‐shapepattern,withthegreatestsensitivityoccurringatboththehighestandlowest
forestcategories.
Table4.3.3.RegionalSimulationResultsforSensitivitytoAgriculturalReturns
Region

Factual
simulation
(scenario1)
99%agriculturalreturns
(scenario2) 101%agriculturalreturns
(scenario3)

Totalforest
loss,
2000‐12
(Ha)

Totalforest
loss,2000‐
12
(Ha)
Difference
fromfactual
simulation
(%)

Total
forestloss,
2000‐12
(Ha)
Difference
fromfactual
simulation
(%)
TotalCountry 1,883,4781,878,961 ‐0.24% 1,888,3600.26%
Northwest
(Region1)
68,97568,629‐0.50%69,3820.59%
Bajio&Northeast
(Region2)
179,624178,142‐0.83%181,3730.97%
West
(Region3)
57,16556,916‐0.44%57,4190.44%
CenterandEast
(Region4)
247,089246,899‐0.08%247,3030.09%
South
(Region5)
456,810455,280‐0.33%458,3460.34%
YucatanPeninsula
(Region6)
873,816873,096‐0.08%874,5380.08%
Note:Resultsinthistablearefromthepreferred“negativebinomial”model.2000‐12forestlossis
throughtheendof2011butdoesnotincludedeforestationoccurringin2012.
Theseresultssuggestthatrelativelysmallerpatchesofforestscouldcontribute
disproportionatelytomarginalchangesinincentives,giventhattheyalreadyaccountfora
disproportionateshareofdeforestationrelativetotheforestarea(seenationalmodeling
appendixformorediscussionofthisissue).Atthesametime,relativelymoreintactforests
insomeregionsappeartobeatapotentialeconomictippingpointfordeforestation,where
changesinnetreturnswillcausethemtobeginadeforestationprocess,producingajumpin


32

annualdeforestation,andperhapsevenmorecumulativedeforestationoverthelonger
term.
Asnotedabove,oursimulationsconsideredvariationsinonevariable,holdingall
elseconstant,includingthestartingforestarea.Tofullycapturetheeffectsonthe
dynamicsofdeforestation,wewouldalsowanttosimulatetherepercussionsof
deforestationinoneyearonforestcoveranditseffectondeforestationinthesubsequent
years.Webegintoexploretheseissuesinthenextsectionwhereweconsideraforward‐
lookingsimulationbasedonanincreaseinagriculturalreturnsaswellaspotentialcarbon
paymentsforavoideddeforestation.
4.4.Futureprojections
Weconductafuture‐orientedsimulationundera“businessasusual”scenarioas
wellasaseriesofpolicycaseswhereweintroduceahypotheticalcomprehensiveincentive
tomaintainforestcarbon.Asdiscussedfurtherbelow,avarietyofpolicyapproachescould
beusedtocapturepotentialfinancialflowsforREDD+andimplementlow‐emissions
practicesinMexico.Ourprojectionsservetoquantifyandmapthepotentialreductions
availableforfutureREDD+policyinMexico,ratherthantomodelaparticularREDD+
implementationstrategyinparticular.Thesesimulationsalsoprovideaninputforlocal
modelingfuturedeforestationatthelevelofeachofthesevenAATRs,asdiscussedin
Section4.
Thefuturesimulationsaccountfortherepercussionsofdeforestationfromone
yeartothenextbymodelingdeforestationateach900mcellandaccountingforitseffecton
startingforestcoverareaandcategoryatthestartofthesubsequentyear.Whileour
alternative(“poisson”model)couldprovidebetterpredictions,wefocusonourmainmodel
(the“negativebionomial”)whichshouldbemoreappropriateforexaminingtherelative
changesbetweentheBAUandpolicycases.Wepresentresultsfromthealternativemodel
intheAppendixforcomparisonpurposes.
Forthebusiness‐as‐usual(BAU)scenario,westartwithobservedforestcoverin
2012(thelastyearofourdatafromtheUniversityofMaryland)andthenmodelits
evolutionforeach900mcellatanannualtimestepthrough2024.Wealsostartwith
agriculturalreturnsasof2012andholdtheseconstantforthescenario.Thisinvolves
almostatriplingofmeanandmedianagriculturalreturnsrelativecomparedtothe2000‐
2012period,thoughthisvariesoverspace.Combiningthedataoveralltheyearsand900m
cells,theaveragepotentialreturnsrisefrom4,003to15,464MXN$/hawhilemedian
returnsrisefrom2,470to9,346MXN$/ha.Theincreaseinmedian(andusuallyaverage
returns)islargerintheNorthwest,BajioandNortheast,andWestregions,relativetointhe
CenterandEast,SouthandYucatanPeninsula.
Duetomissingdataonsomeofthevariables,ourestimationandhistorical
scenarioswerebasedonasub‐sampleofthedatathatcapture87%ofthehistorical
deforestationover2000‐2012.Nevertheless,thereisfewermissingdatainthelateryears
ofthedatabase.Thesampleusedforourfuturepredictionscaptured98%oftheobserved


33

deforestationin2011.Giventhatourdataisthusclosetocomplete,wedidnotmakeany
additionaladjustmentstothefutureforestlossprojectionsforthismissinginformation.
Forthepolicyscenarios,weconductaseriesofsimulationswhereweintroducea
comprehensivecarbonincentivepertonofCO2,startingatUSD$5andrisingprogressively
to$100(assuminganexchangerateofMXN$13/USD).Specifically,weconsider“prices”of
$5,$10,$20,$30,$50,$60,$70,$80,and$100pertonofCO2,soastotraceouta“marginal
cost”curvebasedonestimatedemissionsreductionsfromavoideddeforestationat
differentpricepoints.
Wesimulateaneconomicallyidealormostcomprehensiveincentivewhichcan,in
theory,beviewedasonewherealllandownerseitherreceiveasubsidyforland
preservationorpayataxforlandconversionforinstantaneouslyreleasingthecarbon
contentofallabove‐groundlivebiomass.Morepractically,onecanthinkofthisasapolicy
thatreducesthe“business‐as‐usual”agriculturalbenefits(e.g.byreducinggovernment
subsidies)andtranslatingthemintoeconomicbenefitsforlow‐emissionspracticesthat
avoiddeforestation.Thisisimplementedinoursimulationsbyreducingtheagricultural
returnsbytheamountoftheforegonecarbonrevenueifforestsweretobedeforested.We
donotmodelanypotentialshiftsor“leakage”ofdeforestationinresponsetopossible
inducedchangesinagriculturalreturnsorothereffects.Webaseouranalysisonthe
above‐groundcarbondensitydatafromWHRC/MREDD(Cartus,etal.,2014).For
simplicity,thisinitialanalysisdidnotconsiderbelow‐groundorsoilcarbonlosses.
Whilethisanalysisconsidersanotionalcarbonincentivethatcanbetranslatedinto
aparticular“price”andthoughtaboutasataxorsubsidyforeachlandownerorotherland
user,asalreadynote,theresultsdonotpresupposeaparticularREDD+policybasedon
directpaymentstolandowners,suchasatraditionalpaymentsforenvironmentalservices
(PES)program.Rather,ouranalysisservestoidentifythecost‐effectivepotential
emissionsreductions,andtheirspatialdistribution,giventhe“price”intermsofforegone
agriculturalrevenuesonthelandsnotbeingdeforested.Thisanalysisservestoquantify
andspatiallyidentifythemostcost‐effectivereductionsthatcouldbepotentiallytargeted
underavarietyofpotentialpolicyinterventionsandapproachesforpromotinglow‐
emissionsruraldevelopmentandreduceddeforestationemissionsinMexico.Moreover,
whileagriculturalproductionmightbeforegoneontheparticularlandsnotbeing
deforested,agriculturecouldbeintensifiedandexpandedonnon‐forestlandsunderalow‐
emissionsagriculturaldevelopmentstrategy.Thismeansthatagriculturalproductioncould
bemaintainedorincreasedoverallatthesametimethatexpansionofagricultureintoforest
areasisdecreased.
4.4.1.1.“Business‐as‐usual”projection
Table4.4.1showsour“businessasusual”projectionsfor2014‐2024relativetothe
observedandmodeleddeforestationinannualizedtermsduringthehistoricalperiod
(2000‐2012).ResultsarepresentednationallyaswellasbyAATRreferenceregionsand
differentlandownershipcategories.Basedontheeconomicprofitabilityofagricultureand
startingforestcoverin2012,themodelpredictsanoverall27%increaseinannual


