Mosquito-Borne Diseases: Advances in Modelling Climate-Change Impacts

2018·3.4 MB·English

by  Nils Benjamin Tjaden

Review
[436_TD$DIFF]
Mosquito-Borne Diseases: Advances in
Modelling Climate-Change Impacts
Nils Benjamin Tjaden,1 Cyril Caminade,2,3 Carl Beierkuhnlein,1,4,5 and
Stephanie Margarete Thomas1,4,*
Vector-bornediseasesareontheriseglobally.Astheconsequencesofclimate Highlights
change are becoming evident, climate-based models of disease risk are of Theuseofensemblesofdifferentcli-
growing importance. Here, we review the current state-of-the-art in both matemodelsforfutureprojections,as
wellasmultipledifferentmechanisticor
mechanisticandcorrelativediseasemodelling,thedatadrivingthesemodels,
correlativediseasemodelsperstudy,
thevectorsanddiseasescovered,andclimatemodelsappliedtoassessfuture isincreasing.
risk.Wefindthatmodellingtechniqueshaveadvancedconsiderably,especially
Communicating uncertainties related
in terms of using ensembles of climate models and scenarios. Effects of
to disease models, different climate
extreme events, precipitation regimes, and seasonality on diseases are still models,andemissionandpopulation
pathwaystoendusersisbecominga
poorly studied. Thorough validation of models is still a challenge and is com-
commonthingtodo.
plicated by a lack of field and laboratory data. On a larger scale, the main
challenges today lie in cross-disciplinary and cross-sectoral transfer of data Most models tend to project an
and [448_TD$DIFF]methods. increasedriskforvector-bornedisease
(VBD) transmission at high latitudes
and elevations during the upcoming
century.
Spatiotemporal Models of Vector-[436_TD$DIFF]Borne Diseases under Climate Change: An
Overview While mechanistic models typically
coverthewholechainofinfectionby
Modelling spatial patterns and temporal trends in vector-borne diseases (VBDs[452_TD$DIFF]; see
default, most environmental niche
Glossary) has been done through a diversity of approaches. This field of research is very [449_TD$DIFF]models (ENMs) still focus on vector
activeandshowsarapidmethodologicaldevelopmentwithregardtotheinclusionofvarious distributions alone; they are increas-
inglyappliedtowhole[450_TD$DIFF]diseasesystems
drivers of diseases.
aswell.
Modelsappliedinthisfieldarecommonlydividedintotwogroups.First,‘correlative’models
predict a species’ geographic distribution or support the understanding of why populations
persist at a certain place. Thereby, both vector and pathogen occurrences can be used as
response parameters. For this, various approaches, ranging from simple regression to
advanced machine-learning techniques,are employed [1].
1DepartmentofBiogeography,
Secondly, ‘mechanistic’ or process-based models make explicit assumptions about the UniversityofBayreuth,Germany
2InstituteofInfectionandGlobal
biological [2] and environmental mechanisms that drive species’ distributions or infection Health,UniversityofLiverpool,UK
dynamics.Inepidemiology,thesemodelsaremainlyderivedfromtheRoss–MacDonaldmodel 3NIHR,HealthProtectionResearch
UnitinEmergingandZoonotic
(compare e.g., [3]). These models are based on a system of differential equations depicting
Infections,Liverpool,UK
eachinfectiousstageforvectorsand/orhosts.Importantepidemiologicalparameters,suchas 4BayCEER,BayreuthCenterfor
vector biting rates, vector development, and mortality rates, and the extrinsic incubation EcologyandEnvironmentalResearch,
Bayreuth,Germany
period(EIP),largelydependonrainfallandtemperature.Theempiricalrelationshipbetween
5GIB,GeographischesInstitut
climateandtheseepidemiologicalparametersisderivedfromlaboratoryand,lessfrequently, Bayreuth,Bayreuth,Germany
fieldexperiments.
*Correspondence:
Theindividualstrengthsandweaknesses,aswellastheunderlyingparadigms,ofthesetwo
[email protected]
approacheshaveledtoheateddiscourse(compare[4]and[5]). However,bothapproaches (S.M.Thomas).
TrendsinParasitology,March2018,Vol.34,No.3 https://doi.org/10.1016/j.pt.2017.11.006 227
©2017ElsevierLtd.Allrightsreserved.
exhibitspecific qualities[2,6],andsome authorsexplorepromisinghybridapproaches(e.g., Glossary
[7–9]).
Areaunderthecurve(AUC):the
areaundertheROCcurve(see
Model approaches for VBD risk assessment need to consider both positive and negative below)iscommonlyusedfor
assessingamodel’sperformancein
aspects of altered climatic conditions across different spatial and temporal scales. Global
distinguishingbetween(inthis
warmingmayshiftclimaticallysuitableregionsforvectorestablishmentanddiseasetransmis- context)thepresenceorabsenceof
siontohigherlatitudesandhigherelevations.Conversely,itmaylimittransmissionofVBDinthe aspecies.AnAUCof1is
warmest places, where temperature thresholds for vector or pathogen survival may be considereda‘perfect’model,whilea
valueof0.5indicatesthatthemodel
exceeded [10,11].
isnotbetterthanarandomguess.
