J Anim Ecol. 2019;00:1–16. wileyonlinelibrary.com/journal/jane  | 1© 2019 The Authors. Journal of Animal Ecology © 2019 British Ecological Society

Received:29January2019  |  Accepted:30March2019DOI:10.1111/1365-2656.13036

B I O L O G G I N G : ' H O W T O . . . '

Light‐level geolocator analyses: A user's guide

Simeon Lisovski1  | Silke Bauer1  | Martins Briedis1  | Sarah C. Davidson2,3,4,5  | Kiran L. Dhanjal‐Adams1  | Michael T. Hallworth6  | Julia Karagicheva7  | Christoph M. Meier1  | Benjamin Merkel8  | Janne Ouwehand9  | Lykke Pedersen10  | Eldar Rakhimberdiev9,11  | Amélie Roberto‐Charron12  | Nathaniel E. Seavy13  | Michael D. Sumner14  | Caz M. Taylor15  | Simon J. Wotherspoon16  | Eli S. Bridge17

1DepartmentofBirdMigration,SwissOrnithologicalInstitute,Sempach,Switzerland;2DepartmentofMigration,MaxPlanckInstituteforAnimalBehavior,Radolfzell,Germany;3DepartmentofCivil,EnvironmentalandGeodeticEngineering,TheOhioStateUniversity,Columbus,Ohio,USA;4CentrefortheAdvancedStudyofCollectiveBehaviour,UniversityofKonstanz,Konstanz,Germany;5DepartmentofBiology,UniversityofKonstanz,Konstanz,Germany;6MigratoryBirdCenter,SmithsonianConservationBiologyInstitute,Washington,DistrictofColumbia,USA;7DepartmentofCoastalSystems,NIOZ,RoyalNetherlandsInstituteforSeaResearch,UtrechtUniversity,Texel,TheNetherlands;8NorwegianPolarInstitute,FramCentre,Tromsø,Norway,DepartmentofArcticandMarineBiology,UniversityofTromsø–TheArcticUniversityofNorway,Tromsø,Norway;9ConservationEcologyGroup,GroningenInstituteforEvolutionaryLifeSciences,UniversityofGroningen,Groningen,TheNetherlands;10CenterforMacroecology,EvolutionandClimate,NaturalHistoryMuseumofDenmark,UniversityofCopenhagen,Copenhagen,Denmark;11DepartmentofVertebrateZoology,LomonosovMoscowStateUniversity,Moscow,Russia;12UniversityofManitoba,Winnipeg,Manitoba,Canada;13PointBlueConservationScience,Petaluma,California,USA;14AustralianAntarcticDivision,Kingston,Tasmania,Australia;15EcologyandEvolutionaryBiology,TulaneUniversity,NewOrleans,Louisiana,USA;16InstituteforMarineandAntarcticStudies,UniversityofTasmania,Hobart,Tasmania,Australiaand17OklahomaBiologicalSurvey,UniversityofOklahoma,Norman,Oklahoma,USA

CorrespondenceSimeonLisovskiEmail:simeon.lisovski@gmail.com

Funding informationDanishNationalResearchFoundation,Grant/AwardNumber:DNRF96;BundesamtfürUmwelt,Grant/AwardNumber:UTF254.08.08andUTF400.34.11;SchweizerischerNationalfondszurFörderungderWissenschaftlichenForschung,Grant/AwardNumber:31003A_160265andIZ32Z0_135914⁄1;USNationalScienceFoundation,Grant/AwardNumber:#EF-0553768,EAGER0946685,IDBR-1152356andDBI-0963

HandlingEditor:GarrettStreet

Abstract1. Light-levelgeolocatortagsuseambientlightrecordingstoestimatethewherea-boutsofanindividualoverthetimeitcarriedthedevice.Overthepastdecade,thesetagshaveemergedasanimportanttoolandhavebeenusedextensivelyfortrackinganimalmigrations,mostcommonlysmallbirds.

2. Analysing geolocator data can be daunting to new and experienced scientistsalike.Over thepastdecades, severalmethodswith fundamentaldifferences intheanalyticalapproachhavebeendevelopedtocopewiththevariouscaveatsandtheoftencomplicateddata.

3. Here,weexplaintheconceptsbehindtheanalysesofgeolocatordataandprovideapracticalguideforthecommonstepsencompassingmostanalyses–annotationoftwilights,calibration,estimatingandrefininglocations,andextractionofmove-mentpatterns–describinggoodpracticesandcommonpitfallsforeachstep.

4. Wediscusscriteriafordecidingwhetherornotgeolocatorscananswerproposedresearchquestions,provideguidanceinchoosinganappropriateanalysismethodandintroducekeyfeaturesofthenewestopen-sourceanalysistools.

5. Weprovideadviceforhowtointerpretandreportresults,highlightingparametersthatshouldbereportedinpublicationsandincludedindataarchiving.

6. Finally,weintroduceacomprehensivesupplementaryonlinemanualthatappliestheconceptstoseveraldatasets,demonstratestheuseofopen-sourceanalysis

2  |    Journal of Animal Ecology LISOVSKI et aL.

1  | INTRODUC TION

Light-level data loggers, or geolocators, have been used to trackanimals since the early 1990s (Wilson, Ducamp, Rees, Culik, &Niekamp,1992)but it hasbeen theadventof lightweightdevicessome10yearslaterthatmadethistechnologyatremendoussuccessstory.Theirweight(<0.5g)andrelativelylowcostcompared,forex-ample,tosatellitetransmittershavemadethemapplicableinlargenumbers toanimalsweighingas littleas10g (Bridgeetal.,2011).Althoughtheuseofgeolocatorscomeswithobviousethicalandlo-gisticalchallenges,suchasthepotentialimpactontheanimal(Brlíketal.,2019)andtheneedtoretrievethedevice,thesedeviceshavebeenappliedtothousandsofanimalstoansweradiverserangeofquestionsrelatingtomigrationpatterns(Briedisetal.,2019;Finch,Butler,Franco,&Cresswell,2017;Stutchburyetal.,2009),species–habitatrelationships(e.g.MacPhersonetal.,2018),identificationofbreedinglocations(Johnsonetal.,2010;Lisovski,Gosbell,Hassell,&Minton,2016),delineationofwinteringdistributions(Braceyetal.,2018;Egevangetal.,2010),migrationstopoverbehaviour(Salewskietal.,2013;Yamauraetal.,2017)andgeneralmigrationstrategies(Briedisetal.,2019;Knightetal.,2018).Light-levelgeolocationhasalreadyprovidedunprecedented insights intomigratorybehaviourandannualmovementsofanimals(McKinnon&Love,2018)andwillcontinuetodosointhefuture.

The fundamental underlying principle of light-level geoloca-tors is to estimate geographic locations frompatterns of ambientlight,allowing researchers to infer thegeneral locationandmove-mentpatternsofan individualover the time it carried thedevice.Afundamentalbutoftenunrecognizedchallenge isthetranslationof light levels into geographic locations. The simplest geolocatoranalysisderives latitude fromday lengthand longitude fromsolarnoon,which is relatively straightforward (Hill,1994).Lessobviousishowtominimizeandestimateerrors in locationdatacausedbyvariabilityandambiguityinthelightdatathatoriginatefromavari-etyofcauses,includingweather,habitat,behaviourandtimeofyear(fordetaileddescriptionontheaccuracyandprecisionofgeolocatorestimatessee:Fudickar,Wikelski,&Partecke,2012;Lisovskietal.,2012;Merkeletal.,2016;Phillips,Silk,Croxall,Afanasyev,&Briggs,2004;Rakhimberdievetal.,2016;Shafferetal.,2005).Researchersregularlystrugglewiththeanalyticalstepstoobtain,interpretandpresentlocationestimates.