34

deforestationinMexicooverthenexttenyears,relativetotherecentpast.Estimated
changesreportedarerelativetothemodeleddeforestation(the“factualsimulation”)for
2000‐2012.Mostofthisincreaseisduetoasignificantincreaseindeforestationinthe
SouthandYucatanPeninsularegions,whichassumesanevengreatershareofnational
deforestation,assomeotherregions(WestandCenter/Eastregions)experienceadecrease
inannualdeforestation.Thehigheragriculturalprofitsin2012relativetothehistorical
periodaccountsfortheoverallincreaseindeforestationnationwideandinthemore
forestedareas.
Despitethesignificantincreaseintheaverageandmedianagriculturalreturns
comparedtothehistoricalperiod,theoverallincreaseindeforestationissmallerthan
suggestedbyoursimulationsofsmallermarginalchangesinagriculturalreturnsinsection
2.Thisislikelyduetothefactthatwearecomparingresultsacrossawholehistorical
periodwithawiderangeinreturns,includingreturnssimilartoourprojectedonesatthe
endoftheperiod.Wearealsonowaccountingforthedecliningforestareaswithineach
gridcell,whichfurtherreducespotentialdeforestation.Theprojecteddecreaseinsome
regionsrelativetothehistoricalperiodislikelyduetosmallerremainingareasofforestin
2012relativetothehistoricalperiod.
Asnotedbefore,ourmodelsareintendedfornationalanalysisbutgenerallycapture
regionaldistributions.Map4.4.1showsthespatialdistributionofprojectedaggregate
forestlossunderthebusiness‐as‐usualscenarioforthenext10years(2014untilthestart
of2024).Themapshowsthatthegreatestamountofdeforestationisprojectedtooccur
intheSouthandYucatanPeninsularegions.Table4.4.2showshowtheseregionsarenot
onlyprojectedtocontributethemostdeforestationinabsoluteterms,butarealsoprojected
toexperiencethegreatestpercentageincreasesindeforestation,withprojected
deforestationrisingby72%intheSouthandby26%intheYucatanPeninsula.Incontrast,
deforestationincreasesby17%intheNorthwest,3%intheBajioandNortheastand
percentdecreasesintheWestandCenter/Eastregions.Ouralternativemodel(the
“poisson”)alsopredictsanincreaseinnationaldeforestationof27%,withthegreatest
proportionalincreasesoccurringintheSouthandYucatanPeninsula(AppendixtableA‐11).
Thebreakdownacrossregionsisabitdifferentinabsoluteterms,butthequalitativeresults
arestillgenerallythesame.Thisalternativemodel,whichmaybemoreprecisefor
predictivepurposes,showsrelativelysmallerincreasesintheSouthandYucatan(41and
48%,respectively)andlargerincreases(smallerdecreases)intherestofthecountry.
TherelativelygreaterprojectedincreasesindeforestationintheSouthandYucatan
comparedtotherestofthecountrycontrastwiththehistoricalsimulationresultsinTable
4.3.3formarginalchangesinreturnsofplusorminus1%.Whileotherregionsappear
moresensitivetosmallchangesinreturns,thegreatercumulativedeforestationinthe
SouthandYucataninthefutureprojectionsmaybedueinpartbythemuchlargerchanges
inreturnsbeingconsideredinthebusiness‐as‐usualprojection,whichelicitsalarger
responsefromallforestareas.Theotherpartofthestoryisthatwearenowaccountingfor
howforestareasevolveovertime.Thus,areaswithsmallinitialforestcovermightrespond


35

withalargeproportionalchangesindeforestationintheshortrunbutthenhavelittle
forestcoverremainingtocontinuehavingforestlosses.

Table4.4.1.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,by
AATRreferenceregionsandlandownershipcategory
Region/Land
Category 
Observed
forestloss
(insample),
2000‐12
(Ha/yr)*

Modeled
forestloss
(factual
simulation),
2000‐12
(Ha/yr)
Business‐
as‐usual
(BAU)
forest
loss,
2014‐24
(Ha/yr)

Changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(Ha/yr)
%change
inannual
forestloss,
projected
BAUvs.
modeled
2000‐12
(%)
TotalCountry  160,259 171,225 217,963 46,738 27%
Non‐AATR 110,299 113,679 131,434 17,755 16%
AATRregions 49,960 57,546 86,528 28,982 50%
Mixteca 1,298 1,902 3,180 1,278 67%
SierraNorte 1,827 1,055 1,920 865 82%
SierraPucc 35,471 41,078 57,863 16,785 41%
Chiapas 4,546 7,165 12,847 5,682 79%
Raramuri 1,725 2,107 2,556 449 21%
ValledeBravo 481 498 417 ‐81 ‐16%
Itsmo 4,613 3,739 7,745 4,006 107%
Comunidades10,531 10,288 21,255 10,967 107%
Ejidos89,613 92,800 113,070 20,269 22%
Protectedareas 4,778 6,529 11,542 5,013 77%
Otherlands 55,337 61,608 72,096 10,488 17%
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualnationaldeforestationwas1,997,765haor13%higherthan
thein‐sampleamountasallobservationscouldnotbeusedduetomissingdataonsomevariables.
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐11.Protectedareasarethefederallyprotected
areasconsideredinthisanalysis.2000‐12forestlossisthroughtheendof2011butdoesnotinclude
deforestationoccurringin2012.Similarly,2014‐24forestlossisthroughtheendof2023butdoes
notincludedeforestationoccurringin2024.






36



Table4.4.2.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,by
nationalregionsandAATRReferenceRegions
Region/Land
Category

Observed
forestloss
(in
sample),
2000‐12
(Ha/yr)*

Modeled
forestloss
(factual
simulation),
2000‐12
(Ha/yr)
Business‐
as‐usual
(BAU)
forestloss,
2014‐24
(Ha/yr)

Changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(Ha/yr)
%changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(%)
Northwest(Region1)
Total  5,751 6,270 7,187 917 15%
Non‐AATR 4,025 4,163 4,632 469 11%
AATRregions 1,725 2,107 2,556 449 21%
Bajio&Northeast(Region2)
Total  15,125 16,329 18,154 1,825 11%
Non‐AATR  15,120 16,320 18,151 1,831 11%
AATRregions 4 10 3 ‐7 ‐70%
West(Region3) 
Total  4,969 5,197 5,267 70 1%
Non‐AATR  4,630 4,973 5,048 75 2%
AATRregions  339 224 219 ‐5 ‐2%
CenterandEast(Region4)
Total  22,777 22,463 15,833 ‐6,630 ‐30%
Non‐AATR  21,684 21,176 14,799 ‐6,377 ‐30%
AATRregions  1,093 1,286 1,034 ‐252 ‐20%
South(Region5) 
Total  39,863 41,528 71,262 29,734 72%
Non‐AATR  28,535 28,688 46,408 17,720 62%
AATRregions  11,328 12,840 24,854 12,014 94%
YucatanPeninsula(Region6)
Total  71,776 79,438 100,260 20,822 26%
Non‐AATR 36,304 38,359 42,397 4,038 11%
AATRregions 35,472 41,078 57,863 16,785 41%
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualnationaldeforestationwas1,997,765haor13%higherthan
thein‐sampleamountasallobservationscouldnotbeusedduetomissingdataonsomevariables.
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐12.2000‐12forestlossisthroughtheendof


37

2011butdoesnotincludedeforestationoccurringin2012.Similarly,2014‐24forestlossisthrough
theendof2023butdoesnotincludedeforestationoccurringin2024.

