Ecological niche modeling as a tool for prediction of the potential geographic distribution of Bacillus anthracis spores in Tanzania

Ecological niche modeling as a tool for prediction of the potential geographic distribution of Bacillus anthracis spores in Tanzania

Elibariki R. Mwakapeje

*, Sood A. Ndimuligo

, Gladys Mosomtai

, Samuel Ayebare

, Luke Nyakarahuka

, Hezron E. Nonga

, Robinson H. Mdegela

, Eystein Skjerve

aEpidemiologyandDiseasesControlSection,MinistryofHealth,CommunityDevelopment,Gender,ElderlyandChildren,DaresSalaam,Tanzania

bDepartmentofVeterinaryMedicineandPublicHealth,SokoineUniversityofAgriculture,Morogoro,Tanzania

cFacultyofVeterinaryMedicine,NorwegianUniversityofLifeSciences,Oslo,Norway

dCentreforEcologicalandEvolutionarySynthesis,UniversityofOslo,Oslo,Norway

eEarthObservationUnit,InternationalCentreforInsectPhysiologyandEcology,Nairobi,Kenya

fClimateandBiodiversityUnit,WildlifeConservationSociety,Bronx,NewYork,USA

gDepartmentofBiosecurity,EcosystemsandVeterinaryPublicHealth,MakerereUniversity,Kampala,Uganda

ARTICLE INFO Articlehistory:

Received11July2018

Receivedinrevisedform24November2018 Accepted27November2018

CorrespondingEditor:EskildPetersen,Aar- hus,Denmark

Keywords:

Habitatsuitability Bacillusanthracis Spatialdistribution Ecologicalnichemodeling Tanzania

ABSTRACT

Introduction:Anthraxiscausedbythespore-forming,Gram-positivebacteriumBacillusanthracis.The aimofthisstudywastopredictthepotential distributionofB.anthracisinTanzaniaandproduce epidemiologicalevidenceforthemanagementofanthraxoutbreaksinthecountry.

Methods:TheMaxentalgorithmwasusedtopredictareasatriskofanthraxoutbreaksbasedonthe occurrenceandenvironmentaldatainArushaandKilimanjaroregions;themodelwaslatertransferred topredicttheentirecountry.Seventypercentoftheoccurrencedatawereusedtotrainthemodel,while 30%wereusedformodelevaluation.

Results:FourregionsofnorthernTanzaniaarepredictedtohaveahighriskforanthraxoutbreaks,while thesouthernandwesternregionshadlow-riskareas.Soiltype(56.5%),soilpH(23.7%),andisothermally (10.4%)werethemostimportantvariablesforthemodelprediction,andthemostsignificantsoiltypes weresolonetz,fluvisols,andlithosols.

Conclusions:AstrongrisklevelacrossdistrictsoftheTanzaniamainlandwasidentifiedinthisstudy.A totalof18districtsinTanzaniaMainlandarepredictedtobeatveryhighriskofananthraxoutbreak occurrence. These findings are important for policymakers to effectively mount targeted control measuresforanthraxoutbreaksinTanzania.

ﾩ

ThisisanopenaccessarticleundertheCCBY-NC-NDlicense(http://creativecommons.org/licenses/by- nc-nd/4.0/).

Introduction

Bacillus anthracis, thecausative agent of anthraxin wildlife, livestock,andhumans,isasoil-borne,spore-formingandGram- positive bacterium (Mullins et al., 2011). Upon entry into a susceptiblehost,thesporesgerminateintovegetativecells,which replicate rapidly in the bloodstream resulting in septicemia

(Dragon and Rennie, 1995). The septicemic infection leads to hemorrhageinthehost,whichresultsinbloodoozingintothesoil (Steenkamp et al., 2018). It is speculated that death of the susceptiblehostoccursduetoatripartitetoxinproducedbythis bacterium(SmithandKeppie,1954).Thediseasecanbeperacute, acute,orchronicdependingonthehostsusceptibility,immunity statusofthehost,andsizeofthesporeinoculum;however,the peracuteformisthemostcommoninfectioninherbivores,while scavengers such as dogs may be infected without showing symptoms(AchaandSzyfres,2005).Inhumans,theskinformis themostcommon(FAO-OIE-WHO,2008).

Disease transmission pathways are a complex system that involvesseveralagentsofdispersion.Inwildlifeconservationareas such as Serengeti, Ngorongoro Conservation Area, Kilimanjaro, Arusha,andMkomazinationalparks,itisnotpossibletorapidly and properly dispose of anthrax-infected carcasses. Bloody

*Correspondingauthorat:FacultyofVeterinaryMedicine,NorwegianUniversity ofLifeSciences,Oslo,Norway.

E-mailaddresses:[email protected], [email protected](E.R. Mwakapeje),

[email protected](S.A. Ndimuligo),[email protected] (G.Mosomtai),[email protected](S.Ayebare),[email protected] (L.Nyakarahuka),[email protected](H.E. Nonga),[email protected] (R.H.Mdegela),[email protected](E.Skjerve).

https://doi.org/10.1016/j.ijid.2018.11.367

1201-9712/ﾩ2018TheAuthor(s).PublishedbyElsevierLtdonbehalfofInternationalSocietyforInfectiousDiseases.ThisisanopenaccessarticleundertheCCBY-NC-ND license(http://creativecommons.org/licenses/by-nc-nd/4.0/).

ContentslistsavailableatScienceDirect

International Journal of Infectious Diseases

j o u r n a lh o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / i j i d

vulturescontaminatewaterbodiesthroughbathingafteropening up and feeding on anthrax-contaminated carcasses (Blackburn et al., 2010). It can also be dispersed by insects such as necrophagousflies,whichplayacrucialroleinspreadinganthrax (Blackburnetal.,2014).Hugh-JonesandDeVos(2002)reported the role of blow flies, which feed on fluids of the anthrax- contaminatedcarcassanddeposittheirfecesorregurgitateliquids onleavesofvegetationnearthecarcass,readyfortransmissionto another animal. However, biting/hemophagic flies are also consideredtotransmitanthraxamongwildanimalsandlivestock (FoodandAgricultureOrganizationoftheUnitedNations,2003) andeven tohumans(Fasanella,2013).Theinfectioninhumans occurs when B. anthracis penetrates through skin abrasionsor mucousmembrane when a personcomesinto contactwithan infectedanimal(Klousetal.,2016)oranimalproducts,throughthe inhalation of spores, or through the consumption of raw or undercookedinfectedmeat(BengisandFrean,2014).