Overthe lastdecade,arangeoftoolshavebeendevelopedtoanalysegeolocatordata.Choosingtherighttools,performingasci-entificallysoundanalysisandpresentingtheresultsinatransparent

mannerrequiresathoroughunderstandingoftheconceptsbehindgeolocationandthemethodologicaldifferencesbetweentheavail-abletools.Weaimto(a)describetheconceptsandprovidepracticalguidelinesforthegeneralprocedureofanalysinggeolocatordata,(b)outlinekeycriteriaformatchingresearchquestionstoappropriateanalysis and tools and (c) highlight themost important limitationsandpitfallsofgeolocatoranalysesanddiscusshowtoaddressthem.As an additional aid, we provide a comprehensive supplementaryonlinemanualthatincludesexampledataandcodeforstateoftheartanalyses,examplesofcommonproblemsandtheirsolutions,anddetailsourguidelinesforinterpretation,reportingandarchiving.

2  | FROM LIGHT‐LE VEL RECORDINGS TO LOC ATION ESTIMATES: GENER AL METHODS

Methodsof light-levelgeolocationcanbeclassified into thresholdmethods and curve-fitting methods (also known as “template-fitmethods”;Figure1).

Threshold methodsproduceasingle locationbasedontheesti-mated timesof two successive twilights events, that is, the timesatwhich the recorded light level crosses a predetermined thresh-old.Standardastronomicalformulaethendeterminethepositionofterminatoratthesetwotimes.Inastronomicalterms,the terminator isthetransitionbetweentheilluminated(day)andnon-illuminated(night)facesoftheplanet.Assumingthetaghasbeenstationaryintheinterveningperiod,itmustbelocatedattheintersectionofthesetwocurves(Figure1a).Byconsideringthealternatingsequenceofsunrise/sunsetandsunset/sunrisepairs,twolocationscanbeesti-matedper24hr.

Curve‐fitting methodsproducea locationestimatefromasingletwilight. The location of the terminator is derived from the esti-matedtimeoftwilight,andthelocationofthetagontheterminatorisdeterminedbytherateofchangeinlightlevelduringthetwilightperiod, that is, the interval of rapidly increasing/decreasing light(Figure1b).AlthoughtheEarthitselfrevolvesataconstantrate,sim-plegeometryrequiresthatthespeedatwhichtheterminatorpassesovertheEarth'ssurfaceisgreateratlowlatitudesthanathighlati-tudes.Consequently,variationintherateofilluminationchangecanbeusedtolocatethetagalongtheterminator.

Independentofmethod, twilight timesandperiodsarea focalpoint of geolocator analyses. Defining discrete twilight events– sunrise and sunset times – from raw light recordings (“twilight

toolswithstep-by-step instructionsandcodeanddetailsour recommendationsforinterpreting,reportingandarchiving.

K E Y W O R D S

animaltracking,archivaltags,FLightR,GeoLight,Movebank,probGLS,SGAT,solargeolocation

     |  3Journal of Animal EcologyLISOVSKI et aL.

annotation”) is, therefore, the first importantstep forall analyses.Oncetwilighttimesaredefined,onecanobtainquickbutcrudeloca-tionestimatesandapplymoresophisticatedmodellingapproachestorefinetheseinitiallocationestimates.Particularlyforthelatter,adiverserangeofanalyticaltoolshasbeendevelopedinrecentyears.

3  | WHAT DETERMINES THE CHOICE OF TOOL S FOR MY ANALYSIS?

Wehererecommendasuiteofsixfrequentlyusedopen-sourcetools,forgeolocatoranalysesanddiscusstheirrequirements intermsofdataquality,userexperienceandcomputingpower:theonlineplat-form TAGS (https://tags.shinyapps.io/tags_shiny), the R packagesTwGeos(Lisovski,Sumner,&Wotherspoon,2015),GeoLight(Lisovski&Hahn,2012),probGLS(Merkeletal.,2016),SGAT(Wotherspoon,Sumner, & Lisovski, 2013a) and FlightR (Rakhimberdiev, Saveliev,Piersma,&Karagicheva, 2017).Weomit three other open-sourceRpackagesUkfsst(Nielsen,Bigelow,Musyl,&Sibert,2006),Trackit (Nielsen&Sibert,2007)andPolarGeolocation(Lisovski,2018),asthefirsttwoarespecializedtoolsforanalysisoffishmovementsandthelastoneaddressesarelativelyspecificproblemofestimatingloca-tionsduringpolarsummerunder24-hrdaylight.

Notoolormethodshouldbeconsideredasilverbullet for theanalysisofgeolocatordata.Thechoiceofanalysismethodwillul-timatelydependon theoverall researchquestion, the typeof tagused,thelifehistoryandbehaviouroftheanimalsstudied,andre-searchers' available time andexperience. For instance, is theuserlookingforcoarseestimatesofbreedingandnon-breedinglocationsorfiner-scalestopoverlocationsanddurations?Istheanimallivingindenseforestswhereshadingisexpectedtobehighduringsomeperiodsoftheyear?

Inthefollowingsection,wediscussthesetrade-offsbasedonthedifferent analysis stepsneededand theavailability and limitationsofmethodsforeachstep(Figure2andTable1).Inaddition,thereareim-portantconsiderationstotakeintoaccountbeforeusinggeolocatorsinananimalmovement study.Forexample, light-level geolocators can-notbeusedtoanswerresearchquestionsaboutmovementslessthan~200km,becausethesemovementsarelessthanthetypicalerrorofallmethods(Lisovskietal.,2012).Similarly,ifananimalmainlymoveslati-tudinallyduringtheperiodsaroundtheequinoxes,whenlocationerrorislargest,estimatingmigrationroutesandstopoversitesfortheseani-malsishardlyornotpossible,thoughbreedingornon-breedingrangescanstillbeestimated(Briedisetal.,2016).Thus,therearelimitationstotheuseofgeolocatorstoanswerquestionsaboutmigratoryanimalsthatcannotbeovercomebyeventhemostsophisticatedanalysisandshould,therefore,beconsideredpriortothebeginningofanystudy.