Themodelpredictsthatmostnewdeforestationinabsoluteterms,aswellasmost
absoluteincreasesindeforestation,willoccuronejidolands(Table4.4.2).Inpercentage
terms,however,forestlosseswithinejidosareprojectedtoincreaseby22%orlessthan
thenationalaverage.Incontrast,agrariancommunities(comunidades)areprojectedto
experiencethelargestincrease,followedbydeforestationwithinprotectedareas,with
projectedincreasesof107%and77%,respectively.Ouralternativemodelprojectssimilar
qualitativepatterns,thoughtherelativedifferencesamonglandtypesaresmaller.
Map4.4.1indicatesthattheAATRsarenotalllocatedintheareaswiththehighest
projectedfuturedeforestation.Nevertheless,asshowninTable4.4.1,overalltheAATR
referenceregionshavehigherprojecteddeforestationincreasesthanotherforestedareas
(50%versus16%forthelandsoutsidethesereferenceareas).TheAATRreferenceregions
discussedherearethoseusedinthelocalmodeling(section4),andincludethespecific
REDD+earlyactionareasites,aswellasasurrounding50kmbuffer.Giventhenational
scaleofthemodeling,resultsaremoreappropriateatlargerscalesofanalysis,dictatingour
focusonlargerversussmallerareassurroundingtheAATRs.
LookingspecificallyattheAATRreferenceregions,themodelpredictsthegreatest
increaseintheItsmoandSierraNorteregionandthesmallestincreasesintheRaramuri
andValledeBravoregions,withthelatterregionactuallyexperiencingadeclinein
deforestation.Thealternativemodelgeneratessimilarqualitativeresults,thoughit
predictsasmallerrelativeincreaseindeforestationintheChiapasAATRreferenceregion
(26%versus79%increaseinourpreferredmodel).
TheAATRreferenceregionsnotonlyhavehigherprojecteddeforestationversus
otherlandsonaggregatenationally,buttheyalsogenerallyhavehigherprojected
deforestationrelativetootherlandswithineachregion.ThecomparisonofAATRvs.non‐
AATRlandswithineachregionisshownintable4.4.2.TheAATRreferenceregions
generallyhavehigherprojectedincreasesindeforestation(orsmallerprojecteddecreases
inthecaseoftheCenterandEast),relativetootherforestedlandsinthesameregion.The
exceptionsaretheBajioandNortheast(Region2)andWest(Region3),buttheseresults
arenotindicativegiventhattheseregionscontainedtrivialamountsoflandswithinanyof
theAATRsreferenceareas.Ouralternativemodelgeneratessimilarfindings(TableA‐12).
4.4.1.2.CarbonIncentiveProjections 
Asanexampleofourcarbonincentiveresults,wepresentresultsforahypothetical
carbonincentiveofUSD$10/tCO2intable4.4.3.Thiscarbonincnetivetranslatesintoa
median(average)subsidy/taxofabout4700(5200)MXN$/ha,comparedtomedian
(average)agriculturalreturnsofabout9,300(15,400)MXN$/ha.Thisrepresentsa
medianreductioninagriculturalreturnsof25%,withmorethana100%reductionon
average.Underthissimulatedcarbonincentiveof$10/tC,deforestationfallsnationallyby
anestimated35%.Map4.4.2showsthespatialdistributioninthereductioninforestloss


38

underthe$10carbonincentive(relativetotheBAUcaseshowninmap4.4.1)whilemap
4.4.3showstheremainingdeforestation.Theresultsforthealternativemodelandatthe
levelofeachAATRreferenceregionareshownintheAppendixinmapsA‐15toA‐24.
Ingeneral,theregionsprojectedtohavethegreatestincreasesindeforestationover
thenextdecadearealsoestimatedtobethemostresponsivetoreducingdeforestation
underacarbonincentive.Overall,AATRreferenceregionsareestimatedtoreduce
deforestationby41%comparedtoareductionof32%fornon‐AATRlands.Theanalysis
suggestssignificantreductionsinthespecificAATRs,rangingfrom35%inRaramurito58%
inSierraNorte.AlloftheAATRdemonstrategreaterpotentialreductionsthanthenon‐
AATRregionsofthecountry.However,thegreatestpotentialreductionsoccurinthe
YucatánPeninsulaandSouthasseeninmap4.4.2.Similarly,mostoftheremaining
deforestationisdistributedintheseregions(map4.4.3).
Comunidadesandprotectedareaswerethelandtypesprojectedtohavethebiggest
proportionalincreaseinforestlossesoverthenext10yearsandarealsoestimatedtohave
thegreatestpercentdeclinesinresponsetoacarbonincentive.Inabsoluteterms,
however,thegreatesttotalestimatedreductionsoccuronejidos,aswellasprivateand
otherlandtypesapartfromcomunidadesornationalprotectedareas.
Inadditiontoconsideringchangesinforestareaasaresultofacarbonprice,we
alsoconsiderchangesincarbondioxideemissionsfromlossesinabove‐groundbiomass.
Map4.4.4showsthespatialdistributionofreducedemissionsfromabove‐groundforest
biomass,associatedwiththereducedforestlossscenarioata$10priceshowninmap4.4.2.
Estimatedemissionsreductionsforthe$10carbonincentivearealsocombinedwiththose
fromtheothercarbonincentivesimulationsandareusedtoconstructcostcurvesshownin
Figures4.4.1and4.4.2.Thesefiguresshowtheestimatedabove‐groundforestcarbon
emissionsavoidedannuallyundereachofourcarbonincentivescenarios,relativetothe
business‐as‐usualprojectionoverthe10yearsstartingin2014.
Underthebusiness‐as‐usualscenario,representedbyacarbonincentiveofzero,
averageannualCO2emissionsfromdeforestationareapproximately17milliontonsofCO2
atthenationallevel.Despitereflectingincreasesinfuturedeforestation,theseareabout
37%ofthe45.3MtCO2for2010reportedforland‐usechangeemissionsinthefifthnational
communicationstotheUnitedNationsFrameworkConventiononClimateChange
(SEMARNAT/INECC,2012).Thereareseveralpossibleexplanations.Ourestimatesare
basedonnewsourcesofinformationonbothforestloss,aswellasonabove‐groundforest
carbondensities.Also,thenumbersinthenationalcommunicationsincludeconversionof
grasslands(pastizales),whichwerenotconsideredinouranalysis.Furthermore,our
analysisonlyconsideredemissionsfromabove‐groundforestcarbonstocks,without
consideringpotentiallossesofbelow‐groundforestcarbonorsoilcarbon.Estimatesof
aboveandbelow‐groundforestcarbonstocksinMexicofromFAO(2005)andRueschand
Gibbs(2008)areapproximately95and113tonsofCperhectare.Incontrast,themeanand
medianforesthectarein2012hadanestimated23.6and21.8tonsofC/ha,respectively,
accordingtotheestimatesusedinourstudy(Cartus,etal.,2014).Thecarbondensitiesfor
thedeforestedhectaresinourprojectionsfrom2014‐2024wereabitlower,withameanof