Environmentalnichemodel
Expectations for long-term climatic trends are mostly robust, particularly as far as average (ENM):amodelthatestimatesthe
conditionsinairtemperatureareconcerned.Projectionsonfuturedevelopmentofmedium- ecologicalniche(oraspectsthereof)
ofaspeciesbasedonthe
termvariability[453_TD$DIFF][manifestedinclimaticphenomenasuchastheNorthAtlanticOscillation(NAO)
environmentalconditionsatlocations
or the El Niño Southern Oscillation (ENSO[454_TD$DIFF])] are very uncertain [12]. This is a challenge, wherethespeciesisknowntoexist.
becauseinterannualandevenmultidecadalclimaticfluctuationsareaffectingVBDtransmis- Itcanbeusedtoexaminespecies–
environmentrelationshipsorasa
sion in some parts of the world [13]. Furthermore, long-term climatic trends affect the
speciesdistributionmodel(SDM)in
probabilityofextremetemperatureandrainfallevents,makingthemlessrareinoccurrence
ordertopredict(changesin)species
andmoreelusive[14].Eventhoughheatorcoldwaves,drought,orfloodingareimportantfor occurrencesinspaceandtime.
diseaseemergence,vectorabundance,andpathogentransmissiondynamics,suchevents Extrinsicincubationperiod(EIP):
thetimethatneedstopassaftera
can hardly be predicted [15].
vector’sinfectiousbloodmealbefore
itcantransmitthepathogento
Inthisarticlewereviewrecentadvancesinmodellingtheimpactsofclimatechange(Box1)on anotherhost.
VBD, providing an overview of the literature published since 2014. We discuss primarily Globalclimatemodelsorgeneral
circulationmodels(GCMs):
mosquito-bornediseasesthataremonitoredbytheEuropeanCentreforDiseasePrevention
modelsthatareusedtosimulatethe
and Control (ECDC),focusingon theapplied models aswell asonthe datadrivingthem. earth’sclimateonalargescale(see
Box2fordetails).
Correlative Models IntergovernmentalPanelon
ClimateChange(IPCC):theIPCC
Environmentalnichemodels(ENMs),andtheirspatialapplicationasspeciesdistribu-
definesitselfasthe‘international
tionmodels(SDMs),havebecomeanintegraltoolinthefieldsofbiogeography,ecology, bodyforassessingthescience
andconservationbiology.Differentmodellingtoolsandalgorithmswithindividualstrengths relatedtoclimatechange’.Its
andweaknessesexist,butthegeneralconceptremainsthesame(Figure1).First,locations assessmentreportsaimtomakethe
state-of-the-artinclimateresearch
ofatargetspeciesarecollectedinthefieldorderivedfromexistingdata.Somemodelling
accessibleforpolicymakersand
toolsrequireknowledgeofthelocationswheretargetspeciesaretrulyabsent,butasthisis providethescientificbasisfortheUN
difficult to acquire for various reasons [1], pseudo-absence or background data are ClimateConferences.
commonly generated instead [16]. The difficulty of finding high-quality presence/absence Receiveroperatingcharacteristic
(ROC):theROCcurveisusedwhen
dataofvectorsorpathogensisanimportantlimitationforcorrelativenichemodels.Spatial
acontinuousmodeloutput(e.g.,
data representing environmental parameters relevant to the species in question (such as probabilityofpresenceofaspecies)
climate, land use, soil type etc.) are acquired, usually in the form of continuous grids istranslatedintobinaryinformation
(e.g.,presence/absenceofthe
covering the study area.
species).Itillustrateshowtheratioof
truevs.falsepositivesvarieswith
differentthresholdsforthedistinction
Box1.ClimateChangeinEurope
betweenpositiveandnegative.
Duringthe20thcentury,mostofEuropeexperiencedanincreaseinannualsurfaceairtemperatureofabout[442_TD$DIFF]0.8(cid:1)C, Regionalclimatemodels(RCMs):
mostlywithastrongerwarminginwinterthaninsummer.Atthesametime,somepartsofsouthernEuropehavedried modelsthatcanbeseenas
byasmuchas20%whileprecipitationincreasedby10–40%overnorthernEurope[114].Warmnightanddaytime refinementsofGCMsthatareableto
temperatureextremesincreased,coldtemperatureextremesdecreased,andmanyregionsarefacedmorefrequently betterreflectlocalconditionson
withheavy-raindays[115].Expectationsforthefuturevarybyregionandseason.Whiletemperaturesaregenerally smallerspatialscales(seeBox2for
expectedtoincreaseacrossthecontinent,thiswillbemorepronouncedinthesummerinsouthernEurope(FigureIA) details).
andinthewinterinnorthernandeasternEurope(FigureIB).Projectionsforchangesinprecipitationaresubjectto Representativeconcentration
relativelyhighuncertainties.Thegeneraltrends,however,arereducedrainfallinthesouthandincreasedprecipitationin pathways(RCPs):RCPssucceed
thenorth.Atintermediatelatitudes,thereareopposingeffectsinthedifferentseasons,withdryersummersandwetter theolderSRESscenarios(see
winters(FigureIC,D). below).
