Mexico Deforestation Vulnerability Final Report mwRL

Transcription

Mexico Deforestation Vulnerability Final Report mwRL
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
ALIANZAMÉXICOPARALAREDUCCIÓNDE
EMISIONESPORDEFORESTACIÓNYDEGRADACIÓN
PolíticapúblicaDesarrollodecapacidadesArquitecturafinancieraMRVComunicación
i
www.alianza‐mredd.org.mx
ThisreportwasmadepossiblebythegeneroussupportoftheAmericanpeoplethroughtheUnited
StatesAgencyforInternationalDevelopment(USAID)underthetermsofitsCooperative
AgreementNumberAID‐523‐A‐11‐00001(M‐REDDProgram)implementedbyprimerecipientThe
NatureConservancyandpartners(RainforestAlliance,WoodshallResearchCenterandEspacios
NaturalesyDesarrolloSustentable).Thecontentsandopinionsexpressedhereinarethe
responsibilityoftheM‐REDDPROGRAManddonotnecessarilyreflecttheviewsofUSAID.Views
expressedinthispaperdonotnecessarilyreflectthoseoftheanyoftheauthors’institutions.Any
remainingerrorsaretheauthors’soleresponsibility.
TheauthorsarealsothankfulforvaluableguidancefromBronsonGriscomandPeterEllisofTNC,
helpfulcommentsfromYvesPaiz,JuanFranciscoTorresOrigel,andJoseCantoVergaraofTNC,
earlyadviceandsupportfortheprojectfromLeticiaGutierrezLoranidiofTNC,aswellasvaluable
inputfromparticipantsintheworkshopon“ModelingDeforestationintheYucatanandBeyond”
convenedbyTNCandMREDDfromApril30toMay2,2014inMerida,Mexico.WealsothankJose
CarlosFernandezofINECCforinspirationandinvaluablesupportfromthefirstinceptionofthe
nationalanalysis.Inaddition,theauthorsaregratefulforimportantmodelingcontributionsfrom
RuohongCaiofEDF,GISsupportfromJeremeyProvilleofEDF,andeditorialassistancefromPasha
FeinbergofEDF.Theauthorsgratefullyacknowledgekeydataonmaximumcroprevenuesfrom
JensEngelmannofCGDaswellasdataoncommunalpropertiesfromLeonardGoffandAllen
BlackmanatRFF.
ii
Acknowledgments
Authors
RubenLubowski,MaxWright,KalifiFerretti‐Gallon,A.JavierMirandaArana,MarcSteininger,and
JonahBusch
Contactdetails:
Projectlead
RubenLubowski;rlubowski@edf.org
Literaturereviewandmeta‐analysis
KalifiFerrettiGalon(lead);kalifi.fg@gmail.com
JonahBusch;jbusch@cgdev.org
Nationalmodeling
RubenLubowski(lead)
A.JavierMirandaArana(consultant);javmi@yahoo.com
Localmodeling
MaxWright(lead);twright@conservation.org
MarcSteininger;msteininger@conservation.org
Organization
Core Staff
EnvironmentalDefenseFund
RubenLubowski(ChiefNaturalResourceEconomist)
A.JavierMirandaArana(Consultant)
ConservationInternational
MarcSteininger(ScientificDirector)
MaxWright(RemoteSensingandGeospatialAnalyst)
CenterforGlobalDevelopment
JonahBusch(ResearchFellow)KalifiFerretti‐Gallon
(ResearchAssistant)
iii
Contents
1.ExecutiveSummary
1.1.KeyFindings
2.Introduction
1 1 5 2.1.GlobalGreenhouseGasesandMexico’sForests
5 2.2.ReportOutline
5 3.LiteratureReviewofDriversofDeforestationinMexico
11 3.1.Introduction
11 3.2.Overviewofdeforestation
11 3.2.1.DeforestationinMexico
11 3.2.2.DeforestationintheYucatán
12 3.3.Overviewoflandtenure,ruralagriculturalsupport,andpaymentsfor
ecosystemsservicesinMexico
12 3.3.1.LandTenure
12 3.3.1.1.Privatelands
12 3.3.1.2.PublicLands
12 3.3.1.3.CommunalLands
12 3.3.2.RuralAgriculturalSupport
14 3.3.3.PaymentsforEcosystemsServices
15 3.4.Databaseregressionresults
16 3.4.1.AMeta‐analysisofDriversofDeforestationinMexico:Methods
16 3.4.2.ResultsforMexicoandSEsub‐regions
17 4.Analysisofdeforestationatnationallevel/OSIRIS
21 4.1.Introduction
21 4.2.EmpiricalModel
21 iv
4.2.1.EconometricSpecification
21 4.2.1.1.Relationshipofdeforestationtoavailableforestareawithina900mgridcell
22 4.2.1.2.Observedandunobservedcomponentsofnetreturnsfromlandconversion
24 4.3.HistoricalSimulations
26 4.3.1.SimulationScenario
26 4.3.2.SimulationResults
28 4.3.2.1.ChangesinAgriculturalReturns
28 4.4.Futureprojections
32 4.4.1.1.“Business‐as‐usual”projection
33 4.4.1.2.CarbonIncentiveProjections
37 5.LocalModelingofDeforestation
46 5.1.Introduction
46 5.1.1.Overallapproach
46 5.1.2.Definitionofextents
46 5.2.DataandMethods
46 5.2.1.Deforestationdata
46 5.2.2.Otherdata
48 5.2.3.Spatialmodeling
50 5.3.Results
51 5.3.1.Deforestationsince2000
51 5.3.2.Modeleddeforestationbeyond2012
53 5.4.Predictingdeforestationinthefuture:
56 5.5.Conclusions
66 6.Conclusion
69 6.1.Summaryofreportfindingsanddirectionsforfutureresearch
69 6.1.1.Literaturereviewandmeta‐analysis
69 6.1.2.Nationalmodeling
70 6.1.3.LocalModeling
71 v
7.Workscited
73 Listoftables,figuresandmaps
Tables
Table3.4.1DriversofdeforestationinMexico,bydrivercategory
17 Table4.2.1.Principalexplanatoryvariablesusedinnationalregressions(900mcell)
25 Table4.3.1.Simulationscenariosoverhistoricalperiodindataset,2000‐2012
26 Table4.3.2.NationalSimulationResults
29 Table4.3.3.RegionalSimulationResultsforSensitivitytoAgriculturalReturns
31 Table4.4.1.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,byAATR
referenceregionsandlandownershipcategory
35 Table4.4.2.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,bynational
regionsandAATRReferenceRegions
36 Table4.4.3.FuturePredictions,2014‐2024,Business‐as‐Usualand$10/tonCO2PolicyCase,
forAATRandnon‐AATRregions
40 Table5.2.1.Driverindependentvariablesusedforspatialmodelsatthelocallevel.
49 Table5.3.1.Summaryofforestcoverin2000anddeforestationfrom2000to2012among
AATRs.
52 Table5.3.2.Relativeimportanceofthedifferentdrivervariablesformodelsrunineachof
thelocalstudyareas.SeeTable5.3.1forthelistofvariables.
54 Table5.4.1.Predicteddeforestationfrom2012‐2022
58 Figures
Figure2.2.1.ProjectFlowchart
Figure3.4.1DriversofDeforestationinMexico:ResultsofMeta‐Analysis
9 18 Figure3.4.2.DriversofDeforestationintheYucatánPeninsulaasComparedtotheRestof
Mexico:ResultsofMeta‐Analysis
20 Figure4.4.1.EstimatedcostcurvesforCO2emissionsreductionsfromabove‐groundforest
45 carbonlossesinMexico,byregion
Figure4.4.2.Estimatedcostcurvesforreducingemissionsfromabove‐groundforestcarbon
lossesinMexico,byAATRandnon‐AATRregions.
45 Figure5.4.1.Predicteddeforestation2012‐2022,OaxacaIstmo.AATRsitehighlightedin
yellowthatching.
59 vi
Figure5.4.2.Predicteddeforestation2012‐2022,OaxacaMixteca.AATRsitehighlightedin
yellowthatching.
60 Figure5.4.3.Predicteddeforestation2012‐2022,OaxacaSierraNorte.AATRsite
highlightedinyellowthatching.
60 Figure5.4.4.Predicteddeforestation2012‐2022,SierraChiapas.AATRsitehighlightedin
yellowthatching.
62 Figure5.4.5.Predicteddeforestation2012‐2022,CutzmalaValleBravo.AATRsite
highlightedinyellowthatching.
63 Figure5.4.6.Predicteddeforestation,SierraPUCC.AATRhighlightinyellowthatching
64 Figure5.4.7.Predicteddeforestation,SierraRaramuri.AATRsitehighlightinyellowthatch
65 Maps
Map4.4.1.Projected“BusinessasUsual”(BAU)ForestLoss2014‐2024
41 Map4.4.2.ProjectedAvoidedForestLoss2014‐2024,with$10/tonCO2incentive
42 Map4.4.3.ProjectedRemainingForestLosswith$10/tonCO2incentive,2014‐2024
43 Map4.4.4.ProjectedAvoidedEmissions2014‐2024,with$10/tonCO2incentive
44 vii
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
1. ExecutiveSummary
In2010,Mexicoranked8thamongcountrieswiththelargestareaofprimaryforest(FAO
2010).Mexico’sforests,coveringaboutathirdofthenation,provideanumberofservicesincluding
carbonsinks,highlevelsofendemismandspeciesrichness,andsubsistenceresourcesforlocal
population.TheseservicesarebeingerodedasMexicocontinuestoexperienceforestcoverloss.
Mexicohaslostabouthalfitsforestareasince1950.From2005‐2010,thecountrymaintainedan
averagedeforestationrateof0.24%accordingtoFAO,reducingitscapacityforcarbon
sequestrationandincreasinglandconversionrelatedemissions.Landuse,land‐usechangeand
forestrywasrecentlyestimatedtoemitabout10%ofMexico’stotalGHGemissions.
TheMexico‐REDDAlliance(MREDD)issupportingMexico’seffortstoreduceitsemissions
fromdeforestationandforestdegradationandtoenhanceforestcarbonstocksarecurrently
supportedby.TheprogramhasidentifiedEarlyActionAreas(ÁreasdeAcciónTempranaorAATR),
orhighrisk–highrewardareaslocatedinMexicanstatesthatarerecognizedashavinghigh
biodiversity,culturaldiversityaswellashighratesofdeforestation.Researchrelatedtoforest
coverlossinMexicohassofarfocusedondriversofdeforestation,includingtheimpactofland
ownershiptypesuniquetothecountry(communityforestry,protectedareas,andprivatelands).
Missingfromthesizeableliteraturearetwotopicsofparticularimportancefortheidentificationof
vulnerableregionsandthedesignofconservationstrategiesunderMREDD:thefirstisananalysis
ofdriversofdeforestationthatisspecifictotheAATRs.Thesecondisananalysisoftheeffectof
geographiccharacteristicsorpolicymeasuresthatisdisaggregatedbylandownershiptype.
Tohelpaddressthesegaps,weconductaseriesofanalysesthatcombinebothnationaland
localscalemodelingtoaidtheMREDDAlliancepartnersinassessingthevulnerabilityofMexico’s
foreststodeforestation.TheseanalysesfocusonthevulnerabilityofforestedlandswithinMexico’s
AATRs,accountingforMexico’suniqueforestmanagementdynamicsthroughdisaggregatingthe
resultsbylandownershiptypes.Theseanalysesareultimatelymeanttoinformnationaland
subnationalpolicy,pavingthewayforincentivebasedprograms,andultimatelyreduced
deforestationvulnerabilityinMexico.Ourmethodologyincludesthreedifferentand
complementaryapproaches:(i)reviewingtheexistingliterature,(ii)anationaleconometric
analysisandassociatedscenariosimulationmodeling,and(iii)local‐levelspatialspatialmodeling
foreachAATR.Keyfindingsfromeachofthesethreepartsofthereportaresummarizedbelow.
1.1.KeyFindings
LiteratureReviewofDriversofDeforestationinMexico.

Whiledeforestationhasdecreasedoverthepastdecades,forestlosscontinuesatabout
0.24%peryear,accoridingtoUN‐FAO,generatingabout6%ofthecountry’stotal
greenhousegasemissionsin2010.Deforestationismostlyoccurringinthemoredensely
forestedareasofsouth‐easternMexicandlargelyattributedintheliteraturetocropand
cattledevelopment.
1
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report





Landtenure(communitylandmanagement,includingejidos),ruralagriculturalsupport,
andpaymentsforecosystemsservicesaremajorfocusesoftheliterature.Conclusionson
theroleofthemajorlandtenuretypeinMexico,communitylandmanagement,aremixed.
Studiesarealsoindisagreementontheroleofsuchruralagriculturalsupportprogramsas
PROCAMPO.
Moststudiesagreethatpaymentsforecosystemsservicesdecreasedeforestationrisk,with
somecaveatsrelatedtoregionaldifferencesandstartingdeforestationrisk.
Theserelationshipsweremirroredinthemeta‐analysis:regressionresultsweremixedfor
ejidosandruralincomesupport,whileresultsforPEStendedtobeassociatedwith
decreaseddeforestation.Furthermore,resultsfromthemeta‐analysisrevealedother
variableswithconsistentrelationshipstodeforestationinMexico.Thevariablesmost
associatedwithreduceddeforestationinMexicowereassociatedwithprotectionmeasures
(asproxiedbyprotectedareasandPES),reducedaccessibility(elevation),reducedresource
competition(propertysize)andcommunityforestry.
Thevariablesmostassociatedwithincreaseddeforestationwererelatedtoareaswhere
economicreturnstoagriculturearehigher(proximitytoagricultureandagriculture
returns),biophysicalconditionsforconversionarefavorable(soilsuitability),and
competitionforresourcesarehigh(population).
MostoftheserelationshipswererobustwhenresultsweredisaggregatedtotheYucatán
Peninsula.Notablyhowever,atthenationallevel,povertyappearstobelinkedtoincreases
indeforestation,whileintheYucatánPeninsulapovertyisassociatedwithdecreased
deforestation.Conversely,indigenouspopulationisassociatedwithdecreased
deforestation.
NationalAnalysisofDeforestationinMexico.



Thenationalanalysisrevealsthatacriticaldriverofdeforestationhasbeentheanticipated
economicreturnsfromlandconversion,specificallyfromagricultureasproxiedbycrop
productioninourstudy.Keyfactorsmodulatingdeforestationvulnerabilityincludeland
ownershiptypeandinitialforestareawithinagridcell.
Weestimatetheresponsivenessofgrossdeforestationtochangesinneteconomic
incentivesforlandconversion.A1%decreaseinpotentialagriculturalreturnsover2000‐
2012wouldhavedecreasedcumulativegrossdeforestationnationallyoverthisperiodby
anestimated0.24%.Conversely,a1%increasewouldhaveboostedgrossdeforestationby
anestimated0.26%.Similarly,a10%decreaseinpotentialagriculturalreturnsover2000‐
2012wouldhavedecreasedcumulativegrossdeforestationnationallyoverthisperiodby
anestimated2%.Conversely,a10%increasewouldhaveraisedgrossdeforestationbyan
estimated3.3%.
Apreliminaryexaminationsuggeststhatdecreasingpotentialcropreturns(orincreasing
benefitstolowemissionsactivitiesthatavoiddeforestation)bytheamountofPROCAMPO
subsidiesonejidosandagrariancommunitylandswouldhavedecreaseddeforestationby
about5%over2000‐2012.
2
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report





Basedontheeconomicprofitabilityofagricultureandstartingforestcoverin2012,the
modelpredictsanoverall27%“business‐as‐usual”increaseinannualdeforestationin
Mexicooverthenexttenyears,relativeto2000‐2012.
Ontheonehand,thereisrelativelyhighsensitivitytoagriculturalreturnsandhigh
estimatedfuturevulnerabilitytodeforestationamongforestremantsinareaswith
relativelysparserforestcover,includingintheNorthwestandBajioandNortheastregions.
Comunidadesandprotectedareaswerethelandtypesprojectedtohavethebiggest
proportionalincreaseinforestlossesoverthenext10yearsandarealsoestimatedtohave
thegreatestpercentdeclinesinresponsetoapotentialcarbonincentive.
Themostsensitiveareas,however,arenotthatimportantinabsoluteterms.Thegreatest
amountofdeforestationisprojectedtooccurintheSouthandYucatanPeninsularegion,as
wellaswithinejidosandprivatelandtypes.Theseareas,particularlytheYucatan
Peninsula,holdthelion’sshareofestimatedpotentialforreducingdeforestationand
emissions.
ThesevenAATRsarenotallconcentratedintheareaswiththehighestprojectedfuture
deforestation,andsomeofsitesarelocatedinareaswithlowhistoricalratesofforestloss,
comparedtothenationalaverage.Nevertheless,overallasagroup,AATRsandtheir
surroundingregionshavehigherprojecteddeforestationincreasesthanotherforested
areasnationally,aswellasregionally,aswellasthemajorityofthepotentialtocost‐
effectivelyavoiddeforestation.
Weuseourstatisticalparameterstoestimatenationalandregionalcarbonemissionscost
curves,basedonahypotheticalcarbonincentivefocusingonlyonabove‐groundforest
carbon.Wefindthatthereisrisingpotentialnationallytoreduceemissionsatcostsranging
from$5to$100/tonCO2,atwhichpointabout90%oftheemissionsareavoided.About
halfoftheestimatedreductionsavailableatpricesof$10/tonCO2orbelowandmorethan
twothirdsoftheestimatedreductionsavailableatpricesof$20/tonCO2orbelow.The
nationalandregionalcostcurvesarerisingatanincreasingrate,indicatingthatitcosts
moreandmoretoavoiddeforestationonlandswithgreateragriculturalpotentials. LocalModelingofDeforestationinMexico.


Judgingfromtheprojecteddeforestationscenarios,thegreatestbenefitsfrom
implementingREDD+oranotherincentivebasedconservationactivitywouldbefeltin
AATRsitesthatareprimarilyunfragmentedforest,meaningthattheystillcontainlarge
areasofundisturbedcoreforest,andareexperiencingfrontierexpansion,usuallystemming
frompopulationcentersoraccesspoints.SitessuchasSierraPUCCCheneandOaxaca
IstmodisplaythesecharacteristicsascomparedtositeslikeOaxacaMixtecaorSierra
Raramuriwhicharehighlyfragmentedandexperiencelowerratesofdeforestation.
Variablesrelatedwithaccessibilityandmarketsweremostinfluentialintheless
fragmentedreferenceregions,whilevariablesrelatedtobiophysicalsuitabilityweremost
influentialinthefragmentedsites.Thevariable,distancetomegacites,wasimportantin
3
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report



thetworegionsthatcontainedthem(SierraRararmurinearCuliacunandCutzemalaValle
BravonearToluca).
Theremaybemultiplepatternsofforestchangepresentinthereferenceregions;lossof
primaryforest,lossofsecondaryforest,fallowrotationsandagro‐forestry.Modelscouldbe
strengthenedbyaddressingtheseseparatelyorfocusingonaparticularpattern.
Theinterpretationofthelocalmodelsshouldincludeboththesoftandhardpredictions
underthevariousscenariosaswellasthegeneralpatternofthesofttransitionsurface.
Futureworkcouldincludeamorethoroughexaminationoftheeffectsofland‐usepractices
withincomunidadesandejidos,asthesedesignationshadsomeinfluenceoverthemodels,
howevertheresultsweremixed.
4
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
2. Introduction
2.1.GlobalGreenhouseGasesandMexico’sForests
Greenhousegasemissionsfromagriculture,forestryandotherland‐useactivitiesaccount
foranestimated24%ofglobalemissions,secondonlytoemissionsproducedbyfossilfuel
combustion(IPCC5thAssessmentReport,2014).In1990,officialestimatesarethatdeforestation,
forestdegradation,andotherland‐usechangesinMexicoproducedover100MtCO2eofemissions
peryear,accountingfor18.2%ofnationalemissions.Morerecentlyin2010,forestsandland‐use
changesproducedcloseto47MtCO2eorabout6.3%oftotalemissions(SEMARNAT/INECC,
2012).Mexicoiscurrentlyundertakingeffortstoreduceitsemissionsfromdeforestationand
forestdegradationandtoincreasesequestrationbyenhancingforestcarbonstocks(REDD+),
supportedbytheMexico‐REDD(MREDD)Allianceprogram.1Crucialtothesuccessofanti‐
deforestationpoliciesisanunderstandingofhowspatialvariationingeographiccharacteristics,
landownership,economicprofitability,andpolicymeasuresaffectMexico’svulnerabilitytoforest
coverloss.Itisalsoimportanttounderstandhowpotentialchangesinthesefactorsovertime
mightaffectdeforestationinthefuture.
Mexicohaslostroughlyhalfitsforestareasince1950.From2005to2010,thecountrylost
155,000hectaresofforestcover,anaveragedeforestationrateof0.24%(FAO,2010).Forest
conservationinMexicoprovidesbiodiversityco‐benefitsbeyondclimate,asthecountryboasts
bothhighlevelsofendemismandspeciesrichness(Barsimantov&Kendall,2012).While
deforestationrateshavedecreasedandreforestationeffortsareevident(FAO,2010),widespread
deforestationcontinuestothreatencommunitiesandecosystemsthatdependonforests.Thereisa
needtobetterunderstanddeforestationinMexicotoidentifyvulnerabilitiesandinformpolicies
thataimtoreduceforestloss.
2.2.ReportOutline
Missingfromthesizeableliteratureonland‐usechangeinMexicoaretwotopicsof
particularimportancefortheidentificationofvulnerableregionsandthedesignofandlow‐
emissionsdevelopmentstrategiesunderMREDD:thefirstisananalysisofdriversofdeforestation
thatisspecifictotheREDD+earlyactionareas(AATRs)undertheMREDDprogram.2Thesecondis
ananalysisoftheeffectofgeographiccharacteristicsorpolicymeasuresthatisdisaggregatedby
landownershiptype(e.g.ejido,protectedarea,privatelands).
1TheAllianzaMREDD+isapartnershipofTheNatureConservancy,RainforestAlliance,theWoodsHole
ResearchCenter,Mexico’sgovernment,andcivilsocietytohelplaythebasisforeffortstoreduceemissions
fromdeforestation,forestdegradation,andotherforestryactivities(i.e.REDD+).(see:www.alianza‐
mredd.org)
2
ÁreasdeAcciónTemprana(AATR)areREDD+EarlyActionareaslocatedinMexicanstateswithhigh
biodiversity,culturaldiversityandhighratesofdeforestation,butalsogreatREDD+potential.Lessons
learnedinthesesubnationaltargetareascouldhelpscaleupbestpractices.
5
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
Toaddressthesegaps,weconductaseriesofanalysesthatcombinebothnationalandlocal‐
scalemodelingtosupporttheMREDDAlliancepartnersinassessingthevulnerabilityofforested
landstodeforestationinMexico,focusingonthevulnerabilityofforestedlandswithinMexico’s
AATRs,withtheresultsdisaggregatedbylandownershiptypes.Thegoalistoinformnationaland
subnationalpolicy,pavingthewayforincentivebasedprograms,andultimatelyreduced
deforestationvulnerabilityinMexico.Ouranalysisonlyconsideredforestlosses,ratherthangains,
duetodatalimitations.WhileincreasingforestgainscouldbeanimportantpieceofREDD+
programs,afocusonavoidingdeforestationshouldcapturethelargestnear‐termopportunitiesfor
reducingnetemissionsfromforests.
Thisprojectgeneratedseveralanalyticresultsaswellasdataproducts,including:
Avulnerabilitydataset:aspatiallyexplicitrasterdatasetinwhicheachcellhasavalue
indicatingtherelativeriskoffuturedeforestation,bothatthenationalandregionalscale.
‐ Afuturedeforestationprojection:aspatialdatasetprojectinglocationsoffuture
deforestationasafunctionofthevulnerabilitydatasetandpredictedratesoffuture
deforestation.
‐ Adatabaseofallvariablesinthemodelinganalyses.
‐ AdatabaseofeconometricstudiesofthedriversofdeforestationinMexico(andother countries.
Thisreportdescribesourvulnerabilityanalysisandkeyfindings,alongwiththemethods
usedtogeneratethe“soft”and“hard”deforestationprojections‐‐thevulnerabilitymapand
deforestationprojections,respectively.Ourmethodologyincludesthreedifferentand
complementaryapproaches:(i)reviewingtheexistingliterature,(ii)conductinganational
econometricanalysisandbuildinganassociatedpolicysimulationmodel,and(iii)conductinglocal‐
levelspatialanalyses.TheflowdiagraminFigure2.2.1illustratestheroleofthedifferent
projectcomponentsandassociatedinputsandoutputs.