The disease burden and economic impact of anthrax in domestic animals is not yet fully documented (Lewerin et al., 2010).However,itisestimatedthat2000–20000humananthrax casesarereportedannuallyworldwide(Khomenkoetal.,2013), withmorecasesoccurringinAfrica,Asia,theUSA,andAustralia (Fasanellaetal.,2014).Chinareported112000humancasesduring the years 1956–1997 (Chen et al., 2016). In 2004, the Pollino National Park in Italy reported animal deaths due to anthrax, accountingfor81cattle,15sheep,ninegoats,11horses,andeight deer(Garofoloetal.,2011).InZambia,521humancasesandfive deaths (case fatality rate of 0.95%) were reported in 2011 (Hang'ombe et al., 2012). Kenya reported53 deaths of Grevy’s zebra in 2006 (Muoria et al., 2007) and Uganda reported500 deathsofwildlifeand400deathsofdomestic animalsin2004– 2005(Coffinetal.,2015).In2016,Tanzaniaexperiencedalarge anthrax outbreak in Monduli District in the region of Arusha affecting 21 humans, 109 wildebeest, 21 Grant’s gazelles, one rabbit, 10cattle,26goats,andthreesheep(Mwakapejeetal.,2017).

In favorable environmental conditions, the bacteria from drained blood form spores, which can remain dormant for an extendedperiodoftimeinthesoil,possiblydecades,untilthey affect a new susceptible host (Driks, 2009). Studies on the environmentalsuitabilityforthepersistenceofsporeshaveshown thatsoilparameterssuchasalkalinity,calciumandhighorganic matter contents (Dragon and Rennie, 1995; Hugh-Jones and Blackburn, 2009), elevation, precipitation, temperature, and vegetation biomass(Blackburnet al., 2007; Joyneret al.,2010) maysupporttheextendedsurvival ofB.anthracissporesin the environment. In a previous retrospective study by the present authors’ group, it was established that recurrence of anthrax outbreaks in human, livestock, and wildlife interface areas of northern Tanzania were highly correlated with cycles of short rainfallfollowedbydryandhotweather(Mwakapejeetal.,2018).

However,thespatialecologyandanthraxoutbreakpatterninthe countryarenotwellunderstood.Otherstudieshavereportedthat areaswithanambienttemperatureabove15.5C(Munang’andu etal.,2012), anda cyclicrainfallpatternwithhighevaporation potentialcharacterizedbycalcareoussoil(Winsemiusetal.,2006) tendtofavorlong-termsurvivaloftheB.anthracisspores,causing frequentanthraxoutbreaksinsuchareas.

Ecologicalnichemodeling isatoolforidentifyinggeographicand ecological areas suitable for species persistence based on the environmentalvariablesofknownoccurrencesites(Phillipsand Elith,2010).Duringmodelingofspeciesdistribution,presence-only orbothpresenceandabsencedatamaybeused.Theuseofboth presenceandabsencedatahasbeenshowntoimprovethemodel performancein somecases (Brotons etal., 2013).Nevertheless, absencedataarechallengingtoverify(GuandSwihart,2004)and thereforeamodelingtechniquewithpresencedataonly(Anderson

etal.,2002)canbeemployed.However,thespeciescanbeabsent fromthesuitablehabitatforhistoricalreasonsorduetofailureto disperse to those areas (Holt, 2003). Various presence-only modelingtechniquestopredictthegeographicdistributionofB.

anthracissporeshavebeenemployed,suchasMaxent(maximum entropy)(Chikeremaetal.,2013)andGARP(geneticalgorithmfor therule-setprediction)(Barroetal.,2016).Comparingcorrelative modelssuchasBioclim,GARP,andMaxentusingthesameinput data,Maxentwasfoundtogivethebestpredictions(Tarkeshand Jetschke, 2012).Therefore, inthis study,Maxent (Phillipset al., 2006)wasusedtomodelB.anthracissporespersistenceandits spatialdistributionusingpresence-onlydata.

Theaimofthisstudywastopredictthepotentialgeographic distribution of B. anthracis spores in Tanzania and produce epidemiological evidence for the management of anthrax out- breaks in Tanzania. This information will provide a better ecologicalandepidemiologicalunderstandingoffrequentanthrax outbreaksinthemostat-riskareas.Itwillalsohelptozonethe countrybasedonrisksandinformdecision-makersintheeffective allocation of resources for targeted preventive and control measures,suchasintensifiedsurveillance,communityawareness, improved diagnostic capacity,and livestock vaccination against anthraxintheidentifiedhigh-riskareas(Kracaliketal.,2017a).

Materialsandmethods Studyareas

ThestudywasconductedintheArushaandKilimanjaroregions of northern Tanzania, where the occurrence data for anthrax outbreakswerecollectedforwhole-countrymodeling.Theregionof ArushaliesontheKenyanborder atlatitude3.36667andlongitude 36.683330, with an elevation of 1415m above sea level. The populationsizeofthisregionwasestimatedtobe1 694310in2012 (2012census)(The Minister ofState—Planning andParastatal Sector Reform,1998).Thisregionencompassesthesavannahsandpartof theGreatRiftValley.Ithaswildlifeprotectedareasincludingthe NgorongoroConservationArea(NCA),whichcontainsthemassive NgorongoroCrater,andArushaNationalPark,whichcoversvolcanic MountMeru.ManyaraNationalPark,GrumetiGameReserve,and Lake Natron GameReserve arealsofound in this region.Meru, Ngorongoro,andMondulidistrictswereselectedforthisstudy.

TheKilimanjaro regionislocated inthenorthernpart ofthe Tanzania mainland, south of the equator at 225⁰ and 415⁰; longitudinally it lies between 3625⁰30⁰⁰ and 3810⁰45⁰⁰ east of Greenwichandtheregionhasanelevationof2400mabovesealevel.

The2012censusestimatedapopulationofapproximately1 640087, withanaverageannualpopulationgrowthof1.8%(Tanzania2012 PopulationandHousingCensus).Theregionhasthreeecological zones:lowland(1500mandbelow),highland(1501–3000m),and forest(3001mandabove)(Reform,1998;Census,2013).Kilimanjaro RegionisborderedtothenorthandeastbyKenya,tothesouthby TangaRegion,tothesouthwestbyManyaraRegion,andtothewest byArushaRegion.Theselectedstudydistricts inthisregion wereHai, MoshiRural,Rombo,andSihadistricts.Figure1illustratesthestudy areaandthedistributionoftheoccurrencedata.