39

21.7andmedianof19.8tC/ha.Adetailedcomparisonofthesenumberswasbeyondthe
scopeofouranalysis.
Focusingonlyontheabove‐groundcarbon,wefindthatthereisrisingpotential
nationallytoreduceemissionsatcarbonincentivesrangingfrom$5to$100,atwhichpoint
about90%oftheemissionsareavoided.Closetohalfoftheestimatedreductionsavailable
atpricesof$10/tonorbelowandmorethantwothirdsoftheestimatedreductions
availableatpricesof$20/tonorbelow.Thenationalandregionalcostcurvesarerisingat
anincreasingrate,indicatingthatitcostsmoreandmoretoavoiddeforestationonlands
withgreateragriculturalpotentials.
Whiletherearepotentialreductionsavailablefromallregionsatpricesupto$20‐
$30,thebulkofestimatedreductionsisfromtheSouthandYucatanPeninsula,which
accountforabout35%and60%ofthetotalpotentialupto$100.Reductionsfromthe
otherregionscollectivelyrisesteeplyandareexhaustedatpricesof$20and$30,atwhich
pointthecostcurvesturnvertical,withabout1milliontonsofemissionsavoidedintotal.
Thisreflectsthehigheragriculturalreturnsintheseregionsaswellassmallertotalamount
offorestlossesandcarbonemissionsthatcanbeavoided.Incontrast,thecostcurvesfor
theSouthandYucatanPeninsuladonotbegintoturnupwardssharplyuntilabout$50.At
pricesof$5,theSouthandYucatanPeninsulaaccountfor43%and50%ofthecost‐effective
potential,respectively.ThecostofreductionsintheSouthrisessomewhatfasterthanin
theYucatanPeninsula,withtheSouthrepresentingasmallershareofthecost‐effective
potentialatprogressivelyhigherprices(e.g.37%versus56%fortheYucatanatapriceof
$50).
Figure4.4.2breaksoutthecostcurvesaccordingtolandswithinandoutsideofthe
AATRreferenceregions.TheseshowthatthebroadAATRregionsonaggregatecontain
morethanhalfofthecost‐effectivepotentialreductionsinemissionsateachpricepoint,
withabout55%ofthetotalmodeledpotentialforallofMexico.Asalreadynoted,our
exercisedidnotpresupposetheimplementationofanactualcarbonpriceorpayment
system.Rather,weconsiderahypotheticalcarbonincentivesoastoestimatethemost
cost‐effectivereductionspotentialavailableforagivenreductioninforegoneagricultural
revenuesontheparticularlandsnotbeingdeforested(thoughofcourseagricultural
productionmightstillincreaseonotherlands).Thesecost‐effectivereductionscouldbe
achievedinpracticethroughavarietyofpolicyapproaches.Also,whileouranalysis
consideredanidealizedpolicycase,whichisindicativeofthepotentialforREDD+policies,
additionalanalysiswouldbeneededtoconsiderimpactsondeforestation,including
possible“leakage,”aswellasothereconomicimplicationsundermorerealisticandlikely
lesscomprehensivepolicyapproaches.






40


Table4.4.3.FuturePredictions,2014‐2024,Business‐as‐Usualand$10/tonCO2
PolicyCase,forAATRandnon‐AATRregions
Region/Land
category 
Business‐
as‐usual
(BAU)
forestloss,
2014‐24
(Ha/yr)
Forestloss,
2014‐24
with
$10/tCO2
(Ha/yr)

Changein
annual
forestloss,
$10/tCO2
vs.BAU
(Ha/yr)
%changein
annual
forestloss,
$10/tCO2
vs.BAU
(%)
TotalCountry 217,963 141,106 ‐76,856 ‐35%
Non‐AATR  131,434 89,727  ‐41,707 ‐32%
AATRregions  86,528 51,379  ‐35,149 ‐41%
Mixteca  3,180 1,466  ‐1,714 ‐54%
SierraNorte  1,920 802  ‐1,118 ‐58%
SierraPucc  57,863 37,082  ‐20,781 ‐36%
Chiapas  12,847 6,324  ‐6,523 ‐51%
Raramuri  2,556 1,668  ‐887 ‐35%
ValledeBravo  417 242  ‐175 ‐42%
Itsmo  7,745 3,795  ‐3,950 ‐51%
Comunidades  21,255 9,837  ‐11,418 ‐54%
Ejidos  113,070 77,174  ‐35,895 ‐32%
Protectedareas  11,542 5,804  ‐5,738 ‐50%
Otherlands  72,096 48,290  ‐23,805 ‐33%
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐13.Protectedareasarethefederallyprotected
areasconsideredinthisanalysis.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.













41


Map4.4.1.Projected“BusinessasUsual”(BAU)ForestLoss2014‐2024

Note:Thismapshowsprojecteddeforestationatthe900m(81ha)resolutionover10yearsstarting
in2014,basedoninformationonforestcoverin2012,estimatedmodelparametersfrom2000‐12,
andholdingconstantagriculturalprofitsat2012levels.Projectionsarefromthepreferred“negative
binomial”model.Forcomparison,wereportresultsfromthealternative“poisson”modelin
AppendixMapA‐11.Greenareasindicatenolossofforestcover.Progressivelyredderareasindicate
greateramountsofforestloss.Greyareasarethosewithoutanyforestcoverin2012andhenceno
projectedforestloss.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.



42

Map4.4.2.ProjectedAvoidedForestLoss2014‐2024,with$10/tonCO2incentive

Note:Thismapshowsprojectedreductionsindeforestationatthe900m(81ha)resolutionover10
yearsstartingin2014,basedonintroducinganeconomicallyidealcomprehensivecarbonpriceof
$10t/CO2onforestcarbonlosses.Reductionsarerelativetothe“business‐as‐usual”(BAU)scenario
inMap4.4.1.Thisanalysisdoesnotconsiderpotentialpriceadjustmentsorotherpossiblesourcesof
inducedshiftsindeforestationandemissions(i.e.“leakage”).Projectionsarefromthepreferred
“negativebinomial”model.Forcomparison,wereportresultsfromthealternative“poisson”model
inAppendixMapA‐12.Whitecolorareasindicatenoreductioninforestlossasaresultofthecarbon
price.Lighttodarkyellow,followedbylighttodarkgreen,areasindicateprogressivelygreater
amountsofavoideddeforestationunderthecarbonpricerelativetotheBAUcase.Greyareasare
thosewithoutanyforestcoverin2012andhencenoprojectedreductioninforestloss.2014‐24
forestlossisthroughtheendof2023butdoesnotincludedeforestationoccurringin2024.



43

Map4.4.3.ProjectedRemainingForestLosswith$10/tonCO2incentive,2014‐2024

Note:Thismapshowstheprojectedforestlossatthe900m(81ha)resolutionover10yearsstarting
in2014thatisestimatedtoremainaftertheintroductionofthe$10t/CO2onforestcarbonlosses
(i.e.thismapshowstheremainingforestlossstartingfromtheforestlossinmap4.4.1and
subtractingouttheavoidedforestlossinmap34.4.2).Projectionsarefromthepreferred“negative
binomial”model.Forcomparison,wereportresultsfromthealternative“poisson”modelin
AppendixTableA‐13.Greenareasindicatenolossofforestcover.Progressivelyredderareasindicate
greateramountsofforestloss.Greyareasarethosewithoutanyforestcoverin2012andhenceno
projectedforestloss.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.



44

Map4.4.4.ProjectedAvoidedEmissions2014‐2024,with$10/tonCO2incentive

Note:Thismapshowsprojectedreductionsinabove‐groundcarbonlossesatthe900m(81ha)
resolutionover10yearsstartingin2014,basedonintroducinganeconomicallyidealcomprehensive
carbonpriceof$10t/CO2onforestcarbonlosses.Reductionsarerelativetotheforestlossesinthe
“business‐as‐usual”(BAU)scenarioinMap4.4.1.Thisanalysisdoesnotconsiderpotentialprice
adjustmentsorotherpossiblesourcesofinducedshiftsindeforestationandemissions(i.e.
“leakage”).Projectionsarefromthepreferred“negativebinomial”model.Forcomparison,we
reportresultsfromthealternative“poisson”modelinAppendixMapA‐14.Whitecolorareas
indicatenoreductioninforestlossesandassociatedcarbonemissionsasaresultofthecarbonprice.
Lighttodarkyellow,followedbylighttodarkgreen,areasindicateprogressivelygreateramountsof
avoideddeforestationandassociatedemissionsunderthecarbonpricerelativetotheBAUcase.
Greyareasarethosewithoutanyforestcoverin2012andhencenoprojectedreductioninforest
lossesandassociatedemissions.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.