The open-source tools that are currently available implementthestepsof light-levelgeolocationanalysis inslightlydifferentways(Figure2b,Table1).Forexample,thefirststepforanalysinglight-leveldata is toannotate the twilightevents (step 1) that canbedone inTAGS,TwGeos and GeoLight.Forcurve-fittingmethodsimplementedinTwGeos and FlightR,oneshouldusethetwilighteventspredefinedinoneoftheabove-mentionedpackagestoextracttwilightperiodsfromrawlight-leveldatausingfunctionsinFLigthR or TwGeos.Calibrationfunctions (step 2)are included inall toolsexceptSGAT.Methodstoestimatelocations(step 3),refinetheseestimates(step 4)andextractmovement patterns (step 5) are implemented inGeoLight, probGLS,SGAT and FLightR.Insum,thesetoolsareasfollows:

• TAGSisanonlineplatformthatallowsuserstouploadrawlight-leveldata,annotatetwilights,performcalibrationandperformasimplethreshold-basedanalysis.Althoughmoststepsarebasedon GeoLight,theadvantageofthistoolisagraphicaluserinterface

F I G U R E 1  Theprinciplebehindlocationestimatesfromlightrecordings.Twobroadmethodsexist,thethresholdmethodandthecurvemethod.Anyobserverexperiencingatwilightmustbelocatedontheterminator–theboundarybetweentheilluminatedandthenon-illuminatedsurfaceoftheEarth.Thethresholdmethod(a)producesasinglelocationfortwosuccessivetwilightsestimatedviathetimesatwhichtherecordedlightlevelcrossesapredeterminedthreshold.Thecurve-fittingmethod(b)producesalocationforasingletwilightbasedonthepositionoftheterminatorandtherateofchangeinlightlevels.Thisrateofchangevarieswithlatitude,asindicatedbythearrows(arrowsizeisroughlyproportionaltorateofchange)

Threshold method Curve method(a) (b)

4  |    Journal of Animal Ecology LISOVSKI et aL.

tovisualizetherawlight-levelrecordingswithdifferentzoomop-tions,fastprocessingandauser-friendlyinterfacethatdoesnotrequireuserstowritecode.

• TheRpackageTwGeosprovidestoolstoimportrawlight-levelrecordings from various tags, a set of functions to plot andinvestigate the rawdata,andan interactive interface for twi-light annotation and extraction of twilight times and twilightperiods.

• TheRpackageGeoLightprovidessimplethreshold-basedlocationestimatesandmovementanalysis.Groupedmethods,whichpro-videsinglelocationestimateswithuncertaintybasedonaseriesoftwilighttimesduringstationarybehaviour,haverecentlybeenincluded(Hiemer,Salewski,Fiedler,Hahn,&Lisovski,2018;andsupplementaryonlinemanualChapter5“Movementanalysis”).

• TheRpackageprobGLSusesiterativeprobabilitysamplingtoes-timatenumerous trajectoriesbasedon threshold-based twilight

eventsandadditionalinputssuchasthetwilighterror,movementandspatialmasks.Thesecanbeusedtoderiveamostlikelytrackand location-specific uncertainties (see supplementary onlinemanualChapter6forexampleofaBrünnich'sguillemot(Uria lom‐via)track).

• TheRpackageSGATusesametropolissamplingprocesstocreateMarkovchainMonteCarlosimulationofmovementtrajectories(Sumner, Wotherspoon, & Hindell, 2009). Both threshold andcurvemethodscanbe implemented,and locationestimatesarerefinedusingatwilightandmovementmodelaswellasaspatialmask.

• TheRpackageFLightRusesthecurvemethodandaparticlefiltertoestimateactualmovement.Itsstartingpointisthespatiallyex-plicitlikelihoodforeachtwilight,whichiscalculatedbasedonthevalueandthevariationoftheslopeofobservedversusexpectedlightintensitiesduringthecalibrationperiod,whengeographiclo-cationofatagisknown.Usingamovementmodelandaspatialmask,theparticlefiltergraduallyprogressesthroughthespatiallyexplicit likelihoods in time,weighing possiblemovement trajec-tories to derive a probability distribution, which can ultimatelybeusedtocalculatethemediantrackanditscredibleintervalss(Rakhimberdievetal.,2015).

4  | ANALYSING LIGHT‐LE VEL GEOLOC ATOR DATA

4.1 | Step 1: Twilight annotation

Definingtwilighteventstypicallyrequiresalightintensitythreshold.Thetimewhenlightlevelsexceedthisthresholdinthemorningand

FIGURE 2 Exampleofhigh-andlow-qualitydata(a).Thequalitycanbeaffectedbymanyfactorssuchasthesensitivityofthesensor,thetagsettings(recordingfrequency),theindividual’sbehaviorandthehabitat(grounddwellingvs.openlandscapespecies),aswellasbyweatherandtopography.Thelefthigh-qualitydatawasrecordedonablack-tailedgodwitusingaloggingintervalof5minutesandasensorthatisabletoresolvetheentirelightrange(datafromRakhimberdievetal.,2016).Therightlow-qualitydatahasbeenrecordedonaCommonRosefinch(Carpodacus erythrinus)witharecordingintervalof10minutesandasensorthatonlyresolvesdimlightvalues(early/latesunrise/sunsettimes).Themajordifferenceisthatthereareclearsunriseandsunsettimesinthehigh-qualitydata,whichrecordmorethan5readingsduringthetwilightperiods(unpublisheddata).Theseperiodsarenotresolvedinthelow-qualitydataduetothelongerintervalbetweenrecordingsandpresenceoflightvaluesofzeroevenduringtheday,probablyduetoheavyshadingcausedbytreesandshrubsatthenon-breedinglocation.Thequalityofthedataisamajordeterminantforthemethodandthetoolthatcanbeusedfortheanalysis(b).Whilecurvemethodsrequiregoodqualitydataaroundtwilightandatleastacoupleofrecordingsduringtheincreasinganddecreasinglight,thresholdmethodscandealwith(almost)alldatawithacleardifferencebetweennightandday.Theflowchartshowssubsequentstepsandavailableanalysistoolsforeach

High quality data Low quality dataLi

ght

DateLi

ght

Date

Tim

e

Day Day

extract twilight periodsTAGS, TwGeos, FLightR

2. CalibrationTAGS, TwGeos, GeoLight

3. Simple Threshold Estimates

TAGS, GeoLight, SGAT

1. Twilight annotationTAGS, TwGeos, GeoLight

Movement analysisGeoLight

Curve methods Threshold methods

2. CalibrationSGAT, FLightR

FLightR SGAT probGLS GeoLight

(a)

(b)

     |  5Journal of Animal EcologyLISOVSKI et aL.

fallbelowitintheeveningaredefinedassunriseandsunset,respec-tively.Althoughthesetwilight eventsarethesoleparametersusedforthethresholdmethod,theyarealsousedasanintermediatesteptocalculatetwilight periodstoapplycurve-fittingmethods.Duringthisfirststepoftheanalysis,westronglyrecommendtakingthetimetovisuallyreviewtherawlightdatatoidentifypartsofthetimese-ries thatmightbestronglyaffectedbyshading (Figure2a;dueto,forexample,feathergrowth,monsoonsorbreedingbehaviour)andto identifydata gapsor light recordingerrors associatedwithde-vicemalfunction.Becomingfamiliarwiththedataprovidesaroughideaoftheoveralldataqualityandwhichmethodsarefeasible(seeFigure2andTable1).TheonlineplatformTAGSandtheRpackageTwGeosprovidethemostconvenientprocedurestodefinetwilight

events. BothTAGS and TwGeos allow interactive visualization andeditingoftwilightevents.Inaddition,theykeeptrackofuseractionsandthusdocumentthechoiceoftwilightevents(ortheirexclusion)forrepeatability.