Anthraxoccurrencedata

A database of 192 mixed cases of human (n=68), wildlife (n=21),livestock(n=80),andenvironmental(n=23)sampleswas constructedfromsporadicanthraxoutbreaks,whichoccurredin different placesof Arusha and Kilimanjaro regions in northern TanzaniabetweenOctober2016andMarch2018.Thisinformation was used to map risks of anthrax outbreaks for the whole of Tanzania.

E.R.Mwakapejeetal./InternationalJournalofInfectiousDiseases79(2019)142–151 143

It is standard practice under normal circumstances during outbreaksthat specimens for laboratory analysisare not taken fromallsuspectedhumancases;rather,a fewarecollectedfor confirmationoftheexistenceofanoutbreak.Allspecimenswith errorsingeo-coordinatesandmissinginformationwereomitted fromthedatabase.Therefore, atotalof108(56.25%)specimens weremaintained in the database, of which 44(40.74%) tested positiveforB.anthracis.Thepositivecasesforanthraxwerelinked tothe geo-coordinates (latitude/longitude) that were collected eitherataresidenceofahumansuspectedcaseorataburialpoint ofananimalcarcass(livestockorwildlife)ifasuspectedhuman casereportedanylivestockdeathintheperiod2weekspriorto datacollection,thengeo-coordinateswerecollectedatthecarcass disposalsite using a handheld global positioning system (GPS) machine.Itshouldalsobenotedthatonsomeoccasionsrecording ofthesamegeo-coordinatesofoutbreakswasrepeatedindifferent time periods; this also contributed to the lower number of occurrence data in the final database compared to what was recordedinitially.

ThecollecteddatawerestoredinaMicrosoftExcelspreadsheet.

Thiswaseditedandtherecordswithgeo-coordinateerrors,and thespecimens which were nottested and those with negative laboratoryresultswereremovedfromthedataset.Afinalversion containingthegeo-coordinatesandpositivelaboratoryresultswas savedincommaseparatedvalue(CSV)format,whileenvironmen- talcovariatesweresavedinESRIASCIIformatforfurtheranalysis inQuantum GISsoftware.Figure 1 showsthemap of Tanzania illustratingthe distribution of thegeographical positionof the sampled areas with the laboratory results of the collected specimens, following sporadic anthrax outbreaks in northern Tanzania.However,occurrencedatafrom asmall sampledarea

wereusedformodelcalibrationand thentopredictthehabitat suitabilityofanthraxsporesforthewholecountry.

Environmentalcovariates

Atotalof21climaticvariables,assummarizedinTable1,were obtainedfromthe1-kmgridAfriclimdatabase.Theseconsistedof twocategoriesofdata:(1)temperaturevariablesincludingannual meantemperature(BIO1),meandiurnalrange(BIO2),isothermally (BIO3),temperatureseasonality(BIO4),maximumtemperatureof warmestmonth(BIO5),minimumtemperatureofcoldestmonth (BIO6),temperature annualrange (BIO7),mean temperatureof wettestquarter(BIO8),meantemperatureofdriestquarter(BIO9), meantemperatureofwarmestquarter(BIO10),meantemperature of coldest quarter (BIO11);(2) precipitation variables including annual precipitation (BIO12), precipitation of wettest month (BIO13), precipitation of driest month (BIO14), precipitation seasonality (BIO15), precipitation of wettest quarter (BIO16), precipitationofdriestquarter (BIO17),precipitationofwarmest quarter(BIO18),precipitationofcoldestquarter(BIO19),moisture indexaridquarter(MIAQ),andpotentialevapotranspiration(PET) (obtained from http://doi.org/10.1111/aje.12180). The variables wereobtainedunderthecurrentscenariothatcomprisesmonthly measurementsobtainedfromweatherstationsaroundtheworld between1950to2000modeledunderthe4.5RCP(representative concentration pathways) scenario. Furthermore,1-km grid soil typeandsoilpHdataobtainedfromtheISRICAfricansoildatabase werealsoincluded;soilpHwasincludedasapredictorvariable because it has been shown that epidemics of anthrax are associated withanalkaline pH(Dragon andRennie,1995).Soil type as a categorical variable was also included in themodel, Figure1.MapshowingthestudyareasanddistributionofthelaboratoryresultsintheareassampledinnorthernTanzania.

becausetheinfluenceofsoiltypeonB.anthracissporespersistence is ecologically documented and it is speculated that there is a significantrelationshipbetweenthesoiltypeand theextensive presenceofanthraxoutbreaksincertainareas(Griffinetal.,2014).

Modeldevelopment

A pairwise Pearson correlation analysis for environmental variableswasdoneusingENMTools(Warrenetal.,2010).Thiswas donein orderto reducemulticollinearity of theenvironmental variables, and only variables with a lower than 0.75 were retainedformodelfitting.Afterthisprocedure,thenon-correlated environmental variableswerechosen for thedevelopment of a species distribution model. The candidate variables identified included isothermally (BIO3), temperature seasonality (BIO7), moistureindexaridquarter(MIAQ),potentialevapotranspiration (PET),soiltype(Sanchezetal.,2009),andsoilpH.

AMaxentmodelwasfittedusing100bootstraprunswith70/30 partition percentage for the training/testing datasets. Default Maxent model parameter settings were used (auto features, convergence threshold of 0.00001, maximum number of back- groundpoints=10000,regularizationmultiplier=1)(Phillipsand Dudík,2008).InordertotrainMaxent,onefoldwasusedtofita modelandtheremainingfoldsweretreatedasindependentdata for the evaluation of the predictive ability of the model performance (testing). A masked file was created and used in the model development in order to constrain the selection of backgroundvalues,andtheperformanceofthemodelwasthen evaluated. In each iteration, the contribution of every single variabletothegeneraldistributionwasdeterminedbyJackknife statisticaltechnique,whichallowedthevariableswiththegreatest influence on the probability of persistence of B. anthracis and spatial distribution in Tanzania to be identified. However, the resolutionoftheresultingriskmodelswasoptimallymaintainedat 1km.

Modelevaluation

Severalmethodsexistfortheevaluationofmodelaccuracy,but themostcommonmethodinvolvestheareaunderthereceiver operatingcharacteristicscurve(AUC)(HanleyandMcNeil,1982a).