‐
Wepresentanddiscussthemainresultsfromeachofthesethreeunderlyinganalyses.The
modelingincludedtestingthepredictivepowerofaseriesofindividual“driver”datasets,which
mayormaynotactuallycausedeforestation,butarepotentiallycorrelatedwithit.Wediscussthe
relativepredictivepowerofthedifferentdriverdatasets,withspecialfocusontheircorrelation
withthespatialdistributionofhistoricandprojecteddeforestationwitheachoftheseven
identifiedAATRs.Wealsoseektounderstandhowdeforestationmightchangecausallyinthe
futurewithchangesintheeconomicincentivesgoverningforestcoverloss.
Thefirstapproachisaliteraturereviewandmeta‐analysisofexistingstudiesof
deforestationandland‐usechangeinMexicoandelsewheregloballytoidentifytrends,
contradictions,andtoprovidecontextonland‐usedecision‐makinginMexico,aswellasinother
countries(Ferretti‐Gallon&Busch,2014).Thisreviewuncoversgapsintheliterature,informsthe
selectionofdrivervariablesforthenationalandlocalmodelingdescribedbelow,andprovides
contextforevaluatingthemodelingresults.
6
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
Thesecondapproachmodelstheimpactofdifferentdriversofland‐usechangeatthe
nationalscale,tocomplementandprovideinputstothelocalanalysesconductedusingtheIDRISI‐
SelvaLandChangeModeler(LCM).ThenationalanalysisforMexicoadaptstheapproachofthe
OpenSourceImpactsofREDD+Incentives(OSIRIS)model,whichwasdevelopedforanalyzingthe
impactofalternativeREDD+policiesinBolivia,Madagascar,PeruandIndonesia(Busch,etal.,
2012).3OurnationalanalysisforMexicofocusesonidentifyingtheimpactofonevariablethatis
arguablyofcausalimportancefordeforestation:theneteconomicreturnsperhectarefrom
convertinglandfromforesttonon‐forestlanduses.Usingthislargergeographicscaleisespecially
importanttocapturebroadervariationineconomicvariablesinordertoexplicitlymeasuretherole
ofchangingeconomicreturnsfromcompetinglanduses.Inparticular,wemodeldeforestationin
relationtovariationinestimatedgrossagriculturalrevenuesandproxiesforfixedandvariable
costsusingobservablesitecharacteristics.Theestimatedresponsivenesstotheeconomic
profitabilityofagriculturallanduseprovidesthebasisforsimulatingdeforestationunder
alternativescenarioswithdifferenteconomicincentivesforforestprotection,includingtheeffectof
potentialREDD+policies.
Thenationalsimulationyieldsanestimateddeforestationvulnerabilitymapatthenational
scaleata900mresolution.Weusethenationaleconometricmodeltoconductaseriesof
simulationsthatyieldregionalpredictionsofdeforestationunderabusiness‐as‐usual(BAU)
referencescenarioaswellasasetofhypotheticalpolicycases.Theseregionalpredictionsprovide
aninputtothelocalscaleanalysestomakepredictionsonfuturedynamicsofforestcoveratseven
AATRs.
ThethirdapproachusesLCMinordertodrawonitspredictivespatialmodelingcapacityto
morefinelydisaggregatetheregionalresultsacrossthelandscapeinthelocalstudyareas.Foreach
ofthesevenAATRs,theLCMmodelsexaminetherelationshipbetweenpotentialdrivervariables
andobservedpatternsofdeforestation.Thesemodelsgeneratea“soft”vulnerabilitymapaswellas
a“hard”predictionofdeforestationunderaseriesofhistoricalandalternativescenarios,informed
bythemoreaggregatepredictionsofthenationallevelmodel.
TheempiricalanalysesinthisstudyuseanewglobaldatasetfromtheUniversityof
Maryland,basedonLandsatsatelliteinformation,justreleasedinJanuaryofthisyear(Hansen,et
al.,2013).Toourknowledge,thisstudyisthefirsteconometricstudytoexploittherichspatial
detailandmultipletimeperiodsfromthesenewdata.Assuch,resultsfromouranalysisand
approachforMexicocouldprovideinsightsforanalyzingdeforestationinothercountriesand
regionsaswell.
3TheOpenSourceImpactsofREDD+Incentives(OSIRIS)modelisasuiteoffree,transparent,open‐source,
spreadsheet‐baseddecisionsupporttools.OSIRISgoesbeyondpredictionsofthespatialdistributionandrate
offuturedeforestationtoestimateandmaptheclimate,forestandrevenuebenefitsofalternativepolicy
decisionsforREDD+.See:http://sp10.conservation.org/osiris/Pages/overview.aspx
7
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
Thisreportisdividedinto6sections.Section3describestheliteraturereviewofdriversof
deforestationinMexico.Section4discussesthenational‐scaleeconometricanalysis.Section5
presentsthelocalmodelingfortheAATRs.Section6concludes.
8
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
Figure2.2.1.ProjectFlowchart
9
ALIANZA MÉXICO PARA LA REDUCCIÓN DE
EMISIONES POR DEFORESTACIÓN Y DEGRADACIÓN
Mexico Deforestation Vulnerability Analysis and Capacity Building.
Final Project Report
10
3. LiteratureReviewofDriversofDeforestationinMexico
3.1.Introduction
Wecompiledadatabaseofeconometricstudiesofdeforestation,including117
studiesglobally,ofwhich23studiesfocusonMexico.AppendixTableA‐1providesan
annotatedbibliographyoftheMexicostudies.Fromouranalysis,drivervariables
associatedwithlowerratesofdeforestationinMexicoincludedprotectedareas,community
forestry,andpaymentsforecosystemsservices.Drivervariablesassociatedwithhigher
ratesofdeforestationinMexicoincludeagriculturalactivity,population,soilsuitabilityand
proximitytourbanarea.Theseassociationsbetweendifferent“drivers”anddeforestation
donotnecessarilyindicatecausalrelationships.CausalstudiesofprotectedareasinMexico
havefoundtheseterritoriestobelinkedwithdecreaseddeforestation.Causalstudiesof
ejidoshavenotbeenperformed,suggestingtheneedforfurtherstudy.
3.2.Overviewofdeforestation
3.2.1.DeforestationinMexico
AllknowncategoriesofMexicanforestcover(tropicaldry,tropicalwet,and
montaneforests)havebeensubjecttodeforestation(Vaca,etal.,2012).Deforestationis
occurringmostlyinSouthernMexico,withthehighestratesoccurringinthestatesof
CampecheandQuintanaRoo.Whilerecentstudiesobserveapatternofnetdeforestationin
Mexico(Vaca,etal.,2012),recentlythenation’stotalannualdeforestationhasdecreased.
Between1990and2000,Mexicolost354,000ha/year;from2000to2005,thearea
deforestedannuallyhaddecreasedto235,000;and,from2005‐2010,Mexico’sforestloss
furtherdeclinedto155,000haperyear(FAO,2010).
ReforestationhasoccurredinsomeregionsofMexico(about178,000ha/yearfrom
1990‐2010)(FAO,2010).Thistrendhasbeenattributedtoplantedforestswithproduction
astheirprimaryfunction(FAO,2010).Reforestationthroughtreeplantationsisaresultof
increaseddemandforoilpalm,eucalyptus,andcitrusproducts.Regenerationofforest
coverisalsoseenasaresultofpassivetransition,wherefarmersabandonlandandmigrate
toareaswithbetterpaidfarmjobs.Itcanalsobearesultofactivetransition,inwhichthe
growingscarcityofforestproductsencouragegovernmentsandlandownerstoplanttrees,
i.e.sustainablecommunityforestmanagement(Vaca,etal.,2012).Thereislittleevidenceof
naturalforestregeneration.
AlthoughthedeforestationrateinMexicohasdeclined,widespreadforestcoverloss
persists.Mostdeforestationprocessesareattributedtoagriculture(mainlycoffee,maize,
beans,andsugarcane)andcattledevelopment.Otherhistoricdriversofdeforestationhave
includedhumansettlement,monocultureforestry(inSouthernMexico)andnatural
phenomena(e.g.,hurricanesandfiresintheYucatán)(Vaca,etal.,2012).Population
growth,poverty,andphysiogeographicvariablesareclaimedtobesignificantdriversof
forestlossinMexico(Barsimantov&Kendall,2012).However,literatureonthesubject
rendersconflictingconclusionsontheeffectsondeforestationofotherdrivervariables,
11
includinglandownership,subsidyprograms,roaddensityandpercapitaincome
(Barsimantov&Kendall,2012).
3.2.2.DeforestationintheYucatán
InMexico,mostoftheGulfCoastlowlandshavealreadybeendeforested,and
significantlandclearanceoccurredintheinteriorLacandonforestsofChiapas(TurnerII,et
al.,2001).TheforestsofsouthernCampecheandQuintanaRoohavebeenconsideredthe
lastfrontierinthe“westtoeastmovementoftropicallowlanddevelopment”inMexico
(TurnerII,etal.,2001).TheSouthernYucatánhasbeenidentifiedasadeforestationhot
spot(Rueda,2010).Itisconsideredtobeoneoftheworld’simportantforestedregions,
characterizedbytheCalakmulBisophereReserveandtheMesoamericanBiological
Corridor(Busch&Geoghegan,2010).Itisthereforecrucialtounderstanddriversofland‐
useandland‐coverchangeintheregion.
3.3.Overviewoflandtenure,ruralagriculturalsupport,andpaymentsfor
ecosystemsservicesinMexico
3.3.1.LandTenure
Mexicohasalonghistoryofpolicyreformsfocusedonpropertyrightsandtherole
oflandtenureonlandcoverchange(Bonilla‐Moheno,etal.,2013).Therearethreetypesof
landmanagementinMexico:Private,public(protectedareas,publicenterprises,etc.),and
communal(comunidadesagrariasandejidos).
3.3.1.1.Privatelands
Asof2011,privatelandsthatareownedand/ormanagedbycompanies,
sharecroppers,andlandlessruralpopulationrepresent37%oftheMexicanagrarian
landscape.Theseprivatelands,however,onlyencompass26%ofthecountry’sforests
(Corbera,etal.,2010).
3.3.1.2.PublicLands Publiclands,inturn,belongtofederalorregionalpublicagencies,aswellasto
publicenterprises.Theselandsrepresentjustover8%oftheagrarianlandscapeandcover
only4%offorestedareas,primarilyincludingprotectedareasandbodiesofwater
(Corbera,etal.,2010).
3.3.1.3.CommunalLands
Landsundercommonmanagementisthemostcommontypeofmanagement,
representing52%oftheMexicanagrarianlandscapeand70%oftheforests(Corbera,etal.,
2010).Therearetwomaintypesoftenurearrangements:comunidadesagrarias(agrarian
communities)andejidos.Comunidadesagrariasrefertorepatriatedindigenouslandsand
ejidosarelandsgrantedbythepostrevolutiongovernment(Barsimantov&Kendall,2012).
Botharecommunallyownedlands.NucleosAgrariosisageneraltermforejidosand
comunidadesagrariasinMexico.CarrilloandMota‐Villanueva(2006)explainthatthis
generalizationisbasedonsharedcharacteristicslikelegalstatusandlandownershipgiven
byPresidentialActorbytheHighAgrarianCourtofJustice.
12
A.HistoryofCommunalLands
A.aComunidadesagrarias
TheSpanishCrowngrantedtheselandrightstogroupsconsideredoriginalsettlers.
Thecommunitiesthatdeveloped,therefore,consistofpeoplewhohavehistorically
inhabitedaregionandsharelanguage,traditionsandgoverninginstitutions.Landholder
typesinthisformofmanagementconsistofagrariancommunitiesandindividualrights
holders(comuneros).Forestregulationisgovernedbyacommunalassemblymadeupofall
comuneros(someofwhommaybewomen).Acouncilofauthoritiesisrenewed
periodically,normallyeverythreeyears(Corbera,2010).
A.bEjidos
Ejidos,ontheotherhand,areamorespecificformoflandmanagementthan
comunidadesagrarias.Theywereestablishedwhenagroupoffamiliesclaimsrightsovera
territory,andtheparceloflandgrantedtothesegroupsremainsundercommunal
ownership.Anyrentalorlandsalesareprohibited.Landcanonlybegivenbyoneejido
landholder(ejidatario)toasingledescendant.Forestandlandforpasture(forfuelwood
collection,timberharvestingandgrazing)areusuallymanagedincommon.Forestfor
timberharvesting,inparticular,isorganizedthroughcommunitymembersandgroups,or
throughexternalconcessions.Ejidotimberconcessionsareorganizedthroughextraction
quotasandcorrespondingbenefitsaredefinedanddistributedthroughtheejidoassembly
and/orthecouncilofauthorities.
Bothcomunidadesagrariasandejidoshavemembers(avecindados)whohavebeen
givenaparceltofarmandanothertoliveon,butwhodonothaverightstobenefitsfrom
theforest.Itisestimatedthatthereareover30,000agrariancommunitiesandejidosinthe
country,occupyingover50%ofthetotalnationalterritory(PROCEDE,2010).Community
landmanagementinMexicoisoftenclaimedtohavepositiveenvironmentaland
socioeconomicoutcomes(Barsimantov&Kendall,2012).
B.Historyofcommunallandmanagement
Mexico’scurrentsystemoflandmanagementdevelopedfrompost‐revolution
governmentlandmanagementreform.AftertheMexicanRevolutioninthe1910s,Article
27ofthe1917Constitutiondeclaredthatalllandsandwatersoriginallybelongedtothe
nationandthatthenationwouldgrantprivatepropertyrightsundercertainconditions
(CamaradeDiputados,2008).Article27limitedthesizeofprivateproperties,parceled
largeprivatelandholdingsand,mostimportantly,grantedrightstoruralcommunitiesand
groupsoffamiliestoownlandtomeettheirbasicneedsortorestorecustomaryrightsheld
beforethe1800s(Corbera,etal.,2010).Theshareofcommunallandincreasedupuntilthe
early1980s.Intheearly1990s,Article27wasreformed,legalizingtheformationofjoint
venturesbetweencommunallandholdersandprivatecapital.Thisallowedcommunityland
managementmembersandejidomemberstobecomeprivateowners,andtorentandsell
landtothirdparties.Forests,however,couldnotbesubdividedandsold,excludingthem
fromprivatization(Corbera,etal.,2010).
13
C.Impactofcommunityforestryondeforestation
Amajorityofpublishedacademicstudieshaveconcludedthatcommunityforestry
doesnotinfluencedeforestation.Forinstance,Perez‐Verdinconcludedthatdeforestationis
drivenbyresource‐specificcharacteristics,suchaslocationandsoilproductivity,andnotby
ejidos’attributes(Perez‐Verdin,etal.,2009).However,a2012studyreviewedevidence
relatedtocommunityforestmanagementandforestcover,findingthatcommonproperty
andcommunityforestryaresignificantlyrelatedtoreducedratesofdeforestationand
increasedratesofforestrecoveryofconiferousforestsinMexico(Barsimantov&Kendall,
2012).Theirresultssuggestthatcommonpropertycanleadtogreaterforestconservation
whenthereisaneconomicallyvaluableassettoprotect(coniferousforests)andwhenthere
aremanagementplansinplacetoformalizetheextractionprocessandrevenue
distribution.Anotherstudyconfirmedthatcommunitylandmanagementpracticeshave
resultedinthemaintenanceofforestedlandscapeinsomeareasofMexico(Bray,etal.,
2004).Butotherstudiesconcludedthatcommunitymanagementhasmixed,ifnota
negativeeffectonforestcover(Vance&Iovanna,2006)(Alix‐Garcia,2007).Astudyin
2010demonstratedthatthecharacteristicsoftheejido,ratherthanthepresenceorabsence
anejidalsystem,determinetheimpactondeforestation:populationdensity,agricultural
productionandintensificationwithinejidosaffecteddeforestationrates(Rueda,2010).
VanceandGeoghegan(2002)observedincreasingdeforestationasejidodemographics
change,withageandpopulationdensitybeingsignificantlypositivelyrelatedto
deforestation.Geogheganetal.(2004)supportsthisconclusionandfurtherpositedthat
deforestationprimarilyfollowsagriculturalexpansionbytheejidosector,thepredominate
formoflandtenureinthesouthernYucatán.
3.3.2.RuralAgriculturalSupport
TheroleofgovernmentagriculturalsubsidiesondeforestationinMexicoismixed.
In1999,astudywasdonecontrastingtheeffectswhichtheBancodeDesarrolloRuralor
RuralDevelopmentBank(BANRURAL)creditandtechnicalassistancehaveon
deforestation.Itwasinitiallythoughtthatthistypeofaidwouldincreaseagricultural
intensification,therebyrelievingpressureonnearbyforestsforfutureconversion.The
studyrevealedthat“governmentsubsidizedcreditfailedtospuraprocessofagricultural
intensificationthatcouldhavesubstitutedforcuttingdownforests”(Deininger&Minten,
1999).Thesameauthorsproducedanotherstudyafewyearslaterthatdeterminedthat
BANRURALis,infact,associatedwithsignificantlyhigherlevelsofdeforestation,andthat
thesecreditsubsidies“seemtohaveencouragedthecuttingdownofforests”(Deininger&
Minten,2002).
Asecondstudythatsameyearconfirmedthatanotherruralsubsidyprogram,
ProgramadeApoyosDirectosalCampoorFamersDirectSupportProgram(PROCAMPO),is
alsoassociatedwithhigherlevelsofdeforestation(Vance&Geoghegan,2002).PROCAMPO
isaMexicanruralsupportprogramcreatedtoalleviatethefinancialimpactoftheNAFTA
onagriculturalworkersin1994(Klepeis&Vance,2003).Theprogramwasalso
implementedwiththeintentionofdecreasingenvironmentaldegradationthroughthe
promotionofmoreefficientlanduse,usingfundstointensifyproductionanddecrease
14
pressureonremainingforests(Klepeis&Vance,2003).Theresultingincreasein
deforestationputstheprogramatoddswithitsintent.VanceandGeoghegan(2002)
suggestpoorintegrationoflandownersintomarketsthatwouldotherwiseencourageland‐
intensivechemicalinputsasareasonforincreasedagriculturalexpansionand,
consequently,decreasedforestcover.Thesamestudyalsosuggeststhatthespecificterms
oftheprogram,whichstipulatetheareaandlocationsupportedbyPROCAMPObe
maintainedundercontinuousproduction,discouragesaforest/fallowagriculturalmethod
thatmaintainsthefertilityofsoilsused.LaterstudiesofPROCAMPOreportedmixed
results(Geoghegan,etal.,2004)orinsignificantrelationships(Chowdhury,2006).
Alternatively,anotherstudyfoundthateachhectareregisteredinPROCAMPOactually
decreasedthehazardofdeforestationby2.21%(Vance&Iovanna,2006).
AthirdcreditprogramthatmayaffectdeforestationistheProgramaNacionalde
SolidaridadorMexico’sNationalSolidarityProgram(PRONASOL).Themostrecentstudy
onPRONASOLandforestcoverchangedeterminedthattheprogram’ssubsidiesinnorthern
municipalitiesarecausingaconsiderableincreaseinforestloss,whilesubsidiesinthe
southandeastarenot(Jaimes,2010).TheeffectofMexico’sruralagriculturalsupport
programsondeforestationrequiresfurtherstudyofthetypesofruralagriculturalsubsidies
andwhereandtowhatextenttheyarerelatedtodeforestation.
3.3.3.PaymentsforEcosystemsServices
Mexicohasalreadydesignedandimplementedapaymentsforecosystemsservices
(PES)program,apaymentsforhydrologicalservicesprogram(PSAH),whichisdesignedto
incentivizetheincreasedproductionofhydrologicalservicesthroughforestconservation
(Alix‐Garcia,etal.,2012).ThroughPSAH,theMexicanfederalgovernmentpays
participatingforestownersforthebenefitsofwatershedprotectionandaquiferrechargein
areaswherecommercialforestryisnotcurrentlycompetitive(Munoz‐Pina,2008).Most
studieshavefoundthatthisapplicationofPESinMexicoreducesdeforestationtosome
extent.Anumberofstudiesonprotectedforestsrevealthatacombinationoflegalforest
protectionandfinancialincentiveshashelpedreducedeforestationinMexico(Honey‐
Roses,etal.,2011).In2011,astudyfoundthatacombinationoflegalprotectionandPES
hashelpedprotectforesthabitatforthemonarchbutterflyinMexico.Thestudyestimated
thatwithoutthejointconservationinitiative,lossesofforestwouldhavebeen3%and11%
higherinareaswithjustaloggingbanorwithdensecanopy,respectively(Honey‐Roses,et
al.,2011).In2012,inanotherstudyanalyzingPSAH,resultssuggestedPESinMexico
reduceddeforestationthatwouldhaveoccurredunderBAUscenarios,butresultresults
wereuneven.Itwasfurtherrevealedthattheprogramseemedtobemoreeffectivein
generatingavoideddeforestationwherepovertyislowerandinthesouthernandnorth‐
easternstatesofMexico(Alix‐Garcia,etal.,2012).A2008studyrevealedthatwhilePSAH
isassociatedwithreduceddeforestation,theprogram’spaymentshavebeeninareaswith
lowdeforestationrisk,suggestingthattheselectioncriteriabemodifiedtobettertarget
higherriskareas(Munoz‐Pina,2008).Thereisroomforfurtherstudyonsocio‐economics
oftheareaunderPSAHaswellasotherpotentialPESprogramdesigns.
15
3.4.Databaseregressionresults
3.4.1.AMeta‐analysisofDriversofDeforestationinMexico:Methods
Recenttechnologicalandmethodologicaladvancementshaveencouragedthe
proliferationofeconometricstudiesofdeforestationgroundedinremotelysensedevidence
offorestcoverloss.Wehavecompiledacomprehensivedatabaseof117econometric
studiesofdeforestation,including23studiesinMexico,publishedbetween1996and2014.
Tobeincludedinthedatabase,studieshadtomeetfivecriteria:(1)thedependentvariable
mustmeasureforestcoverorforestcoverchange;(2)thedependentvariablemustbe
remotelysensed;(3)thedependentvariablemusthaveresulted,inpart,from
anthropogeniccauses;(4)thearticlemustincludeatableofmultivariateregression
outputs;and,(5)thearticlemusthavebeenpublishedinapeer‐reviewedjournal.The
databaseismeanttobeasinglesourceforalleconometricstudiesofdeforestation,allowing
easyaccessandanalysisofdeforestation.Thisdatabasewascreatedtoprovidean
overviewofcurrentscientificunderstandingofforestcoverloss,toimprovepolicy
implementationaimedatdeforestationmitigation,andtoidentifygapsinscientific
evidencerequiringfurtherresearch.
Fromtheindividualstudieswecategorizeddrivervariables(n=1159)into“meta‐
variables”suchaselevation,proximitytoroad,oragriculturalactivity,ofwhich33were
includedinthestudiesofdeforestationinMexico(Table3.4.1).Asinglemeta‐variableis
thesumofallregressionresultsfromindicatorsmeasuringthesamephenomenon.For
instance,themeta‐variableElevationiscomprisedofvariableslabelled“Elevation,”“Mean
Elevation,”“Altitude”etc.WhileTable3.4.1presentsacomprehensivelistofdriver
variablescollectedinthedatabasefromstudiesinMexico,somevariableshaveyettobe
analyzedduetothecomplexityofinterpretingthevariable(e.g.SoilType).
Foreachmeta‐variable,withineachstudy,wesummedthenumberofregression
outputsormatchingoutputsthatfoundtheassociationbetweenthatmeta‐variableand
deforestationtobenegativeandsignificant,notsignificant,orpositiveandsignificant.
Theseresultswerethenorganizedintoadatabaseuponwhichwebasedouranalysis.We
termedthemeta‐variabletobeconsistentlyassociatedwithlower(orhigher)deforestation
iftheratioofpositiveandsignificantoutputstonegativeandsignificantoutputswas
statisticallysignificantlylessthan(orgreaterthan)1:1inatwo‐tailedt‐testatthe95%
confidencelevel.Wetermedthemeta‐variabletobenotconsistentlyassociatedwithlower
orhigherdeforestationiftheratioofpositiveandsignificantoutputstonegativeand
significantoutputswasnotstatisticallysignificantlydistinguishablefrom1:1.