Two frequentquestions troubling researchers are (a)what is agood threshold? and (b) should one edit potentially false twilighttimes?Regardingthethreshold,oneshouldchoosethelowestvaluethatisconsistentlyabovenoiseinthenight-timelightlevels.Itisim-portanttorealizethatanythresholdistag-andspecies-specific.Theprospectofeditingtwilight isafarmoresensitiveissue.Thereareprocedures inTAGS and TwGeos thatautomaticallydetermine twi-lightevents(seeBox1andChapter4inthesupplementaryonlinemanual).However,shadingduringthetwilightperiodsmayleadto

TA B L E 1  Abriefcharacterizationofsixopen-sourcetoolsfortheanalysisofgeolocatordata

TAGS TwGeos*  GeoLight probGLS SGAT**  FLightR

Softwarecapability

1.Twilightannotation

Definetwilightevents ✓ ✓ ✓

Extracttwilightperiods ✓ ✓ ✓

2.Calibration

Thresholdmethod ✓ ✓ ✓ ✓

Curvemethod ✓ ✓ ✓

3.Locationestimation

Method(Threshold/Curve) T T T T/C C

Uncertaintyestimates (✓) ✓ ✓ ✓

4.Refinementoflocations

GroupedAnalysis***  ✓ ✓

Trackoptimization Probabilitysampling

MCMC Particlefilter

Inputs:Twilighterror ✓ ✓ ✓

Movementmodel ✓ ✓ ✓

Spatialmask ✓ ✓ ✓ ✓

5.Movementanalysis

MovementversusResidency

✓ ✓

Requirements

Dataquality

Tolerancetoshading ••• •• •• •

Minimumrecordingfrequency

1/10 min 1/10 min 1/10 min 1/5min**** 

Fulllightrangerequired ✓

Codingexperience •• •• ••• •

Calculationtime(trackestimation)

Minutes Minutes Minutes Hours

Note:Thetoolsdifferinwhichanalyticalstepstheycover,complexityandrequirementsastodataqualityandresearcherexperience(seetextfordetail).*basedonBAStag(Wotherspoon,Sumner,&Lisovski,2013b)withextrafunctionalities.**furtherdevelopmentoftheRpackagetripEstimation(Sumneretal.,2009).***Singlelocationestimatesforstationaryperiods(seriesoftwilightevents).****Insomecases,notablyunderlowshading,FLightRcandealwith1/10minfrequency.

6  |    Journal of Animal Ecology LISOVSKI et aL.

atimeserieswhereinthe lightvaluescross thethresholdmultipletimesduringashortperiodoftime,andautomatedprocessesmaypickthewrongsunrise/sunsettime.Onehastheoptionhereofad-justingordeletingsuchtwilightevents.Werecommendaconserva-tiveapproachtoediting,meaningthatitisbettertohavesomenoiseinthedatasetthantoerroneouslyremovegooddatapointsorinfor-mativedatathatmightconveyrapidlong-distancemovements.Also,the criteria for editing or removing poorly classified twilightswilldependonthemethodthatisfurtherusedtoinferlocations.Duringperiodsoflimitedmovement,forcurve-fittingmethods,day-to-dayconsistency in the shapeof the curve around sunriseor sunset is

mostimportant,whileforthresholdmethodstheday-to-dayconsis-tencyinthetwilighteventsisimportant(Figure3).

4.2 | Step 2: Calibration

Geolocatorsvary in theway theymeasure light and other informa-tion.Most importantly, tags can have different data logging sched-ulesand(arbitrary)unitsfor light-leveldata.Therefore,calibration isneededtorelateobservedandexpectedlightlevels.Thisprocedureisthemostimportantstepinanygeolocatoranalysisasitaccountsfortwotypesoferrors:(a)thedevice-specificaccuracytomeasuretrue

BOX 1 Annotating twilight times using the R package TwGeos

ThefunctionlightImageprovidesanoverviewplotoflightdatarecorded(data:Europeanbee-eater,seesupplementaryonlinemanualChapter1).

library(TwGeos)

raw <- readMTlux(̀ LightRecordings.lux̀ )

offset = 12 # the offset for the vertical axis in hours.

lightImage(raw, offset = offset, zlim = 12)

Inthisplot,eachdayisrepresentedbyathinhorizontallinethatplotslightvaluesasgrayscalepixels(dark=lowlightandwhite= maximumlight).

threshold <- 2.5 # slightly above night recordings

twl <- preprocessLight(raw, threshold=threshold, offset=offset)

Tovisualizethequalityofthedata(e.g.cleanlighttransitionsduringthetwilighttimes)andtodefinealight-levelthresholdwecanplotpartsofthelightrecordings.

with(raw[2000:5000,], plot(Date, Light))

abline(h=threshold, col="grey80", lty = 2, lwd = 2)

Hou

r

Sep Nov Jan Mar May Jul

Jul 18 Jul 20 Jul 22 Jul 24 Jul 26

0

5

10

15

Ligh

t

     |  7Journal of Animal EcologyLISOVSKI et aL.

lightintensitiesand(b)thelevelofshadingananimalexperiencedinits typical habitat andperforming its typical behaviour.To this end,aperiodisoftenselectedwhentheanimalwasataknownlocationin its preferred habitat for at least a couple ofweeks, for example,at the site of deployment, and light levels recorded during this pe-riodarecomparedtothoseexpectedfortheknownlocation.Forthe“thresholdmethod”,thecalibrationprocessdeterminesasinglebestreferencesunelevationangle,whichisthepositionofthesun(e.g.5degreesbelowthehorizon)duringthetwilighteventsacrossallcali-brationrecordings.Notethatsomeanalyses(e.g.SGAT)requirezenithangle,definedastheanglefromthezenith, insteadofsunelevationanglewhichistheangleofthesuninrelationtothehorizon.Forthe“curve-fittingmethod”,thecalibrationprovidestheslopeofrecordedversusexpectedchangeinlightduringthetwilightperiods.

Giventheimportanceofcalibration,itiscrucialtoplanthestudy,forexampledeploymenttiming,toensurethatgeolocatorrecordingsincludeanappropriatecalibrationperiodataknownlocation.Suchacalibration isbestperformedas “in-habitat calibration”,wherein thedeviceisattachedtotheanimaltobetrackedwhileitisinitstypicalenvironment.Thisformofcalibrationaccountsfortheinfluenceofbe-haviourandhabitatondataqualityandmightbeparticularlyusefulforspecies that live inshadedhabitatsor thosewhosebehaviour leadstofrequentshading.Inreality,however,manybirdsaretaggedduringthebreedingperiod,wherebehaviourandhabitatmaydifferfromtherestof theannual cycle.Forexample,birdsmayusenestboxes re-sultinginmisdetectionsofthetruetimeofsunriseorsunset,causingthelightpatternrecordedinthebreedingseasontobedifferentfromthelightpatternrecordedthroughouttherestoftheyear,whenthe

birdisnotusinganestbox(Meieretal.,2018).Thoughthebreedinglocationmaythereforenotbequalifiedas“typical”behaviour,inmanycasesitwillbetheonlyknownlocationforthetaggedbirdthatcanbeusedtocalibratethedata.Ifcavityorburrownestingisapparentinthetaggedbird,itisnotrecommendedtousethisnestingperiodforcalibration.Similarly,whengeolocatorsaremountedonthe leg, thismayalso introduceshadingeventsduring thecalibrationperiod, forexamplewhenthelegsarecoveredorsubmerged.