AsuccessfulmodelhasanAUCvaluecloseto1.0,andthehigherthe AUC,thebetterthemodeldistinctionofthepresencefromabsence ofaspecies;modelswithnocleardistinctionhaveanAUCcloseto 0.5 (Hanley and McNeil, 1982b). Evaluation of the critical individual environmental predictors in themodel development wasdonebyjackknifetest,andresponsecurveswerealsousedto show how each environmental variable affects the Maxent prediction and how the logistic prediction changes as each environmental variableis varied,bykeeping allother variables attheiraveragesamplevalue.Figure2illustratesasummaryofthe modeldevelopmentprocessandmethodsusedtoobtainsignifi- cantvariablesformodeling.

Results

IntheMaxentmodel,anAUCof0.93wasobtained,indicating thatthemodelhad‘excellent’abilitytopredictthepresenceofB.

anthracissporesinthemostat-riskareasoftheTanzaniamainland (Figure3).ThetestindicatedthatthedifferencebetweentheAUC frommodelpredictionand theAUCatrandomwasstatistically significant,showingthatthemodelperformedbetterthanrandom prediction.

Table1

BioclimaticvariablesusedformodelingbyMaxentsoftware.

Variablecode Variabledescription Unit

1.Temperaturevariables

Bio1 Annualmeantemperature C

Bio2 Meandailytemperature C

Bio3 Isothermally(Bio2/Bio7)100 –

Bio4 Temperatureseasonality(standarddeviation100) C Bio5 Maximumtemperatureofwarmestmonth C Bio6 Minimumtemperatureofcoldestmonth C Bio7 Temperatureannualrange(Bio5–Bio6) C

Bio8 Meantemperatureofwettestquarter C

Bio9 Meantemperatureofdriestquarter C

Bio10 Meantemperatureofwarmestquarter C Bio11 Meantemperatureofcoldestquarter C

PET Potentialevapotranspiration mm

2.Precipitationvariables

Bio12 Annualprecipitation mm

Bio13 Precipitationofwettestmonth mm

Bio14 Precipitationofdriestmonth mm

Bio15 Seasonalrainfall(coefficientofvariation) mm

Bio16 Precipitationofwettestquarter mm

Bio17 Precipitationofdriestquarter mm

Bio18 Precipitationofwarmestquarter mm

Bio19 Precipitationofcoldestquarter mm

MIMQ Moistureindexmoistquarter N/A

MIAQ Moistureindexaridquarter N/A

DM Numberofdrymonths Months

LLDS Lengthoflongestdryseason Months

N/A,notapplicable.

Figure2.Flowchartindicatingthemodelbuildingprocessandpotentialvariablesincluded.

E.R.Mwakapejeetal./InternationalJournalofInfectiousDiseases79(2019)142–151 145

Outofthe23environmentalvariables,thefollowingvariables wereidentifiedasnon-collinear:isothermally(BIO3),temperature seasonality(BIO7),moistureindexaridquarter(MIAQ),potential evapotranspiration(PET),soiltype(Sanchezetal.,2009),andsoil pH.Table2indicatesthepercentagecontributionofeachofthese variables, withsoil type demonstrating the highest percentage contribution (56.5%), followed by pH (23.7%); hence the two variables (soil type and pH) in total contributed 80.2%. The Jackknife test helped to identifythe variables contributing the mostinthepersistenceandgeographicdistributionofB.anthracis sporesinTanzania.

TheJackknifetestof variablesindicatedthatomittinganyof thesesixvariablesaffectedtheregularizationgain,AUC,andtest gaininthemodel(Figure4).Amongthevariablesretainedinthe model,thesoiltypewasthemostimportantcontributingvariable tothemodel,followedbypH.However,pHdecreasedthegainthe mostwhenremovedfromthemodel.Therefore,bylookingatthe AUCoftheJackknifetest,themostsignificantvariableswithscores of >0.7 (above fair) were soil type, soil pH, BIO3, and BIO7.

Responsecurvesforthesevariableswithregardtotheirsuitability forthepredictionofB.anthracissporegeographicdistributionare showninFigure5.

TheresponsecurveforsoilpHshowedthattheprobabilityof geographic suitability increased with the level of alkalinity (correspondingtohighlevelsofcalcium)inthesoil.

Thesoilcharacteristicsforthesoiltypesthatwereidentifiedas havingthehighestpredictivepowerforB.anthracissporesurvival (asshownintheresponsecurveforsoiltypesinFigure5Dand Table3)werecalciccambisols(2),lithosols(9),eutricfluvisols(11),

eutrichistosols(16),andorthicsolonetz(20).Thesoiltype,soilpH, andisothermallywerethemostimportantvariables;however,soil typewas thesinglemostimportantvariablethataccountedfor 56.5% of the model prediction, with the following soil types identifiedasthemostsignificant:solonetz,fluvisols,andlithosols.

Figure6 showsariskmapindicating regionswithveryhigh, high,medium,andlowprobabilityofenvironmentalsuitabilityfor persistence and spatial distribution of B. anthracis spores in Tanzania.Theregionswithstableareasofhighandveryhighrisk were Arusha and Kilimanjaro from the northern part of the country, while other regions like Mara, Manyara, Simiyu, and Singidahadafewpatchesofhighandveryhighriskareas.Regions likeDodoma,Mwanza,DaresSalaam,Lindi,Mbeya,Rukwa,Katavi, and Kigoma were predicted to have a medium risk in a few locations,andtheremainingregionsinthecountryhadalowrisk ofgeographicsuitabilityforB.anthracissporepersistence.

Discussion

Despitethefactthat anthraxis adiseaseofboth publicand livestockimportanceinTanzania,risk-mappingofthediseasehas notbeenusedpreviouslyinthecountry.Consequently,thereisno evidence-based allocation of resources for the prevention and controlofthedisease—bearinginmindthat therearealotof competingprioritiesforthedistributionoffinancialresourcesin thecountry.Therefore,thefindingsofthisstudyprovideimportant insights for spatially allocating and prioritizing resources for anthrax surveillance, prevention, and control based on the predictedlevelofrisk(veryhigh,high,medium,andlow)within eachdistrict.Thedistrictisanimportantadministrativelevelfor diseasepreventionandcontrolpolicyimplementationinTanzania.