16
Table3.4.1DriversofdeforestationinMexico,bydrivercategory
Biophysical
Elevation
(n=15)
Slope(n=16)
Wetness(n=8)
ForestArea
(n=3)
SoilSuitability
(n=6)
Proximityto
Clearing(n=9)
Proximityto
Water(n=3)
Built
Infrastructure
Proximityto
Road(n=13)
Proximityto
UrbanArea
(n=12)
Agriculture,Pasture,
andWorkingForests
AgriculturalActivity
(n=9)
ProximitytoAgriculture
(n=8)
AgriculturalPrices(n=4)
EconomicActivity(n=2)
LivestockActivity(n=2)
TimberActivity(n=1)
TimberPrice(n=1)
UseofFuelwood(n=1)
Demographics,
Poverty,and
Income
Population(n=10)
Poverty(n=14)
Education(n=8)
Indigenous
Population(n=8)
Age(n=1)
Presenceof
Females:(n=1)
PropertySize(n=7)
RuralIncome
Support(n=8)
Off‐Farm
Employment(n=3)
Land
Management
TenureSecurity
(n=6)
ProtectedAreas
(n=6)
PlotSize(n=4)
LandUse(n=4)
Logging
Activities(n=3)
PES(n=2)
Community
Forestry/Ejidos
(n=15)
Note:“n”indicatesthenumberofstudiesthathaveanalyzedthemeta‐variableinrelationto
deforestationinMexico,outofatotal23studies.Wecategorizedeveryregressionresultreportedin
theincludedstudiesintooneofthreecategories.Regressionresultsshowinganegativeand
significantrelationshipbetweenadrivervariableanddeforestationwerecodedas“‐“;regression
resultsshowingapositiveandsignificantrelationshipbetweenadrivervariableanddeforestation
werecodedas“+“;regressionresultsshowingnosignificantrelationshipbetweenadrivervariable
anddeforestationwerecodedas“n.s.“
3.4.2.ResultsforMexicoandSEsub‐regions
Theresultsforhoweachmeta‐variableisassociatedwithdeforestationacross
statisticalstudiesofdeforestation,areshowninFigures3.4.1and3.4.2attheendofthis
sectionandFigureA‐1intheAppendix.Figure3.4.1presentsthedatabaseresultsforall
studiesfocusedonMexico.InMexico,variablesmostassociatedwithdecreasesin
deforestation,includeprotectedarea,propertysize,elevation,communityforestry,and
paymentsforecosystemsservices(PES).Therearesomepredictableresults:thatprotected
areasandPESareassociatedwithdecreaseddeforestationisnotsurprising.Forestsin
areasofhigherelevationmaywellbemoreremoteandhavemorelimitedaccess.That
increasedpropertysizeisassociatedwithlowerdeforestationcouldreflectthatbigger
propertiesimplyfewerlandusers,andconsequentlyreducedcompetitionforforest
resources.
Variablesassociatedwithincreaseddeforestationincludeproximitytoagriculture,
population,agriculturalactivityandsoilsuitability.Again,theserelationshipsareprobably
notsurprising:deforestationinMexicooccurswhereeconomicreturnstoagricultureare
higher(asproxiedbyproximitytoclearedlandandagriculturalactivity)andwhere
biophysicalconditionsarefavourable(asindicatedbysoilsuitability).Populationisalso
generallyassociatedwithincreaseddeforestation,asitsuggestsincreasedcompetitionfor
forestresources.
17
Figure3.4.1DriversofDeforestationinMexico:ResultsofMeta‐Analysis
Note:ThisgraphpresentsregressionresultsfromstudiesondeforestationinMexico.Resultsare
orderedbyratioofnegativetopositiveassociationwithdeforestation.
Mostvariablesthatarenotconsistentlysignificantareperhapsalsonotsurprising.
Asexpected,resultsforruralincomesupportaremixed.Surprisingly,however,community
forestryismoreconsistentlyassociatedwithlessdeforestation,whereastheeffectofejidos
ondeforestationismixed.Weseparatedvariablesreferringspecificallytoejidosandthose
referringtothebroadertermofcommunityforestry.Thisdiscrepancysuggestsmorestudy
isneededofthedifferencesbetweenvariouscommunitylandtenuresinMexicoandtheir
respectiverelationshipswithdeforestationrates.Alsosurprising,variablesindicating
indigenousterritoryarenotsignificantlyrelatedtodeforestation,eitherpositiveor
negative,inMexico.Inourglobalstudywefoundindigenouslandtenureiscommonly
associatedwithdecreaseddeforestation(Ferretti‐Gallon&Busch,2014).
18
FigureA‐1intheAppendixcomparesresultsontherelationshipsbetweenthe
variablesanddeforestationatthegloballevelandinMexico.Duetospacelimitations,the
figureonlyincludesthe15topvariablesthathavebeenmostincludedinregression
analysesattheMexicolevel.Still,thefiguresuggeststhatvariablesaffectingdeforestation
aregenerallythesameinMexicoasatthegloballevel.Protectedareaextentandelevation
arebothassociatedwithdecreasedratesofdeforestationandarerobustatbothlevelsof
study.Ontheotherhand,globally,communalforestmanagementisassociatedwith
increaseddeforestation,whileattheMexicolevel,thecommunityforestry(includingboth
ejidosandothervariablesrelatedtocommunallandownership)isassociatedwithlower
deforestation.Similarly,ruralincomesupportisassociatedwithincreasesindeforestation
atthegloballevel,buttheresultsaremoremixedattheMexicolevel.Finally,attheglobal
level,povertyisassociatedwithlowerdeforestation,whileinMexicoincreasedpoverty
appearstobeassociatedwithhigherdeforestation.
Figure3.4.2comparesresultsdisaggregatedfromtheMexicoleveltotheYucatán
Peninsula(includingtheYucatán,QuintanaRoo,andCampeche,butexcludingTabasco).
Duetospacelimitations,thegraphagainonlyincludesthe15topvariablesthathavebeen
mostregressedattheYucatánlevel.Variablesassociatedwithlessdeforestation(property
sizeandelevation)andvariablesassociatedwithmoredeforestation(population,proximity
toagricultureandpopulation)arerobustatthislevelofdisaggregation.Notably,poverty
againhasaninconsistentassociationwithdeforestation.Atthenationallevel,poverty
appearslinkedtoincreasesindeforestation,whileintheYucatánPeninsulapovertyis
associatedwithdecreaseddeforestation.Asimilarinconsistencyisnotedwithindigenous
populations.WhileatthenationallevelIndigenousterritoryisassociatedwithdecreased
deforestation,thesamevariableisassociatedwithincreaseddeforestationattheYucatan
Peninsulalevel.TheseinconsistenciesperhapssupportthewidelyheldviewthatMexico’s
landscapeandtherelateddriversofdeforestationvarygreatlybyregion.
Itisimportanttoemphasizethedistinctionbetweencorrelation,orassociation,and
causation.Toprovideonewell‐knownexample,ratesofdeforestationmightbelower
withinprotectedareasbecauseprotectedareasarepreventingdeforestationfrom
occurring(causality).Thisrelationshipmightalsobebecauseareasthathavelowratesof
deforestationforotherreasonssuchasgeographicremotenesshavegreaterintact
biodiversity,whichledtoprotectedareasbeingdesignatedinthoselocations(anexample
ofreversecausality).Disentanglingtheseeffectsrequiresspecializedtechniquessuchas
matchingmethods,whichhavebeenperformedinMexicoforprotectedareasandpayments
forecosystemservices(Honey‐Roses2011),butnotyetforejidos,suggestinganavenuefor
furtheranalysis.
19
Figure3.4.2.DriversofDeforestationintheYucatánPeninsulaasComparedtothe
RestofMexico:ResultsofMeta‐Analysis
Note:ThisgraphdisplaysregressionresultsfromstudiesfocusedontheYucatánPeninsula
(includingCampeche,QuintanaRooandYucatán)ascomparedtoresultsfromthestudiesfocusedon
therestofMexico.Foreachmeta‐variable,twosetsofresultsarereported:thefirstsetrepresents
resultsfortheYucatánPeninsulainlightercolors,whilethesecondsetrepresentsresultsforMexico
indarkercolors.Resultspermeta‐variableareorderedbyratioofaveragenegativetoaverage
positiveassociationwithdeforestation.
20
4. Analysisofdeforestationatnationallevel/OSIRIS
4.1.Introduction
WeconductedaneconometricanalysisofdeforestationinMexicoatthenational
scaleinordertocalibrateasimulationmodeltoexploretheimpactofalternativeeconomic
andpolicyscenarios.Inparticular,weanalyzeddetailedspatially‐explicitdataonannual
forestcoverlossesacrossallofMexicoover2000‐2012.Oureconometricanalysisisbased
ontheideathatlandowners4willchoose,fromasetofpotentiallanduses,theoptionthat
bringsthehighestexpecteddiscountedreturnsThegoalistoexplicitlycapturethe
influenceoftheeconomicnetbenefitsfromconvertinglandfromforesttonon‐forestuses
forthepurposesofcalibratingapolicy‐simulationmodelthatcan,forexample,analyzethe
impactofdifferentREDD+policystructures,orotherpotentialpaymentsforecosystem
services.
Thenationalmodelservesto1)measuretheimpactofdifferenthistoricaldriversof
land‐usechange2)generateaspatialdistributionofprobabilityoffuturedeforestation
underalternativepolicyandmarketscenarios,3)helptoidentifycost‐effectivemitigation
opportunitiesandestimatetheopportunitycostsofabatingcarbonemissionsfrom
deforestation,and4)provideabasisforexaminingpolicydesignelementssoastocreate
economicincentivesfortheimplementationofREDD+inMexico.Inparticular,resultsfrom
aneconometricanalysisservetocalibratethesimulationandestimationonthedistribution
andtotalrateofdeforestationacrossMexicounderasetofeconomicandpolicyscenarios
thataltertheeconomiccalculusforlandconversion,lookingretrospectivelyover2000‐12
aswellasoutintothefutureoverthenext10years.Thenationalmodelpredictssite‐level
deforestationbasedonfittedvaluesfromtheeconometricmodel,estimatedusingobserved
deforestation.Inparticular,wemodeldeforestationinrelationtovariationinestimated
grossagriculturalrevenuesandproxiesforfixedandvariablecostsusingobservablesite
characteristics.Theresultsfromthesimulationprovideregionaldeforestationratesasan
inputtotheLCMmodelingofthesevenAATRs.
4.2.EmpiricalModel
4.2.1.EconometricSpecification
Severalchallengesariseindevelopinganempiricallytractablespecificationto
identifytheroleofeconomicreturnsindrivingdeforestationinMexico.Oureconometric
approachfocusesonaddressingtwomainsetsofissues.Thefirstsetofissuesrelatestothe
structureofourdependentvariable,whichisanaggregationofthenativedataatthe30m
cellresolution.Theaggregationintroducesthechallengeofmodelingarangeofpotential
changesinforestareawithinalargergridcell,wherethepotentialmagnitudeofchangesis
4InMexico,approximately70%offorestlandhasacommunalformofownership(Corbera,etal.,
2010).Therefore,forouranalysisbothprivateindividualsandcommunitiesaretherelevantland
ownersormanagers.
21
linkedtotheamountofforestareawithineachgrid.Thesecondsetofissuesrelatestothe
factthatweonlyhaveimperfectobservationsofeconomicreturnsforourunitsof
observation,asmentionedabove.Afulldiscussionofournationalmodel,econometric
approach,data,andestimationresultsareprovidedinAppendixI.
4.2.1.1.Relationshipofdeforestationtoavailableforestareawithina900mgridcell
Ourunderlyingdatasourcefordeforestation,ourdependentvariableinterest,
providesbinaryinformationonthepresenceornon‐presenceofforestsatthe30mcelllevel
foreachyearbetween2000and2012,providingatotalof11observedannualchanges
(Hansen,etal.,2013).Whileweconductthelocalscaleanalysesatthismostdetailedlevel
ofresolution,ananalysisatthislevelofdetailisnotcomputationallytractableforallof
Mexicoasthiswouldinvolveover1billionpointsperyearoralmost13billiondatapoints
acrossall11observedyearlychangeperiods.Tomakethenationalanalysis
computationallyfeasible,weaggregateour30mx30mcellsintolarger900mx900mcells,
eachofwhichcontain900potentiallyforestedsmallercellsatthe30mresolution.This
procedurereducesthesizeofthedatasettoabout1.39millionobservationsannually,after
eliminatingany900mgridcellsnotcontaininganyofthesmaller30mforestedcellsinthe
year2000.5Atthisscale,ourpreferredspecificationstilltookabout24hourstorunonour
mostpowerfulcomputerwith24GBofRAM.
Ourconstructeddependentvariableisthustheannualchangeinforestcoverfrom
2000through2012oneach900mcellcontainingforests,spanningallthecontinentalland
areaofMexico(i.e.,islandswereexcluded).Ourunitsofanalysisthusmeasure900mx
900mor810,000m2(equivalentto81haor0.81km2).Werestrictattentionto900mcells
thatcontainatleastoneforested30mcell.Thechangewithineachoftheseunitsis
measuredintermsofthenumberofconstituent30mcellsthatareforestedatthestartof
theyearbutthenchangefromforesttonon‐forestcoverovertheyear.Whileweassignthe
sameexplanatoryvariablestoallthesmaller30mcellswithineachofour900munits,we
thusmodelchangesin30mcellincrements.Thesechangesmightrepresentdecisionsby
oneormorelandownerswithineach900mcell.Wedonothavecomparableannualdata
forpossibleforestgainsonthesecells,soonlyconsiderforestlossesinourmodel.6Thus,if
a900mcelllosesallofitsforestcoverinaparticularyear,thatcelldoesnotenterintoour
econometricanalysisinanysubsequentyears.
5Givenavailabledata,weonlyexaminelossesofforestcoverinareasthatwereforestedin2000.
Thusourdeforestationanalysiscannotconsiderdeforestationonareasthatwerenotforestedin
2000butcouldhavesubsequentlygainedandlostforestbetween2000and2012.Thisis
appropriategivenourfocusontheREDD+policyandthegreatercarbonandbiodiversityvalues
associatedwithmorematureforests,ratherthanrecentlyregeneratingforests.
6While(Hansen,etal.,2013)doprovidedataoncumulativeforestgainsfrom2000to2012,an
analysisofthesedatawouldhaverequiredaseparateanalysisandwasbeyondthescopeofthe
currentstudy.
22
Thestructureofourdependentvariableraisesseveralissues.Thefirstissueisthat
ourdatahasa“count”structure,asforestareaandchangesinareaaremeasuredindiscrete
units,rangingfrom0upto900,themaximumnumberof30mcellswithinalarger900m
gridcell.Giventhiscountstructure,oureconometricestimationmethodisaPoissonquasi‐
maximumlikelihoodestimator(QMLE)whichisconsistentwithestimatingacountvariable
generatedbyindependent,binarydecisionsatthe30mcellresolution(Wooldridge,2002).
Forrobustness,wealsoconducttheanalysisusinganegativebinomialmodel,which
modifiesthePoissonregressionmodelwithamultiplicativerandomeffecttorepresent
unobservedheterogeneity.Thisisawaytoaddresspotential“over‐dispersion,”whichisa
commonsituationinanalysesofcountdata,wheretheobservedvarianceofthedependent
variableexceedsthevarianceofthetheoreticalmodel,indicatingthemodelisnotagood
representationoftheunderlyingphenomenon.
Thereisanotherimportantissuetoconsiderwhenestimatingthemagnitudeof
changesinforestareawithinarelativelysmallfixedgeographicboundary:theamountof
deforestationoveragivenperiodiscloselylinkedtotheamountofforestavailabletobe
deforestedwithineachcellatthebeginningoftheperiod.Oneissueisthatthereisasimple
physicalconstraint.Theamountofforestthatcanbelostinanygivenyearislimitedbythe
availabilityofforestwithinthegridcell.Givenourdatasetwithoutforestgains,moreforest
cannotbelostoverayearthanexistsatthestartoftheyear.Rather,whendeforestation
progressesovertime,theavailableforestdeclinesand,insomecases,iscompletely
exhaustedwithina900mgridcell.
Althoughthestartingforestcoversetsaphysicallimitonthepotential
deforestationwithineach900mcell,therearealsoeconomicfactorsatwork.Thedifficulty
ofaccessinganddeforestinga30mforestcellislikelytobegreaterthefartherawaythat
cellisfromnon‐forestareas,includingpreviouslyforestedlandthathasalreadybeen
cleared,givengreatercostsintermsoftraveltimeandefforttotransportpeopleand
machinerythroughforestsascomparedtomoreopenareas.Asaresult,asacellis
progressivelydeforested,moreandmoreofthecell’sforestedareasbecomeaccessibleand
easier(lowercost)tocutdown.Thus,generallyspeaking,thecostsofconvertingahectare
offorestwithina900mcellarelikelytobeinverselyrelatedtothetotalamountofforest
areainthecell.Thisignores,forthetimebeing,thedispositionofthesurroundingcellsas
wellasdifferencesinthespatialconfigurationoftheforestareaatthe30mresolution
withinthe900mcell.
Anothereconomicconsiderationisthefactthatforestlosswithina900mgridcellis
notlikelytobedistributedinacompletelyrandommanner.Peopleshouldhavean
incentivetopreferentiallydeforestthoseareasyieldingahighernetreturn,eitherbecause
ofhighernetrevenuesorbecauseoflowercostsofconversion.Thus,onewouldexpect
peopletotendtofirstcutthoseareasthataremosteasilyaccessibleorbestsuitedfor
agriculture.Asaresult,thefactthatwhileacertainshareoftheforesthasbeencleared,
anothershare(oneminusthedeforestedshare)stillremainsinforestcovermayconvey
certaininformationabouttherelativeprofitabilityofconvertingthoseremainingforests.
Forexample,iffivepercentoftheoriginalforestextent(e.g.45outof900possible30m
23
cells)remainsstanding,whiletheotherninety‐fivepercenthasbeencutdown,thismay
indicatethatthelastfivepercentisrelativelydifficultorotherwiseunprofitabletoconvert.
Thismayalsoprovidesomeinformationregardingthelikelydegradationandpotential
timbervalueoftheremainingforestcover.
Wetakethesedynamicsintoaccountinourmodelbydirectlycontrollingforthe
startingforestareaineach900mgridcell.Inparticular,westratifythesampleinto20
startingforestareacategories,withthebinschosentocontainroughlysimilarnumbersof
900mgridcells(giventhattheseobservationsareourunitofanalysis).Thisincludesabin
forcellswith100%forestcover(themaximum900countofforested30mcells).Wethen
includedummyvariablesforeachofthesestartingforestareacategoriesaswellas
additionalmultiplicativetermsthatcapturetheinteractionsbetweenthisinitialsetof
dummyvariableandeachofourkeyexplanatoryvariablesintheregression.Thisallowsus
toestimatehoweachofthesedifferentvariablesaffectthelikelihoodandscaleof
deforestationwithinagridcell,dependingonthestartingareaoftheforest.Inthisway,we
cancaptureboththephysicalconstraintsimposedbythedifferentavailablequantitiesof
forestaswellasthedifferenteconomicdynamicsofforestclearingatdifferentstagesof
deforestationwithina900mcell.
Untilnow,thediscussionhasfocusedonhowdeforestationwithina900mcell
dependsontheextentofforestclearancewithinthegridcellitself.Thesurroundingarea
outsidethecellshouldmatterbothintermsofmakingthecellmoreorlessaccessibleand
thusincreasingordecreasingthecostsofconversion,asdiscussedearlier.Wecontrolfor
thesurroundinglandscapebycalculatingameasureoftheaveragedistanceofagridcellto
allofthenon‐forest30mcellsinthesurroundingarea,withina2.5kmradius.Weusea
“kerneldensity”tointerpolatetheinfluenceofthenon‐forestareaoverspace,assuming
decreasing“gravity”oftheseareasasdistanceincreases,uptothechosen2.5kmradius,at
whichpointtheinfluenceofnon‐forestareaisconsideredzero.
4.2.1.2.Observedandunobservedcomponentsofnetreturnsfromlandconversion
Theprincipalchallengeindevelopingamodelforempiricalestimationisthatwe
onlyhavepartialinformationonthepotentialnetreturnsthatlandownerscouldobtain
fromthemostprofitablenon‐forestlanduse.Weproxyforsomedifferencesinthecostsof
conversionandheterogeneousqualityofagriculturallandwithinagridcellbyaccounting
forthestartingforestareaonitsownaswellasininteractionwithourkeyexplanatory
variables.Ourmainexplanatoryvariableofinterestisanestimateofthepotential
economicreturnsperhectarefromcropproduction,whichweconsiderasaproxyforthe
potentialreturnsfromconvertingland.Wedonothavedataonthecostsofproducing
cropsintermsoflabor,fertilizer,chemicals,andanyotherinputsnordowehavedataon
thecostsoftransportinganyproductstothemarket.Wealsodonothavedataonthe(one‐
time)costsofconversion(aswellasanypotentialone‐timebenefitsofconversionsuch
salesoftimber).Bothfixedandvariablecostsaswellasrevenueswilldeterminethe
economicrationaleforconvertingforests.
24
Toaccountforthesedifferentcosts,ourapproachistointroduceadditionalcontrol
variablesatthelevelofthe900mcellthatweexpectwillbecorrelatedwithproductionand
conversioncosts.Alltime‐varyingexplanatoryvariablesarelaggedoneyearsoasnottobe
contemporaneouswiththedependentvariable.Thestartingforestareacategories,
describedabove,provideoneproxyforpotentialconversioncostsaswellaspotential
differencesinagriculturalreturnswithinthegridcell.Aswiththestartingforest
categories,eachoftheothercontrolvariablesinourmodelisincludedindependentlyandin
interactionwithourmeasureofpotentialrevenuesforeachgridcell.Whenthesevariables
areincludedindependently,theestimatedparametersontheseadditionalvariableswill
adjusttheinterceptinthemodel,capturingpotentialone‐timeconversionorotherfixed
costs(orbenefits).Whenthevariablesareincludedininteractionswiththeagricultural
revenues,theestimatedeconometricparameterswillscaletheresponsetotheestimated
economicreturnsbasedontheproxiesforadditionalcostfactors.
OurprincipalvariablesarelistedinTable4.2.1.Whilethesevariableshelptoadjust
thefixedcostsandtoscaletheeffectsoftheagriculturalreturns,theremaystillbe
significantunobservedfactorsaffectingeconomicprofitabilityoflandconversion.Asa
result,giventhespecificinterestoftheMREDDprogramintheYucatánandSouthern
regions,wealsointroducedregionaldummyvariables,singlyandmultiplicatively(i.e.in
interaction)withagriculturalreturns,toaccountforotherfactors,suchasgovernment
policies,thatmayaffectagriculturalprofitabilityatthebroadregionallevel.
Table4.2.1.Principalexplanatoryvariablesusedinnationalregressions(900mcell)
Units
VariationoverSpace
Variationover
Time
MXN$/ha
Yes
Yes
Startingforestareacategory
0/1
Yes
Yes
Non‐forestinfluence
km2
Yes
Yes
Urbaninfluence
km2
Yes
No
Protectedareaextent
m2
Yes
Yes
Ejidoareaextent
m2
Yes
No
Comunidadesareaextent
m2
Yes
No
Slope
%
Yes
No
Lat/long
Yes
No
Variable
PotentialCropRevenue
Spatialtrendsurface
WecompileddetailedinformationonPROCAMPOpayments.However,wedidnot
directlyincludePROCAMPOpaymentsinoureconometricmodelbecausereceiptof
paymentsfromPROCAMPO(andothergovernmentprograms)isnotrandom.These
paymentsareafixedamountperhectarebasedonthesizeoffarms,andpaymentsare
concentratedinejidosandagrariancommunityareas.Asaresult,theconstanttermsinour
modelandthevariableonejidosandagrariancommunitylandswithinagridcellsmay
alreadycapturetheroleofthegovernmentpayments.Includingtheseexplicitlyislikelyto
25
capturecharacteristicsofthelandowners(notablyfarmsize)ratherthantheimpactofthe
paymentsthemselves.EconometricallyidentifyingtheroleofPROCAMPOandother
governmentpaymentswouldrequireadistinctempiricalstrategy,exploitingchangesinthe
programcriteria,andwasbeyondthescopeofthisstudy.Nevertheless,weareableto
simulatethepotentialroleofeliminatingagriculturalsubsidiesfromPROCAMPO,building
ontheideathattheroleoftheseprogramsisalreadycapturedinourestimatedparameters.