228 TrendsinParasitology,March2018,Vol.34,No.3
Theseclimaticchangeshaveanimpactonvectors’habitats.Winterwarmingmaypromoteoverwinteringofvectors. SEIR/SIR:susceptible,exposed,
Increased precipitation could lead to increased habitat availability due to increased soil moisture, humidity, and infectiousandrecoveredarethe
availabilityofnaturalponds.Extremefloodingcanleadtothedestructionofvectors’habitats,throughflushingof stagesofaninfectionanindividual
stagnantwaterbodies,butatthesametimeitcancreatenewbreedinggroundswhenthewaterrecedes[116].Lower cantypicallygothrough.Theymake
summerrainfallintheMediterraneancouldmakesuitablebreedingsitesscarce(cid:3)orhavetheoppositeeffectifitleadsto upthedifferentcompartmentsofa
moreopencontainersbeingusedforwatersupplyandirrigation. typicalmechanisticdiseasemodel.
SpecialReportonEmissions
Scenarios(SRES):thisreport,
publishedbytheIPCC,introduceda
(A) (B)
rangeofscenariosforhowemissions
ofgreenhousegasesmaychangein
thefuture,dependingonhow
mankindreactstothechallengesof
climatechange.Thesescenarios
havebeenthebasisforIPCC
assessmentreports,policymaking,
andclimatemodelling,butbynow
havebeensupersededbythe
representativeconcentration
pathways(RCPs).
Speciesdistributionmodel(SDM):
SDMsareusedtoestimatethe
1 1.25 1.5 1.75 2 2.25 2.5 2.75 3 1 1.25 1.5 1.75 2 2.25 2.5 2.75 3
°C °C geographicaldistributionofaspecies
(orothertaxonomicrank).Theyare
(C) (D)
often,butnotalways,basedonan
environmentnichemodel(ENM).
Vector-bornediseases(VBDs):
illnessesinhumansorother
vertebratesthataremainly
transmittedbyotheranimals(cid:3)often
bloodsuckinginsectssuchas
mosquitoes.
-250 -200 -150 -100 -50 0 50 100 150 200 250 -250 -200 -150 -100 -50 0 50 100 150 200 250
mm mm
FigureI.ClimateChangeinEurope.Simulateddifferencesintemperature(A,B)andrainfall(C,D)between2065and
2085,andbetween1961and1999,undertherepresentativeconcentrationpathway(RCP)4.5scenariobasedon16
globalclimatemodel(GCM).A,C.Borealsummer(June–August).B,D.Borealwinter(December–February).Cross
hatchingindicatesareasofhighuncertaintyduetolowagreementofthedifferentGCMs.
In the second step, a multivariate regression model is created. From an ecological per-
spective, loosely following the Hutchinsonian niche concept [17], this can be seen as
constructingavirtualspacerepresentingallpossiblecombinationsofvaluesofthechosen
environmental parameters. From the location of the presence (and, when available,
absence)ofrecordswithinthisenvironmentalspace,theenvironmentalnicheofthetarget
species is constructed using methods ranging from simple multiple linear regression
models to advanced machine-learning techniques. This model of the species’ preferred
environmentalconditionsisthenprojectedbackintogeographicalspace,producingamap
depictinghowsuitabletheenvironmentalconditionsareforthespeciesineachgridcellof
the study area. Since functional features, such as dispersal barriers, cannot be directly
included in this kind of model, environmental suitability cannot be easily translated into
probability of occurrence. Instead, it should be perceived as an indicator for a particular
species’abilitytosurviveatagivenlocationifsomeindividualsweretoreachthisplace.On
TrendsinParasitology,March2018,Vol.34,No.3 229
(A)Occurrences (B) Predictors
pp11
p1 p2 p3
(D)Spa(cid:2)al projec(cid:2)on and model valida(cid:2)on
(C) Model fit
1.0
0.8
p1
0.6
0.4
0.6
0.2
y0.0
bilit −10 0 10 20 30 40°C
(E) Climate scenario al suita00..46 p2
ent0.2
mperature change 6420....0000 °°°°CCCC POeps(cid:2)smimisis(cid:2)(cid:2)cc 222000333000sss 222000337000sss Environm00100.....08046 −40 −20 0p320°C
Te-2.0 °C 00..20
1950 2000 2050 2100 0 2000 6000mm
Year
(F) Spa(cid:2)otemporal change
Op(cid:2)mis(cid:2)c 2030s Op(cid:2)mis(cid:2)c 2070s Pessimis(cid:2)c 2030s Pessimis(cid:2)c 2070s
Figure1.CharacteristicWorkflowofanEnvironmentalNicheModel(ENM).(A)Occurrencerecordsforthe
vector,host,ordiseaseinquestionareacquired.(B)Asetofpredictorvariablesselected(hereexemplarily:p1=summer
temperaturein(cid:1)C,p2=wintertemperaturein(cid:1)C,p3=annualprecipitationinmm).(C)Basedonthesefactors,thebestfit
model that describes the probability of occurrence of the species in (multivariate) dependence to its environmental
conditions is developed (here simplified: environmental suitability in dependence of p1–3). Ideally, several different
algorithmsareutilized.(D)Aspatialprojectionofthemodelismadebasedonthepredictorvariables.(E)Asetoffuture
climatechangescenariosandrelevanttimeframesischosen.Shownhereistheobservedpastandexpectedfuture
averagetemperaturechangeovertimeforanoptimistic(blue)andapessimistic(red)climatechangescenario.Color
shadingsaroundtheblacklinesshowanestimateoftheuncertainties.(F)Usingdatafromglobalorregionalclimate
models,furtherprojectionsfortheselectedscenariosandtimeframes(greyverticalbarsinpanelE)aremade.Ideally,
differentclimatemodelsareusedtodrivetheENM.