Inthisstudy,anecologicalnichemodelingtechniquewasused topredictpotentialsuitablehabitatdistributionofanthraxspores inTanzania.Thisis thefirst studytopresentthe potentialrisk distributionassociatedwithB.anthracissporesinTanzaniausing climaticand abioticfactors,such assoil type and soil pH. The regionsofArushaandKilimanjarohadahigherrisk(veryhighand highrisk)ofB.anthracissporehabitatsuitability,asillustratedin the national prediction (risk map shown in Figure 6). The observations in the current study are in line with a previousreport by Mwakapeje et al. (2018): in a retrospective Figure3. Averagereceiveroperatingcharacteristics(ROC)andrelatedareaunderthecurve(AUC)ofthe100bootstrapreplicates.

Table2

Thepercentagecontributionandpermutationimportanceofthevariablesusedin Maxentmodeling.

Variable Percentagecontribution Permutationimportance

Soiltype 56.5 37.4

SoilpH 23.7 20.6

Bio3 10.4 26.8

MIAQ 5.2 9.4

PET 3.2 2.9

Bio7 1.1 2.9

Mask 0 0

study(2006–2016),theincidencerateofhumananthraxcaseswas foundtobe7.88per100000populationinArushaRegion,followed by6.64per100000populationinKilimanjaroRegion(Mwakapeje etal.,2018).

Inthesamepredictedmap,streaksofpredictedveryhighand highriskwereobservedinTanga,Coastal,Manyara,andSingida

regions. From the predicted at-risk regions, the corresponding districtspredictedwithveryhighandhighrisksareindicatedin Figure7;theseareArushaRegion(Ngorongoro,Monduli,Longido, Arusha Rural, and Meru), Kilimanjaro Region (Hai, Siha, Moshi Rural, and Rombo), Mara Region (Serengeti), Manyara Region (Simanjiro,Hanang,andBabatiurban),SimiyuRegion(Bariadiand Figure4.Jackknifetestsofvariableimportancewith(A)regularizedtraininggain,(B)testgain,and(C)AUC.Thelightbluebarsillustratethemodelgainwithoutvariable inclusion,whilethesolidbluebarsshowitsgainwiththevariableonly.

Figure5.ResponsecurvesforthemostsignificantvariablesintheMaxentmodelforthepersistenceofBacillusanthracisinTanzania:(A)BIO3(isothermal),(B)BIO7(annual temperaturerange),(C)soilpH,and(D)soiltype.Theredlinesindicatethemeanvalues,whileblueareasdenote1standarddeviationlimits,resultingfromcross-validation modelruns.

E.R.Mwakapejeetal./InternationalJournalofInfectiousDiseases79(2019)142–151 147

Itilima),TangaRegion(Kilindi),andSingidaRegion(Mkalamaand Iramba).The areas withinthese regionsare color-codedin the orderofriskrecognition.Theabilitytoidentifytheriskassociated with specific places and areas is extremely important for an efficientdiseasesurveillanceandcontrolprogram.

Infact,someofthepredicteddistrictswithahighrisksuchas Hanang,Simanjiro, Itilima,Serengeti, Bariadi, Kilindi,Mkalama, and Iramba have had no reported anthrax cases through the surveillancesystems.Thismightbeattributedtothepoorhuman and animal surveillance systems, leading to severe under- reporting and hence misleading disease burden information (Gibbons etal., 2014).Monduliand Ngorongoroare amongthe districtswithapredictedhighandveryhighriskofsuitabilityfor anthraxspores in Arusha Region, which corresponds withthe recentfrequentanthraxoutbreaksinMonduliDistrict;forinstance inlate2016,atotalof130wildlifecarcasses,39livestockcarcasses, and 21humancases wereconfirmedtohave beeninfected by anthrax(Mwakapejeet al.,2017).It istherefore envisagedthat implementingtargetedcontrolmeasuresbasedonthediseaserisk

mappingwillbemorecost-effective,duetothereducedcostfor carcassdisposal,costforlaboratoryreagents,andcostforoutbreak managementingeneral.Itmayalsohelpintheimplementationof a targetedlivestock vaccinationprogramandintensifiedhuman andanimaldiseasesurveillance,byfocusingmorecloselyonthe predictedhighandveryhighriskdistrictsusingtheOneHealth approach(Cleavelandetal.,2017;Baumetal.,2017).

Thismodeldemonstratedthattheenvironmentalsuitabilityfor thepersistenceofB.anthracissporeswashighlyinfluencedbythe soiltype, pH,BIO3, BIO7,MIAQ, andPETvariables,respectively.

ApartfromsoiltypeandsoilpH,othervariablesarecategorized intotemperature(BIO3–isothermally,BIO7–annualtemperature range)andprecipitation(MIAQ–moistureindexaridquarter,PET –potentialevapotranspiration)variables.Environmentalvariables suchassoilandclimate arepostulatedtofavorand extendthe survivalofB. anthracisspores inthesoilfora longperiod.This findingsupportstheresultsofotherstudies,whichhaveshown thatanthraxoutbreaksareexacerbatedbywarmertemperatures, moistsoils,andhighorganicmattercontent(humus)—thesefavor B.anthracissporeamplification(Deyetal.,2012).

ItwasfoundinthisstudythatsoiltypeandsoilpHwerethe most significant variables for long-term persistence of anthrax spores in the identified high and very highrisk areas. This is supportedbyotherstudies,whichhaveshownthatsoilwithhigh moisture,alkalinepH,andhumusaresuitableforanthraxspore germinationandsporulationoutside amammal host;theseare someofthecriticalvariablesthatleadtotheoccurrenceofanthrax outbreaksinanimals,withspillovertohumans(KreuderJohnson etal.,2015).OtherstudieshavedocumentedthatsoilpHofmore than 6.1 (alkalinity) in combination with calcium levels are Table3

Summaryofsoiltypeswithastrongassociationtopersistenceandenvironmental suitabilityforBacillusanthracissporesinTanzania.

Soilcodenumber Soiltypekey Soiltype

20 So Orthicsolonetz

11 Je Eutricfluvisols

9 I Lithosols

2 Bk Calciccambisols

16 Oe Eutrichistosols

Figure6.RiskmapforthepredictedenvironmentalsuitabilityofBacillusanthracissporeswithinregionsoftheTanzaniamainland.

important variables for the long-term survival of B. anthracis spores(Kracalik etal.,2017b).Thistypeof soilisregardedasa naturalreservoirforB.anthracisspores(Barroetal.,2016).This information is important for regional and district veterinary offices,aswellasselectedfarmingcommunities,toincreasetheir awarenessandthelocalrelevanceofpredictedrisks,theneedto report unexpected livestock deaths, and also for livestock vaccinationpolicychange.