45

Figure4.4.1.EstimatedcostcurvesforCO2emissionsreductionsfromabove‐ground
forestcarbonlossesinMexico,byregion

Figure4.4.2.Estimatedcostcurvesforreducingemissionsfromabove‐groundforest
carbonlossesinMexico,byAATRandnon‐AATRregions.



46

5. LocalModelingofDeforestation
5.1.Introduction
5.1.1.Overallapproach
Thenational‐levelmodelingcapturesdriversandpossibledeforestationoutcomes
atonescalethatisonlyarguablyveryrelevanttothelocalscale.Onecouldalsoclaimthat
dynamicsatthelocalscalehavealocalcharacter,andthatamodeloftheselocaldynamics
shouldbeindependentofrelationshipsderivedfromdistantlands.Thus,weadda
componenttothisstudythatmodelsdeforestationbasedsolelyonlocaldata.Wedothis
forthesevenfocusareasselectedbyTNC.Also,whileboththenational‐andlocal‐level
analysesarebasedonspatialmodeling,thenational‐leveloneisviamodelingeconomic
incentivesthatvarywithspatialpatternsofopportunitycost.Atthelocallevel,
opportunitycostislessvariableandinformationisscarcer.Forthesereasons,wetakea
differentapproachtospatialmodelinginthelocalcasestudies.
Theoverallapproachwetakeistofollowthefundamentalstepsfoundinthemost‐
widelyusedmethodologiesforestimatingreferenceemissionslevels(RELs)forREDD+
initiativesapprovedbytheVoluntaryCarbonStandardsgroup(VCS).Howeveritis
importanttonotethatthiswasnotareferencelevelsettingexercise.Wedonotconduct
themethodtothelevelofdetailthatwouldbeexpectedforaVCSProjectDescription(PD)
document,sincethatwouldrequirelocalfielddataonbiomassandlocalimprovementof
GISdatausedinmodels.NonethelesswefollowtheoveralllogicoftheVCSmethodologies
andtheirfundamentalstepsinspatialmodeling.
5.1.2.Definitionofextents
First,somespatialextentsaredefined.Thisincludesthesiteitself,whichineachof
thesevencasesisanexistingprotectedarea(PA).EachPAisoneofMexico’sREDD+early
actionsites(ÁreasdeÁccionTempranaREDD+;AATR).WeusedtheofficialPAboundary
filesprovidedtousbyTNC.Secondisthedefinitionofareferenceareaformodelingland
useinsideandaroundeachsite.
Eachreferenceareawasdefinedbyfirstcreatinga50kmbufferaroundtheAATR
site.Thisbufferwascombinedwithmunicipalityboundaries,andtheentireextentofany
municipalitythatintersectedthebufferwasincludedinthereferenceregion.
5.2.DataandMethods
5.2.1.Deforestationdata
Withineachsite’sreferencearea,weobtaineddataonforestcoverand
deforestationfrom2000to2012.Weusedthesamedatathatwereusedforthenational‐
levelanalysesfromthelatestUniversityofMaryland(UMD)assessment.Correspondingly,
thesedataarebasedontheanalysisofLandsatimagesandhaveaspatialresolutionof
30m.However,wedidnotconductanyspatialdegradation(coarseningofspatial
resolution),aswasdoneforthenational‐levelanalyses,sinceeachreferenceisnot
prohibitivelylargetoconductanalysesatfullresolution.


47

TheUMDsourcedatacanbeseenascomposedoftwoparts.Firstisamapoftree‐
coverpercentforeach30mcellinyear2000.Treecoverisnotthesameasforestcover.
Onecanassumethattreecoveraspresentedinthisproductisrelatedtocrowncoveras
estimatedinthefieldandusedinnationaldefinitions.However,thetwoconceptsarenot
theexactsame,andsuchanassumptioncanleadtoproblems.Forboththenationaland
local‐levelmodelingweusedthisassumptionforsimplicity.
Thenationaldefinitionofforesthas,ascriteria,aminimumcrown‐coverof25
percent.Weappliedthisvalueasathresholdtothepercenttree‐covermapfromUMDto
createamapofforestin2000.Thisleadstoagenerousestimateofthedistributionof
forestinthemodelingareas.Webelievethatmostsecondaryforestfallowsandshrub
fallowsassociatedwithrotationalagricultureorrecently‐abandonedfarmlandisincluded
intheestimationofforestextentin2000.Themappedandmodeledpatternsofforest
coveranddeforestationlikelyincludesiteswithsignificanttreecoverandareasof
clearanceoftreecoverthatarenotmatureforestortheclearanceof“mature”forest.We
believethatplantationsandselectively‐loggedforestarealsoincludedintheforestclass.
Thusthisdefinitionshouldbekeptinmindwheninterpretingresultsofthisstudy.
ThesecondpartoftheUMDdatafocusesonestimatesofthelocationsoflossof
treecoverforeachyearfrom2001to2012.Webelievethattheseshouldberobustdata,
sincethetemporal‐spectralsignalofsuchclearingeventsisstrong,andthemethodsof
UMDmaximizethepotentialfortheirdetectionbyminingtheentireLandsatarchiveover
thestudyperiodandemployaneffectivedecision‐treestatisticalapproach.Thus,we
expectthatthemajorityofdeforestationiscaptured,aswellasmuchoftheotherformsof
clearanceoftreecover,becauseofthegenerousdefinitionofforestextentin2000,as
notedabove.
Incontrasttothenationalanalysisthatconsideredforestlossesonanannualbasis,
forthelocalanalyses,theannuallossdataweregroupedtocreatemapsofforestlossover
twotimeperiods:2000to2006and2007to2012.Wethencombinedthemapsoflossfor
thesetwoperiodswiththatofderivedforestextentin2000tocreateathree‐dateproduct.
Inanefforttolimittheeffectofsmall‐scalechangesintreecoverthatmightnottruly
representforestlosses,wefilteredtheoutputtominimizeverysmallartifactsandtoseta
minimumpatchsizeforboththebaselineforestdistributionandpatternsofloss.First,a
three‐by‐threecellmajorityfilterwasappliedtothemergedproduct.Second,wefiltered
theoutputusingaone‐hectaresieve.Thiseliminatesanypatchofforestorforestlossthat
issmallerthanonehectareandreplacesthecellswiththedominantclassborderingthe
eliminatedpatchofcells.10

10Webelievedthisfilteringwasprudentinthelocalanalyses,butnotnecessaryinthenational
analysis,asthelattercontrolledforstartingforestareaintheeconometricprocedure.Inaddition,
themuchlargeramountofdatausedforthenationalstudyreducesthepotentialinfluenceof
spuriousforestlossobservations.


48

5.2.2.Otherdata
WeobtainedasuiteofdatafromTNCandpartnerstoexplorethespatialrelationships
betweenpossible“drivers”anddeforestation.Thedataaremoreaccuratelydescribedas
geographicalparametersratherthandrivers.Theseparametersareindicativeofwherethe
driversoccurandaremostlikelytobelinkedtodeforestationpatterns.Forexample,roads
themselvesarenotdrivers,buttheirdistributionindicateswherepeoplehaveeasier
accesstoforestsandcanrapidlymovetotheirhomesormarkets.Thus,roadsarea
geographicalparameterthatallowsustounderstandwheretheinteractionsamong
people,forestsandmarketsoccur,andtheythustypicallyarevaluableinpredictingwhere
deforestationwillmostlikelyoccur.However,theterm“driver”iscommonlyusedinsuch
modelingcontextstorefertodataonsuchgeographicalparameters,andwewilldosohere
forsimplicity.
Thespatialdataondriversweobtainedareofthreedatatypes.Firsttypeisraster
dataoncontinuousvariables,suchasdistancetoroadsandelevation.Thesecondtypeis
polygondatathatwereusedasclassvariables.Theseincludesoiltype,community,etc.
Thethirdtypeisadatasetcreatedspecificallyforthismodelingexercise.Torepresenthow
“marginal”acommunityis,i.e.howisolatedandlackinginresourcesand/orsubsidies,we
assignedathree‐classvariablebasedona“marginalization”indextoamapoflocationsof
communitycenters.Wethencreatedamapofdistancetoeachclassofcommunity.
Allofthesedatawererasterizedandcreatedorresampledtomatchthe30mcell
arrayofthedeforestationmap.Thefulllistofpotentialdriverdatatouseinthelocal
modelsisreportedinTable5.2.1.
