4.3.HistoricalSimulations
4.3.1.SimulationScenario
Weuseourestimatedmodelparameterstoconductaseriesofsimulationsto
explorealternativescenarios,lookingbackretrospectivelyoverthe2000‐2012period.In
thenextSection4.4,weconsiderforwardlookingscenariosover2014‐2024.Webegin
withanalysesthatarewithinthesampleperiodtobeasconsistentaspossiblewiththedata
usedtoestimatethemodel.Thegoalofthesescenariosistounderstandtherelativeeffect
ofdifferentvariables,aswellastoexploresomealternativepolicyscenarios.Wethen
conductaforward‐lookingsimulationtopredictdeforestationinthefutureinthenext
section(Section3.6).Weconductsixsimulationsoverourhistoricalperiodofanalysis,as
summarizedintable4.3.1.below.
Table4.3.1.Simulationscenariosoverhistoricalperiodindataset,2000‐2012
Scenarioname
Description
1)Factualsimulation
Allvariablesheldathistoricallevelsfrom2000to2012.
2)99%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenues fromconvertingforestlands
reducedby1%relativetohistoricallevelsinallyears.
3)101%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenuesfromconvertingforestlands
increasedby1%relativetohistoricallevelsinallyears.
4)90%PotentialAgricultural
ReturnsonForestLands
Potnatialagriculturalrevenuesfromconvertingforestlands
reducedby10%relativetohistoricallevelsinallyears.
5)110%PotentialAgricultural
ReturnsonForestLands
Potentialagriculturalrevenues fromconvertingforestlands
increasedby10%relativetohistoricallevelsinallyears.
6)NoPROCAMPOpaymentson
Potentialagriculturalreturnsfromconvertingforestlands
forestedlandsinejidosoragrarian withinagrariancommunityandejidosreducedbyvalueof
communities.
PROCAMPOpaymentsperprogramhectareinmunicipality
*Thesimulationsregardingchangestoagriculturalreturnsareaimedatrevealingtheestimated
sensitivityofdeforestationtochangesinthenetbenefitsfromconvertingforeststoagriculturaluses.
First,weestablishabaselineforcomparingoursimulationresultsbyconductinga
“factual”simulationusingtheactualhistoricalvaluesofallthevariablesusedinthe
estimation.Thenextfoursimulationsexaminetheimpactofourprimaryvariableof
interest,theestimatedagriculturalreturns.Thisvariableisourbestguessofthepotential
netbenefitsofconvertingforestlandstoanon‐forestuse.Theestimatedsensitivitytothis
variablewillbeusedinourmodelingtoexaminethepossibleimpactsofalternativepolicies
thatcouldchangethenetbenefitsfromconvertingforestedland.Suchchangesinthenet
26
benefitscouldcomethroughchangesintheprofitabilityofthenon‐forestuse(e.g.because
ofchangesingovernmentagriculturalsubsidiesonconvertedforestlands),orthrough
changesintherelativevalueofmaintaininglandinforestcover(e.g.becauseofdifferent
otentialincentivesforforestprotection).
Scenarios2and3explorethesensitivityofdeforestationtoourpotential
agriculturalreturnsvariableby,respectively,decreasingandincreasingestimatedpotential
agriculturalreturnsby1%relativetotheirfactualvalues.Thisprovidesanestimated
elasticityforchangesindeforestationwithrespecttochangingeconomicincentives,as
capturedbyourmodel.Thesesimulationsaregenerallymoreindicativeofthemodel
findingsforsmallerchangesinthevariablesthatarewithintherangeofthedatausedinthe
analysis.Nevertheless,inordertoseehowtheseresultsmightscalewithlargerchangesin
returns,scenarios3and4repeattheexercisewithasomewhatlargerchangeinreturns,
decreasingandincreasingestimatedagriculturalreturnsby10%relativetotheirfactual
values.
Thefourthscenariousestheestimatedparametersonagriculturalreturnsto
simulatechangesintheeconomicincentivesforconvertinglands.Scenario4isa
preliminaryexplorationofthepotentialinfluencehistoricalimpactsofthePROCAMPO
agriculturalsupportprogramundertheassumptionsthatfarmersweighingthepotential
benefitsofconvertinglandtocroplandrespondtoexpectedPROCAMPOpaymentsfromthe
governmentinthesamewayastheyrespondtoexpectedcroprevenuesreceivedfromthe
market.Inreality,farmersmayrespondtothesepotentialincomestreamsindifferent
potentialwaysgivendifferentperceptionsovertheirrelativeuncertaintyandfuture
evolution,forexample.Nevertheless,wemaintainthisassumptionasafirst
approximation.WhilewedidnotexplicitlyincludePROCAMPOpaymentsinthemodel,the
estimatedparametersimplicitlyreflecttheeffectsofthesepayments.Thus,reducing
potentialagriculturalreturnsbytheamountofthesepaymentswillreflecttheeffectof
reducingtheexpectedbenefitsfromcropproduction,takingintoaccountallofthepolicies
inplacefrom2000‐2012.
Aquestionisbywhatamounttoreducepotentialagriculturalreturnsgiventhatnot
allcroplandareaswereeligibletoreceivePROCAMPOpayments.Approximately80%of
currentlyplantedacresoverbothgrowingseasonsreceivedPROCAMPOsupportlastyear.
Between2000and2012,thesepaymentswentlargelytolandsinejidooragrarian
communitydesignations.FromouranalysisofthePROCAMPOdatafrom1999to2011,
about85%oflandsreceivingpaymentsnationallywereclearlyidentifiedasbeingwithin
ejidosoragrariancommunities,whileabout8%wereclearlyidentifiableasprivately
owned.Therelevantissue,however,isnotwhatshareofcurrentcroplandiseligiblefor
PROCAMPOpaymentsbutwhatshareofforestareasthatmightbeconvertedtocropswas
eligibletoreceivepaymentsinthepastandwouldbeeligibletoreceivepaymentsinthe
future.Theshareoflandseligibleforpaymentscouldbesignificantlyhigherinforested
areasifpotentialfarmsizesaresmallerthaninotherareas,whichmightespeciallybecase
onejdooragrariancommunitylands.Givenlackofadditionalinformation,asapreliminary
exploration,ourscenario4assumesthatPROCAMPOpaymentsonlywenttolandsinejidos
27
andagrariancommunities,andthatallnewcroplandacresinthesedesignationswere
entitledtofulllevelofpayments.Inparticular,wereducethepotentialagriculturalreturn
onejidoandagrariancommunitylandsbytheaveragePROCAMPOpaymentreceivedonthe
PROCAMPOprogramhectaresinthemunicipalityintheprioryear.7Whileitisnotthecase
thatnoforestedlandsoutsideejidosandcomunidadeswouldhavebeeneligibletoreceive
payments,itisalsolikelynotthecasethatalllandswithintheselandstypeswouldhave
receivedpayments.Wesimulateascenariowherenolandsoutsideofejidosand
comunidadesreceivedpaymentsinordertobeconservativeinnotoverstatingtheimpacts
oftheprogram.
RemovingthefullamountofPROCAMPOpaymentsperhectarerepresentsabouta
35%reductionintheestimatedpotentialagriculturalrevenuesonforestedlandsoverthe
historicalperiodforthemediangridcellintheejidosandagrariancommunities.This
scenariowilllikelyunderestimatetheeffectofPROCAMPOoutsideofcommunallandareas,
asweareassumingzeroeffectatfirstapproximation,butwilllikelysomewhat
overestimatetheprogram’seffectswithinthecommunalareasbyassumingallnew
croplandsinthosedesignationsareeligibletoreceivePROCAMPOprogrampayments,
despitethelimitsonpaymentsaccordingtothesizeoffields.
Thesesimulationsexploretheeffectsofchangingjustonevariableinthemodel,
holdingallothersconstant.Inreality,allothervariableswouldnothavebeenconstant,
mostspecificallythestartingforestarea.Forexample,ifdeforestationin2000islower
(higher)duetolower(higher)agriculturalreturns,thenstartingforestareawouldhave
beenhigher(lower)inthesubsequentyear.Wedonottakethisintoaccountinour
historicalsimulationssincethegoalisjusttoexaminethesensitivitytotheonevariable.
Nevertheless,forthepurposesofthefuturepredictions,describedinthenextsection,we
updatethestartingforestareaineachyeartoreflectthedeforestationinthepreviousyear.
4.3.2.SimulationResults
4.3.2.1.ChangesinAgriculturalReturns
ResultsfromthesimulationatthenationallevelaresummarizedinTable4.3.2
below.Wepresentresultsfromourpreferredmodel(the“negativebinomial”),butinclude
resultsfromouralternativemodel(the“poisson”withoutfixedeffects)intheAppendix.8
7Whena900mcellwasonlypartiallyincommunallandownership,weestimatedaweighted
averageofthePROCAMPOpaymentassumingtheejidoandagrariancommunityportionswere
eligibleforthefullpayment,whiletheremainderwasnot.
8Forthepurposesofevaluatingchangesinresponsetoparticularvariables,wepreferthenegative
binomialspecificationasPearsontestindicatesthedataarenotagoodfittothepoissonmodel,even
thoughthelatterhasabetterfittothehistoricaldata.Wereportresultswithbothmodelsfor
comparison.Weonlyreportresultsforthepoissonmodelwithoutfixedeffectsasweareunableto
conductsimulationswiththe“fixedeffects”modelgiventhatwewereonlyabletoestimate
“conditional”fixedeffectsmodel,whichdoesnotactuallyestimatethefixedeffectsforeachofthe
900mcells.Estimatesoftheseeffectsarenecessarytomakeabsolutepredictionsofthedependent
28
Ouralternativemodel(the“poisson”modelwithoutfixedeffects)replicatestheobserved
quantityofdeforestationpreciselyatthenationalaswellasregionallevels.Ourpreferred
modelhasasomewhatlessprecisefit,overestimatingnationaldeforestationoverthe2000‐
2012periodbyabout120thousandhectaresor6.8%,withapredictedtotalforestlossof
1.88millionhectaresversusanobservedlossof1.76million.9Althoughthismodel
providesasomewhatlessprecisefittothedatainabsoluteterms,wefocusonresultsfrom
thismodelasitisourpreferredspecificationforestimatingrelativechangesinforestlossin
responsetochangesinparticularvariables.
Table4.3.2.NationalSimulationResults
Observed(within
sample)*
1)Factualsimulation
2)99%agricultural
returns
3)101%agricultural
returns
4)90%agricultural
returns
5)110%agricultural
returns
6)NoPROCAMPO
payments
Totalforestloss,
2000‐12
(Ha)
Differencefrom
factualsimulation
(Ha)
Differencefrom
factualsimulation
(%)
1,762,854
‐120,624
‐6.4%
1,883,478
0
0.0%
1,878,961
‐4,517
‐0.24%
1,888,360
4,882
0.26%
1,845,771
‐37,707
‐2.0%
1,946,100
62,623
3.3%
1,789,400
94,078
‐5.0%
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualdeforestationwas1,997,765haor13%higher,aswecould
notusealltheobservationsduetomissingdataforsomeofthevariables.Note:2000‐12forestlossis
throughtheendof2011butdoesnotincludedeforestationoccurringin2012.Resultsinthistable
arefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfromthe
alternative“poisson”model(withoutfixedeffects)inAppendixTableA‐10.
Attheregionallevel,thepreferredmodelcapturesthegeneraldistributionofforest
loss,byregion,aswellasareaswithinandoutsidetheAATRreferenceregions.A
comparisonoftheobservedversusmodeledforestloss(the“factualsimulation”)for
differentregionsandlandtypesisshownintables4.4.1and4.4.2.Themodelvariesinits
precisionbyregion,underestimatingdeforestationbyalmost10%intheYucatánPeninsula
(region6),byabout4%intheSouthandWest(regions5and3),by7‐8%intheNorthwest
variable.Estimatingactualfixedeffectsprovedcomputationallyimpossibleevenwithdistrict‐level
fixedeffects.
9 Forthepurposesofcomparingtotheestimatesfromourmodels,the“observed”forestlossfigure
representstheobserveddeforestationfor900mcellswithinthesampleusedforourestimation.
Actualdeforestationwas1,997,765haor13%higher,aswecouldnotusealltheobservationsdueto
missingdataforsomeofthevariables. 29
(region1)andBajioandNortheast(region2),andbyjust1%intheCenterandEast(region
4).Suchvariationsarenotsurprisinggiventhatwearepredictingregionalandsub‐
regionalforestlossesbasedonanempiricalestimationofdeforestationresponsesacross
thewholecountry,withonlyafewregion‐specificdummiestocaptureregion‐specific
particularities.
Theresultsexaminingthesensitivityofdeforestationtothepotentialnetbenefits
fromconvertingforeststocroplanduseconfirmthatgreaterexpectedpotentialagricultural
returnswereassociatedwithincreasesinannualdeforestation,asexpectedbytheory.The
simulationsbasedonourpreferredmodelindicatethata1%decreaseinpotential
agriculturalreturnsover2000‐2012wouldhavedecreasedcumulativedeforestation
nationallyoverthisperiodby0.24%.Conversely,a1%increasewouldhaveboosted
deforestationby0.26%.Thesimulationsfromthealternativemodelsuggestaverysimilar
deforestationresponse,withdeforestationdecreasing0.26%fora1%fallinagricultural
returns,andincreasing0.27%fora1%increaseinreturns(seeAppendixtableA‐10).
Resultsforthe10%changesinreturnsareroughlyproportional,butshowamore
asymmetricresponse,withforestlossesdecreasing2.0%forat10%decreasein
agriculturalreturnsandincreasingby3.3%fora10%increase.
Ourfinalsimulationsuggeststhatdecreasingcropreturnsbytheamountof
PROCAMPOsubsidiesonejidosandagrariancommunitylandswouldhavedecreased
deforestationbyabout5%.Giventhatthisrepresentsarounda35%decreaseinreturns,
thisisabitlessthanproportionaltoourfindingthata10%decreasewouldhavereduced
deforestationbyabout2%.Mostoftheestimatedreductionsfromeliminatingthe
PROCAMPOpaymentsoncommunallandcategoriesoccurintheYucatánPeninsulaand
Southregions.About46%ofthereductionsoccurintheYucatánPeninsulaandabout
25%intheSouth.
Thefindingthatdeforestationincreasesmorethanitdecreasesforanequivalent
percentincreaseanddecreaseinagriculturalreturns,respectively,isperhapssurprisingif
oneimaginesthatprogressivelymoreandmoremarginalagriculturallandisentering
production,makingitmoreandmoredifficultforlandtocomein.Inpart,thisresult
reflectsthefactthatoureconometricmodelsarenon‐linearcountdatamodels,wherethe
coefficientsarecontributionstoaratesuchthattheydonothaveasimplelinear
interpretationintermsofabsoluteimpacts.
Thesensitivitytomarginalchangesinagriculturalreturnsvariesbyregion,as
showninTable4.3.3.ThemostsensitiveregionsaretheNorthwestandBajioand
Northeast,withtheleastsensitiveregionsbeingtheCenterandEastandtheYucatan
Peninsula.Whiletheformerregionsareestimatedtorespondabout0.5‐0.6%and0.8‐
1.0%,respectively,forevery1%changeinagriculturalreturns,thelatterregionisonly
estimatedtorespondabout0.08%.Inpartthisreflectsthenatureofoursimulations,which
consideredpercentageratherthanabsolutechanges.Asaresult,areaswithlargerabsolute
levelsofreturns,experiencelargerchangesinabsolutereturns,forthesamepercentage
change.Thelargerresponseinregions1and2reflectsthefactthattheseregionshave
higherpotentialagriculturalreturnsandthuslargerabsoluteincreasesanddecreasesin
30
deforestationunderthesescenarios(whichsimulatedpercentage,ratherthanabsolute
changes)and,consequently,havemorenon‐linearchangesinthedeforestationratefora
givenpercentageincreaseinnetreturns.
Theseregionalresultsshouldnotbetakentooliterallygiventhatthemodelismost
appropriatetoreflectnational‐averageresponses.However,themodelisalsopickingup
somedifferencesintheresponsivenesstodeforestationassociatedwithforestcategories.
Thelargerpercentresponseforanincreaseinreturnsinregions1and2alsoreflectsthe
factthatregionscontainmoresmallareasofforest.Breakingoutthesimulationresultsby
startingforestcategorywithineachregionindicatesthattheresponsivenessto1%changes
inagriculturalreturnsgenerallyincreasesasforestcoverdeclines.Thismightindicate
loweraccesscoststothesegridcells,makingthemmoresensitivetochangesingross
revenues.However,insomeregions,notablytheSouth,West,andYucatanPeninsula,there
isaU‐shapepattern,withthegreatestsensitivityoccurringatboththehighestandlowest
forestcategories.
Table4.3.3.RegionalSimulationResultsforSensitivitytoAgriculturalReturns
Region
TotalCountry
Factual
simulation
(scenario1)
Totalforest
loss,
2000‐12
(Ha)
1,883,478 99%agriculturalreturns
(scenario2)
Totalforest
loss,2000‐
12
(Ha)
1,878,961
Difference
fromfactual
simulation
(%)
‐0.24%
101%agriculturalreturns
(scenario3)
Total
Difference
forestloss, fromfactual
simulation
2000‐12
(%)
(Ha)
1,888,360 0.26%
Northwest 68,975 68,629 ‐0.50% 69,382 0.59% (Region 1) Bajio & Northeast 179,624 178,142 ‐0.83% 181,373 0.97% (Region 2) West 57,165 56,916 ‐0.44% 57,419 0.44% (Region 3) Center and East 247,089 246,899 ‐0.08% 247,303 0.09% (Region 4) South 456,810 455,280 ‐0.33% 458,346 0.34% (Region 5) Yucatan Peninsula 873,816 873,096 ‐0.08% 874,538 0.08% (Region 6) Note:Resultsinthistablearefromthepreferred“negativebinomial”model.2000‐12forestlossis
throughtheendof2011butdoesnotincludedeforestationoccurringin2012.
Theseresultssuggestthatrelativelysmallerpatchesofforestscouldcontribute
disproportionatelytomarginalchangesinincentives,giventhattheyalreadyaccountfora
disproportionateshareofdeforestationrelativetotheforestarea(seenationalmodeling
appendixformorediscussionofthisissue).Atthesametime,relativelymoreintactforests
insomeregionsappeartobeatapotentialeconomictippingpointfordeforestation,where
changesinnetreturnswillcausethemtobeginadeforestationprocess,producingajumpin
31
annualdeforestation,andperhapsevenmorecumulativedeforestationoverthelonger
term.
Asnotedabove,oursimulationsconsideredvariationsinonevariable,holdingall
elseconstant,includingthestartingforestarea.Tofullycapturetheeffectsonthe
dynamicsofdeforestation,wewouldalsowanttosimulatetherepercussionsof
deforestationinoneyearonforestcoveranditseffectondeforestationinthesubsequent
years.Webegintoexploretheseissuesinthenextsectionwhereweconsideraforward‐
lookingsimulationbasedonanincreaseinagriculturalreturnsaswellaspotentialcarbon
paymentsforavoideddeforestation.
4.4.Futureprojections
Weconductafuture‐orientedsimulationundera“businessasusual”scenarioas
wellasaseriesofpolicycaseswhereweintroduceahypotheticalcomprehensiveincentive
tomaintainforestcarbon.Asdiscussedfurtherbelow,avarietyofpolicyapproachescould
beusedtocapturepotentialfinancialflowsforREDD+andimplementlow‐emissions
practicesinMexico.Ourprojectionsservetoquantifyandmapthepotentialreductions
availableforfutureREDD+policyinMexico,ratherthantomodelaparticularREDD+
implementationstrategyinparticular.Thesesimulationsalsoprovideaninputforlocal
modelingfuturedeforestationatthelevelofeachofthesevenAATRs,asdiscussedin
Section4.
Thefuturesimulationsaccountfortherepercussionsofdeforestationfromone
yeartothenextbymodelingdeforestationateach900mcellandaccountingforitseffecton
startingforestcoverareaandcategoryatthestartofthesubsequentyear.Whileour
alternative(“poisson”model)couldprovidebetterpredictions,wefocusonourmainmodel
(the“negativebionomial”)whichshouldbemoreappropriateforexaminingtherelative
changesbetweentheBAUandpolicycases.Wepresentresultsfromthealternativemodel
intheAppendixforcomparisonpurposes.
Forthebusiness‐as‐usual(BAU)scenario,westartwithobservedforestcoverin
2012(thelastyearofourdatafromtheUniversityofMaryland)andthenmodelits
evolutionforeach900mcellatanannualtimestepthrough2024.Wealsostartwith
agriculturalreturnsasof2012andholdtheseconstantforthescenario.Thisinvolves
almostatriplingofmeanandmedianagriculturalreturnsrelativecomparedtothe2000‐
2012period,thoughthisvariesoverspace.Combiningthedataoveralltheyearsand900m
cells,theaveragepotentialreturnsrisefrom4,003to15,464MXN$/hawhilemedian
returnsrisefrom2,470to9,346MXN$/ha.Theincreaseinmedian(andusuallyaverage
returns)islargerintheNorthwest,BajioandNortheast,andWestregions,relativetointhe
CenterandEast,SouthandYucatanPeninsula.
Duetomissingdataonsomeofthevariables,ourestimationandhistorical
scenarioswerebasedonasub‐sampleofthedatathatcapture87%ofthehistorical
deforestationover2000‐2012.Nevertheless,thereisfewermissingdatainthelateryears
ofthedatabase.Thesampleusedforourfuturepredictionscaptured98%oftheobserved
32
deforestationin2011.Giventhatourdataisthusclosetocomplete,wedidnotmakeany
additionaladjustmentstothefutureforestlossprojectionsforthismissinginformation.
Forthepolicyscenarios,weconductaseriesofsimulationswhereweintroducea
comprehensivecarbonincentivepertonofCO2,startingatUSD$5andrisingprogressively
to$100(assuminganexchangerateofMXN$13/USD).Specifically,weconsider“prices”of
$5,$10,$20,$30,$50,$60,$70,$80,and$100pertonofCO2,soastotraceouta“marginal
cost”curvebasedonestimatedemissionsreductionsfromavoideddeforestationat
differentpricepoints.
Wesimulateaneconomicallyidealormostcomprehensiveincentivewhichcan,in
theory,beviewedasonewherealllandownerseitherreceiveasubsidyforland
preservationorpayataxforlandconversionforinstantaneouslyreleasingthecarbon
contentofallabove‐groundlivebiomass.Morepractically,onecanthinkofthisasapolicy
thatreducesthe“business‐as‐usual”agriculturalbenefits(e.g.byreducinggovernment
subsidies)andtranslatingthemintoeconomicbenefitsforlow‐emissionspracticesthat
avoiddeforestation.Thisisimplementedinoursimulationsbyreducingtheagricultural
returnsbytheamountoftheforegonecarbonrevenueifforestsweretobedeforested.We
donotmodelanypotentialshiftsor“leakage”ofdeforestationinresponsetopossible
inducedchangesinagriculturalreturnsorothereffects.Webaseouranalysisonthe
above‐groundcarbondensitydatafromWHRC/MREDD(Cartus,etal.,2014).For
simplicity,thisinitialanalysisdidnotconsiderbelow‐groundorsoilcarbonlosses.
Whilethisanalysisconsidersanotionalcarbonincentivethatcanbetranslatedinto
aparticular“price”andthoughtaboutasataxorsubsidyforeachlandownerorotherland
user,asalreadynote,theresultsdonotpresupposeaparticularREDD+policybasedon
directpaymentstolandowners,suchasatraditionalpaymentsforenvironmentalservices
(PES)program.Rather,ouranalysisservestoidentifythecost‐effectivepotential
emissionsreductions,andtheirspatialdistribution,giventhe“price”intermsofforegone
agriculturalrevenuesonthelandsnotbeingdeforested.Thisanalysisservestoquantify
andspatiallyidentifythemostcost‐effectivereductionsthatcouldbepotentiallytargeted
underavarietyofpotentialpolicyinterventionsandapproachesforpromotinglow‐
emissionsruraldevelopmentandreduceddeforestationemissionsinMexico.Moreover,
whileagriculturalproductionmightbeforegoneontheparticularlandsnotbeing
deforested,agriculturecouldbeintensifiedandexpandedonnon‐forestlandsunderalow‐
emissionsagriculturaldevelopmentstrategy.Thismeansthatagriculturalproductioncould
bemaintainedorincreasedoverallatthesametimethatexpansionofagricultureintoforest
areasisdecreased.