230 TrendsinParasitology,March2018,Vol.34,No.3
smallerspatialscales,itmaybefeasibletoconductextensivefieldcollectionsofspeciesthat
allow for an estimate of abundance rather than just simple presence or absence. Based on
thesedata,speciesabundancemodelscanbecreatedinaverysimilarfashion(e.g.,[18]).
Inathirdstep,thepreparedmodelcanbeusedtoidentifythespecies’potentialtobecome
establishedinotherpartsoftheworldbyusingthesamesetofenvironmentalpredictorsbutfor
adifferentregionasareferencefortheprojection.Similarly,projectionsovertimecanbemade
byusingenvironmentaldatathatfollowhistoricalemissionsforthepastoremissionscenarios
forthefuture[455_TD$DIFF][suchasglobalclimatemodels(GCMs)orregionalclimatemodels(RCMs);
Box 2[456_TD$DIFF]].
Inepidemiology,ENMsarecommonlyusedtomapthepotentialdistributionofvectorspecies
(Table1).Forsimplediseasesystems,thisalonecangiveareasonablygoodestimateofregions
that could be affected by pathogen transmission [1], although abundance models should be
preferred whenever possible due to the more differentiated picture they provide. For more
complexsystems,suchasthoseconsistingofmultipledifferenthosts,reservoirsand/orvectors,
focusing on a single target species is often insufficient, asdifferent species are likely to have
different environmental requirements and competence to transmit diseases. In this particular
case,thepotentialdistributionsofthedifferentspeciesinvolvedcanbemodelledasseparate
components.Theexpectedgeographicalrangesofthesespeciescanthenbeoverlaidinorderto
deriveareasofelevatedriskoftransmission[19].Fordiseaseswheretheinvolvedspecies,their
contributiontodiseasetransmission,ortheirspatialdistributionareunknown,thedevelopmentof
anENMbasedonobservedoccurrenceofthediseasecanbehelpful[1].Inaway,thisapproach
considersapathogenanditstransmissionrangeasaspeciesanditsestablishedpopulations.
Regarding ectotherm arthropod vectors, this approach has the advantage of being able to
additionallyaccountforthermalimpactsonthepathogenitself,suchasthetemperature-depen-
denceoftheEIPobservedforseveralviraldiseases[20].
While free and open-source software packages (like dismoi[446_TD$DIFF] or biomod2ii for R; see https://
www.r-project.org/)makethedevelopmentofENMsrelativelyeasyfromatechnicalpointof
view.Thereareseveralaspectsthatneedtobeconsideredcarefullyinordertogainmeaningful
results [1,21].Theseinclude,forinstance,sampling biasintheoccurrencerecords[22], the
regionswherepseudo-absencelocationsaredrawnfrom[23],potentialniche-shiftsofinvasive
species[24],and thechoiceof meaningfulenvironmentalpredictors[25,26].
Thorough out-of-sample validation of a model is crucial; and there are numerous evaluation
methods available for different kinds of ENM. While these evaluation methods have been
reviewedelsewhere[27],itseemsworthpointingoutthattheareaunderthecurve(AUC)of
the receiver operating characteristic (ROC), one of the most widely applied evaluation
metrics, hasbeencriticized forbeing potentiallymisleading(e.g., [28]).
Among the techniques available, Maxent [29] has been by far the most popular choice for
studiesofclimatechangeimpactonvector-bornediseasesoverthepastfewyears(Table1).
Thisissomewhatsurprisingasthereareseveralotherestablishedmethodsavailable[458_TD$DIFF][suchas
Bioclim, Boosted Regression Trees (BRT), Random Forest (RF), Generalized Linear Models
(GLM), Generalized Additive Models (GAM), or Genetic Algorithm for Rule-set Production
(GARP)] and from the numerous studies comparing their performances (e.g., [30–32]) no
preferential method has emerged so far. Consequently, there is a new trend towards using
an ensemble of different ENMs to make up for the uncertainties inherent to the individual
algorithms [31,33].