49

Table5.2.1.Driverindependentvariablesusedforspatialmodelsatthelocallevel.
Variable
labelVariablenameDatasourceNotes
Var1 var_dist_hi_marginalized_villages Conabio Themarginalizationindex
isasummaryfor
differentiatingcensus
townsinthecountry,
accordingtotheglobal
impactofdeficienciesthat
affectthepopulationasa
resultoflackofaccessto
education,residencein
inadequatehousingand
lackofassets.
Var2 var_dist_low_margin_villages Conabio
Var3 var_dist_medium_margin_villages Conabio
Var4 var_dist_primary_road Conabio
Var5 var_dist_railroad Conabio
Var6 var_dist_rivers Conabio
Var7 var_dist_secondary_road Conabio
Var8 var_dist_small_medium_cities Conabio Regionallyimportant
urbancenters,including
statecapitals
Var9 var_dist_trail Conabio
Var10 var_elev_30_30m INEGI Originalresolutionof60
meters
Var11 var_slope_30m INEGI Derivedfromthedigital
elevationmodel(DEM)
Var12 var_pop_dens
GlobalRural
Urban
Mapping
Project
(GRUMP)
Thisvariablewasnot
inlcudedintheSierra
Ramarurimodel
Var13 var_protected_areas_dummy Conabio Presenceorabsenceof
Federalprotectedareas
Var14 var_dist_non_forest_2006 UMD/Hansen
data
DerivedfromtheinputLC
maps
Var15 var_dist_megacities Conabio
Thisvariablewasnot
includedinmodelsfor
AATRswithoutamegacity
(populationGT75,000)
withinthereferenceregion



50

5.2.3.Spatialmodeling
WeusedtheIDRISILandChangeModelertool(LCM)forallspatialmodelingatthe
locallevel.ThisisdevelopedbyClarkLabsandoneofthestrongermodelingtools
availableforland‐usemodeling.Documentationonthetoolandtermsusedinthis
descriptioncanbefoundat:http://www.clarklabs.org/products/Land‐Change‐Modeling‐
IDRISI.cfm.
Theyearlydeforestationdatafrom2001to2012weregroupedintothreedates
andtwotimeperiods:2001‐2005‐2012.Thefirsttimeperiodisusedtocalibrateeach
localmodel.Thecalibratedmodelisthenusedtopredictdeforestationoverthefollowing
timeperiod.Sincedataonactualobservationsofdeforestationforthelatterperiodexist,a
validationofthemodelispossiblebycomparingthemodeledtoactualpatternsof
deforestation.
Forclassvariables,wecreated“evidencelikelihood”mapsforinputintomodels.
Theseassigntheproportionalimportanceofaparticularpolygontothestudyarea’s
overalldeforestationrate.Thisisthenusedasapotentialweightingfactorinthemodeling
algorithm.
TheLCMtoolandmethodsapprovedbytheVCScomparethespatialpatternsof
drivervariableswiththoseofhistoricaldeforestation.Statisticalrelationshipsarethen
usedtoproduceestimatesofthe“potential”fordeforestationineachmodelcell.These
valuesofpotentialcouldbere‐scaledtobevaluesoflikelihood,wheretheirsumequalsa
definedtotalrateforthemodeledperiod.Ifthisisdone,thentheoutputwouldbesimilar
tothenationalmodelinthatcellsareassignedacontinuousvalue.Thelikelihoodvalues,
rangingfromzerotoone,couldbeusedasiftheywereestimatesoftheproportionofthe
cellthatisdeforested.Thiscouldbecalleda“continuous”approach.
Anotherapproachistoassigncompletedeforestationtothecellswiththehighest
valuesofpotential,whichcouldbecalleda“discrete”approach.Thisproducesamap
wherecellsareeitherdeforestedornot.Thisassumesthatdeforestationentirelyoccursin
thesitesofgreatestpotentialorrisk.Whilethiscouldbearguedarealisticapproach,
thereareproblemswithitsassumptions,i.e.thatthereisnofinerscalevariationinrisk
duetounobservablereal‐worldfactors.Thus,highrisksitesarefullydeforestedandzero
deforestationhappensinallplacesotherthanthestrictlymostthreatenedsites.
Regardless,themethodsapprovedbyVCSallrequirethisdiscreteapproach,andthisisthe
approachthatweappliedinthelocalmodels.Wedo,however,maintainthecontinuous
dataondeforestationpotential,andfurtherstudycouldexplorethedifferencesbetween
theresultsofthetwoapproaches.
Theapproachofthistool,andofmostothersusedinsuchapplications,isto
calibratewithasubsetofthedata,whetherselectingaparticulartimeperiodorspatial
subset,thentorunthemodelandvalidateitwithalatertimeperiodorseparatespatial
subset.Weusedonetimeperiodinordertoallowtheoptionofvalidationoverthesecond
timeperiod.Differentalgorithmsformodelingtherelationshipsbetweendriversand
deforestationexist.WeselectedtheMulti‐LayeredPerceptron(MLP)algorithmwithin


51

IDRISI’sLCMbecauseofitsefficiencyandrelativelystrongperformancecomparedtoother
algorithms,suchasmultipleregression,etc.(Eastman,2005).TheMLPisaformofa
neuralnetworkthatcantakecontinuousandclassvariablesasinputsandisnotdependent
onassumptionsofnormaldatadistributions.
Weranmultiplemodelsforeachstudyareatogetageneralsenseofperformance
andimpactsofdifferenttypeofdataondrivers.Wetriedexcludingdifferentindividual
driversorsetsofdrivers.Amongthesevensites,wefoundthatthedataintheformof
polygonsalmostalwaysledtoresultswithconspicuousartifacts.Thesewerebothinthe
formofsharpchangesinthevaluesofpotentialalongboundariesofpolygons.Also,subtle
differencesamongpolygonshadexaggeratedimpactsontheresultingdiscretemapsof
predicteddeforestation.Ingeneral,wefoundthatthemodel,especiallythediscrete
predictionsoflocationsofdeforestation,werehighlysensitivetotheclassvariables.
Becauseofthis,ourfinalmodelsexcludedallpolygon‐typeclassvariablesotherthan
protectedareas.Thelatterwaskeptsincethisdatalayeryieldedrealisticimpactson
outputs,consideringthetrendsindeforestationratesinprotectedversusnon‐protected
landevidencedbythehistoricaldeforestationmaps.Asaresult,themostimportantsocio‐
economicparameterusedasaninputtothefinalmodelsisthedistancetocommunities
stratifiedbylevelofmarginalization.
Withtheselectionoffinalmodels,wehaveoutputsofestimatesofthepotentialfor
deforestation.Tocreatemapsofdiscretelocationsofpredicteddeforestation,werequired
asourceforthetotalrateofeachreferencearea.Weusedtheratesderivedfromthe
nationalmodelswithineachreferencearea.Therateforthereferenceregionisthen
appliedtothevalueofpotentialgeneratedbytheLCMmodel,assigningdeforestationto
thehighestpotentialcellsuntilthetotalchangeareaobtainedfromthenationalmodelis
reached.Wedidthisforthreedifferentscenarios:thealternative(Poisson)regression
modelofthe“business‐as‐usual”ornon‐REDD+scenario(AlternativeBAU),theratefrom
thepreferred(negative‐binomial)modelofthenon‐REDD+scenario(PreferredBAU),and
theratefromthepreferredmodeloftheREDD+scenario(PreferredBAU).Modelswere
runtosimulatedeforestationfrom2012through2022andoutputsweretabulatedfor
eachsiteandeachreferencearea.
5.3.Results
5.3.1.Deforestationsince2000
Deforestation,asdefinedbya25%thresholdappliedtotheUMDforestcoverin
2000andyearlytreecoverlosssincethen,hasbeensignificantinmostsites,especiallythe
Yucatánsite.Forestcoverin2000andaggregateddeforestationfrom2000to2012are
reportedinTable4.2.Annualizedratesarehighlyvariableamongsites.Twosites,
ComunidadesForestalesdeOaxacaMixtecaandSierraRaramuri,haveratesnearzero.
Twoothersites,ComunidadesForestalesdeOaxacaIstmoandSierraPucclosCheneshave
relativelyhighratesthatinareasapproach0.5percentperyear.Inmostcasesejidoshad
higherdeforestationratesthantherestofthelocalreferencearea,howeverinSierra