4.4.1.1.“Business‐as‐usual”projection
Table4.4.1showsour“businessasusual”projectionsfor2014‐2024relativetothe
observedandmodeleddeforestationinannualizedtermsduringthehistoricalperiod
(2000‐2012).ResultsarepresentednationallyaswellasbyAATRreferenceregionsand
differentlandownershipcategories.Basedontheeconomicprofitabilityofagricultureand
startingforestcoverin2012,themodelpredictsanoverall27%increaseinannual
33
deforestationinMexicooverthenexttenyears,relativetotherecentpast.Estimated
changesreportedarerelativetothemodeleddeforestation(the“factualsimulation”)for
2000‐2012.Mostofthisincreaseisduetoasignificantincreaseindeforestationinthe
SouthandYucatanPeninsularegions,whichassumesanevengreatershareofnational
deforestation,assomeotherregions(WestandCenter/Eastregions)experienceadecrease
inannualdeforestation.Thehigheragriculturalprofitsin2012relativetothehistorical
periodaccountsfortheoverallincreaseindeforestationnationwideandinthemore
forestedareas.
Despitethesignificantincreaseintheaverageandmedianagriculturalreturns
comparedtothehistoricalperiod,theoverallincreaseindeforestationissmallerthan
suggestedbyoursimulationsofsmallermarginalchangesinagriculturalreturnsinsection
2.Thisislikelyduetothefactthatwearecomparingresultsacrossawholehistorical
periodwithawiderangeinreturns,includingreturnssimilartoourprojectedonesatthe
endoftheperiod.Wearealsonowaccountingforthedecliningforestareaswithineach
gridcell,whichfurtherreducespotentialdeforestation.Theprojecteddecreaseinsome
regionsrelativetothehistoricalperiodislikelyduetosmallerremainingareasofforestin
2012relativetothehistoricalperiod.
Asnotedbefore,ourmodelsareintendedfornationalanalysisbutgenerallycapture
regionaldistributions.Map4.4.1showsthespatialdistributionofprojectedaggregate
forestlossunderthebusiness‐as‐usualscenarioforthenext10years(2014untilthestart
of2024).Themapshowsthatthegreatestamountofdeforestationisprojectedtooccur
intheSouthandYucatanPeninsularegions.Table4.4.2showshowtheseregionsarenot
onlyprojectedtocontributethemostdeforestationinabsoluteterms,butarealsoprojected
toexperiencethegreatestpercentageincreasesindeforestation,withprojected
deforestationrisingby72%intheSouthandby26%intheYucatanPeninsula.Incontrast,
deforestationincreasesby17%intheNorthwest,3%intheBajioandNortheastand
percentdecreasesintheWestandCenter/Eastregions.Ouralternativemodel(the
“poisson”)alsopredictsanincreaseinnationaldeforestationof27%,withthegreatest
proportionalincreasesoccurringintheSouthandYucatanPeninsula(AppendixtableA‐11).
Thebreakdownacrossregionsisabitdifferentinabsoluteterms,butthequalitativeresults
arestillgenerallythesame.Thisalternativemodel,whichmaybemoreprecisefor
predictivepurposes,showsrelativelysmallerincreasesintheSouthandYucatan(41and
48%,respectively)andlargerincreases(smallerdecreases)intherestofthecountry.
TherelativelygreaterprojectedincreasesindeforestationintheSouthandYucatan
comparedtotherestofthecountrycontrastwiththehistoricalsimulationresultsinTable
4.3.3formarginalchangesinreturnsofplusorminus1%.Whileotherregionsappear
moresensitivetosmallchangesinreturns,thegreatercumulativedeforestationinthe
SouthandYucataninthefutureprojectionsmaybedueinpartbythemuchlargerchanges
inreturnsbeingconsideredinthebusiness‐as‐usualprojection,whichelicitsalarger
responsefromallforestareas.Theotherpartofthestoryisthatwearenowaccountingfor
howforestareasevolveovertime.Thus,areaswithsmallinitialforestcovermightrespond
34
withalargeproportionalchangesindeforestationintheshortrunbutthenhavelittle
forestcoverremainingtocontinuehavingforestlosses.
Table4.4.1.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,by
AATRreferenceregionsandlandownershipcategory
%change
inannual
forestloss,
projected
BAUvs.
modeled
2000‐12
(%)
27%
Observed
forestloss
(insample),
2000‐12
(Ha/yr)*
TotalCountry
160,259
171,225
217,963
Changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(Ha/yr)
46,738
Non‐AATR
110,299
113,679
131,434
17,755
16%
AATRregions
49,960
57,546
86,528
28,982
50%
Mixteca
1,298
1,902
3,180
1,278
67%
SierraNorte
1,827
1,055
1,920
865
82%
SierraPucc
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%
ValledeBravo
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%
Protectedareas 4,778
6,529
11,542
5,013
77%
55,337
61,608
72,096
10,488
17%
Region/Land
Category
Otherlands
Modeled
forestloss
(factual
simulation),
2000‐12
(Ha/yr)
Business‐
as‐usual
(BAU)
forest
loss,
2014‐24
(Ha/yr)
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualnationaldeforestationwas1,997,765haor13%higherthan
thein‐sampleamountasallobservationscouldnotbeusedduetomissingdataonsomevariables.
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐11.Protectedareasarethefederallyprotected
areasconsideredinthisanalysis.2000‐12forestlossisthroughtheendof2011butdoesnotinclude
deforestationoccurringin2012.Similarly,2014‐24forestlossisthroughtheendof2023butdoes
notincludedeforestationoccurringin2024.
35
Table4.4.2.Comparisonofhistoricalchangeandfuturepredictions,2014‐2024,by
nationalregionsandAATRReferenceRegions
Observed
forestloss
(in
sample),
2000‐12
(Ha/yr)*
Region/Land
Category
Modeled
forestloss
(factual
simulation),
2000‐12
(Ha/yr)
Business‐
as‐usual
(BAU)
forestloss,
2014‐24
(Ha/yr)
Northwest(Region1)
Changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(Ha/yr)
%changein
annualforest
loss,
projected
BAUvs.
modeled
2000‐12
(%)
Total
5,751
6,270
7,187
917
15%
Non‐AATR
4,025
4,163
4,632
469
11%
1,725
2,107
2,556
449
21%
AATRregions
Bajio&Northeast(Region2)
Total
15,125
16,329
18,154
1,825
11%
Non‐AATR
15,120
16,320
18,151
1,831
11%
AATRregions
4
10
3
‐7
‐70%
West(Region3)
Total
4,969
5,197
5,267
70
1%
Non‐AATR
4,630
4,973
5,048
75
2%
AATRregions
339
224
219
‐5
‐2%
CenterandEast(Region4)
Total
22,777
22,463
15,833
‐6,630
‐30%
Non‐AATR
21,684
21,176
14,799
‐6,377
‐30%
AATRregions
1,093
1,286
1,034
‐252
‐20%
South(Region5)
Total
39,863
41,528
71,262
29,734
72%
Non‐AATR
28,535
28,688
46,408
17,720
62%
AATRregions
11,328
12,840
24,854
12,014
94%
YucatanPeninsula(Region6)
Total
71,776
79,438
100,260
20,822
26%
Non‐AATR
36,304
38,359
42,397
4,038
11%
AATRregions
35,472
41,078
57,863
16,785
41%
*This“observed”forestlossfigurerepresentstheobserveddeforestationfor900mcellswithinthe
sampleusedforourestimation.Actualnationaldeforestationwas1,997,765haor13%higherthan
thein‐sampleamountasallobservationscouldnotbeusedduetomissingdataonsomevariables.
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐12.2000‐12forestlossisthroughtheendof
36
2011butdoesnotincludedeforestationoccurringin2012.Similarly,2014‐24forestlossisthrough
theendof2023butdoesnotincludedeforestationoccurringin2024.
Themodelpredictsthatmostnewdeforestationinabsoluteterms,aswellasmost
absoluteincreasesindeforestation,willoccuronejidolands(Table4.4.2).Inpercentage
terms,however,forestlosseswithinejidosareprojectedtoincreaseby22%orlessthan
thenationalaverage.Incontrast,agrariancommunities(comunidades)areprojectedto
experiencethelargestincrease,followedbydeforestationwithinprotectedareas,with
projectedincreasesof107%and77%,respectively.Ouralternativemodelprojectssimilar
qualitativepatterns,thoughtherelativedifferencesamonglandtypesaresmaller.
Map4.4.1indicatesthattheAATRsarenotalllocatedintheareaswiththehighest
projectedfuturedeforestation.Nevertheless,asshowninTable4.4.1,overalltheAATR
referenceregionshavehigherprojecteddeforestationincreasesthanotherforestedareas
(50%versus16%forthelandsoutsidethesereferenceareas).TheAATRreferenceregions
discussedherearethoseusedinthelocalmodeling(section4),andincludethespecific
REDD+earlyactionareasites,aswellasasurrounding50kmbuffer.Giventhenational
scaleofthemodeling,resultsaremoreappropriateatlargerscalesofanalysis,dictatingour
focusonlargerversussmallerareassurroundingtheAATRs.
LookingspecificallyattheAATRreferenceregions,themodelpredictsthegreatest
increaseintheItsmoandSierraNorteregionandthesmallestincreasesintheRaramuri
andValledeBravoregions,withthelatterregionactuallyexperiencingadeclinein
deforestation.Thealternativemodelgeneratessimilarqualitativeresults,thoughit
predictsasmallerrelativeincreaseindeforestationintheChiapasAATRreferenceregion
(26%versus79%increaseinourpreferredmodel).
TheAATRreferenceregionsnotonlyhavehigherprojecteddeforestationversus
otherlandsonaggregatenationally,buttheyalsogenerallyhavehigherprojected
deforestationrelativetootherlandswithineachregion.ThecomparisonofAATRvs.non‐
AATRlandswithineachregionisshownintable4.4.2.TheAATRreferenceregions
generallyhavehigherprojectedincreasesindeforestation(orsmallerprojecteddecreases
inthecaseoftheCenterandEast),relativetootherforestedlandsinthesameregion.The
exceptionsaretheBajioandNortheast(Region2)andWest(Region3),buttheseresults
arenotindicativegiventhattheseregionscontainedtrivialamountsoflandswithinanyof
theAATRsreferenceareas.Ouralternativemodelgeneratessimilarfindings(TableA‐12).
4.4.1.2.CarbonIncentiveProjections Asanexampleofourcarbonincentiveresults,wepresentresultsforahypothetical
carbonincentiveofUSD$10/tCO2intable4.4.3.Thiscarbonincnetivetranslatesintoa
median(average)subsidy/taxofabout4700(5200)MXN$/ha,comparedtomedian
(average)agriculturalreturnsofabout9,300(15,400)MXN$/ha.Thisrepresentsa
medianreductioninagriculturalreturnsof25%,withmorethana100%reductionon
average.Underthissimulatedcarbonincentiveof$10/tC,deforestationfallsnationallyby
anestimated35%.Map4.4.2showsthespatialdistributioninthereductioninforestloss
37
underthe$10carbonincentive(relativetotheBAUcaseshowninmap4.4.1)whilemap
4.4.3showstheremainingdeforestation.Theresultsforthealternativemodelandatthe
levelofeachAATRreferenceregionareshownintheAppendixinmapsA‐15toA‐24.
Ingeneral,theregionsprojectedtohavethegreatestincreasesindeforestationover
thenextdecadearealsoestimatedtobethemostresponsivetoreducingdeforestation
underacarbonincentive.Overall,AATRreferenceregionsareestimatedtoreduce
deforestationby41%comparedtoareductionof32%fornon‐AATRlands.Theanalysis
suggestssignificantreductionsinthespecificAATRs,rangingfrom35%inRaramurito58%
inSierraNorte.AlloftheAATRdemonstrategreaterpotentialreductionsthanthenon‐
AATRregionsofthecountry.However,thegreatestpotentialreductionsoccurinthe
YucatánPeninsulaandSouthasseeninmap4.4.2.Similarly,mostoftheremaining
deforestationisdistributedintheseregions(map4.4.3).
Comunidadesandprotectedareaswerethelandtypesprojectedtohavethebiggest
proportionalincreaseinforestlossesoverthenext10yearsandarealsoestimatedtohave
thegreatestpercentdeclinesinresponsetoacarbonincentive.Inabsoluteterms,
however,thegreatesttotalestimatedreductionsoccuronejidos,aswellasprivateand
otherlandtypesapartfromcomunidadesornationalprotectedareas.
Inadditiontoconsideringchangesinforestareaasaresultofacarbonprice,we
alsoconsiderchangesincarbondioxideemissionsfromlossesinabove‐groundbiomass.
Map4.4.4showsthespatialdistributionofreducedemissionsfromabove‐groundforest
biomass,associatedwiththereducedforestlossscenarioata$10priceshowninmap4.4.2.
Estimatedemissionsreductionsforthe$10carbonincentivearealsocombinedwiththose
fromtheothercarbonincentivesimulationsandareusedtoconstructcostcurvesshownin
Figures4.4.1and4.4.2.Thesefiguresshowtheestimatedabove‐groundforestcarbon
emissionsavoidedannuallyundereachofourcarbonincentivescenarios,relativetothe
business‐as‐usualprojectionoverthe10yearsstartingin2014.
Underthebusiness‐as‐usualscenario,representedbyacarbonincentiveofzero,
averageannualCO2emissionsfromdeforestationareapproximately17milliontonsofCO2
atthenationallevel.Despitereflectingincreasesinfuturedeforestation,theseareabout
37%ofthe45.3MtCO2for2010reportedforland‐usechangeemissionsinthefifthnational
communicationstotheUnitedNationsFrameworkConventiononClimateChange
(SEMARNAT/INECC,2012).Thereareseveralpossibleexplanations.Ourestimatesare
basedonnewsourcesofinformationonbothforestloss,aswellasonabove‐groundforest
carbondensities.Also,thenumbersinthenationalcommunicationsincludeconversionof
grasslands(pastizales),whichwerenotconsideredinouranalysis.Furthermore,our
analysisonlyconsideredemissionsfromabove‐groundforestcarbonstocks,without
consideringpotentiallossesofbelow‐groundforestcarbonorsoilcarbon.Estimatesof
aboveandbelow‐groundforestcarbonstocksinMexicofromFAO(2005)andRueschand
Gibbs(2008)areapproximately95and113tonsofCperhectare.Incontrast,themeanand
medianforesthectarein2012hadanestimated23.6and21.8tonsofC/ha,respectively,
accordingtotheestimatesusedinourstudy(Cartus,etal.,2014).Thecarbondensitiesfor
thedeforestedhectaresinourprojectionsfrom2014‐2024wereabitlower,withameanof
38
21.7andmedianof19.8tC/ha.Adetailedcomparisonofthesenumberswasbeyondthe
scopeofouranalysis.
Focusingonlyontheabove‐groundcarbon,wefindthatthereisrisingpotential
nationallytoreduceemissionsatcarbonincentivesrangingfrom$5to$100,atwhichpoint
about90%oftheemissionsareavoided.Closetohalfoftheestimatedreductionsavailable
atpricesof$10/tonorbelowandmorethantwothirdsoftheestimatedreductions
availableatpricesof$20/tonorbelow.Thenationalandregionalcostcurvesarerisingat
anincreasingrate,indicatingthatitcostsmoreandmoretoavoiddeforestationonlands
withgreateragriculturalpotentials.
Whiletherearepotentialreductionsavailablefromallregionsatpricesupto$20‐
$30,thebulkofestimatedreductionsisfromtheSouthandYucatanPeninsula,which
accountforabout35%and60%ofthetotalpotentialupto$100.Reductionsfromthe
otherregionscollectivelyrisesteeplyandareexhaustedatpricesof$20and$30,atwhich
pointthecostcurvesturnvertical,withabout1milliontonsofemissionsavoidedintotal.
Thisreflectsthehigheragriculturalreturnsintheseregionsaswellassmallertotalamount
offorestlossesandcarbonemissionsthatcanbeavoided.Incontrast,thecostcurvesfor
theSouthandYucatanPeninsuladonotbegintoturnupwardssharplyuntilabout$50.At
pricesof$5,theSouthandYucatanPeninsulaaccountfor43%and50%ofthecost‐effective
potential,respectively.ThecostofreductionsintheSouthrisessomewhatfasterthanin
theYucatanPeninsula,withtheSouthrepresentingasmallershareofthecost‐effective
potentialatprogressivelyhigherprices(e.g.37%versus56%fortheYucatanatapriceof
$50).
Figure4.4.2breaksoutthecostcurvesaccordingtolandswithinandoutsideofthe
AATRreferenceregions.TheseshowthatthebroadAATRregionsonaggregatecontain
morethanhalfofthecost‐effectivepotentialreductionsinemissionsateachpricepoint,
withabout55%ofthetotalmodeledpotentialforallofMexico.Asalreadynoted,our
exercisedidnotpresupposetheimplementationofanactualcarbonpriceorpayment
system.Rather,weconsiderahypotheticalcarbonincentivesoastoestimatethemost
cost‐effectivereductionspotentialavailableforagivenreductioninforegoneagricultural
revenuesontheparticularlandsnotbeingdeforested(thoughofcourseagricultural
productionmightstillincreaseonotherlands).Thesecost‐effectivereductionscouldbe
achievedinpracticethroughavarietyofpolicyapproaches.Also,whileouranalysis
consideredanidealizedpolicycase,whichisindicativeofthepotentialforREDD+policies,
additionalanalysiswouldbeneededtoconsiderimpactsondeforestation,including
possible“leakage,”aswellasothereconomicimplicationsundermorerealisticandlikely
lesscomprehensivepolicyapproaches.
39
Table4.4.3.FuturePredictions,2014‐2024,Business‐as‐Usualand$10/tonCO2
PolicyCase,forAATRandnon‐AATRregions
Region/Land
category
TotalCountry
Business‐
as‐usual
(BAU)
forestloss,
2014‐24
(Ha/yr)
217,963
Forestloss,
2014‐24
with
$10/tCO2
(Ha/yr)
141,106
Changein
annual
forestloss,
$10/tCO2
vs.BAU
(Ha/yr)
‐76,856
%changein
annual
forestloss,
$10/tCO2
vs.BAU
(%)
‐35%
Non‐AATR
131,434
89,727
‐41,707
‐32%
AATRregions
86,528
51,379
‐35,149
‐41%
Mixteca
3,180
1,466
‐1,714
‐54%
SierraNorte
1,920
802
‐1,118
‐58%
SierraPucc
57,863
37,082
‐20,781
‐36%
Chiapas
12,847
6,324
‐6,523
‐51%
Raramuri
2,556
1,668
‐887
‐35%
ValledeBravo
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%
Protectedareas
11,542
5,804
‐5,738
‐50%
Otherlands
72,096
48,290
‐23,805
‐33%
Note:TheAATRregionsinthistablearetheAATR“referenceregions”usedinthelocalmodeling
discussedinsection4.ThereferenceregionsincludetheAATRsiteplusa50kmbuffer.Resultsin
thistablearefromthepreferred“negativebinomial”model.Forcomparison,wereportresultsfrom
thealternative“poisson”modelinAppendixTableA‐13.Protectedareasarethefederallyprotected
areasconsideredinthisanalysis.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.
40
Map4.4.1.Projected“BusinessasUsual”(BAU)ForestLoss2014‐2024
Note:Thismapshowsprojecteddeforestationatthe900m(81ha)resolutionover10yearsstarting
in2014,basedoninformationonforestcoverin2012,estimatedmodelparametersfrom2000‐12,
andholdingconstantagriculturalprofitsat2012levels.Projectionsarefromthepreferred“negative
binomial”model.Forcomparison,wereportresultsfromthealternative“poisson”modelin
AppendixMapA‐11.Greenareasindicatenolossofforestcover.Progressivelyredderareasindicate
greateramountsofforestloss.Greyareasarethosewithoutanyforestcoverin2012andhenceno
projectedforestloss.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.
41
Map4.4.2.ProjectedAvoidedForestLoss2014‐2024,with$10/tonCO2incentive
Note:Thismapshowsprojectedreductionsindeforestationatthe900m(81ha)resolutionover10
yearsstartingin2014,basedonintroducinganeconomicallyidealcomprehensivecarbonpriceof
$10t/CO2onforestcarbonlosses.Reductionsarerelativetothe“business‐as‐usual”(BAU)scenario
inMap4.4.1.Thisanalysisdoesnotconsiderpotentialpriceadjustmentsorotherpossiblesourcesof
inducedshiftsindeforestationandemissions(i.e.“leakage”).Projectionsarefromthepreferred
“negativebinomial”model.Forcomparison,wereportresultsfromthealternative“poisson”model
inAppendixMapA‐12.Whitecolorareasindicatenoreductioninforestlossasaresultofthecarbon
price.Lighttodarkyellow,followedbylighttodarkgreen,areasindicateprogressivelygreater
amountsofavoideddeforestationunderthecarbonpricerelativetotheBAUcase.Greyareasare
thosewithoutanyforestcoverin2012andhencenoprojectedreductioninforestloss.2014‐24
forestlossisthroughtheendof2023butdoesnotincludedeforestationoccurringin2024.
42
Map4.4.3.ProjectedRemainingForestLosswith$10/tonCO2incentive,2014‐2024
Note:Thismapshowstheprojectedforestlossatthe900m(81ha)resolutionover10yearsstarting
in2014thatisestimatedtoremainaftertheintroductionofthe$10t/CO2onforestcarbonlosses
(i.e.thismapshowstheremainingforestlossstartingfromtheforestlossinmap4.4.1and
subtractingouttheavoidedforestlossinmap34.4.2).Projectionsarefromthepreferred“negative
binomial”model.Forcomparison,wereportresultsfromthealternative“poisson”modelin
AppendixTableA‐13.Greenareasindicatenolossofforestcover.Progressivelyredderareasindicate
greateramountsofforestloss.Greyareasarethosewithoutanyforestcoverin2012andhenceno
projectedforestloss.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.
43
Map4.4.4.ProjectedAvoidedEmissions2014‐2024,with$10/tonCO2incentive
Note:Thismapshowsprojectedreductionsinabove‐groundcarbonlossesatthe900m(81ha)
resolutionover10yearsstartingin2014,basedonintroducinganeconomicallyidealcomprehensive
carbonpriceof$10t/CO2onforestcarbonlosses.Reductionsarerelativetotheforestlossesinthe
“business‐as‐usual”(BAU)scenarioinMap4.4.1.Thisanalysisdoesnotconsiderpotentialprice
adjustmentsorotherpossiblesourcesofinducedshiftsindeforestationandemissions(i.e.
“leakage”).Projectionsarefromthepreferred“negativebinomial”model.Forcomparison,we
reportresultsfromthealternative“poisson”modelinAppendixMapA‐14.Whitecolorareas
indicatenoreductioninforestlossesandassociatedcarbonemissionsasaresultofthecarbonprice.
Lighttodarkyellow,followedbylighttodarkgreen,areasindicateprogressivelygreateramountsof
avoideddeforestationandassociatedemissionsunderthecarbonpricerelativetotheBAUcase.
Greyareasarethosewithoutanyforestcoverin2012andhencenoprojectedreductioninforest
lossesandassociatedemissions.2014‐24forestlossisthroughtheendof2023butdoesnotinclude
deforestationoccurringin2024.
44
Figure4.4.1.EstimatedcostcurvesforCO2emissionsreductionsfromabove‐ground
forestcarbonlossesinMexico,byregion
Figure4.4.2.Estimatedcostcurvesforreducingemissionsfromabove‐groundforest
carbonlossesinMexico,byAATRandnon‐AATRregions.
45
5. LocalModelingofDeforestation
5.1.Introduction
5.1.1.Overallapproach
Thenational‐levelmodelingcapturesdriversandpossibledeforestationoutcomes
atonescalethatisonlyarguablyveryrelevanttothelocalscale.Onecouldalsoclaimthat
dynamicsatthelocalscalehavealocalcharacter,andthatamodeloftheselocaldynamics
shouldbeindependentofrelationshipsderivedfromdistantlands.Thus,weadda
componenttothisstudythatmodelsdeforestationbasedsolelyonlocaldata.Wedothis
forthesevenfocusareasselectedbyTNC.Also,whileboththenational‐andlocal‐level
analysesarebasedonspatialmodeling,thenational‐leveloneisviamodelingeconomic
incentivesthatvarywithspatialpatternsofopportunitycost.Atthelocallevel,
opportunitycostislessvariableandinformationisscarcer.Forthesereasons,wetakea
differentapproachtospatialmodelinginthelocalcasestudies.