Studylimitations

Duringmodelbuilding,factorssuchaslivestockdensity,number ofdrymonths,elevation,andlengthofthelongestdryseasonwere highly correlated with themost significant variables identified.

Therefore, we are scientifically convinced that, apart from the identified mostsignificant variablesfavoring the persistence of anthraxoutbreaksintheareaswithhighandveryhighrisk,thereare otherfactorsthatcontributetoanthrax-relateddeathsinTanzania;

forexampletheGainer–Koloninhypothesisofhyperacutedeaths involving latent infections, climate stress, and severe seasonal biting-flyactivityin theabsenceofsuitablesoils (Gainer,2018).

However,westilltrustthattheidentifiedsuitableenvironmentsfor anthraxoutbreaksareimportantregionsand/ordistrictsthatshould be given more attention, because they have been identified as hotspotareasforanthraxoutbreaksinpreviousstudies.

Conclusions

Thisstudy modeled theoccurrence data and environmental variablestocreateriskmapswithcategorizedrisks,whichassisted inestablishingdistrictswithveryhigh,high,medium,andlowrisk ofanthraxoutbreakemergenceinTanzania.Theresultsshowed

thatnorthernTanzaniahasahigherprobabilityoftheoccurrence of anthraxoutbreaksthanotherpartsof thecountry.Themost significantfactorsidentifiedforanthraxpersistenceweresoiltype, soilpH,isothermality,mean temperaturerange,moistureindex aridquarter,andpotentialevapotranspiration.

The categorized risks are important and will help to direct decision-makerswithregardtoresourceallocationinamostcost- effectiveapproach.Theidentifiedhigh-riskdistrictshavetoreduce mortalitiesinlivestockandpreventthediseaseinhumansthrough continued pre-outbreak targeted livestock vaccination, safe carcass disposal (preferably incineration) of animals that have diedfromanthrax,publicawarenesscampaigns,theprovisionof relevantdiagnostics forlivestockand humancarefacilities, and intensified human and animal surveillance systems. These activities, if implemented effectively, will help to significantly controltheexistingdevastatingfrequentanthraxoutbreaksinthe predictedhigh-riskareasofTanzania.

Acknowledgements

Theauthors’greatappreciationgoestoDrAugustinoMarmo,the MonduliDistrictVeterinaryOfficer,forassistinginthesampling exercise during the response to anthrax outbreaks in human, livestock, and wildlife species in the hotspot areas of Monduli District. Theauthors areindebted to thehuman, livestock, and wildlife authorities ofthe Tanzanian government for their supportive logisticsontheground,whichhelpedinthecompletionofthisstudy.

Funding

ThisstudywasfundedbytheWorldBankundertheEastAfrican Public Health Laboratory Networking Project (EAPHLN; Project Figure7.RiskmapforthepredictedenvironmentalsuitabilityofBacillusanthracissporesdistributionwithindistrictsintheveryhighriskregionsoftheTanzaniamainland.

E.R.Mwakapejeetal./InternationalJournalofInfectiousDiseases79(2019)142–151 149

NumberP153665)incollaborationwiththeNorwegianUniversity ofLifeSciences(NMBU).

Conflictofinterest

Theauthorsdeclarenoconflictsofinterest.

Authorcontributions

ERM:conceptualizedthestudy, collectedspecimens, partici- patedinthelaboratoryconfirmationofanthrax,participatedinthe dataanalysisandinterpretationoftheresults,draftedthearticle, andaddressedcommentsfromtheco-authorsuntilthearticlewas readyforsubmissiontothepeerreviewjournal;SAN:participated in the conceptualization of the study, participated in the data analysis and interpretation of the results, and reviewed and commentedonthearticle;GM:participatedintheconceptualiza- tionofthestudy,participatedinthedataanalysisandinterpreta- tionoftheresults,andreviewedandcommentedonthearticle;SA:

participatedintheconceptualizationofthestudy,participatedin thedataanalysisandinterpretationoftheresults,andreviewed andcommentedonthearticle;LN:participatedintheconceptu- alization of the study, participated in the data analysis and interpretationoftheresults,andreviewedandcommentedonthe article;HEN:supervisedthedesignofthestudy,andreviewedand commentedon thearticle; RHM:supervised the design of the study,andreviewedandcommentedonthearticle;ES:supervised theinitialdesignofthestudy,participatedinthedataanalysisand interpretationoftheresults,participatedintheconceptualization ofthestudy,participatedindraftingthearticle,andreviewedand commentedonthearticle.

AppendixA.Supplementarydata

Supplementarydataassociatedwiththisarticlecanbefound,in theonlineversion,athttps://doi.org/10.1016/j.ijid.2018.11.367.

References

AchaPN,SzyfresB.Zoonosesandcommunicablediseasescommontomanand animals: volumeI.Bacteriosesandmycoses.PAHOscientificandtechnical publication no. 580, I. 3rd ed. , doi:http://dx.doi.org/10.1590/S1135- 57272005000300012378pp..

AndersonRP,Gómez-LaverdeM,PetersonAT.Geographicaldistributionsofspiny pocket mice in South America:insightsfrom predictive models. Glob Ecol Biogeogr 2002;11(2):131–41,doi:http://dx.doi.org/10.1046/j.1466-822X.2002.00275.x.

BarroAS,FeganM,MoloneyB,PorterK,MullerJ,WarnerS,etal.Redefiningthe AustralianAnthraxBelt: modelingtheecologicalnicheandpredictingthe geographicdistributionofBacillusanthracis.PLoSNeglTropDis2016;10(6),doi:

http://dx.doi.org/10.1371/journal.pntd.0004689.

BaumSE,MachalabaC,DaszakP,SalernoRH,KareshWB.Evaluatingonehealth:are wedemonstratingeffectiveness?.OneHealth2017;3:5–10,doi:http://dx.doi.

org/10.1016/j.onehlt.2016.10.004.

BengisRG,FreanJ.AnthraxasanexampleoftheOneHealthconcept.RevSciTechde l’OIE2014;33(2):593–604,doi:http://dx.doi.org/10.20506/rst.33.2.2309.

BlackburnJK,McNysetKM,CurtisA,Hugh-JonesME.Modelingthegeographic distributionofBacillusanthracis,thecausativeagentofanthraxdisease,forthe contiguous United States using predictive ecological [corrected] niche modeling.AmJTropMedHyg2007;77(6)1103–1077/6/1103[pii].