TrendsinParasitology,March2018,Vol.34,No.3 231
Box2.ClimateModelsandScenarios
Generalcirculationmodels(GCMs),alsocalledglobalclimatemodels,areusedtosimulatetheearth’sclimateatlarge
spatialscalesandestimateitslong-termfuturedevelopment.Theyusuallyconsistofseveralcoupledcomponentssuch
astheatmosphere,oceandynamics,seaice,andvegetation.Internationaleffortsinclimatemodellingarecoordinated
throughtheCoupledModelIntercomparisonProject(CMIP,http://pcmdi-cmip.llnl.gov),andGCMoutputdataare
madeavailablethroughthevariousdatanodesoftheEarthSystemGridFederation(ESGF,https://esgf.llnl.gov).Since
runningGCMsimulationsiscomputationallyexpensive,somefiner-scaleprocesses,suchasconvection,havetobe
heavilyparameterizedinglobal-scalemodels.However,regionalclimatemodels(RCMs)drivenbyaGCMcanbeused
tomakeupforthisonsmallerspatialscales.TheCoordinatedRegionalDownscalingExperiment(CORDEX,http://
www.cordex.org)providesacommonframeworkforsuchinitiatives.
TheIntergovernmentalPanelonClimateChange(IPCC)providesthescientificbasistoassessclimatechange,suggest
adaptationandmitigationstrategies,andhighlightimpactsandfuturerisksfordecision-makers.TheIPCCassessments
arecompiledbyhundredsofleadingandvolunteeringscientists,theyundergomultipleroundsofreviewtoensure
objectivity, and underlie negotiations at the Conferences of the Parties (COP) of the United Nations Framework
Conventionon Climate Change(UNFCCC). Inthe latestassessment, theFifthAssessment Report in 2013, new
climate-changescenarios,so-calledrepresentativeconcentrationpathways(RCPs),weredeveloped.Theydescribea
widerangeofpossiblemagnitudesofclimatechangebyspecifyingconcentrationsandcorrespondingemissions.
Although not directly based on socioeconomic storylines like the former IPCC Special Report on Emissions
Scenarios(SRES),theyareadditionallybasedonshort-livedgasesandland-usechanges[117].Thestart-point
forallfourRCPscenariosis2006,withabaselinehistoricalperiodfrom1986to2005.TheseRCP[443_TD$DIFF]xscenariosleadtoa
definedadditionalradiativeforcingby2100(increaseby[444_TD$DIFF]xW/m2[441_TD$DIFF]),whichcanalsobeexpressedasanincreaseinthe
globalmeansurfacetemperaturesfor2081–2100.ThisincreaseforthedifferentRCPsisexpectedtorangebetween
[445_TD$DIFF]0.3(cid:1)Cand1.7(cid:1)C(RCP2.6),1.1(cid:1)Cand2.6(cid:1)C(RCP4.5),1.4(cid:1)Cand3.1(cid:1)C(RCP6.0),2.6(cid:1)Cand4.8(cid:1)C(RCP8.5)[118].
Bothglobalandregionalclimatemodelsrelyonthesepathwaysforfutureprojectionsofclimatechange.
MostENM-typemodelsappliedindiseasemodellingfocusonvectordistributions(Table1).
Among these, the most studied vector genus is Aedes (competent mosquito vector for
diseases such as dengue, chikungunya, or Zika) followed by Anopheles (malaria mosquito
vectors)andLutzomyia(sandflies,vectorsofleishmaniasis).WhileseveralENM-typemodels
for complete disease systems have been published recently [7,34–36], only a few of them
featurefutureprojectionsunderclimate-changescenarios[20].Theprojectedfuturechangesin
vector ranges vary among species and regions. However, there is a general trend of range
expansion towards higher latitudes and altitudes, while some of the regions that are most
affected by VBD today may benefit from a decline in environmental suitability under climate
change (Table1).
ENM-typemodelsarecommonlyappliedacrossallspatialscales.Whenitcomestofuturerisk
mapping, however, they are mostly used on larger global to continental scales (Table 1).
Consequently,moststudiesuseglobalratherthanregionalclimatemodels.Almosthalfofthe
studiesinTable1incorporatedatafrommorethanoneclimatemodel.Thisisgoodpractice,as
this leadsto betterestimatesof uncertainty in finalmodeloutput[37].
Regardingthepredictorsbeingused,mostmodelsrelymainlyonvariousmetricsappliedto
temperatureandprecipitation(andcombinationsthereof,suchas‘precipitationofthewarmest
month’), both of which have been identified as important drivers of VBD transmission [38].
Some models additionally use other input parameters that may influence host or vector
distribution, such as measures of air moisture [39–42], soil moisture [43], topography [42–
45], or land cover/land use [42,43,46]. Socioeconomic factors, such as human population
density or vulnerability indicators, can be included as well (e.g., [7]), but continuous future
projections of theseare oftennot available,andthey are subjectto largeuncertainties.