Theoverallapproachwetakeistofollowthefundamentalstepsfoundinthemost‐
widelyusedmethodologiesforestimatingreferenceemissionslevels(RELs)forREDD+
initiativesapprovedbytheVoluntaryCarbonStandardsgroup(VCS).Howeveritis
importanttonotethatthiswasnotareferencelevelsettingexercise.Wedonotconduct
themethodtothelevelofdetailthatwouldbeexpectedforaVCSProjectDescription(PD)
document,sincethatwouldrequirelocalfielddataonbiomassandlocalimprovementof
GISdatausedinmodels.NonethelesswefollowtheoveralllogicoftheVCSmethodologies
andtheirfundamentalstepsinspatialmodeling.
5.1.2.Definitionofextents
First,somespatialextentsaredefined.Thisincludesthesiteitself,whichineachof
thesevencasesisanexistingprotectedarea(PA).EachPAisoneofMexico’sREDD+early
actionsites(ÁreasdeÁccionTempranaREDD+;AATR).WeusedtheofficialPAboundary
filesprovidedtousbyTNC.Secondisthedefinitionofareferenceareaformodelingland
useinsideandaroundeachsite.
Eachreferenceareawasdefinedbyfirstcreatinga50kmbufferaroundtheAATR
site.Thisbufferwascombinedwithmunicipalityboundaries,andtheentireextentofany
municipalitythatintersectedthebufferwasincludedinthereferenceregion.
5.2.DataandMethods
5.2.1.Deforestationdata
Withineachsite’sreferencearea,weobtaineddataonforestcoverand
deforestationfrom2000to2012.Weusedthesamedatathatwereusedforthenational‐
levelanalysesfromthelatestUniversityofMaryland(UMD)assessment.Correspondingly,
thesedataarebasedontheanalysisofLandsatimagesandhaveaspatialresolutionof
30m.However,wedidnotconductanyspatialdegradation(coarseningofspatial
resolution),aswasdoneforthenational‐levelanalyses,sinceeachreferenceisnot
prohibitivelylargetoconductanalysesatfullresolution.
46
TheUMDsourcedatacanbeseenascomposedoftwoparts.Firstisamapoftree‐
coverpercentforeach30mcellinyear2000.Treecoverisnotthesameasforestcover.
Onecanassumethattreecoveraspresentedinthisproductisrelatedtocrowncoveras
estimatedinthefieldandusedinnationaldefinitions.However,thetwoconceptsarenot
theexactsame,andsuchanassumptioncanleadtoproblems.Forboththenationaland
local‐levelmodelingweusedthisassumptionforsimplicity.
Thenationaldefinitionofforesthas,ascriteria,aminimumcrown‐coverof25
percent.Weappliedthisvalueasathresholdtothepercenttree‐covermapfromUMDto
createamapofforestin2000.Thisleadstoagenerousestimateofthedistributionof
forestinthemodelingareas.Webelievethatmostsecondaryforestfallowsandshrub
fallowsassociatedwithrotationalagricultureorrecently‐abandonedfarmlandisincluded
intheestimationofforestextentin2000.Themappedandmodeledpatternsofforest
coveranddeforestationlikelyincludesiteswithsignificanttreecoverandareasof
clearanceoftreecoverthatarenotmatureforestortheclearanceof“mature”forest.We
believethatplantationsandselectively‐loggedforestarealsoincludedintheforestclass.
Thusthisdefinitionshouldbekeptinmindwheninterpretingresultsofthisstudy.
ThesecondpartoftheUMDdatafocusesonestimatesofthelocationsoflossof
treecoverforeachyearfrom2001to2012.Webelievethattheseshouldberobustdata,
sincethetemporal‐spectralsignalofsuchclearingeventsisstrong,andthemethodsof
UMDmaximizethepotentialfortheirdetectionbyminingtheentireLandsatarchiveover
thestudyperiodandemployaneffectivedecision‐treestatisticalapproach.Thus,we
expectthatthemajorityofdeforestationiscaptured,aswellasmuchoftheotherformsof
clearanceoftreecover,becauseofthegenerousdefinitionofforestextentin2000,as
notedabove.
Incontrasttothenationalanalysisthatconsideredforestlossesonanannualbasis,
forthelocalanalyses,theannuallossdataweregroupedtocreatemapsofforestlossover
twotimeperiods:2000to2006and2007to2012.Wethencombinedthemapsoflossfor
thesetwoperiodswiththatofderivedforestextentin2000tocreateathree‐dateproduct.
Inanefforttolimittheeffectofsmall‐scalechangesintreecoverthatmightnottruly
representforestlosses,wefilteredtheoutputtominimizeverysmallartifactsandtoseta
minimumpatchsizeforboththebaselineforestdistributionandpatternsofloss.First,a
three‐by‐threecellmajorityfilterwasappliedtothemergedproduct.Second,wefiltered
theoutputusingaone‐hectaresieve.Thiseliminatesanypatchofforestorforestlossthat
issmallerthanonehectareandreplacesthecellswiththedominantclassborderingthe
eliminatedpatchofcells.10
10Webelievedthisfilteringwasprudentinthelocalanalyses,butnotnecessaryinthenational
analysis,asthelattercontrolledforstartingforestareaintheeconometricprocedure.Inaddition,
themuchlargeramountofdatausedforthenationalstudyreducesthepotentialinfluenceof
spuriousforestlossobservations.
47
5.2.2.Otherdata
WeobtainedasuiteofdatafromTNCandpartnerstoexplorethespatialrelationships
betweenpossible“drivers”anddeforestation.Thedataaremoreaccuratelydescribedas
geographicalparametersratherthandrivers.Theseparametersareindicativeofwherethe
driversoccurandaremostlikelytobelinkedtodeforestationpatterns.Forexample,roads
themselvesarenotdrivers,buttheirdistributionindicateswherepeoplehaveeasier
accesstoforestsandcanrapidlymovetotheirhomesormarkets.Thus,roadsarea
geographicalparameterthatallowsustounderstandwheretheinteractionsamong
people,forestsandmarketsoccur,andtheythustypicallyarevaluableinpredictingwhere
deforestationwillmostlikelyoccur.However,theterm“driver”iscommonlyusedinsuch
modelingcontextstorefertodataonsuchgeographicalparameters,andwewilldosohere
forsimplicity.
Thespatialdataondriversweobtainedareofthreedatatypes.Firsttypeisraster
dataoncontinuousvariables,suchasdistancetoroadsandelevation.Thesecondtypeis
polygondatathatwereusedasclassvariables.Theseincludesoiltype,community,etc.
Thethirdtypeisadatasetcreatedspecificallyforthismodelingexercise.Torepresenthow
“marginal”acommunityis,i.e.howisolatedandlackinginresourcesand/orsubsidies,we
assignedathree‐classvariablebasedona“marginalization”indextoamapoflocationsof
communitycenters.Wethencreatedamapofdistancetoeachclassofcommunity.
Allofthesedatawererasterizedandcreatedorresampledtomatchthe30mcell
arrayofthedeforestationmap.Thefulllistofpotentialdriverdatatouseinthelocal
modelsisreportedinTable5.2.1.
48
Table5.2.1.Driverindependentvariablesusedforspatialmodelsatthelocallevel.
Variable
label
Variablename
Datasource
Notes
Themarginalizationindex
isasummaryfor
differentiatingcensus
townsinthecountry,
accordingtotheglobal
impactofdeficienciesthat
affectthepopulationasa
resultoflackofaccessto
education,residencein
inadequatehousingand
lackofassets.
Var1
var_dist_hi_marginalized_villages
Conabio
Var2
var_dist_low_margin_villages
Conabio
Var3
var_dist_medium_margin_villages Conabio
Var4
var_dist_primary_road
Conabio
Var5
var_dist_railroad
Conabio
Var6
var_dist_rivers
Conabio
Var7
var_dist_secondary_road
Conabio
Var8
var_dist_small_medium_cities
Conabio
Var9
var_dist_trail
Conabio
Var10
var_elev_30_30m
INEGI
Originalresolutionof60
meters
Var11
var_slope_30m
INEGI
Derivedfromthedigital
elevationmodel(DEM)
Var12
var_pop_dens
GlobalRural
Urban
Mapping
Project
(GRUMP)
Thisvariablewasnot
inlcudedintheSierra
Ramarurimodel
Var13
var_protected_areas_dummy
Conabio
Presenceorabsenceof
Federalprotectedareas
Var14
var_dist_non_forest_2006
UMD/Hansen DerivedfromtheinputLC
data
maps
Var15
var_dist_megacities
Conabio
Regionallyimportant
urbancenters,including
statecapitals
Thisvariablewasnot
includedinmodelsfor
AATRswithoutamegacity
(populationGT75,000)
withinthereferenceregion
49
5.2.3.Spatialmodeling
WeusedtheIDRISILandChangeModelertool(LCM)forallspatialmodelingatthe
locallevel.ThisisdevelopedbyClarkLabsandoneofthestrongermodelingtools
availableforland‐usemodeling.Documentationonthetoolandtermsusedinthis
descriptioncanbefoundat:http://www.clarklabs.org/products/Land‐Change‐Modeling‐
IDRISI.cfm.
Theyearlydeforestationdatafrom2001to2012weregroupedintothreedates
andtwotimeperiods:2001‐2005‐2012.Thefirsttimeperiodisusedtocalibrateeach
localmodel.Thecalibratedmodelisthenusedtopredictdeforestationoverthefollowing
timeperiod.Sincedataonactualobservationsofdeforestationforthelatterperiodexist,a
validationofthemodelispossiblebycomparingthemodeledtoactualpatternsof
deforestation.
Forclassvariables,wecreated“evidencelikelihood”mapsforinputintomodels.
Theseassigntheproportionalimportanceofaparticularpolygontothestudyarea’s
overalldeforestationrate.Thisisthenusedasapotentialweightingfactorinthemodeling
algorithm.
TheLCMtoolandmethodsapprovedbytheVCScomparethespatialpatternsof
drivervariableswiththoseofhistoricaldeforestation.Statisticalrelationshipsarethen
usedtoproduceestimatesofthe“potential”fordeforestationineachmodelcell.These
valuesofpotentialcouldbere‐scaledtobevaluesoflikelihood,wheretheirsumequalsa
definedtotalrateforthemodeledperiod.Ifthisisdone,thentheoutputwouldbesimilar
tothenationalmodelinthatcellsareassignedacontinuousvalue.Thelikelihoodvalues,
rangingfromzerotoone,couldbeusedasiftheywereestimatesoftheproportionofthe
cellthatisdeforested.Thiscouldbecalleda“continuous”approach.
Anotherapproachistoassigncompletedeforestationtothecellswiththehighest
valuesofpotential,whichcouldbecalleda“discrete”approach.Thisproducesamap
wherecellsareeitherdeforestedornot.Thisassumesthatdeforestationentirelyoccursin
thesitesofgreatestpotentialorrisk.Whilethiscouldbearguedarealisticapproach,
thereareproblemswithitsassumptions,i.e.thatthereisnofinerscalevariationinrisk
duetounobservablereal‐worldfactors.Thus,highrisksitesarefullydeforestedandzero
deforestationhappensinallplacesotherthanthestrictlymostthreatenedsites.
Regardless,themethodsapprovedbyVCSallrequirethisdiscreteapproach,andthisisthe
approachthatweappliedinthelocalmodels.Wedo,however,maintainthecontinuous
dataondeforestationpotential,andfurtherstudycouldexplorethedifferencesbetween
theresultsofthetwoapproaches.
Theapproachofthistool,andofmostothersusedinsuchapplications,isto
calibratewithasubsetofthedata,whetherselectingaparticulartimeperiodorspatial
subset,thentorunthemodelandvalidateitwithalatertimeperiodorseparatespatial
subset.Weusedonetimeperiodinordertoallowtheoptionofvalidationoverthesecond
timeperiod.Differentalgorithmsformodelingtherelationshipsbetweendriversand
deforestationexist.WeselectedtheMulti‐LayeredPerceptron(MLP)algorithmwithin
50
IDRISI’sLCMbecauseofitsefficiencyandrelativelystrongperformancecomparedtoother
algorithms,suchasmultipleregression,etc.(Eastman,2005).TheMLPisaformofa
neuralnetworkthatcantakecontinuousandclassvariablesasinputsandisnotdependent
onassumptionsofnormaldatadistributions.
Weranmultiplemodelsforeachstudyareatogetageneralsenseofperformance
andimpactsofdifferenttypeofdataondrivers.Wetriedexcludingdifferentindividual
driversorsetsofdrivers.Amongthesevensites,wefoundthatthedataintheformof
polygonsalmostalwaysledtoresultswithconspicuousartifacts.Thesewerebothinthe
formofsharpchangesinthevaluesofpotentialalongboundariesofpolygons.Also,subtle
differencesamongpolygonshadexaggeratedimpactsontheresultingdiscretemapsof
predicteddeforestation.Ingeneral,wefoundthatthemodel,especiallythediscrete
predictionsoflocationsofdeforestation,werehighlysensitivetotheclassvariables.
Becauseofthis,ourfinalmodelsexcludedallpolygon‐typeclassvariablesotherthan
protectedareas.Thelatterwaskeptsincethisdatalayeryieldedrealisticimpactson
outputs,consideringthetrendsindeforestationratesinprotectedversusnon‐protected
landevidencedbythehistoricaldeforestationmaps.Asaresult,themostimportantsocio‐
economicparameterusedasaninputtothefinalmodelsisthedistancetocommunities
stratifiedbylevelofmarginalization.
Withtheselectionoffinalmodels,wehaveoutputsofestimatesofthepotentialfor
deforestation.Tocreatemapsofdiscretelocationsofpredicteddeforestation,werequired
asourceforthetotalrateofeachreferencearea.Weusedtheratesderivedfromthe
nationalmodelswithineachreferencearea.Therateforthereferenceregionisthen
appliedtothevalueofpotentialgeneratedbytheLCMmodel,assigningdeforestationto
thehighestpotentialcellsuntilthetotalchangeareaobtainedfromthenationalmodelis
reached.Wedidthisforthreedifferentscenarios:thealternative(Poisson)regression
modelofthe“business‐as‐usual”ornon‐REDD+scenario(AlternativeBAU),theratefrom
thepreferred(negative‐binomial)modelofthenon‐REDD+scenario(PreferredBAU),and
theratefromthepreferredmodeloftheREDD+scenario(PreferredBAU).Modelswere
runtosimulatedeforestationfrom2012through2022andoutputsweretabulatedfor
eachsiteandeachreferencearea.
5.3.Results
5.3.1.Deforestationsince2000
Deforestation,asdefinedbya25%thresholdappliedtotheUMDforestcoverin
2000andyearlytreecoverlosssincethen,hasbeensignificantinmostsites,especiallythe
Yucatánsite.Forestcoverin2000andaggregateddeforestationfrom2000to2012are
reportedinTable4.2.Annualizedratesarehighlyvariableamongsites.Twosites,
ComunidadesForestalesdeOaxacaMixtecaandSierraRaramuri,haveratesnearzero.
Twoothersites,ComunidadesForestalesdeOaxacaIstmoandSierraPucclosCheneshave
relativelyhighratesthatinareasapproach0.5percentperyear.Inmostcasesejidoshad
higherdeforestationratesthantherestofthelocalreferencearea,howeverinSierra
51
Rairumiprotectedareascategoryhadthehighestrate,andinsierraPucclosChenesthe
AATRhadthehighestrate.
PatternsofhistoricaldeforestationareshowninMapsA‐25throughA‐31inthe
Appendix.Inallthefigures,forestcoverisdefinedbya25%thresholdappliedtothe
Hansen,etal.(2014)data,anddeforestationisasumofalllossdatawithinthatdefined
forestarea.
Table5.3.1.Summaryofforestcoverin2000anddeforestationfrom2000to2012
amongAATRs.
Totalland
area(ha)
Forest
area,
2000(ha)
Forested
fraction
(2000)
Forest
area,
2012(ha)
Defor
Defor
00‐12
(ha/yr)
00‐12
(%/yr)
ComunidadesForestalesdeOaxacaIstmo
AATRsite
265,382
213,844
Land‐use:Ejidos
784,033
383,469
Land‐use:
Comunidades
1,755,724
1,334,791
Land‐use:Protected
Areas(federal)
na
na
Comunidades_ForestalesdeOaxacaMixteca
AATRsite
Land‐use:Ejidos
Land‐use:
Comunidades
Land‐use:Protected
Areas(federal)
0.81
206,217
636
0.30%
0.49
362,324
1,762
0.46%
0.76
1,308,087
2,225
0.17%
na
na
na
na
471,624
203,659
0.43
203,555
9
0.00%
1,090,966
418,825
0.38
415,866
247
0.06%
2,696,932
1,148,990
0.43
1,145,713
273
0.02%
435,452
67,044
0.15
67,021
2
0.00%
0.93
383,997
210
0.05%
0.45
389,478
1,108
0.28%
0.65
1,237,395
1,349
0.11%
0.30
55,456
12
0.02%
0.62
2,966,421
4,017
0.13%
ComunidadesForestalesdeOaxacaSierraNorte
AATRsite
417,588
386,522
Land‐use:Ejidos
903,043
402,775
Land‐use:
Comunidades
1,932,267
1,253,583
Land‐use:Protected
Areas(federal)
187,212
55,597
CuencasInterioresdelaSierradeChiapas
Referenceregion
4,897,982
AATRsite
1,058,629
3,014,625
611,574
0.58
606,088
457
0.07%
Land‐use:Ejidos
1,975,301
Land‐use:
Comunidades
775,639
Land‐use:Protected
Areas(federal)
639,767
CutzamalaValledeBravo
1,174,757
0.59
1,157,540
1,435
0.12%
637,955
0.82
630,996
580
0.09%
526,698
0.82
523,044
304
0.06%
Referenceregion
3,008,360
1,011,168
0.34
1,007,582
299
0.03%
263,333
117,204
0.45
116,480
60
0.05%
1,219,455
320,089
0.26
319,202
74
0.02%
229,479
111,390
0.49
110,940
38
0.03%
273,411
151,909
0.56
150,921
82
0.05%
AATRsite
Land‐use:Ejidos
Land‐use:
Comunidades
Land‐use:Protected
Areas(federal)
52
SierraPuccLosChenes
AATRsite
1,535
1,429
0.93
1,332
8
0.57%
Land‐use:Ejidos
Land‐use:
Comunidades
Land‐use:Protected
Areas(federal)
7,109
6,480
0.91
6,091
32
0.50%
1
1
0.59
1
0
0.55%
1,608
1,252
0.78
1,241
1
0.08%
SierraRaramuri
AATRsite
1,883,875
984,941
0.52
984,242
58
0.01%
Land‐use:Ejidos
Land‐use:
Comunidades
Land‐use:Protected
Areas(federal)
5,933,054
2,783,030
0.47
2,776,926
509
0.02%
1,560,860
871,818
0.56
870,021
150
0.02%
70,206
47,239
0.67
46,935
25
0.05%
Note:Forestcoverisdefinedbya25%thresholdappliedtotheHansen,etal(2014)data,and
deforestationisasumofalllossdatawithinthatdefinedforestarea.Notethatdeforestationvalues
differfromthoseintheglobalanalysissincethehistorical‐deforestationmapswerefilteredforthe
localanalysis.Thefilteringremovedanypatchesofforest,non‐forestordeforestationforagiven
timeperiodsmallerthanonehectare.
5.3.2.Modeleddeforestationbeyond2012
Weranmultiplemodelswithdifferentcombinationsofdrivervariables.Among
these,thegenerallyconsistentresultwasthatthebestperformingmodelsweretheones
usingall15inputvariables.Also,wefoundthattheinclusionofdistancetoanon‐forested
edgedidnotimprovemodels.Thisparametertendedtoleadtoanover‐fittingof
deforestationalongexistingedges,andexclusionofthisparameterdidnotresultinun‐
realisticallyremotedeforestationinthemodeloutputs.
Thus,ourfinalmodelsallwerebasedontheMLPmodelsusingallinputsexcept
distancetoanon‐forestededge.Finally,MLPrandomlyselects“seedcells”tobeginmodel
calibration,andmodeloutputsmayvarymodestlyinarandommannerdependingonthe
selectionoftheseseeds.Thus,wereporttwomodeliterationsforeachfinalmodel.We
appliedthesensitivityanalysisincludedinIDRISI’sMLPtooltoestimatetherelative
importanceofdifferentinputvariables.ImportancevaluesarereportedinTable5.3.2.
Inputvariablesmostimportanttothemodelvariedamongthestudyareas.
Distancetomega‐citieswashighlyimportantforthestudyareaswheretheyoccurred,
CutzamalaValledeBravoandSierraRaramuri.Forregionsthatareexemplaryoffrontier
areas,accessibility,i.e.distancetoroads,trailsandrivers,wasmostimportant.Forregions
thatareexemplaryofheavily‐fragmentedforest,biophysicalvariables,e.g.slope,were
mostimportant.Therewasoverallnoconsistenttrendontheimportanceofthevariable
distancetohighly‐marginalizedvillages.Insomeareassitesnearhighly‐marginalized
villageshadhigherdeforestationrateswhileinotherareasthetrendwasreversed.Inall
butonestudyarea,includingdistancetonon‐forestlandincreasedmodelskill.Only
CutzamalaValledeBravo,whichishighlyfragmentedforest,didn’thavethiseffect.
53
Table5.3.2.Relativeimportanceofthedifferentdrivervariablesformodelsrunineachofthelocalstudyareas.SeeTable5.3.1for
thelistofvariables.
Region
Comunida
des
Forestales
deOaxaca
(Istmo)
Comunida
des
Forestales
deOaxaca
(Mixteca)
Comunida
des
Forestales
deOaxaca
(Sierra
Norte)
Cuencas
Interiores
dela
Sierrade
Chiapas
Cutzamala
Vallede
Bravo
var1
var2
var3
var4
Dist.hi
margin
alized
villages
Dist.
low
margin.
villages
Dist.
medium
margin.
villages
Dist
primar
yroad
MLPrun1
12
8
5
10
1
MLPrun2
11
7
13
4
3
MLPrun
w/odist.to
non‐forest
8
10
5
9
MLPrun1
11
4
13
MLPrun2
13
4
10
Model
Run
MLPrun
without
distanceto
non‐forest
MLPrun1
MLPrun2
var5
var8
Dist.
Second
ary
road
Dist.
small/
medium
cities
13
2
3
9
4
7
11
n/a
6
n/a
0.4935
8
6
1
9
12
5
10
n/a
2
n/a
0.5348
2
11
3
1
6
4
7
12
n/a
n/a
n/a
0.4931
8
5
14
7
3
2
1
12
10
12
9
n/a
0.7419
11
5
9
7
2
3
1
14
12
14
8
n/a
0.7405
10
n/a
n/a
0.6802
12
4
n/a
0.5593
9
1
n/a
0.5958
9
n/a
n/a
0.5351
Dist.
rivers
var9
Dist.
trail
var10
var11
var12
Elev30
30m
Slope
30m
Pop
density
var7
Dist.
railroa
d
var6
var13
var14
var15
Protect
ed
areas
dummy
Dist
nonfor
est:
2006
Dist
Megacit
ies
Skill
11
4
12
8
6
7
13
2
3
1
10
9
14
6
7
2
5
13
11
5
10
1
3
8
14
5
13
2
5
12
8
6
10
4
3
7
MLPrun
w/odist.to
non‐forest
13
5
4
3
6
7
11
8
12
1
2
10
MLPrun1
13
5
9
6
3
10
12
4
11
2
1
13
7
8
n/a
0.4124
MLPrun2
11
14
7
10
6
13
8
4
12
2
1
9
3
5
n/a
0.4504
7
n/a
n/a
0.4006
MLPrun
w/odist.to
non‐forest
12
4
8
MLPrun1
8
6
MLPrun2
6
5
MLPrun
w/odist.to
non‐forest
9
7
5
3
13
10
6
9
2
1
11
10
9
12
7
3
4
15
2
13
14
5
11
1
0.6282
15
10
13
8
3
4
14
2
11
12
9
7
1
0.677
11
6
10
8
2
5
13
4
12
14
3
n/a
1
0.71
54
Sierra
PuccLos
Chenes
Sierra
Raramuri
MLPrun1
9
11
10
4
2
12
8
3
7
14
6
5
MLPrun2
9
2
14
3
11
4
5
5
10
12
7
6
13
1
n/a
0.5661
13
1
n/a
0.5715
10
n/a
n/a
0.3906
MLPrun
w/odist.to
non‐forest
6
2
1
8
5
3
13
7
9
4
12
11
MLPrun1
13
12
5
14
6
8
7
2
9
1
4
n/a
11
10
3
0.6178
MLPrun2
14
12
5
13
9
8
11
2
7
1
4
n/a
10
6
3
0.6115
13
n/a
2
0.6224
MLPrun
w/odist.to
non‐forest
11
10
6
7
8
9
5
3
12
1
4
n/a
Note:Lownumericalvaluesindicatehigherimportancelevels,i.e.thesearerankscores.Themostimportantthreevariablesforeachmodelarehighlighted
inyellow.