Blackburn JK,Curtis A, Hadfield TL, O’Shea B, Mitchell MA, Hugh-Jones ME.

ConfirmationofBacillusanthracisfromflesh-eatingfliescollectedduringaWest Texas anthraxseason.J WildlDis2010;46(3):918–22,doi:http://dx.doi.org/

10.7589/0090-3558-46.3.918.

BlackburnJK,VanErtM,MullinsJC,HadfieldTL,Hugh-JonesME.Thenecrophagous flyanthraxtransmissionpathway:empiricalandgeneticevidencefromwildlife epizootics.VectorBorneZoonoticDis2014;14(8):576–83,doi:http://dx.doi.org/

10.1089/vbz.2013.1538.

BrotonsL,ThuillerW,AraújoMB,HirzelAH.Presence-absenceversuspresence-only methodsforpredictingbirdhabitatsuitability.Ecography2013;27(4):437–48, doi:http://dx.doi.org/10.1111/j.0906-7590.2004.03764.x.

CensusH.TheUnitedRepublicofTanzania,II.2013.

ChenWJ,LaiSJ,YangY,LiuK,LiX,Lou,etal.Mappingthedistributionofanthraxin mainlandChina,2005–2013.PLOSNeglTropDis2016;10(4),doi:http://dx.doi.

org/10.1371/journal.pntd.0004637.

ChikeremaSM,MurwiraA,MatopeG,PfukenyiDM.SpatialmodellingofBacillus anthracisecologicalnicheinZimbabwe.PrevVetMed2013;111(1–2):25–30, doi:http://dx.doi.org/10.1016/j.prevetmed.2013.04.006.

CleavelandS,SharpJ,Abela-RidderB,AllanKJ,BuzaJ,CrumpJA,etal.OneHealth contributionstowardsmoreeffectiveandequitableapproachestohealthin low-andmiddle-incomecountries.PhilosTransRSocBBiolSci2017;372(1725), doi:http://dx.doi.org/10.1098/rstb.2016.0168.

Coffin JL, Monje F,Asiimwe-KarimuG, Amuguni HJ,Odoch T. AOne Health, participatoryepidemiologyassessmentofanthrax(Bacillusanthracis)manage- mentinWesternUganda.SocSciMed2015;129:44–50,doi:http://dx.doi.org/

10.1016/j.socscimed.2014.07.037.

DeyR,HoffmanPS,GlomskiIJ.Germinationandamplificationofanthraxsporesby soil-dwelling amoebas. Appl Environ Microbiol 2012;78(22):8075–81, doi:

http://dx.doi.org/10.1128/AEM.02034-12.

DragonDC,RennieRP.Theecologyofanthraxspores:toughbutnotinvincible.Can VetJ1995;36(5):295–301.

DriksA.TheBacillusanthracisspore.MolAspectsMed2009;30(6):368–73,doi:

http://dx.doi.org/10.1016/j.mam.2009.08.001.

FAO-OIE-WHO.Anthraxinhumansandanimals..p.219,doi:http://dx.doi.org/

10.2105/AJPH.30.3.299.www.who.int.

FasanellaA,AdoneR,Hugh-JonesM.Classificationandmanagementofanimal anthraxoutbreaksbased onthesource ofinfection.AnnIst SuperSanita 2014;50(2):192–5,doi:http://dx.doi.org/10.4415/ANN_14_02_14.

FasanellaA.Bacillusanthracis,virulencefactors,PCR,andinterpretationofresults.

Virulence2013;4(8),doi:http://dx.doi.org/10.4161/viru.26517.

FoodandAgricultureOrganizationoftheUnitedNations.EMPREStransboundary animaldiseasesbulletin.EMPRESTransboundaryAnimDisBull2003;24:2–7.

GainerRS.Sporeconcentrationandmodifiedhostresistanceascauseofanthrax outbreaks:practitioners’CornerLecoindespraticiens.CanVetJ2018;59:185–7, doi:http://dx.doi.org/10.1590/S0074-02761985000200007.

GarofoloG,SerrecchiaL,CorròM,FasanellaA.Anthraxphylogeneticstructurein Northern Italy. BMC ResNotes 2011;4:,doi:http://dx.doi.org/10.1186/1756- 0500-4-273.

GibbonsCL, Mangen MJJ,PlassD, HavelaarAH, Brooke RJ,Kramarz P, etal.

Measuring underreporting and under-ascertainment in infectious disease datasets:acomparisonofmethods.BMCPublicHealth2014;14(1):1–17,doi:

http://dx.doi.org/10.1186/1471-2458-14-147.

GriffinD, SilvestriE,Bowling C, BoeT, Smith D,Nichols T. Anthraxand the geochemistryofsoils inthecontiguousUnitedStates.Geosciences2014;4 (3):114–27,doi:http://dx.doi.org/10.3390/geosciences4030114.

GuW,SwihartRK.Absentorundetected?Effectsofnon-detection ofspecies occurrenceonwildlife-habitatmodels.BiolConserv2004;116(2):195–203,doi:

http://dx.doi.org/10.1016/S0006-3207(03)00190-3.

Hang’ombeMB,MwansaJ,MuwowoS,MulengaP,KapinaM,MusengaE,etal.

Human–animalanthraxoutbreakintheLuangwavalleyofZambiain2011.Trop Doct2012;42(3):136–9.

HanleyJA,McNeilBJ.Themeaninganduseoftheareaunderareceiveroperating characteristic(ROC)curve.Radiology1982a;143(1):29–36,doi:http://dx.doi.

org/10.1148/radiology.143.1.7063747.

HanleyJA,McNeilBJ.Themeaninganduseoftheareaunderareceiveroperating characteristic(ROC)curve.DiagnRadiol1982b;143:29–36,doi:http://dx.doi.

org/10.1148/radiology.143.1.7063747.

Holt RDR. On the evolutionary ecology of species’ ranges. Evol Ecol Res 2003;5:159–78.

Hugh-JonesM,BlackburnJ.TheecologyofBacillusanthracis.MolAspectsMed 2009;30(6):356–67,doi:http://dx.doi.org/10.1016/j.mam.2009.08.003.

Hugh-JonesME,DeVosV.Anthraxandwildlife.RevSciTechdel’OIE2002;21 (2):359–83,doi:http://dx.doi.org/10.20506/rst.21.2.1336.