OnemainadvantageofENMscomparedtomechanisticapproachesisthenon-necessityof
detailed knowledge about the complex interplay between environment, vectors, hosts, and
232 TrendsinParasitology,March2018,Vol.34,No.3
Table1.RecentStudiesUsingEnvironmentalNicheModels(ENMs)toAssessVector-[436_TD$DIFF]BorneDiseaseRiskunderClimateChange
Vector/pathogen ENM Climatemodela/scenariob/ Environmental Country,region Mainfindings Refs
modelled futuretimeperiodc variables
Aedesaegypti MaxLike NAd/RCP4.5,RCP8.5/ te,pf Veracruz,Mexico Datafromtheedgesofthevector’sdistributionis [97]
current,2020s,2080s valuablefor
-monitoringchangesindistribution
-understandinglinksbetweenanthropogenic
driversandclimatechange
Climateenvelope GCM:CCCma-CGCM2, t,p Global -macroclimateisthemaindriverofthespecies [98]
CSIRO-MK2,NIES99,UKMO- rangelimits
HadCM3/A2a,B2b/2020s -anthropogenicinfluencecanhelpthespeciesto
surviveinotherwiseunsuitableclimate
Maxent NA/A2a/2050s t,p Brazil -thevector’srangeinBrazilwilldecreaseinthe [99]
future,butwillspreadfurthersouth
Maxent GCM:NA/NA(CMIP5)/2020s, t,p Tanzania -riskfordengueiscurrentlyconcentratedinthe [100]
2050s coastalareas
-large-scalespreadisprojectedfor2050s
Aedesaegypti,Aedes Biomod2 GCM:HadGEM2-ES/RCP8.5/ t,p,NDVIg Global -Zika’sdistributionmaybefarmoreconstrained [80]
africanus,Aedesalbopictus ensemblemodel 2050s thandengue
-Zikaisunlikelytobecomecosmopolitanin
temperateregions
Aedesaegypti,Aedes Maxent 6GCM/B1,A1B,A2/2050s t,p Global -complexglobalrearrangementsofpotential [101]
albopictus distributionalareasunderclimatechange
-digitizationandsharingofexistingdistributional
dataforvectorsneedstobeapriority
Aedesalbopictus GARP GCM:MPI-ESM-LR/RCP4.5/ t,p Mexico,US, -ENMfitonoccurrencedatafromdifferentregions [102]
2050s,2070s Italy,Brazil,Asia transfertoMexicowell
Maxent RCM:COSMO-CLM/A1B/ t,p,cargo Europe -combiningENMwithmeasuresofcargo [103]
2020s,2050s,2080s movement movementcanhelptoidentifyhotspotsfor
potentialareasofintroduction
T -availabilityoftransportdatainEuropeneedstobe
ren improved
d
s
in Maxent GCM:CSIRO-Mk3.6.0/RCP t,p Germany -establishmentinGermanyispossible [104]
P
ara 2.6,4.5,6.0,8.5/ -northwardrangeexpansionunderclimatechange
sito 2030s,2050s,2070s
lo
gy, Anophelesarabiensis LOBAG-OC GCM:HadleyCM3/A1B,A2A, t,p Africa -thesuitablerangeforthevectorinAfricawillbe [105]
M B2A/2050s stronglyreducedunderclimatechange
a
rc
h2 Anophelesdarlingi,Anopheles Maxent GCM:GISS-E2-R,HadGEM2- t,p, SouthAmerica -vectorsareprojectedtoexperiencerange [43]
018 nuneztovari AO/RCP2.6/2050s,2070s topoh,soilmoisture, expansionunderclimatechange
,V popi,lcovj
o
l.