55
5.4.Predictingdeforestationinthefuture:
Thepatternsofpotentialfordeforestationvariedamongthesites,althoughare
understandablegiventhedifferingimportanceofdrivervariablesinthedifferentstudyareas.Itis
thususefultorefertoTable5.3.2wheninterpretingthepatternsofpotential.Mapsofpotential
deforestationor“soft”deforestationtransitionpotentialareshowninthemapsinfigures5.4.1b‐
5.4.7bbelow.Onealsoseesthatthereismoreinformationinthesemapsthanthehard
classificationsshownaboveeachmap(5.4.1a‐5.4.7a),andonecaninterpretthepatternsofrelative
potentialbeyondconsideringonlythesitesofstrictlygreatestpotential,asisthecaseinthehard
classificationspresentedlaterinthissection.
ThehardclassificationoffuturedeforestationforeachAATRwasbasedonthetransition
potentialsurfacescreatedcombinedwiththetotalratesforeachreferenceregionaccordingthethe
differentscenariosofthenationalmodels.Theseharddeforestationpredictionsareshowninmaps
in5.4.1a‐5.4.7a.Thetransitionpotentialchosenforthefinalpredictionwerebasedonthemodels
withthehighestskillmeasure(shownintable5.3.2).Theratesoftransitionthatwereappliedto
thetransitionpotentialsurfacesareshownbelowinTable5.4.1,aswellasthetotalamountof
deforestationpredictedforeachreferenceregionandAATRsite.Thequantityofdeforestation
predictedwithineachreferenceregionswasbasedontheinputratesshownbelow,andthe
allocationofthedeforestationwasdeterminedbyselectingtheforestcellswiththehighestvalues
inthetransitionpotentialsurfacescreatedinLCM.Thismethodfordeforestationallocationhasthe
advantageofbeingabletocreatethematicland‐covermapsatthenativeresolutionoftheinput
dataset,howeveritalsoassumesthatdeforestingagentknowwhichpixelsareoptimalfor
deforestationandthereforeonlythemosthighlyvulnerablepixelswillbetransitioned.
RatesfortransitionwerederivedfromboththeobservedhistoricalrateintheLCMmodel
andthemodelednationalratesfromthenationalmodel.Thetransitionsshownrepresentthetotal
transitionsfrom2012to2022,andinthecaseofthenationalmodels,wepresentresultsforbotha
businessasusualscenario(BAU)andREDD+scenarioassuminga$10/tCprice.Thehistoricalrate
fromderivedfrom2000‐2012isalsoshowandprojectedlinearlyto2022,however,oneshould
notethatthehistoricalratecannotbedirectlycomparedwiththemodeledratesfromthenational
modeldueinparttothefilteringprocessthatwasusedontheforestcoveranddeforestationdata
fromHansenetal.2013.ThereforethehistoricalrateisconsistentlylowerthantheBAUscenarios.
Thehistoricalratesalsodonottakeintoaccountlargernationaltrendwhichwouldhaveaneffect
onboththemodeledratesandthefuturerates(seetable4.4.1).Thereforethehistoricalratesare
providedascontextonhowmuchdeforestationmightbepredictedwithouttheuseofanexternal
model,usingthemostsimplisticapproach.Amoreusefulcomparisonforunderstandingtheeffect
ofnationalpolicyonthelocalmodelsisthedifferencebetweenthepreferred(NegativeBinomial)
BAUscenarioandthepreferred(NegativeBinomial)REDDscenario.Comparingthetwonegative
binomialscenariosshowsthatatthereferenceregionscalethereweredecreasesbetween23–
58%,whichisconsistentwiththenationallevelpredictionsshowninTable4.4.3.
AttheAATRsitescaletherangeismuchmorevariationintheamountofdeforestation
predictedbetween2012and2022.InthecaseoftwoAATRsites,OaxacaMixtecaandSierra
Raramuri,therewasnodeforestationpredictedwithinthesite,regardlessofthescenario.The
reasonthatthiswasthecaseisduetothemethodwhichwasusedtoassignchange.Asdescribed
56
above,becauseonlythehighestrankedpixelsaretransitionedandthesesiteshaveverylow
deforestationrates(rangingfrom0.23%–2.12%overthe10yearperiod).Similarly,the
extremelyhighreductionindeforestationintheOaxacaSierraNorteAATRsitecanalsobe
attributedtothemethodofallocation.InthecaseofthesethreeAATRsites,REDD+initiatives
wouldhaveaminimaleffectbecausethebaselineratesunderallscenariosareverylow.The
remainingfourAATRsitesobservedareductionindeforestationrangingfrom20%‐67%,andin
mostcasesthesereductionsweresimilartothoseexperiencewithinthereferenceregions.
Thehardclassificationsofpredicteddeforestationindicatedifferentconclusionsamongthe
differentstudyareas(Table5.4.1).Forexample,forthreestudyareas,Mixteca,Chiapasand
Raramuri,bothBAUmodelspredictedratesofdeforestationofovertwicethehistoricalrates.The
remainingsiteshadpredictedratesofwithin50percentofthehistoricalrates.Inalmostallcases,
modeledfutureratesfortheBAUscenariosweregreaterthanhistoricalrates.Mostofthepredicted
ratesforthestudyareasweresimilarforbothBAUmodels,althoughChiapasandValleBravo
modelsdidproducequitedifferenttotalrates.
OneresultcommontoallstudyareaswasthattheREDDscenarioyieldedlowerratesthan
bothBAUscenarios.Thisdifferencewasuptotwo‐foldforfourofthesevenareas.ValleBravohad
thesmallestdifference,andactuallyhadarateforthePoissonBAUscenarioslightlylowerthanthe
REDDscenario.Likewise,inmostcasestherateswithintheAATRsiteswerelowerintheREDD
scenariothanineitherBAUscenario.InChiapastheREDD‐scenarioratewasclosetothePoisson
BAUrateandinValleBravotheREDDratewasslightlyhigherthanthePoissonBAUrate.However,
itissafesttocomparetheREDDscenario,basedonthenegative‐binarymodel,withthesimilarly‐
modeledBAUscenario.Forthese,allREDDscenarios’rateswerelowerthanthoseoftheBAU
scenarios,thepercentreductionsreportedinthelastcolumninTable5.4.1.Thisisexcludingthe
twocaseswithnear‐zeroratesinanyscenariowithintheAATRsite,MixtecaandRaramuri.
Figures5.4.1a‐5.4.7.ashowthepatternsofdeforestationfromthehardclassificationsofthe
predictions.Inthesefigures,redareasareplaceswheretheBAUmodelpredictsdeforestation
whiletheREDDmodeldoesnot;blueareasarewherebothmodelspredictdeforestation.
Twositeshadclosetozerohistoricaldeforestationandnodeforestationinmodeled
scenarios,MixtecaandRaramuri.OaxacaSierraNorteandSierraChiapasdohavedeforestationthat
enterstheAATRsitesinbothscenariosmodeled,howeverthepatternisverydisperse.Atthescale
ofpresentationinthisreport,theserelativelysmallpatchesofdeforestationdonotappear.
However,explorationofthefull‐resolutiondigitalrasterfilesofthemodeloutputswillshow
deforestationinsidethesesites,especiallyinthelowervalleys.Thepatternofmodeled
deforestationinCutzmalaValleBravoisratherunique.Allofthedeforestationinthesiteis
concentratedinonelargepatch.Thispatternissuspicious,andofallthespatialmodelsinthis
study,thisoneappearsthemostsuspectandworthyofre‐assessment.
ThemodelsforboththeBAUandREDDscenariosforOaxacaIstmoandPuccpredict
relativelyhighratesofdeforestationinsidetheAATRsites.InOaxacaIstmo,thisisalmostentirely
inthenorth‐eastofthesite,fromwheretheroadsprovideaccessibilityandtheelevationnois
favourable.InPuccthepredicteddeforestationinbothscenariosisgreatestamongallAATRsites.
Predicteddeforestationisalsoverywelldistributedthoughoutthesite.Thesesitesstandout
57
amongthegroupbothintermsofBAUdeforestationandthepotentialforreductionsin
deforestationunderREDD.
Table5.4.1.Predicteddeforestationfrom2012‐2022
Deforestationperscenario2012‐2022
Historical
Rate*
Alternative
(BAU)
Preferred
(BAU)
12‐24
12‐24
Preferred
(REDD)
12‐24
AATR_RT1‐Mixteca
Decreasein
deforestation
between
BAUand
REDD(%)
Referenceregion
0.36%
2.10%
2.12%
0.98%
GrossDeforestationperRR,2012‐2022(ha)
6,672
38,779
39,138
18,132
54%
‐
‐
‐
‐
0%
Referenceregion
1.83%
1.94%
1.72%
0.72%
GrossDeforestationperRR,2012‐2022(ha)
42,428
44,938
39,787
16,699
58%
226
269
180
3
98%
5.78%
8.50%
7.63%
5.01%
GrossDeforestationperRR,2012‐2022(ha)
448,446
659,371
591,949
388,716
34%
GrossDeforestationinthesite,2012‐2022(ha)
101,249
140,439
128,215
89,539
30%
Referenceregion
1.60%
4.40%
7.83%
3.90%
GrossDeforestationperRR,2012‐2022(ha)
47,433
130,466
232,207
115,629
50%
GrossDeforestationinthesite,2012‐2022(ha)
5,101
22,180
43,245
19,139
56%
0.23%
0.58%
0.57%
0.38%
996
2,510
2,493
1,628
35%
‐
‐
‐
‐
0%
Referenceregion
0.35%
0.39%
0.52%
0.40%
GrossDeforestationperRR,2012‐2022(ha)
3,573
3,963
5,219
4,032
23%
GrossDeforestationinthesite,2012‐2022(ha)
2,223
2,441
3,087
2,481
20%
Referenceregion
3.13%
4.57%
4.19%
2.10%
GrossDeforestationperRR,2012‐2022(ha)
73,122
106,819
97,975
49,128
50%
GrossDeforestationinthesite,2012‐2022(ha)
3,779
6,429
5,729
1,912
67%
GrossDeforestationinthesite,2012‐2022(ha)
AATR_RT2‐SierraNorte
GrossDeforestationinthesite,2012‐2022(ha)
AATR_RT3‐SierraPucc
Referenceregion
AATR_RT4‐Chiapas
AATR_RT5‐Raramuri
Referenceregion
GrossDeforestationperRR,2012‐2022(ha)
GrossDeforestationinthesite,2012‐2022(ha)
AATR_RT6‐ValledeBravo
AATR_RT7‐Itsmo
Note:Allratesaregrossoverthe10yearperiod,percentwhereindicatedotherwisehectares.Alternative(BAU)isthe
rateofdeforestationfromthealternative(negativebinomial)regressionmodelofthe“business‐as‐usual”ornon‐REDD+
scenario;Preferred(BAU)istheratefromthepreferred(negativebinomial)modelofthenon‐REDD+scenario;Preferred
(REDD)istheratefromthepreferredmodeloftheREDD+scenario.Notethathistoricaldeforestationvaluesdifferfrom
thoseintheglobalanalysissincethehistorical‐deforestationmapswerefilteredforthelocalanalysis.Thefiltering
removedanypatchesofforest,non‐forestordeforestationforagiventimeperiodsmallerthanonehectare.
58
Figure5.4.1a.“Hard”predictionofdeforestation,2012‐2022,OaxacaIstmo.AATRsitehighlightedin
yellowthatching.Theredareasindicatethepredicteddeforestationthatwouldoccurinabusiness‐as‐usual
scenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbonincentive.Areas
thatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblackareasfalloutsidethe
boundaryofthereferenceregion.
Figure5.4.2b.“Soft”transitionpotentialsurface,OaxacaIstmoAATRsiteandreferenceregion.Areas
inbluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicatehigh
transitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidetheboundary
ofthereferenceregion.
59
Figure5.4.3a.“Hard”predictionofdeforestation,2012‐2022,OaxacaMixteca.AATRsitehighlightedin
yellowthatching.Theredareasindicatethepredicteddeforestationthatwouldoccurinabusiness‐as‐usual
scenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbonincentive.Areas
thatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblackareasfalloutsidethe
boundaryofthereferenceregion.
Figure5.4.2b.“Soft”transitionpotentialsurface,OaxacaMixtecaAATRsiteandreferenceregion.
Areasinbluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicate
hightransitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidethe
boundaryofthereferenceregion.
60
Figure5.4.4a.“Hard”predictionofdeforestation,2012‐2022,OaxacaSierraNorte.AATRsite
highlightedinyellowthatching.Theredareasindicatethepredicteddeforestationthatwouldoccurina
business‐as‐usualscenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbon
incentive.Areasthatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblack
areasfalloutsidetheboundaryofthereferenceregion.
Figure5.4.3b.“Soft”transitionpotentialsurface,OaxacaSierraNorteAATRsiteandreferenceregion.
Areasinbluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicate
hightransitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidethe
boundaryofthereferenceregion.
61
Figure5.4.5a.“Hard”predictionofdeforestation,2012‐2022,SierraChiapas.AATRsitehighlightedin
yellowthatching.Theredareasindicatethepredicteddeforestationthatwouldoccurinabusiness‐as‐usual
scenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbonincentive.Areas
thatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblackareasfalloutsidethe
boundaryofthereferenceregion.
Figure5.4.4b.“Soft”transitionpotentialsurface,SierraChiapassiteandreferenceregion.Areasin
bluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicatehigh
transitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidetheboundary
ofthereferenceregion.
62
Figure5.4.6a.”Hard”predictionofdeforestation,2012‐2022,CutzmalaValleBravo.AATRsite
highlightedinyellowthatching.Theredareasindicatethepredicteddeforestationthatwouldoccurina
business‐as‐usualscenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbon
incentive.Areasthatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblack
areasfalloutsidetheboundaryofthereferenceregion.
Figure5.4.5b.“Soft”transitionpotentialsurface,CutzmalaValleBravoAATRsiteandreference
region.Areasinbluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinred
indicatehightransitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutside
theboundaryofthereferenceregion.
63
Figure5.4.7a.”Hard”predictionofdeforestation,2012‐2022,SierraPUCC.AATRhighlightinyellow
thatching.Theredareasindicatethepredicteddeforestationthatwouldoccurinabusiness‐as‐usual
scenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbonincentive.Areas
thatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblackareasfalloutsidethe
boundaryofthereferenceregion.
Figure5.4.6b.“Soft”transitionpotentialsurface,SierraPUCCsiteandreferenceregion.Areasinblue
representlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicatehightransition
potential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidetheboundaryofthe
referenceregion.
64
.
Figure5.4.8a.”Hard”predictionofdeforestation,2012‐2022,SierraRaramuri.AATRsitehighlightin
yellowthatch.Theredareasindicatethepredicteddeforestationthatwouldoccurinabusiness‐as‐usual
scenario,whiletheblueareaisthedeforestationthatwouldoccurwitha$10/tCO2carbonincentive.Areas
thatarebluearedeforestedunderbothscenarios.Yellowareasarenon‐forestandblackareasfalloutsidethe
boundaryofthereferenceregion.
Figure5.4.7b.“Soft”transitionpotentialsurface,SierraRaramuri.AATRsiteandreferenceregion.
Areasinbluerepresentlowertransitionpotential,orareaslesslikelytotransition,andareasinredindicate
hightransitionpotential,orareasmorelikelytotransition.Blackareasarenon‐forestorfalloutsidethe
boundaryofthereferenceregion.
65
5.5.Conclusions
Amongthelocalregionsstudied,deforestationisofgreatestconcerninSierraPuccLos
Chenes,thenfollowedbyOaxacaIstmo,OaxacaSierraNorteandSierraChiapas.Theremaining
regionshadverylowannualdeforestationratesduring2000through2012.SierraPuccLosChenes
isoffurthernotebecausewithinthisregionthepercentdeforestationratewasgreatestinsidethe
AATRs.
Intheregionswithrelativelyhighdeforestationrates,deforestationwasnotaslocally
concentratedasisoftenfoundinotherareas.ThedeforestationpatternsinOaxacaIstmo,Oaxaca
SierraNorteandSierraChiapasweremostlyinasub‐region,althoughquitespreadthroughoutthe
sub‐regionasopposedtoveryconcentratedaroundtownsandmajorroads.InSierraPuccLos
Chenes,deforestationwasspreadovermostofthestudyregion.
Land‐usedesignationssuchasejidosandcomunidadeswerenotincludedinthefinal
models.Howeverwhentestedtheydidhavealargeimpactonthetransitionpotentialsurfaces.
Theeffectsofindividualejidosandcomunidadesvariedalotbothbetweensitesandwithinsites.In
somecasesejidoswouldreducethelikelihoodofdeforestationandinothercasestheywould
greatlyincreasethelikelihood.Thisisanindicationthatindividualland‐useandeconomic
decisionsattheejidoorcomunidadeslevelmayhaveastronginfluenceonpatternsofdeforestation.
However,furtherresearchwouldberequiredtofullyexploretheserelationshipsandperhaps
groupejidoorcomunidadesbasedontheirland‐usepractices.
ThesepatternsofhistoricaldeforestationpointMREDDAlliancepartnerstoareas
wherefurtherexplorationmaybewarranted.First,were‐notethatthedefinitionofforestandthe
mannerinwhichtheforest2000benchmarkwasdefinedissignificant.Muchofthedeforestation
weseemayactuallybethere‐clearanceoffallowsand/orplantations,ratherthanclearingof
matureforest.Howmuchdeforestationintheseareasisactuallythatofmatureforestcouldbe
assessedseveralways.First,onecouldconductavisualqualitativeassessment.Thiscouldbeby
superimposingthedeforestationsitesoverthereflectancemosaicfor2000fromUMD.Thiscould
bedoneonlineviatheUMD‐Googlesite,orlocallyiffilesaredownloaded.
Second,amapofabestestimateofthedistributionofmatureforestin2000couldbe
combinedwiththedeforestationdatainaGIS.Thiscouldbeoneofseveralanationalvegetation
map,althoughthatmapitselfshouldbeassessedvisuallywiththeLandsatmosaic,sincemany
nationalforestcovermapsalsohavemostfallowscombinedintotheforestclass.FortheChiapas
andYucutansites,aswellasanotherothersitesofinterestwithinthefivesouthernstatesof
Mexico,theConservationInternational(CI)deforestationmapcouldbeused.Thismapreports
deforestationforthreedates,including2000(Vacaetal.2005).The2000forestcoverestimate
couldbeusedasanexamplebenchmarkmapforthusevaluation,andCIandpartnersattemptedto
minimizeanyinclusionoffallowsorplantationsinitsmatureforestclass.
Eventhoughourforestdefinitionandbenchmarkmapincludesecondaryforestsand
plantations,thepatternsoftree‐coverlossamongthesecovertypesplusforestarerevealing.The
findingsmaybeinterpretedasrevealingbothmatureforestclearanceandaformofagricultural
intensification,viafallowclearing,thatisoftenassociatedwiththesameland‐usepressuresthat
arelinkedtomatureforestclearing.
66
Onthemodels,oneshouldinterpretboththemapofdeforestationpotentialandthehard
classificationofpredicteddeforestation.Theformercanbeseenasadistributionoftherelative
potentialfordeforestation,nottoodissimilartothecontinuousdeforestationlikelihoodmaps
outputfromthenationalanalysis.Thelatterisstrictlythecellsofgreatestpotentialthataddupto
theareaofdeforestationpredictedforthestudyregionbythenationalmodel.
Ourrecommendationistofollowupthisanalysiswithasecondpassthatattemptsto
stratifyhistoricaldeforestationintothatofmatureforestversusothercovertypes.Ifdone,the
spatialmodelscouldbere‐runefficiently,especiallysincethemodelingdatasetsareorganizedand
readyforadditionaliterations.Ifdone,onecouldconsidercharacterizingdeforestationforspecific
landuses,howeverinformationonthosewouldbederivedfromremotesensingproducts,rather
theywouldneedtobebasedonexpertopinionandancillarydata.
Finally,municipallevelvariables,suchasagriculturalyieldandevidencelikelihoodwere
notincludedintheLCMmodelbecausetheyhadanoverlypowerfuleffectonthemodels.These
werethepolygon‐levelvariablesthatwenotewereexcludedbecauseofextrememodelsensitivity
inthemethodssection.Essentially,allofthepredicteddeforestationwouldlocateintoasingle
municipality.However,theremayberealandstrongeffectsofmunicipal‐leveldeforestationatthe
sub‐nationallevelinMexico.Furtherexplorationintootherwaystoincorporatethisinformation
intothemodelsiswarranted.
Foreitherthesemodelsorlateriterationsofmodels,thereareseveraloptionsforvalidation
ofthem.Onecouldcalibrateeachmodelwiththedeforestationpatternsonlythrough2005or
2010,forexample,andpredictdeforestationthrough2012.Theresultingdistributionof
deforestationwouldthenbecomparedtoobserveddeforestationintheUMDproduct.Thisisthe
mostcommonapproachtovalidationrecommendedinVCS‐approvedmethods.
ThereareotheraspectsoftheVCS‐approvedmethodsthatwefindproblematicand
recommendconsideringaltering.First,theseveralVCSmethodsrecommendusingstatisticthatis
calculatedatthecelllevelisfineifthegoalofamodelistopredictdeforestationatthatlevel.
However,thatisusuallynotthegoal.Forexample,ifonecellispredictedasdeforestation,andthe
actualdeforestationmapdoesnotshowchangeinthatexactcellbutdoesshowchangeina
neighbouringcell,thenthestatisticwouldimplythatthemodelhaszeroaccuracy.Thisiswhya
“successful”modelaccordingtothisstatisticisonethathasanaccuracyoffivepercentormore.
Whileofacademicinteresttomodeldevelopersinterestedinperformanceatthecelllevel,itis
difficulttoacceptthisaslogicalforREDD+projectsormostregional‐levelapplications.
Forthisstudy,andforallREDD+projects,thequestionofinterestiswhethermodels
predictaccuratelydeforestationforcertainmanagementorstudyunits,suchasproposedREDD+
sites,leakagezones,politicaldistricts,etc.Theseunitsarerepresentedbypolygons,andthusthe
comparisonshouldbemadeatthepolygonleveloratascalesimilartothepolygonsofinterest.
Oncethisisunderstood,thereareseveralstatisticsthatcouldbeusedtovalidationmodelsatthe
polygonlevel.
Second,VCS‐approvedmethodsrequireproducingthehardclassificationofpredicted
deforestation.Thisforcesalldeforestationintostrictlythecellsofhighestpotential.While
interesting,thisisunrealisticandaformofover‐fitting.Thebestevidenceofthelackofrealismof
67
thisapproachisthatthehistoricaldatathemselvesindicatethatmuchdeforestationoccursinsites
thatareofmoderatedeforestation.Thenationalmodeldoesnotsufferfromthisproblemsinceit
reportsacontinuousestimateofsub‐pixeldeforestationordeforestationlikelihood.Thesame
couldbedonewiththelocalmodelsthatwehaverunbyskippingthestepofproducingthehard
classificationofpredicteddeforestation.Instead,onecantakethemapofdeforestationpotential
andrescalethevaluessuchthattheyadduptothedefinedregionalratepredictedbythenational
model.Thisessentiallyproducesacontinuoussub‐celldeforestationoutput.Thesecanthenbe
summedforanysetofpolygonsandcomparedtothepolygon‐levelratesderivedfromobserved
deforestationinordertovalidatemodels.
68
6. Conclusion
6.1.Summaryofreportfindingsanddirectionsforfutureresearch
Weconductedaseriesofanalysesthatcombinebothnationalandlocalscalemodelingto
aidtheMREDDAlliancepartnersinassessingthevulnerabilityofMexico’sforeststodeforestation.