Toaccountforsuchsourcesoferror,thecalibrationcanpoten-tiallybedoneusing recordingsofa shortperiodafter thenestingwhile individualsremainatthebreeding location.However, ifcali-brationofgeolocatorsonanimals isnotpossible,asanalternativeor in addition to an “in-habitat” calibration, it is recommended toperforma“rooftopcalibration”,whereeachdeviceissimplyplacedin a stationary location (e.g. a rooftop), for one or 2 weeks priorto, or after, the deployment. This form of calibration helps to es-timate thedevice-specific inaccuracy.FlightR provides anoptionalprocess inwhich calibration canbe calculated for stationaryperi-odsinunknownlocations(supplementaryonlinemanualChapter8“Calibration”).Forthresholdmethods,thiscanalsobedoneusinga“Hill-Ekstromcalibration”(Box3andsupplementaryonlinemanualChapter5“Hill-Ekstromcalibration”).

4.3 | Step 3: Simple threshold estimates

Afteridentifyingtwilighteventsandperformingcalibration,thenextstep is to estimate geographic locations.We recommend startingwithasimpleprocedurebasedonthethreshold-basedanalysis(see

WecancalltheinteractiveprocesstodefinetwilighttimesusingthefunctionpreprocessLight(seesupplementarymaterialfordetails).Finally,wecanplotthesunriseandsunsettimesintothelightimage:

lightImage( tagdata = raw, offset = offset, zlim = 12)

tsimagePoints(twl$Twilight, offset = offset, pch = 16, cex = 1.2,

col = ifelse(twl$Deleted, "grey20", ifelse(twl$Rise, "firebrick", "cornflowerblue")))

AndextractthetwilightperiodsusingthefunctionextractCrepuscular.

Detailedexplanationonhowtoannotatetwilight,commonpitfallsandbestpracticerecommendationsareavailableinsupplementaryonlinemanualChapter4.

Hou

r

Sep Nov Jan Mar May Jul

BOX 1 (Continued)

8  |    Journal of Animal Ecology LISOVSKI et aL.

Box2andFigure2).Evenifamoresophisticatedanalysisisplanned,aquickthreshold-basedanalysisprovidesafirst impressionofthetrackandcanrevealdata-qualityissuesthatmayhaveescapedear-lierscrutiny.Aninitialthresholdanalysiscanalsoserveasasanitycheckforfurtheranalysis. It isunlikelythatanysuccessivemodel-lingapproachwill dramatically change thegeneralmovementpat-ternrevealedbyasimplethresholdanalysis.Infact,ifthemodellingdoes produce dramatically different results, then somethingwentwrongwiththeanalysis(e.g.erroneouscalibration,poormovement/behavioural model, inappropriate spatial mask). Finally, a simplethreshold-basedanalysisprovidesanopportunitytoevaluatethein-fluenceofthecalibrationparameters,particularlythereferencesunangle.Althoughexperimentingwithparametersettingssoundsverysubjective,we recommendperforming several preliminary thresh-oldanalysestoevaluatetheeffectsofchangesinthereferencesunelevation angle on the results/track. For example, the amount ofshadingexperiencedbyananimaloftenvariesacross the trackingperiodasthehabitatsusedforbreeding,stopoverandnon-breedingmaydiffer in topography, foliage and localweather that canhavedifferentially influence on location estimates during distinct peri-ods. Insuchcases,goingbacktothecalibrationstepandapplyingalternativemethodssuchasthe“Hill-Ekstromcalibration”(Lisovskietal.,2012)mightberequired.SeeBox3andsupplementaryonlinemanualChapter5(“Hill-Ekstromcalibration”)forexamples investi-gatingsuchissuesaswellasoptionsthatmayallowaccountingforandevencorrectingthem.Theresultsofasimplethresholdanalysisprovideapilotoverview,butoftenshowlargeandunrealisticmovements(Box2)overshorttimeperiodsduetovariableshading.Furthermore,aroundtheequi-noxesdaylengthissimilaracrosstheglobe,andevensmalldeviationsinsunriseorsunsettimesorslopewillresultinlargespatialerrors,particularlyinlatitude.Therefore,itisoftendesirabletorefinethesecrude location estimates by incorporating knowledge about thetrackedanimal,suchasitsmovementcapabilitiesandwhathabitat(e.g.oceanvs.land)itislikelytouse.

Sep/04 Sep/05

Ligh

t (a

rbitr

ary

units

)

Jul Sep Nov Jan Mar May

24 h

rRaw Geolocator Data

1. Twilight annotation

Light intensity threshold

2. CalibrationTwilight events (sunrise/sunset)

Twilight period

Freq

uenc

y

– 6° –5° –4° –3° –2° –1°Sun elevation angle at sunrise/sunset

(degrees below horizon)

Threshold method(twilight events)

Curve method(twilight periods)

Mea

sure

d lig

ht

Expected light–3 –2 –1 0 1 2

2

4

8

3. Location estimation

Calibrationperiod

Median

Reference sun elevation angle Reference slope parameter

F I G U R E 3  Aschematicoverviewoftheanalyticalstepsrequiredtotranslaterawlight-levelgeolocatorrecordingstosimplelocationestimates.Geolocatorrecordambientlightlevelsovertime(toppanel)resultinginatimeseriesofdaylightpatterns.Inaninitialstep(1),discretesunriseandsunsettimes(twilighttimes)aredefinedusingalightintensitythreshold.Whilethesetwilighteventsareusedtoobtainlocationestimateswiththe“thresholdmethod”,twilightperiodsthatfollowasunriseandprecedeasunsetareusedin“curvemethods”.Usinglightrecordingsfromaknownlocation(e.g.afterthedeploymentofthelogger),thecalibrationprocess(2)establishestherelationshipbetweenobservedandexpectedlightlevels.Thiscalibrationprovidesareferencesunelevationangleforthe“thresholdmethod”andareferenceslopeparameterforthe“curvemethods”thatcanthenbeusedtoestimatetwodiscretelocationsperday.TheexemplarydatashownhavebeenrecordedonaBar-tailedgodwit(Limosa limosa)andweredescribedinRakhimberdievetal.(2016).SimplelocationestimateswerecalculatedusingtheRpackageGeoLight

     |  9Journal of Animal EcologyLISOVSKI et aL.

Box 2 Simple location estimation using GeoLight

Startingwiththetablethatcontainstheannotatedtwilighttimes(seeBox1),wecanestimatethereferencesunelevationangle.Here,weuseaspecificrangeoftherecordings,aftermakingsurethatthebirdhasbeenstationaryatthedeploymentsite(51.32°N,11.96°E)duringthisperiod(e.g.breedingtime).