34 Anophelesspp. Maxent,BRT GCM:GISS-E2-R,HadGEM2- t,p, SouthAmerica -currentmainvectorwillexperiencereduced [44]
,
No ES/RCP8.5/2070s topo,terrestrial habitatsuitabilityunderclimatechange
.3 biomes -otherspeciesofthegenusshowsignificant
2 potentialforexpansion
3
3
2
3 Table1.(continued)
4
Vector/pathogen ENM Climatemodela/scenariob/ Environmental Country,region Mainfindings Refs
T
ren modelled futuretimeperiodc variables
d
s
in Anopheles,An.dirus,An. Maxent GCM:BCC-CSM1-1, t,p,lcov China -thedifferentvectorspecies’rangeswillreact [46]
P
a minimus,An.lesteri,An. CCCma_CanESM2,CSIRO- differentlytoclimatechange
ra
sito sinensis Mk3.6.0/RCP2.6,4.5,8.5/ -anoverallnetincreaseinthepopulationexposed
lo 2030s,2050s tothevectorsisexpected
g
y,
M Culicoidesimicola,C.insignis, Maxent 62GCM/RCP2.6,4.5,6.0, t,p Global -potentialdistributionisprojectedtobroaden [106]
a
rc C.variipennis,C.sonorensis, 8.5/2050s underclimatechange,especiallyincentralAfrica,
h
20 C.occidentalis,C.brevitarsis UnitedStatesandwesternRussia
1
8
,V Culicoidesimicola CLIMEX GCM:CSIRO-MK3.0,Miroc- t,p,rhk[437_TD$DIFF],irrigation Global -vector’spotentialdistributionunderclimate [41]
o
l. h/A1B,A2/1975,2030s, changeisprojectedtoexpandnorthwardinthe
3
4, 2070s northernhemisphere
N
o. -potentialdistributionmaycontractinAfrica
3
Culicoidessonorensis Maxent GCM:CanESM2/RCP2.6, t,p,topo,lcov,VPDl NorthAmerica -thecurrentnorthernrangelimitofthevectoris [42]
4.5,8.5/ expectedtoshiftnorthwardunderclimatechange
2030s,2050s
Lutzomyiaevansi, Maxent GCM:CSIRO/A2,B2/2020s, t,p, Colombia -therangeofthevectorsisprojectedtodecrease [45]
Lutzomyialongipalpis 2050s,2080s topo insizeunderclimatechange
Lutzomyiaintermedia, GLM,MaxEnt, GCM:HadGEM2-ES/RCP t,p SouthAmerica -thedifferentvectorspeciesshowadifferent [81]
Lutzomyianeivai RF,SVM,GARP 4.5,8.5/2050s responsetoclimatechange
-‘Ecologicalnichemodelsshouldbespecies
specific,carefullyselectedandcombinedinan
ensembleapproach.’
Lutzomyiaflaviscutellata 6SDM 17GCM/RCP4.5,8.5/2050s t,p [N/n]orthern -thesuitableareaforthevectorisprojectedto [82]
SouthAmerica expandtowardshigherlatitudesandaltitudes
underclimatechange
Lutzomyiamajor, biomod2 GCM:MPI-ESM-LR/RCP4.5/ t,p Libya -coastalregionsofLibyashowhigherriskbecause [107]
Lutzomyiatropica ensemblemodel 2050s ofmoresuitableclimate
-riskofcutaneousleishmaniasisisprojectedto
increaseunderclimatechange
Chikungunya Maxent 5GCM/RCP4.5,8.5/2030s, t,p,pop Global -transmissionpotentialisprojectedtoincrease [20]
2050s,2070s acrosstheglobeunderclimatechange
-somepartsofIndiamayseearelativedecreasein
transmission
aNamesoftheclimatemodelsbeingused(RCMorGCM,seeGlossaryandBox1),unlesstheirnumberexceeds5.
bClimatechangescenario:typicallyRCPsand/orscenariosfollowingSREP,seeGlossary.
c2030setc.=marksthecenterofthe30-yeartimeperiodcovered,standsfor2021–2040or2020–2039dependingondatasource.
dNA=informationnotavailable.
et=temperature.
fp=precipitation.
gNDVI=normalizeddifferencevegetationindex.
htopo=topography(elevation,altitude,slope,aspectratio).
ipop=humanpopulationdensity.
jlcov=landcover,landuse.
krh=relativehumidity.
lVPD=vapourpressuredeficit.
pathogens[1].Thismakesthemapracticaltoolforunderstudied,thatis,‘neglected’VBDs.
However, this comes at the price of accuracy, and consequently ENMs are most useful on
mediumtolargespatialscales.Ifatleastsomeoftheenvironment-dependentmechanismsare
known,thosecanbeusedtorefinetheresults[47].Andfinally,estimatesofdistributionsor
abundancesofhostandvectorspeciesderivedfromENMscanalsoserveasinputdatafor
mechanisticmodels [9].
Mechanistic Models
Mechanistic models are built on biophysical relationships between environmental factors
vectors, pathogens, and hosts (Figure 2). These relationships are generally derived from
laboratory-orfield-basedstudies(seealsosection‘AdaptionandEvolution’).Differentmecha-
nistic approaches can be applied. The most common methodology is derived from the
standardRoss–MacDonaldmodel[3]oritsgeneralization.Asetofdifferentialequationsdefine
thedifferentcompartmentalstagesofvectorsandhosts(susceptible,exposed,infectious,and
recoveredforSEIRmodels,orsusceptible,infectious,andrecoveredforSIRmodels).Thisset
ofequationscanbedirectlyutilizedtomodelthepopulationsizeineachcompartment,based
on their relationship to climatic factors [48]. The steady-state solution of this system of
differentialequationsalsoyieldsthebasicreproductionnumber,R R iscommonlyemployed
0. 0
inepidemiologytoestimatethepropensityofanoutbreaktoexpand(R >1)ortoshrink(R <1)
0 0
inafullysusceptiblepopulation.MathematicalformulationsofR areavailableforseveralVBDs.