52

Rairumiprotectedareascategoryhadthehighestrate,andinsierraPucclosChenesthe
AATRhadthehighestrate.
PatternsofhistoricaldeforestationareshowninMapsA‐25throughA‐31inthe
Appendix.Inallthefigures,forestcoverisdefinedbya25%thresholdappliedtothe
Hansen,etal.(2014)data,anddeforestationisasumofalllossdatawithinthatdefined
forestarea.
Table5.3.1.Summaryofforestcoverin2000anddeforestationfrom2000to2012
amongAATRs.
Totalland
area(ha)
Forest
area,
2000(ha)
Forested
fraction
(2000)
Forest
area,
2012(ha)
Defor
00‐12
(ha/yr)
Defor
00‐12
(%/yr)
ComunidadesForestalesdeOaxacaIstmo
AATRsite 265,382 213,844 0.81 206,217 636 0.30%
Land‐use:Ejidos 784,033 383,469 0.49 362,324 1,762 0.46%
Land‐use:
Comunidades 1,755,724 1,334,791 0.76 1,308,087 2,225 0.17%
Land‐use:Protected
Areas(federal) na na na na na na
Comunidades_ForestalesdeOaxacaMixteca
AATRsite 471,624 203,659 0.43 203,555 9 0.00%
Land‐use:Ejidos 1,090,966 418,825 0.38 415,866 247 0.06%
Land‐use:
Comunidades 2,696,932 1,148,990 0.43 1,145,713 273 0.02%
Land‐use:Protected
Areas(federal) 435,452 67,044 0.15 67,021 2 0.00%
ComunidadesForestalesdeOaxacaSierraNorte
AATRsite 417,588 386,522 0.93 383,997 210 0.05%
Land‐use:Ejidos 903,043 402,775 0.45 389,478 1,108 0.28%
Land‐use:
Comunidades 1,932,267 1,253,583 0.65 1,237,395 1,349 0.11%
Land‐use:Protected
Areas(federal) 187,212 55,597 0.30 55,456 12 0.02%
CuencasInterioresdelaSierradeChiapas
Referenceregion 4,897,982 3,014,625 0.62 2,966,421 4,017 0.13%
AATRsite 1,058,629 611,574 0.58 606,088 457 0.07%
Land‐use:Ejidos 1,975,301 1,174,757 0.59 1,157,540 1,435 0.12%
Land‐use:
Comunidades 775,639 637,955 0.82 630,996 580 0.09%
Land‐use:Protected
Areas(federal) 639,767 526,698 0.82 523,044 304 0.06%
CutzamalaValledeBravo
Referenceregion 3,008,360 1,011,168 0.34 1,007,582 299 0.03%
AATRsite 263,333 117,204 0.45 116,480 60 0.05%
Land‐use:Ejidos 1,219,455 320,089 0.26 319,202 74 0.02%
Land‐use:
Comunidades 229,479 111,390 0.49 110,940 38 0.03%
Land‐use:Protected
Areas(federal) 273,411 151,909 0.56 150,921 82 0.05%


53

SierraPuccLosChenes
AATRsite 1,535 1,429 0.93 1,332 8 0.57%
Land‐use:Ejidos 7,109 6,480 0.91 6,091 32 0.50%
Land‐use:
Comunidades 1 1 0.59 1 0 0.55%
Land‐use:Protected
Areas(federal) 1,608 1,252 0.78 1,241 1 0.08%
SierraRaramuri
AATRsite 1,883,875 984,941 0.52 984,242 58 0.01%
Land‐use:Ejidos 5,933,054 2,783,030 0.47 2,776,926 509 0.02%
Land‐use:
Comunidades 1,560,860 871,818 0.56 870,021 150 0.02%
Land‐use:Protected
Areas(federal) 70,206 47,239 0.67 46,935 25 0.05%
Note:Forestcoverisdefinedbya25%thresholdappliedtotheHansen,etal(2014)data,and
deforestationisasumofalllossdatawithinthatdefinedforestarea.Notethatdeforestationvalues
differfromthoseintheglobalanalysissincethehistorical‐deforestationmapswerefilteredforthe
localanalysis.Thefilteringremovedanypatchesofforest,non‐forestordeforestationforagiven
timeperiodsmallerthanonehectare.
5.3.2.Modeleddeforestationbeyond2012
Weranmultiplemodelswithdifferentcombinationsofdrivervariables.Among
these,thegenerallyconsistentresultwasthatthebestperformingmodelsweretheones
usingall15inputvariables.Also,wefoundthattheinclusionofdistancetoanon‐forested
edgedidnotimprovemodels.Thisparametertendedtoleadtoanover‐fittingof
deforestationalongexistingedges,andexclusionofthisparameterdidnotresultinun‐
realisticallyremotedeforestationinthemodeloutputs.
Thus,ourfinalmodelsallwerebasedontheMLPmodelsusingallinputsexcept
distancetoanon‐forestededge.Finally,MLPrandomlyselects“seedcells”tobeginmodel
calibration,andmodeloutputsmayvarymodestlyinarandommannerdependingonthe
selectionoftheseseeds.Thus,wereporttwomodeliterationsforeachfinalmodel.We
appliedthesensitivityanalysisincludedinIDRISI’sMLPtooltoestimatetherelative
importanceofdifferentinputvariables.ImportancevaluesarereportedinTable5.3.2.
Inputvariablesmostimportanttothemodelvariedamongthestudyareas.
Distancetomega‐citieswashighlyimportantforthestudyareaswheretheyoccurred,
CutzamalaValledeBravoandSierraRaramuri.Forregionsthatareexemplaryoffrontier
areas,accessibility,i.e.distancetoroads,trailsandrivers,wasmostimportant.Forregions
thatareexemplaryofheavily‐fragmentedforest,biophysicalvariables,e.g.slope,were
mostimportant.Therewasoverallnoconsistenttrendontheimportanceofthevariable
distancetohighly‐marginalizedvillages.Insomeareassitesnearhighly‐marginalized
villageshadhigherdeforestationrateswhileinotherareasthetrendwasreversed.Inall
butonestudyarea,includingdistancetonon‐forestlandincreasedmodelskill.Only
CutzamalaValledeBravo,whichishighlyfragmentedforest,didn’thavethiseffect.