First,weexportthetableintotherequiredGeoLightformat:library(GeoLight)

twl.gl <- export2GeoLight(twl) # export the table into the required GeoLight format

Next,weperformcalibration:lon.calib <- 11.96

lat.calib <- 51.32

tm.calib <- as.POSIXct(c("2015-07-20", "2015-08-29"), tz = "GMT")

twl.calib <- subset(twl.gl, tFirst>=tm.calib[1] & tSecond<=tm.calib[2])

gE <- getElevation(twl = twl.calib,

known.coord = c(lon.calib, lat.calib))

Usingthereferencezenithangle(93.50°),wecanestimateandplotlocations:crds <- coord(twl.gl, degElevation = 90 - gE[1])

tripMap(crds, ylim = c(-35, 60),

xlim = c(0, 20), equinox = FALSE)

Longitude

Latitude

0 5 10 15 20

−20

0

20

40

60

10  |    Journal of Animal Ecology LISOVSKI et aL.

4.4 | Step 4: Refinement of locations

Location estimates are typically refined via a statisticalmodellingeffort that incorporates at least three components: (a) a twilightmodel, (b) amovementmodel and (c) a spatialmask. The twilight model isbasicallyanextensionofthecalibrationthatnotonlyde-terminesoneoverallbestsunelevationangleorsingleslopeacrossthewholecalibrationperiod,butalsoconsidersvariationofthedailylightpatternwithinthecalibrationperiod.Incorporatingthetwilight

errordistributionallowscalculationofaspatial likelihoodforeachtwilight(Figure4)viatheknowndistributionofthetwilighteventsorfromthevariationintheslopeparameterofthetwilightperiods(e.g.standarddeviationofslope).

Amovement modelaimsto limittheoccurrenceofunrealistic lo-cationsbyconstraininglengthandfrequencyoflong-distancemove-mentsinaccordancewithwhatisknown(orassumed)aboutaspecies'propensity for movement. Various possibilities for specific move-mentmodelsexistthatdifferinlevelofcomplexityandassumptions;

Box 3 Movement analysis and “Hill‐Ekstrom” calibration

GeoLightofferstoolstodistinguishbetweenperiodsofmovementandperiodsofresidency.ThechangeLightfunctionusesthetwi-lighttimes(notthelocationestimates)thatareunaffectedby,forexample,theequinoxandsearchesforsuddenchangesthatindicatechangesinthelocationoftheanimal.

library(GeoLight)

cL <- changeLight(twl = twl.gl, quantile = 0.85)2.

53.

03.

54.

04.

55.

05.

5S

unris

e (r

ed)

16.5

17.0

17.5

18.0

18.5

19.0

19.5

Sun

set (

blue

)

0.00

0.04

0.08

0.12

0.00

0.04

0.08

0.12

Pro

babi

lity

of c

hang

e

Sep Nov Jan Mar May

     |  11Journal of Animal EcologyLISOVSKI et aL.

for example, models can assume a constant velocity or a bimodalspeed distribution to distinguish resident andmovement behaviour(Schmaljohann,Lisovski,&Bairlein,2017;Yamauraetal.,2017).

Aspatial maskdefinesspatial likelihoodswithintheanalysis toreduceorevenpreventunrealisticlocationestimates,thatis,wherethe specific animal is very unlikely to occur. Spatialmasks can be

Theoutputplotshowsthetwilighttimes(sunriseinredandsunsetinblue)inthecenterplot,theprobabilitiesofchangeinthelowerplotandtheresultingsequenceofmovementsandperiodsofresidency.

Withthisinformation,wecanperformacalibrationforaperiodwithunknownlocation(e.g.thelongeststationaryperiodnr.5).ThefunctionfindHEZenithfromtheRpackageTwGeosanalysesthisperiodandcalculatesthezenithangle(sunelevationangle)thatresultsinthelowestvariationintheestimatesoflatitudes(fordetailssee:Lisovskietal.,2012andsupplementaryonlinematerialChapter5,“Hill-Ekstromcalibration”).

library(TwGeos)

StartEnd <-range(which(twl$Twilight>=min(twl.gl$tFirst[cL$site==8]) &

twl$Twilight<=max(twl.gl$tFirst[cL$site==8])))

HE <- findHEZenith(twl, range = StartEnd)

−60

−40

−20

0

20

40

60

Latit

ude

15−Jul 01−Oct 18−Dec 05−Mar 22−May

90 92 94 96 98

4

6

8

10

12

14

zenith angle

sd in

latit

ude

(with

in ra

nge)

zenith = 94.25

BOX 3 (Continued)

12  |    Journal of Animal Ecology LISOVSKI et aL.

as simple (binary) as restricting land animals to land by excludingoceans or large waterbodies, or more complex when consideringspecies-specific habitat, altitudes or other preferences. Theymayalso include measurements of other variables, such as tempera-ture, that are sometimes recorded by multi-sensor loggers (Lam,Nielsen,&Sibert,2010;Lavers,Lisovski,&Bond,2019;Merkeletal.,2016;Shafferetal.,2005;Sumneretal.,2009).Thefullmodel-basedanalysiscombinesthesesourcesofinformationusingtheprior

distributionsofprobabilitiesfromthetwilightmodel,themovementmodeland the spatialmask togeneratemultiple (often thousandsof) simulated locations for each tracking interval (see Figure 4).Thesesimulatedpointsarethenregardedasaposteriorprobabilitydistributionfromwhichonecanderivethe“best”path(mostprob-able the path that is usually themean ormedian of all estimatedlocationswithineachtimeinterval)andtheassociatedcredible in-tervals. “Best” in thiscasemeans that theresult is themost likely

F I G U R E 4  Theprinciplesbehindtherefininglocationprocess.Usingknowledgeonthevariation(errordistribution)inthedetectionofsunriseandsunsettimesforthe“thresholdmethod”andthevariationintheslopeparameterforthe“curvemethod”allowscalculationofspatiallyexplicitlikelihoodsforeachtwilight(thetwilightmodel).Furtherassumptionsinthemovementbehaviourofthespecies(themovementmodel)andthespatialoccurrencecanthenbeincorporatedinamodellingapproachtoestimatethemostlikelymovementpathandtheassociatedcredibleintervals.ThedatausedareagainfromtheBar-tailedgodwitandthefinalresultscomefromananalysisusingtheRpackageFLightR(Rakhimberdievetal.2016)

Input 1: Twilight model Input 2: Movement model

Input 3: Spatial mask

Binary Continuousor

0 km/h 100 0 km/h 100

Residency

ore.g.

e.g.

e.g. probability sampling, MCMC, particle �lter

Median track95% credible interval

5

0

–5

50

45

40

35

March/01 May/01

Model:

Movement

Threshold method Curve method

Twilight error (min)

Sunrise (N = 1)

Likelihood maps

Sunset (N = 1) Sunrise (N = 1)Sunset (N = 1)

Sunset & Sunrise Sunset & Sunrise

0 50 –3 2Expected light

Den

sity

Long

itude

Latit

ude

Den

sity

Mea

sure

d lig

fht

     |  13Journal of Animal EcologyLISOVSKI et aL.

representation of the recorded geolocator data and the specifiedpriorassumptionsontwilighterrordistribution,movementcapaci-tiesandspatialrestrictions.