0
Theydependonthenumberofvectorsandhostsconsideredinthemodel;see,forexample,
[49] for a one-host–one-vector formulation of R for malaria, [50] for a two-host–one-vector
0
formulationofR forAfricantrypanosomiases,or[51]foraone-host–two-vectorformulationof
0
R forZika.Otherempiricalmechanisticmodelsarebasedonenvironmentalriskfactors,such
0
asfuzzylogicmodelstosimulatetheriskofmalaria[52],orempirical-rule-basedmodels,to
assess the risk of helminth infections in ruminants based on soil moisture availability and
temperature conditions [53].
MechanisticmodelscanbeutilizedtomodeltheriskofVBDsbackwards(usingpastenviron-
mental data) or forwards (using demographic, economic, and climate-change scenarios) in
time. They are generally driven by daily or monthly climate data to simulate the burden of a
particularVBD.Theircomplexityvaries;somemodelsincludeadditionaleffectsofpopulation
density,surface hydrology [54],and herdimmunity factors[55].
In terms of methodology, the first necessary step (which is common to all modelling
approaches) is model validation. For this purpose, mechanistic models are run for the past,
andtheoutputiscomparedtoobservedVBDburdenindicatorsinspaceandtime.Thiscanbe
adauntingtaskasthisstepdependsonthequalityandspatiotemporalcoverageofobserved
diseaseburdeninformation(prevalence,incidence,numberofconfirmedcasesetc.).Different
skillscores(likeAUC,correlations,orreliabilitydiagrams)areemployedtoestimatethemodel
capability in reproducing past observed outbreaks and mean seasonality of a VBD. The
mechanisticmodelisthenprojectedforwardinspaceandtime,usingcalibratedclimatemodel
data outputs and population scenarios, to estimate future human populations at risk (see
[56,57] formalaria).
AnothersignificantprogressliesinthestudyofhistoricalVBDoutbreaksandtheirrelationship
withclimatevariability.AnR modelshowedoptimalclimaticconditionswhenanoutbreakof
0
bluetongueoccurredinnorthernEuropein2006[58].Asimilarmodellingframeworkhighlighted
optimalenvironmentalconditionsformosquito-bornetransmissionriskofZikavirusoverSouth
Americain2015,whenthelargestoutbreakoccurred[51].Thesefindingsareconsistentwith
TrendsinParasitology,March2018,Vol.34,No.3 235
(A)Dynamical model framework
> 1 epidemic expands
< 1 epidemic shrinks
(B)Epidemiological parameters
0.30 a(T) [p.day] 115800 EIP(T) [days] 01..80 μ (T) 23..00 (T)
0.20 120 2.0
0.6
90 1.5
0.4
0.10 60 1.0
30 0.2 0.5
0.00 0 0.0 0.0
0 10 20 30 40 0 10 20 30 40 0 10 20 30 40 0 10 20 30 40
(D)Climate scenario (C)Integra(cid:2)on + valida(cid:2)on
6.0 °C Pessimis(cid:2)c 222000333000sss 222000337000sss
mperature change420...000 °°°CCC Op(cid:2)mis(cid:2)c
Te
-2.0 °C
1950 2000 2050 2100
Year
(E) Future maps - >1

Op(cid:2)mis(cid:2)c 2030s Op(cid:2)mis(cid:2)c 2070s Pessimis(cid:2)c 2030s Pessimis(cid:2)c 2070s
Figure2.TypicalWorkflowofaMechanisticDiseaseModelDerivedfromtheRoss–MacDonaldFramework
[R0(T)Model].(A)Dynamicalmodelframework.T=temperature((cid:1)C);b=vector–hosttransmissionprobability;b=host-
–vectortransmissionprobability;m=vector-to-hostratio;r=recoveryrate;d=infectiousrecoveryrate;a(T)=vector
bitingrateperday;EIP(T)=1/y(T)=extrinsicincubationperiodindays;m(T)=vectormortalityrate.(B)Epidemiological
parametersderivedfromlaboratoryexperimentsorfielddataarefedintothemodeltogainanestimateofR(T).(C)Arisk
0
mapisderivedfromthemodel.(D)Asetoffutureclimate-changescenariosandrelevanttimeframesischosen.Shown
hereistheobservedandexpectedfutureaveragetemperatureincreaseovertimeforanoptimistic(blue)andapessimistic
(red)climate-changescenario.Colorshadingsaroundtheblacklinesshowanestimateoftheuncertainties.(E)Usingdata
fromglobalorregionalclimatemodels,furtherprojectionsfortheselectedscenariosandtimeframes(greyverticalbarsin
panelE)aremade.Ideally,differentclimatemodelsareusedtodrivethemodel.
former results by Patz et al. [59], who showed the capability of a mechanistic model to
reproducepastdengueoutbreaksoverNicaragua,Honduras,andThailand.Anotheradvan-
tageofmechanisticmodelsliesintheirintegrationwithoperationalseasonalclimateforecasting
systemsto anticipatetheriskposedby a particularVBD forthe upcomingseason [60–62].
236 TrendsinParasitology,March2018,Vol.34,No.3