JoynerTA,LukhnovaL,PazilovY,TemiralyevaG,Hugh-JonesME,AikimbayevA, etal.ModelingthepotentialdistributionofBacillusanthracisundermultiple climatechangescenariosforKazakhstan.PLoSOne2010;5(3),doi:http://dx.doi.

org/10.1371/journal.pone.0009596.

KhomenkoS,RozstanlnyyA,GoginA,KolbasovD,PintoJ,LubrothJ,etal.Anthrax outbreaks: a warning for improved prevention, control and heightened awareness.EmpresWatch2013;37(September)Availablefrom:http://www.

fao.org/3/a-i6124e.pdf.[Accessed26August2018].

KlousG,HussA,HeederikDJJ,CoutinhoRA.Human-livestockcontactsandtheir relationshiptotransmissionofzoonoticpathogens,asystematicreviewof literature. One Health 2016;2:66–76, doi:http://dx.doi.org/10.1016/j.

onehlt.2016.03.001.

KracalikI, MalaniaL,BroladzeM,NavdarashviliA,ImnadzeP,RyanSJ,etal.

Changing livestock vaccination policy alters the epidemiology of human anthrax, Georgia, 2000–2013. Vaccine 2017a;35(46), doi:http://dx.doi.org/

10.1016/j.vaccine.2017.09.081.

KracalikIT,KenuE,AyamdoohEN,Allegye-CudjoeE,PolkuuPN,FrimpongJA, et al.Modeling theenvironmental suitabilityof anthraxin Ghana and estimatingpopulationsatrisk:implications forvaccinationandcontrol.

PLoSNeglTropDis2017b;11(10):1–17,doi:http://dx.doi.org/10.1371/jour- nal.pntd.0005885.

KreuderJohnsonC,HitchensPL,SmileyEvansT,GoldsteinT,ThomasK,ClementsA, etal.Spilloverandpandemicpropertiesofzoonoticviruseswithhighhost plasticity.SciRep2015;5:1–8,doi:http://dx.doi.org/10.1038/srep14830.

LewerinSS,ElvanderM,WestermarkT,HartzellLN,NorströmAK,EhrsS,etal.

AnthraxoutbreakinaSwedishbeefcattleherd—1stcasein27years:case report. Acta Vet Scand 2010;52(1):7, doi:http://dx.doi.org/10.1186/1751- 0147-52-7.

MullinsJ,LukhnovaL,AikimbayevA,PazilovY,VanErtM,BlackburnJK.Ecological NichemodellingoftheBacillusanthracisA1.asub-lineageinKazakhstan.BMC Ecol2011;11:,doi:http://dx.doi.org/10.1186/1472-6785-11-32.

Munang’anduHM,BandaF,SiamudaalaVM,MunyemeM,KasangaCJ,HamududuB.

TheeffectofseasonalvariationonanthraxepidemiologyintheupperZambezi floodplainofwesternZambia.JVetSci2012;13(3):293–8,doi:http://dx.doi.org/

10.4142/jvs.2012.13.3.293.

MuoriaPK, MuruthiP, KariukiWK,HassanBA, MijeleD, OgugeNO.Anthrax outbreakamongGrevy’szebra(Equusgrevyi)inSamburu,Kenya.AfrJEcol 2007;45(4):483–9,doi:http://dx.doi.org/10.1111/j.1365-2028.2007.00758.x.

MwakapejeER,AssengaJA,KundaJS,MjingoEE,MakondoZE,NongaHE,etal.

Prevention,detection,andresponsetoanthraxoutbreakinNorthernTanzania usingonehealthapproach:acasestudyofSelelawardinMondulidistrict.IntJ OneHealth2017;3:66–76,doi:http://dx.doi.org/10.14202/IJOH.2017.66-76.

MwakapejeER,HøgsetS,FyumagwaR,NongaHE,MdegelaRH,SkjerveE.Anthrax outbreaksinthehumans—livestockandwildlifeinterfaceareasofNorthern Tanzania: a retrospective record review 2006–2016. BMC Public Health 2018;18:106,doi:http://dx.doi.org/10.1186/s12889-017-5007-z.

PhillipsSJ,DudíkM.ModelingofspeciesdistributionswithMaxent:newextensions andacomprehensiveevaluation.Ecography2008;31(2):161–75,doi:http://dx.

doi.org/10.1111/j.0906-7590.2008.5203.x.

PhillipsSJ,ElithJ.POCplots:calibratingspeciesdistributionmodelswithpresence- only data. Ecology 2010;91(8):2476–84, doi:http://dx.doi.org/10.1890/09- 0760.1.

PhillipsSJ,Anderson RP,Schapire RE.Maximum entropymodeling ofspecies geographicdistributions.EcolModel2006;190(3–4):231–59,doi:http://dx.doi.

org/10.1016/j.ecolmodel.2005.03.026.

Reform,M.ofS.-P.andP.S.Kilimanjaroregionsocio-economicprofile.1998.

SanchezPA,AhamedS,CarréF,HarteminkAE,HempelJ,HuisingJ,etal.Digitalsoil mapoftheworld.Science2009;325:680–1.

SmithH,KeppieJ.Observationsonexperimentalanthrax:demonstrationofa specificlethalfactorproducedinvivobybacillusanthracis[3].Nature1954;, doi:http://dx.doi.org/10.1038/173869a0.

SteenkampPJ,VanHeerdenH,vanSchalkwykOL.Ecologicalsuitabilitymodeling foranthraxintheKrugerNationalPark,SouthAfrica.PLoSOne2018;13(1):1–13, doi:http://dx.doi.org/10.1371/journal.pone.0191704.

TarkeshM,JetschkeG.Comparisonofsixcorrelativemodelsinpredictivevegetation mappingonalocalscale.EnvironEcolStat2012;19(3):437–57,doi:http://dx.

doi.org/10.1007/s10651-012-0194-3.

TheMinisterofState—PlanningandParastatalSectorReform.TheArushaRegion Socio—economicprofile.1998.

WarrenDL,GlorRE,TurelliM.ENMTools:atoolboxforcomparativestudiesof environmentalnichemodels.Ecography2010;33(3):607–11,doi:http://dx.doi.

org/10.1111/j.1600-0587.2009.06142.x.

WinsemiusHC,SavenijeHHG,GerritsAMJ,ZapreevaEA,KleesR.Comparisonoftwo modelapproachesintheZambeziriverbasinwithregardtomodelreliability andidentifiability.HydrolEarthSystSci2006;10(3).

E.R.Mwakapejeetal./InternationalJournalofInfectiousDiseases79(2019)142–151 151