4.5 | Step 5: Movement analysis

Inadditiontoderivinglocationdata,werecommendaformalanaly-sisofmovementbehaviour todelineateperiodsofmovementandresidency.Thisanalysiscanbeafinalstep,basedonposteriorprob-abilities,oran intermediatestepbasedontwilight times. Ineithersituation, determiningmovement dates is not trivial, given the in-herent limitsontheaccuracyandtheprecisionofthe locationes-timates.Differentmethodshavebeenproposedand implementedinthedifferenttools,someofwhichallowrefiningearliersteps inthe analysis (e.g. calibration). An advantage to determiningmove-mentandstationaryperiodsasanintermediatestepisthatitallowsforgroupingandclassifyingsegmentsofthelight-leveltimeseries.Withthesedatacategorizedinthismanner,wecancombinethedatafromseveral successivedays toderivea single location for apre-sumedstationaryperiod(seesupplementaryonlinemanualChapter5“MovementAnalysis”andChapter7“TheGroupModel”fordetailsandexamples).

5  | AF TER THE ANALYSIS

5.1 | Interpretation

Alwaysconsider the limitationsof theresults.Regardlessof theinitial research aims, the geolocation data quality andmigrationpathoftheanimaloftenrequireresearcherstoadjustthescaleatwhichtheydrawconclusions.Ageneralawarenessoftheprecisionandaccuracyofgeolocatortracksisalwayskey,anditisimportanttorealizethatprecisionandaccuracyfrequentlychangethrough-out the trackingperiod.Thehighprecisionexpectedduringonelongshade-freeperiodwilllikelynotapplytoashortshadyperiod.Generally, inferencebecomesmorechallengingoreven impossi-blewhenmigrationpathsaremoregradual rather thandiscrete,whenstopsareofshortdurationanddistancesaresmall,andforlatitudinalmovementsclosetoequinoxperiods.Inpractice,withnoisygeolocationdatayoumaynotalwaysbeabletoaccuratelyand reliably recover stopover sites and timingduring themigra-tionperiod, and it is important to communicate thisuncertaintythroughtheinclusionoferrorrangesintheresults.Notably,suchlimitationsarenotperseuniformacrossanalyses,sincedataqual-ityandmigrationpathscanvary, forexample,betweenseasons,devices,sexes,populationsandspecies.Thelargerthevariabilityindataquality,devicesandmigrationpaths(timingandlocations),themorethecautioniswarrantedtopreventoverestimatingpre-cision and accuracy and over-interpreting data. Caution is par-ticularlywarranted in studies investigatingmultiple specieswithdifferenttypesoftagsandspeciesthatoccupyareaswithhighlyvariablehabitats.

5.2 | Presentation

Geolocator data are often presented inmaps,which are an intui-tiveway to show locations andmigrationpaths.Unfortunately, inmany publications,maps of the results of geolocator analysis ne-glect the uncertainty and assumptions in the underlying analysis,yettheyareimportantforthevalidation,perceptionandinterpreta-tionofresults.Therefore,westronglyencouragevisualizationsthatshowtheprecisionandaccuracy (uncertainty)associatedwith thedata(e.g.Rakhimberdievetal.,2016).Forinstance,summarizedlo-cationsshouldalsoindicatetheirerror.Providingseparateplotsoflatitude versus time and longitude versus time is a simpleway tomakethispresentation(seesupplementaryonlinemanualChapter8“Explorationofresults”).

5.3 | Reporting

While conducting analyses, it is crucial to clearly and unambigu-ouslydocumentallstepstakensothatyourworkwillbereproduc-ible.Althoughinmanycasesthestandardproceduresandpathwaysoutlinedaboveand in thesupplementarymaterialwill suffice, thewealth of options in each analysis requires that methodologicalchoicessuchasparametersettings,filteringorsimilaranalyticalde-cisionsbedocumented.Thesameappliestodeviationsfromstand-ard procedures. The R packages described in this paper facilitatethe creation of annotated and clearly documented scripts, whichwerecommendarchivingassupplementstoanypublishedresearch.Werecommendcompleteandseparateanalysisforeachtag,ratherthansettingupgeneral analysesand looping throughseveral tagsusingtheverysameparametersettings.Thisrecommendationespe-ciallyappliestomoresophisticatedanalysesthatareoftensubjecttotag-specificassumptions.Thesupplementaryonlinemanualgivesanextensiveoverviewofhowtoreportmethodologicalchoicesandpresentresults.

5.4 | Archiving

Archiving geolocator data is not only good scientific practice,it is a precondition for publication in a growing number of jour-nals.Properarchivingretainsthevalueofthedataforfuture(re-)analyses,allowing for re-analysiswithnewanalyticalmethodsoranoverarchingsyntheses(Finchetal.,2017).Asthereareseveraltoolsthatcouldpotentiallybeused,andeachrequiresspecificval-uesforasuiteofparameters,itiscrucialtoarchivebothrawdataandlocationestimatesinconjunctionwiththecodethatledtothespecificresults.

Although several databasesexist for sharing andarchivingan-imalmovementdata, the foremost as a general researchplatformfor animal movement data isMovebank (https://www.movebank.org/).Movebank isparticularlywellsuitedforgeolocatordataas itincludes features tailored to tracking based on ambient light lev-els,supportingstorageofraw lightrecordings, twilightannotation

14  |    Journal of Animal Ecology LISOVSKI et aL.

tables,locationestimateswithcredibilityintervalsandcustomfilessuch as R scripts or model output, along with deployment- andstudy-levelmetadata.Researchersareencouragedtopublishtheirdata inthepublicdomain,whichcanbedonebypublishing intheMovebank Data Repository,wherein a dataset can acquire a digitalobjectidentifier(DOI),persistentlink,licenceandcitationafterun-dergoingreview.Forin-progresswork,Movebankprovidesarangeofsharingoptions,forexampleallowingownerstostoreandman-agedataprivately,giveaccesstospecificresearchersandallowthepublictodiscoverbutnotaccessthedataortodownloaddataafteracceptingowner-definedtermsofuse.Werecommendstoringdatain Movebankorsimilarrepositoriesevenduringpreliminaryanalysisstageswhentheresultsandanalysesremainunpublished.Inthesesituations,theonlinearchiveprovidesasecurebackupandcanbediscoveredbyotherresearcherswhocancontacttheowneraboutpotentialcollaborationusingthedata.

6  | SUPPLEMENTARY MATERIAL: THE MANUAL

Thispublicationisaccompaniedbyacomprehensiveonlinemanualthatapplies theconceptstoseveraldatasets,analysesthemusingthe open-source tools discussed above and provides step-by-stepcode as well as recommendations for addressing common issues(https://geolocationmanual.vogelwarte.ch/).We consider this sup-plementaryonlinemanualtobeacommunityendeavourforwhichwegivehereastartingpoint.Userscanaddtheirownexperiencesandanalysispathwayssimplybycontributingtothesourcecodeofthemanualvia itsGitHubrepository (https://github.com/slisovski/TheGeolocationManual). In addition, we recommend the onlineforum “Geolocator Discussion & Support” hosted by “ornithologyexchange” (http://ornithologyexchange.org/forums/forum/259-geolocator-discussion-support/)asaplatformfordiscussionandtheopportunitytoexchangequestionsandanswerswiththegeolocatorusercommunity.

7  | OUTLOOK

Overthenextdecadeorso,geolocatorswilllikelyremainanimpor-tant tool inmigration research, thanks to their low cost and lightweight,whichareasyetunmatchedbyanyofthemorepreciseGPStagsorothertrackingdevices.