54

Table5.3.2.Relativeimportanceofthedifferentdrivervariablesformodelsrunineachofthelocalstudyareas.SeeTable5.3.1for
thelistofvariables.
RegionModel
Run
var1var2var3var4var5var6var7var8var9var10var11var12var13var14var15
Skill
Dist.hi
margin
alized
villages
Dist.
low
margin.
villages
Dist.
medium
margin.
villages
Dist
primar
yroad
Dist.
railroa
d
Dist.
rivers
Dist.
Second
ary
road
Dist.
small/
medium
cities
Dist.
trail
Elev30
30m
Slope
30m
Pop
density
Protect
ed
areas
dummy
Dist
nonfor
est:
2006
Dist
Megacit
ies
Comunida
des
Forestales
deOaxaca
(Istmo)
MLPrun1 12 8 5 10 1 13 2 3 9 4 7 11 n/a 6 n/a 0.4935
MLPrun2 11 7 13 4 3 8 6 1 9 12 5 10 n/a 2 n/a 0.5348
MLPrun
w/odist.to
non‐forest
8 10 5 9 2 11 3 1 6 4 7 12

n/a n/a n/a 0.4931
Comunida
des
Forestales
deOaxaca
(Mixteca)
MLPrun1 11 4 13 8 5 14 7 3 2 1 12 10 12 9 n/a 0.7419
MLPrun2 13 4 10 11 5 9 7 2 3 1 14 12 14 8 n/a 0.7405
MLPrun
without
distanceto
non‐forest
11 4 12 8 6 7 13 2 3 1 10 9

10 n/a n/a 0.6802
Comunida
des
Forestales
deOaxaca
(Sierra
Norte)
MLPrun1 14 6 7 2 5 13 11 5 10 1 3 8 12 4 n/a 0.5593
MLPrun2 14 5 13 2 5 12 8 6 10 4 3 7 9 1 n/a 0.5958
MLPrun
w/odist.to
non‐forest
13 5 4 3 6 7 11 8 12 1 2 10

9 n/a n/a 0.5351
Cuencas
Interiores
dela
Sierrade
Chiapas
MLPrun1 13 5 9 6 3 10 12 4 11 2 1 13 7 8 n/a 0.4124
MLPrun2 11 14 7 10 6 13 8 4 12 2 1 9 3 5 n/a 0.4504
MLPrun
w/odist.to
non‐forest
12 4 8 5 3 13 10 6 9 2 1 11

7 n/a n/a 0.4006
Cutzamala
Vallede
Bravo
MLPrun1 8 6 10 9 12 7 3 4 15 2 13 14 5 11 1 0.6282
MLPrun2 6 5 15 10 13 8 3 4 14 2 11 12 9 7 1 0.677
MLPrun
w/odist.to
non‐forest
9 7 11 6 10 8 2 5 13 4 12 14

3 n/a 1 0.71


55

Sierra
PuccLos
Chenes
MLPrun1 9 11 10 4 2 12 8 3 7 14 6 5 13 1 n/a 0.5661
MLPrun2 9 2 14 3 11 4 5 5 10 12 7 6 13 1 n/a 0.5715
MLPrun
w/odist.to
non‐forest
6 2 1 8 5 3 13 7 9 4 12 11

10 n/a n/a 0.3906
Sierra
Raramuri
MLPrun1 13 12 5 14 6 8 7 2 9 1 4 n/a 11 10 3 0.6178
MLPrun2 14 12 5 13 9 8 11 2 7 1 4 n/a 10 6 3 0.6115
MLPrun
w/odist.to
non‐forest
11 10 6 7 8 9 5 3 12 1 4 n/a

13 n/a 2 0.6224
Note:Lownumericalvaluesindicatehigherimportancelevels,i.e.thesearerankscores.Themostimportantthreevariablesforeachmodelarehighlighted
inyellow.


56

5.4.Predictingdeforestationinthefuture:
Thepatternsofpotentialfordeforestationvariedamongthesites,althoughare
understandablegiventhedifferingimportanceofdrivervariablesinthedifferentstudyareas.Itis
thususefultorefertoTable5.3.2wheninterpretingthepatternsofpotential.Mapsofpotential
deforestationor“soft”deforestationtransitionpotentialareshowninthemapsinfigures5.4.1b‐
5.4.7bbelow.Onealsoseesthatthereismoreinformationinthesemapsthanthehard
classificationsshownaboveeachmap(5.4.1a‐5.4.7a),andonecaninterpretthepatternsofrelative
potentialbeyondconsideringonlythesitesofstrictlygreatestpotential,asisthecaseinthehard
classificationspresentedlaterinthissection.
ThehardclassificationoffuturedeforestationforeachAATRwasbasedonthetransition
potentialsurfacescreatedcombinedwiththetotalratesforeachreferenceregionaccordingthethe
differentscenariosofthenationalmodels.Theseharddeforestationpredictionsareshowninmaps
in5.4.1a‐5.4.7a.Thetransitionpotentialchosenforthefinalpredictionwerebasedonthemodels
withthehighestskillmeasure(shownintable5.3.2).Theratesoftransitionthatwereappliedto
thetransitionpotentialsurfacesareshownbelowinTable5.4.1,aswellasthetotalamountof
deforestationpredictedforeachreferenceregionandAATRsite.Thequantityofdeforestation
predictedwithineachreferenceregionswasbasedontheinputratesshownbelow,andthe
allocationofthedeforestationwasdeterminedbyselectingtheforestcellswiththehighestvalues
inthetransitionpotentialsurfacescreatedinLCM.Thismethodfordeforestationallocationhasthe
advantageofbeingabletocreatethematicland‐covermapsatthenativeresolutionoftheinput
dataset,howeveritalsoassumesthatdeforestingagentknowwhichpixelsareoptimalfor
deforestationandthereforeonlythemosthighlyvulnerablepixelswillbetransitioned.
RatesfortransitionwerederivedfromboththeobservedhistoricalrateintheLCMmodel
andthemodelednationalratesfromthenationalmodel.Thetransitionsshownrepresentthetotal
transitionsfrom2012to2022,andinthecaseofthenationalmodels,wepresentresultsforbotha
businessasusualscenario(BAU)andREDD+scenarioassuminga$10/tCprice.Thehistoricalrate
fromderivedfrom2000‐2012isalsoshowandprojectedlinearlyto2022,however,oneshould
notethatthehistoricalratecannotbedirectlycomparedwiththemodeledratesfromthenational
modeldueinparttothefilteringprocessthatwasusedontheforestcoveranddeforestationdata
fromHansenetal.2013.ThereforethehistoricalrateisconsistentlylowerthantheBAUscenarios.
Thehistoricalratesalsodonottakeintoaccountlargernationaltrendwhichwouldhaveaneffect
onboththemodeledratesandthefuturerates(seetable4.4.1).Thereforethehistoricalratesare
providedascontextonhowmuchdeforestationmightbepredictedwithouttheuseofanexternal
model,usingthemostsimplisticapproach.Amoreusefulcomparisonforunderstandingtheeffect
ofnationalpolicyonthelocalmodelsisthedifferencebetweenthepreferred(NegativeBinomial)
BAUscenarioandthepreferred(NegativeBinomial)REDDscenario.Comparingthetwonegative
binomialscenariosshowsthatatthereferenceregionscalethereweredecreasesbetween23–
58%,whichisconsistentwiththenationallevelpredictionsshowninTable4.4.3.
AttheAATRsitescaletherangeismuchmorevariationintheamountofdeforestation
predictedbetween2012and2022.InthecaseoftwoAATRsites,OaxacaMixtecaandSierra
Raramuri,therewasnodeforestationpredictedwithinthesite,regardlessofthescenario.The
reasonthatthiswasthecaseisduetothemethodwhichwasusedtoassignchange.Asdescribed


57

above,becauseonlythehighestrankedpixelsaretransitionedandthesesiteshaveverylow
deforestationrates(rangingfrom0.23%–2.12%overthe10yearperiod).Similarly,the
extremelyhighreductionindeforestationintheOaxacaSierraNorteAATRsitecanalsobe
attributedtothemethodofallocation.InthecaseofthesethreeAATRsites,REDD+initiatives
wouldhaveaminimaleffectbecausethebaselineratesunderallscenariosareverylow.The
remainingfourAATRsitesobservedareductionindeforestationrangingfrom20%‐67%,andin
mostcasesthesereductionsweresimilartothoseexperiencewithinthereferenceregions.
Thehardclassificationsofpredicteddeforestationindicatedifferentconclusionsamongthe
differentstudyareas(Table5.4.1).Forexample,forthreestudyareas,Mixteca,Chiapasand
Raramuri,bothBAUmodelspredictedratesofdeforestationofovertwicethehistoricalrates.The