TheseanalysesfocusonthevulnerabilityofforestedlandswithinMexico’sAATRs,accountingfor
Mexico’suniqueforestmanagementdynamicsthroughdisaggregatingtheresultsbyland
ownershiptypes.Theseanalysesareultimatelymeanttoinformnationalandsubnationalpolicy,
pavingthewayforincentivebasedprograms,andultimatelyreduceddeforestationvulnerabilityin
Mexico.Ourmethodologyincludesthreedifferentandcomplementaryapproaches:(i)reviewing
theexistingliterature,(ii)anationaleconometricanalysisandassociatedscenariosimulation
modeling,adaptingtheapproachoftheOpenSourceImpactsofREDD+Incentives(OSIRIS)model
and(iii)local‐levelspatialspatialmodelingforeachAATR,conductedusingtheIDRISI‐SelvaLand
ChangeModeler(LCM).Keyfindingsfromeachofthesethreepartsofthereportaresummarized
below,alongwithsomediscussionofnextstepsforfutureresearch. 6.1.1.Literaturereviewandmeta‐analysis
Theliteraturereviewyieldedinsightsbasedonanoverviewofdeforestationaswellasa
meta‐analysisonstatisticalstudiesofdriversofdeforestation.Theoverviewsuggeststhat,while
deforestationratesinMexicohavedecreased,thetrendpersists,leadingtomorebiodiversityloss,
increasedgreenhousegasemissions,andreducedsubsistenceopportunitiesforlocalpopulations.
Landtenure(communitylandmanagement,includingejidos),ruralagriculturalsupport,and
paymentsforecosystemsservicesaremajorfocusesoftheliterature.Conclusionsontheroleofthe
majorlandtenuretypeinMexico,communitylandmanagement,aremixed.Studiesarealsoin
disagreementontheroleofsuchruralagriculturalsupportprogramsasPROCAMPO.However
moststudiesagreethatpaymentsforecosystemsservicesdecreasedeforestationrisk,withsome
caveatsrelatedtoregionaldifferencesandstartingdeforestationrisk.Theserelationshipswere
mirroredinthemeta‐analysis:regressionresultsweremixedforejidosandruralincomesupport,
whileresultsforPEStendedtobeassociatedwithdecreaseddeforestation.Furthermore,results
fromthemeta‐analysisrevealedothervariableswithconsistentrelationshipstodeforestationin
Mexico.ThevariablesmostassociatedwithreduceddeforestationinMexicowereassociatedwith
protectionmeasures(asproxiedbyprotectedareasandPES),reducedaccessibility(elevation),
reducedresourcecompetition(propertysize)andcommunityforestry.Thevariablesmost
associatedwithincreaseddeforestationwereassociatedwithareaswhereeconomicreturnsto
agriculturearehigher(proximitytoagricultureandagriculturereturns),biophysicalconditionsfor
conversionarefavorable(soilsuitability),andcompetitionforresourcesarehigh(population).
MostoftheserelationshipswererobustwhenresultsweredisaggregatedtotheYucatanPeninsula.
Notablyhowever,atthenationallevel,povertyappearstobelinkedtoincreasesindeforestation,
whileintheYucatánPeninsulapovertyisassociatedwithdecreaseddeforestation.Conversely,
indigenouspopulationisassociatedwithdecreaseddeforestationatthenationallevel,butis
associatedwithhigherdeforestationintheYucatánPeninsula.
ThesediscrepanciessupportthewidelyheldviewthatMexico’slandscapeandtherelated
driversofdeforestationvarygreatlybyregion.Theinconsistenciesalsosuggestfurtherstudyof
69
thedynamicsandimpactofcertainvariablesondeforestation,includingcommunityforestry,land
tenuretype,indigenouspopulations,ruralagriculturalsupport,PES,andpoverty.Thenegative
relationshipbetweencommunityforestryanddeforestationisn’tintuitive;itinvitesfurther
researchintothesustainabilityandbiodiversityretentionofplantedforestsandtheirimpacton
primaryadjacentforestsovertime.Themixedrelationshipsofvariablesassociatedwith
communitylandmanagementinviteadeeperunderstandingofthequalitativedifferencesbetween
suchlandtenuretypesascommunidadesandejidosathouseholdandcommunitylevels,andhow
thesedifferencesimpactlocaltreecover.Theapparentregionaldiscrepancyofindigenous
populations’influenceontreecoveratthenationallevelandintheYucatanPeninsulasuggestsa
similarqualitativeinvestigation.ThecaveatstoPEShighlightedbytheliteratureandthemixed
meta‐analysisresultsforruralsupportprogramshighlighttheneedtounderstandtherelationship
betweenincomeanddeforestation,promptingresearchintotheadvantagesoftyingsupportfor
ruralincomestothemaintenanceofforestresourcesinhigherriskareas.Theregionaldifferences
oftheimpactofpovertyondeforestationsuggestthatperhapsdeforestationcannotbedirectly
attributedtopoverty,highlightingtheneedtounderstandthisdynamicwithinconcurrent
geographicalortemporaltrendssimultaneouslyaffectingdeforestation.
6.1.2.Nationalmodeling
Westatisticallyanalyzeddetailedspatially‐explicitdataonannualforestcoverlossesacross
allofMexicoover2000‐2012inrelationtovariationinestimatedgrossagriculturalrevenuesand
proxiesforfixedandvariablecostsusingobservablesitecharacteristics.Thegoalwastocapture
theinfluenceoftheeconomicnetbenefitsfromconvertinglandfromforesttonon‐forestusesfor
thepurposesofcalibratingapolicy‐simulationmodelthatcan,forexample,analyzetheimpactof
differentpolicystructurestocreateincentivesforlow‐emissionspracticesforreducing
deforestation.
Weaggregatedatafrom(Hansen,etal.,2013)tothe900mresolutionandmodel
deforestationinrelationtovariationinestimatedgrossagriculturalrevenuesandproxiesforfixed
andvariablecostsusingobservablesitecharacteristics.Wequantifytheeffectsofagricultural
revenueondeforestationinMexicobasedonhistoricaldata,andthensimulatedeforestationfor
alternativeagriculturalrevenuescenariosforourstudyperiod(2001‐2011).Wefurtherproject
futuredeforestation(2014‐2023)underabusiness‐as‐usualscenario,basedon2012conditions,
andalternativecarbonincentivesforpracticestoreducedeforestation.Theresultsfromthe
simulationprovideregionaldeforestationratesasaninputtotheLCMmodelingoftheseven
AATRs.
Theultimategoalofthisanalysisistohelpinformcost‐effectivepolicyapproachesto
reducedeforestation.Tohelpachievethisgoal,thereareseveralfutureextensionsofthis
research,includingfurthermodelvalidationandexploringtheimplicationsofadditionalvariables.
Thecurrentanalysisfocusedontheroleofeconomicreturnstocropproduction.Additional
analyseswouldbeneededtoidentifythespecificroleofdifferentlandownershipcategories,such
asdifferenttypesofejidosandothercommunallands,aswellastoexplicitlyidentifytheroleof
PROCAMPO,aswellasotheragriculturalandforestrydevelopmentprograms,alongwiththeroleof
existingconservationincentives.Withadditionalpotentialdata,wealsomightbeabletoconsider
changesbetweenforestsandotherland‐usesbeyondcropproduction.Intermsofthecarbon
70
emissionsreductionscostestimates,apriorityistoextendtheanalysistoincludebelow‐ground
carbonstocks,aswellasfurthercompareourestimateswithotherdatasources.Whilethis
analysisconsideredonlyforestlosses,animportantextensionwouldbetoconductananalysisof
thedatafromUMDandotherdatasourcesonforestgains.Thiswouldprovideamorecomplete
pictureoftheforestandcarbondynamicsinMexico.
Also,withadditionalcomputationalpower,wecouldimprovethespatialresolutionand
furtherrefinetheeconometricestimationtofurthermodelthespatio‐temporalprocessesdriving
deforestation.Wecouldexplicitlyestimateandconductsimulationsusingafullfixed‐effectsmodel,
aswellasalternativespatialpaneldatamodels,whichhelpcontrolforunobservedvariationsthat
onlyexistbetweenneighboringregions(orspatialautocorrelation),usingspatialweightingmatrix.
Thesevariationsmaynotbecontrolledforwithanon‐spatialpaneldatamodel.Anotherextension
wouldbetoexplicitlyconsiderthedynamicdecision‐processbasedonsurvivalmodeling
approaches.Inaddition,wecanimprovehowweconductsimulations,updatingthespatialpattern
ofthesurroundinglandscapeateachtimestep.
Inadditiontopossibleextensionsfortheunderlyinganalysis,amainpriorityforfuture
researchistousetheeconometricestimationtocalibrateaversionoftheOpenSourceImpactsof
REDD+Incentives(OSIRIS)modeltoanalyzealternativeREDD+andagriculturalpolicyscenariosin
Mexico.Thiswillrequirelinkingtheeconometricmodeltoageneralequilibriummodeltoaccount
forpossiblepricefeedbacks,whichcouldproducedeforestationshiftsor“leakage.”Thiswillalso
entailbuildinganopen‐sourceinterfacethatcanmakethemodeluser‐friendlyandmorebroadly
usable.Thesestepswillenablemorerealisticexaminationofalternativepolicydesignsforcreating
economicincentivesforpromotinglow‐emissionsagriculturaldevelopmentandreducing
deforestationinMexico.
6.1.3.LocalModeling
Theresultsofthedeforestationpredictionmapsprovidesinterestingusefulinformationon
areasmostvulnerabletotransitionandtheinclusionofthenationalmodelfordeterminingtherate
oftransitionswhichaccountfornationallevelpolicydecisions.ThecombinationoftheLCMand
nationalmodelprovidesasignificantimprovementtousingeithermodelinisolation.One
weaknessofusingtheLCMinisolationisthattherateofdeforestationiseithersolelybasedonthe
historicalratesorbasedonsubjectiveanalystopinion.Includingdeforestationratesderivedfrom
thenationalmodelprovidesaquantitativerationalforpickingaparticularratebasedonnational
andregionalpolicydecisions.Whileusingnationalmodelinisolationdoesnothavethefinespatial
resolutionthatwouldbenecessaryforlocalandsitelevelanalysis.
Thecombinationofthetwomodelingapproachescouldbefurtherstrengthenedby
applyingthesamefilteringandpre‐processingmethods.Thiswasnotdoneinthisstudyasthetwo
modelsweredevelopedinparallelusingtheinformationmostapplicableforeachindividual
approach.Thecohesionofthemodelscouldalsobeimprovedbycreatingtransitionpotentialsat
thereferenceregionlevelandapplyingtheratesfromthenationalmodelattheAARTscale,which
couldhelptomitigatetheunevenallocationofdeforestation,whichleadtonopredicted
deforestationwithin2oftheAATRsites.
71
Afinalconsiderationforstrengtheningtheallocationofdeforestationwouldbetousea
continuousornon‐discreteallocationofdeforestation.Meaningthatthedeforestationwouldbe
allocatedatasub‐pixellevel,inwhichallpixelsavailablefortransitionwouldbeassignedasmall
amountofdeforestationbasedontherelativevalueofthatpixel.Thedisadvantagetothismethod
isthatmapofpredicteddeforestationwouldnotbeaseasilyvisualized.Howeverthearea
estimateswithinAATRsorotherland‐usecategorieswouldbemorelikelytorepresentreality.
Thismethodrequiresfurtherresearch,butprovidesanalternativetomitigatingtheuneven
allocationofdeforestation,especiallyinsiteswithlowtransitionrates.
Westressthatthesemodeloutputscannotbeusedforreferenceemissionslevelsfor
projects.Thediscussionsectionindicatesseveraloftheissuesthatarisefromusingthesourcesof
dataandmethodsthatwehaveused,andsuggestssomepossiblefollow‐upanalysesthatcouldbe
doneasnextstepstofurtherexploretherelativethreatsofdeforestationamongthesesites.
However,themodelresultsprovidedheredostronglyindicateseveralfindings.First,somesites
haveverylittlehistoricaldeforestationandthreatoverthecomingdecade.Second,thereisa
consistenttrendforreductionsindeforestationratesamongallsitesifREDDwereimplementedat
a$10carbonpriceandwiththeotherassumptionsofthenationalOSIRISREDDmodel.Third,the
studyindicateswhichsiteslikelyhavethehighestmaximumpotentialforemissionsreductions.In
termsofREDDpotential,thesetwositeshavethegreatestifonlyconsideringthemaximum
emissionsreductionsobtainable,asbothwouldhavehighreferenceemissionslevels.ForOaxaca,
anadvantagemaybethattheworktoreduceemissionscanbeconductedinafairlysmallareaand
focussedonfewcommunitiesorejidos.InSierraPucc–LosChenes,whilethemaximumreduction
possibleisgreater,aREDDprojectwouldneedtoworkovermostofthesite,whichmayprove
muchmorecostlyandriskier.
72
7. Workscited
Alix‐Garcia,J.,2007.Aspatialanalysisofcommonpropertydeforestation.JournalofEnvironmnetal
EconomicsandManagement,Volume53,pp.141‐157.
Alix‐Garcia,J.,deJanvry,A.&Sadoulet,E.,2008.TheRoleofDeforestationRiskandCalibrated
CompensationinDesigningPaymentsforEnvironmental.EnvironmentandDevelopmentEconomics,
13(3),pp.375‐394.
Alix‐Garcia,J.M.,Shapiro,E.N.&Sims,K.R.E.,2012.ForestConservationandSlippage:Evidence
fromMexico'sNationalPaymentsforEcosystemServicesProgram.LandEconomics,88(4),pp.613‐
638.
Anderson,J.,2010.TheGravityModel.NBERWorkingPaperSeries,Issue16576.
Barsimantov,J.&Kendall,J.,2012.CommunityForestry,CommonProperty,andDeforestationin
EightMexicanStates.JournalofEnvironmentandDevelopment,pp.414‐437.
Blackman,A.,Albers,H.J.,Murphy,L.C.&Avalos‐Sartorio,B.,2008.LandCoverinaManagedForest
Ecosystem:MexicanShadeCoffee.AmericanJournalofAgriculturalEconomics,90(1),pp.216‐231.
Bonilla‐Moheno,M.etal.,2013.VegetationChangeandLandTenureinMexico:ACountry‐Wide
Analysis.LandUsePolicy,30(1),pp.355‐364..
Bray,D.B.,Ellis,E.A.,Armijo‐Canto,N.&Beck,C.T.,2004.Theinstitutionaldriversofsustainable
landscapes:acasestudyofthe.LandUsePolicy,Volume21,pp.333‐346.
Bray,D.B.etal.,2003.Mexico’scommunity‐managedforestsasaglobalmodelforsustainable
landscapes.ConservationBiology,17(3),pp.672‐677.
Busch,C.&Geoghegan,J.,2010.Laborscarcityasanunderlyingcauseoftheincreasingprevalence
ofdeforestationduetocattlepasturedevelopmentinthesouthernYucatanregion.Regional
EnvironmentalChange,Volume10,pp.191‐203.
Busch,J.etal.,2012.Structuringeconomicincentivestoreduceemissionsfromdeforestation
withinIndonesia.PNAS.
CamaradeDiputados,2008.ConstitucionPoliticadelosEstadosUnidosMexicanos.[Online]
Availableat:http://www.diputados.gob.mx/LeyesBiblio/ref/cpeum.htm
Carrillo,J.&Mota‐Villanueva,J.,2006.GuíaLegalparaDueñosdeBosquesenMéxico.,s.l.:WWF‐
México.
Cartus,O.etal.,2014.ANational,DetailedMapofForestAbovegroundCarbonStocksinMexico.
RemoteSensing,Volume6,pp.5559‐5588.
Chowdhury,R.,2006.DrivingForcesofTropicalDeforestation:TheRoleofRemoteSensingand
SpatialModels.SingaporeJournalofTropicalGeography,27(1),pp.82‐101.
Chowdhury,R.,2006.LandscapechangeintheCalakmulBiosphereReserve,Mexico:Modelingthe
drivingforcesofsmallholderdeforestationinlandparcels.AppliedGeography,26(2),pp.129‐152.
Corbera,E.etal.,2010.RightstoLand,ForestsandCarboninREDD+:InsightsfromMexico,Brazil
andCostaRica.Forests,2(1),pp.301‐342.
73
Deininger,K.&Minten,B.,2002.DeterminantsofDeforestationandtheEconomicsofProtection:
AnApplicationtoMexico.AmericanJournalofAgriculturalEconomics,84(4),p.943.
Deininger,K.W.&Minten,B.,1999.Poverty,Policies,andDeforestation:TheCaseofMexico.
EconomicDevelopmentandCulturalChange.EconomicDevelopmentandCulturalChange,47(2),pp.
313‐344.
Eastman,J.R.M.E.V.F.a.L.A.S.,2005.Transitionpotentialmodelingforlandcoverchange.In:M.
G.a.M.B.e.D.Maguire,ed.GIS,spatialanalysisandmodeling.Redlands,California:ESRIPress,pp.
357‐386.
Ellis,E.A.,2010.LandUse/LandCoverChangeDynamicsandDriversinaShallow‐GradeMarginal
CoffeeGrwoingREgionofVeracruz,Mexico.AgroforestrySystems,80(1),pp.61‐84.
Ellis,E.&Porter‐Bolland,L.,2008.Iscommunity‐basedforestmanagementmoreeffectivethan
protectedareas?:Acomparisonoflanduse/landcoverchangeintwoneighboringstudyareasof
theCentralYucatanPeninsula,Mexico.Forestecologyandmanagement,256(11),pp.1971‐1983.
FAO,2010.GlobalForestResourcesAssessment2010,Rome,Italy:s.n.
Ferretti‐Gallon,K.&Busch,J.,2014.WhatDrivesDeforestationandWhatStopsIt?AMeta‐Analysisof
SpatiallyExplicitEconometricStudies,WorkingPaper361:CenterforGlobalDevelopment,
Washington,DC.
FoodandAgricultureOrganizationofTheUnitedNations,2010.GlobalForestResourceAssessment,
s.l.:s.n.
Geoghegan,J.,Schneider,L.&Vance,C.,2004.TemporalDynamicsandSpatialScales:Modeling
DeforestationintheSouthernYucatánPeninsularRegion.GeoJournal,61(4),pp.353‐363.
Geoghegan,J.etal.,2001.ModelingtropicaldeforestationinthesouthernYucatánpeninsular
region:comparingsurveyandsatellitedata.Agriculture,Ecosystems&Environment,85(1‐3),pp.25‐
46.
Hansen,M.etal.,2013.High‐ResolutionGlobalMapsof21st‐CenturyForestCoverChange.Science,
15November,Volume342,pp.850‐853.
Hansen,M.etal.,2013.SupplementaryMaterialsforHigh‐ResolutionGlobalMapsof21st‐Century
ForestCoverChange.Science,342(850).
Honey‐Roses,J.,Baylis,K.&Ramirez,I.,2011.ASpatiallyExplicitEstimateofAvoidedForestLoss.
ConservationBiology,25(5),pp.1032‐1043.
IPCC5thAssessmentReport,2014.Chapter11:Agriculture,ForestryandOtherLandUse(AFOLU)..
WorkingGroupIII‐ClimateChange2014:MitigationofClimateChange..
Jaimes,P.,2010.ExploringthedrivingforcesbehinddeforestationinthestateofMexico(Mexico)
usinggeographicallyweightedregression.AppliedGeography,20(4),pp.576‐591.
Klepeis,P.&Vance,C.,2003.NeoliberalPolicyandDeforestationinSoutheasternMexico:An
AssessmentofthePROCAMPOProgram.EconomicGeography,79(3),pp.221‐240.
Lubowski,R.,2002.Determinantsofland‐usetransitionsintheUnitedStates:econometricanalysisof
changesamongthemajorland‐use,Cambridge,MA:s.n.
74
Lubowski,R.,Plantinga,A.&Stavins,R.,2006.Land‐usechangeandcarbonsinks:Econometric
estimationofthecarbonsequestrationsupplyfunction.JournalofEnvironmentalEconomicsand
Management,pp.135‐152.
Munoz‐Pina,C.,2008.PayingforthehydrologicalservicesofMexico'sforests:.Ecological
Economics,pp.725‐736.
Naidoo,R.&Iwamura,T.,2007.Global‐scalemappingofeconomicbenefitsfromagriculturallands:
Implicationsforconservationpriorities.BiologicalConservation,Volume140,p.40–49.
NationalGreenhouseGasInventory,2007.Mexico’sThirdNationalCommunicationtotheUNFCCC,
s.l.:s.n.
Nelson,G.C.&Hellerstein,D.,1997.DoRoadsCauseDeforestation?UsingSatelliteImagesin
EconometricAnalysesofLandUse.AmericanJournalofAGriculturalEconomics,79(1),pp.80‐898.
Nelson,G.&Geoghegan,J.,2002.Deforestationandlandusechange:sparsedataenvironments.
AgriculturalEconomics,Volume27,pp.201‐206.
Perez‐Verdin,G.,Kim,Y.S.,Hospodarsky,D.&Tecle,A.,2009.Factorsdrivingdeforestationin
common‐poolresourcesinnorthernMexico.JournalofEnvironmentalManagement,90(1),pp.331‐
340.
Petterson,J.,2003.TheWashingtonPost.[Online]Availableat:
http://www.washingtonpost.com/wp‐
srv/liveonline/03/special/nation/sp_nation_petterson050603.htm
[Accessed11August2014].
Plantinga,A.,1996.TheEffectsofAgriculturalPoliciesonLandUseandEnvironmentalQuality.
AmericanJournalofAgriculturalEconomics,78(4),pp.1082‐1091.
PROCEDE,2010.REgistroAgrarioNacional:PresidenciaDeLaRepublica,Mexico.[Online]
Availableat:http://www.ran.gob.mx/ran/programas_sustantivos/ran_procede.html
Rueda,X.,2010.UnderstandingDeforestationintheSouthernYucatan:InsightsfromaSub‐
Regional,MultiTemporalAnalysis.RegionalEnvironmentalChange,10(3),pp.175‐189.
SAGARPA,2012.ProgramadeApoyoalIngresoAgropecuarioPROCAMPOParaVivirMejor.[Online]
Availableat:http://www.sagarpa.gob.mx/agricultura/Programas/proagro/procampo/Resultados‐
Indicadores/Documents/2012/PROCAMPO_4Inf_trimestral_dic2012.pdf
[Accessed2September2014].
SAGARPA,2014.Procampo.[Online]
Availableat:
http://www.sagarpa.gob.mx/agricultura/Programas/proagro/procampo/Paginas/procampo.aspx
[Accessed20April2014].
SAGARPA,n.d.http://www.siap.gob.mx/.[Online]
Availableat:http://www.siap.gob.mx/
[Accessed15December2013].
75
SEDESOL‐CONAPO‐INEGI,2004.DelimitacióndelaszonasmetropolitanasdeMéxico,México,D.F.:
s.n.
SEMARNAT/INECC,2012.Mexico:QuintaComunicacionNacionalantelaConvencionMarcodelas
NacionesUnidassobreelCambioClimatico,D.F.,Mexico.:SecretariadeMedioAmbienteyRecursos
Naturales(SEMARNAT);InstitutoNacionaldeEcologiayCambioClimatico(INECC).
Stavins,R.&Jaffe,A.,1990.UnintendedImpactsofPublicInvestmentsonPrivateDecisions:The
DepletionofForestedWetlands.AmericanEconomicReview,80(337‐352).
TurnerII,B.L.etal.,2001.DeforestationinthesouthernYucatanpeninsularregion:anintegrative
approach.ForestEcologyandManagement,Volume154,pp.353‐370.
UnitedNationsFrameworkConventiononClimateChange,2009.Mexico:Challenges&
OpportunitiesforMitigationintheAgriculturalSector,s.l.:s.n.Vaca,R.G.D.C.L.H.J.S.M.,2012.
EvidenceofIncipientForestTransitioninSouthernMexico.PLoSONE,7(8).
vanderWerf,G.etal.,2009.CO2EmissionsfromForestLoss.NatureGeoscience,Volume2,pp.737‐
738.
Vance,C.&Geoghegan,J.,2002.TemporalandSpatialModellingofTropicaldeforestation:a
survivalanalysislinkingsatelliteandhouseholdsurveydata.AgriculturalEconomics,27(3),pp.
317‐332.
Vance,C.&Iovanna,R.,2006.AnalyzingSpatialHierarchiesinRemotelySensedData:Insightsfrom
aMultilevelModelofTropicalDeforestation.LandUsePolicy,Volume23,pp.226‐236.
Wooldridge,J.,2002.Econometricanalysisofcrosssectionandpaneldata.Cambridge,MA:MIT
Press.
76
www.alianza‐mredd.org.mx
ALIANZAMÉXICOPARALAREDUCCIÓNDE
EMISIONESPORDEFORESTACIÓNYDEGRADACIÓN
77
78