There are several recent technological advancements thatwillperhaps even increase their appeal. The most important amongthem isprobably thecombinationof light recordingwithotheren routedata, suchasairpressure,acceleration,magnetismandtem-perature(Bäckmanetal.,2017;Dhanjal-Adamsetal.,2018;Sjöbergetal.,2018).Theseadditionaldatacanfacilitatetherefinementoflocationestimates (seestep4 in theanalysis).For instance,accel-eration data or changes in air pressure can distinguishmovementandresidencyperiodsrelativelyeasily,andsuchpriorknowledgecan

thenbefed intothestationary locationestimation(seeBox3andChapter7;“TheGroupModel”inthesupplementaryonlinemanual).

Theseadditional sensordataalsoprovideawealthof comple-mentary information on fine-scale behaviour: air pressure andtemperature canbeused to infer flight altitude and initiation andterminationofmigratoryor foraging flights; accelerationdata canbeusedtodeterminedailyactivitybudgetsand,givensupplemen-tarymeasurements, be related to energy expenditure and energybudgets;magneticfielddatacanbeusedtoinferstrategiesinorien-tationandnavigationduringmigration.Combiningsuchbehaviouralinformationwith refined, reliable and accurate location estimatescan assist in identifying habitat associations of migrants or theirresponses to evolutionarily novel factors such as artificial light atnight,sensorypollutionorwindfarms.Thisknowledgewillimproveourunderstandingofthefatesofmigrantsandthebottleneckstheymightexperienceatsensitivetimesandplaces,whichwillallowustoimproveourconservationandmanagementstrategiesformigratorypopulations.

ACKNOWLEDG EMENTS

Wewant to acknowledge allwhohavebeen involved in thede-velopment of geolocator tools as well as all participants of themany international geolocator workshops. Furthermore, we liketo acknowledge Steffen Hahn and Felix Liechti for organizing afirstworkshopontheanalysisofgeolocatordatafromsongbirdsback in 2011, financially supported by the Swiss OrnithologicalInstitute and the Swiss National Science Foundation (Grant#IZ32Z0_135914⁄1). Initial collaboration of the authorswas alsofacilitatedby theMIGRATE research coordinationnetwork (NSFIOS-541740). The National Centre for Ecological Analysis andSynthesis, a centre funded by NSF (Grant #EF-0553768), theUniversityofCalifornia,SantaBarbara,andtheStateofCalifornia,supportedtwomeetingsin2012and2013andmanyofthetoolsthat are discussed in the supplementary online manual werekick started at these meetings. We want to thank James FoxfromMigrate Technology Ltd., the Swiss Federal Office for theEnvironment(UTF254.08.08,UTF400.34.11),theSwissNationalScience Foundation (31003A_160265), the SEATRACK project,aswellastheUSNationalScienceFoundation(EAGER0946685,IDBR-1152356, DBI-0963969 and IDR-1014891, EPSCoR-0919466)andtheDanishNationalResearchFoundationforsup-portingtheCenterforMacroecology,EvolutionandClimate(Grantno.DNRF96)forcontinuingfinancialsupportinthedevelopmentoftoolsandorganizationofworkshops.

AUTHOR ' S CONTRIBUTIONS

S.L.andS.B.wrotethefirstdraftwithsignificantcontributionsfromK.L.D.-A.,J.O.andE.S.B.Allauthorscontributedtorevisionsandap-provedthefinalversion.S.L.,K.L.D.-A.,M.T.H.,B.M.,J.K.,E.R.andE.S.B.wrote the initial draft of the supplementary onlinemanual,andallauthorsaddedandcommentedonthecurrentversion.

     |  15Journal of Animal EcologyLISOVSKI et aL.

DATA ACCE SSIBILIT Y

Code and data of online supplementarymaterial are available viaZenodo (Lisovski et al., 2019) and viaGitHub: https://github.com/slisovski/TheGeolocationManual.Theonlinesupplementarymanualisreleasedonhttps://geolocationmanual.vogelwarte.ch.

ORCID

Simeon Lisovski https://orcid.org/0000-0002-6399-0035

Silke Bauer https://orcid.org/0000-0002-0844-164X

Martins Briedis https://orcid.org/0000-0002-9434-9056

Sarah C. Davidson https://orcid.org/0000-0002-2766-9201

Kiran L. Dhanjal‐Adams https://orcid.org/0000-0002-0496-8428

Michael T. Hallworth https://orcid.org/0000-0002-6385-3815

Julia Karagicheva https://orcid.org/0000-0002-2404-6826

Christoph M. Meier https://orcid.org/0000-0001-9584-2339

Benjamin Merkel https://orcid.org/0000-0002-8637-7655

Janne Ouwehand https://orcid.org/0000-0003-2573-6287

Lykke Pedersen https://orcid.org/0000-0001-7359-6288

Eldar Rakhimberdiev https://orcid.org/0000-0001-6357-6187

Amélie Roberto‐Charron https://orcid.org/0000-0001-5917-4919

Nathaniel E. Seavy https://orcid.org/0000-0003-0292-3987

Michael D. Sumner https://orcid.org/0000-0002-2471-7511

Caz M. Taylor https://orcid.org/0000-0002-8293-1014

Simon J. Wotherspoon https://orcid.org/0000-0002-6947-4445

Eli S. Bridge https://orcid.org/0000-0003-3453-2008

R E FE R E N C E S

Bäckman, J., Andersson, A., Pedersen, L., Sjöberg, S., Tøttrup, A. P., &Alerstam,T. (2017).Actogramanalysisof free-flyingmigratorybirds:Newperspectivesbasedonaccelerationlogging.Journal of comparative physiology. A, Neuroethology, Sensory, Neural, and Behavioral Physiology,203,543–564.https://doi.org/10.1007/s00359-017-1165-9

Bracey,A.,Lisovski,S.,Moore,D.,McKellar,A.,Craig,E.,Matteson,S.,…Cuthbert,F. (2018).Migratory routesandwintering locationsofdeclininginlandNorthAmericanCommonTerns.The Auk,135,385–399.https://doi.org/10.1642/AUK-17-210.1

Bridge,E.S.,Thorup,K.,Bowlin,M.S.,Chilson,P.B.,Diehl,R.H.,Fleron,R.W.,…Wikelski,M.(2011).Technologyonthemove:Recentandforthcominginnovationsfortrackingmigratorybirds.BioScience,61,689–698.https://doi.org/10.1525/bio.2011.61.9.7

Briedis,M.,Bauer,S.,Adamík,P.,Alves,J.A.,Costa,J.S.,Emmenegger,T., …Hahn, S. (2019). A full annual perspective on sex-biasedmi-gration timing in long-distance migratory birds. Proceedings of the Royal Society B: Biological Sciences, 286, 20182821. https://doi.org/10.1098/rspb.2018.2821

Briedis,M.,Hahn,S.,Gustafsson,L.,Henshaw,I.,Träff,J.,Král,M.,&Adamík,P.(2016).Breedinglatitudeleadstodifferenttemporalbutnotspatialorganizationoftheannualcycleinalong-distancemigrant.Journal of Avian Biology,47,743–748.https://doi.org/10.1111/jav.01002

Brlík,V.,Koleček, J., Burgess,M.,Hahn, S.,Humple,D.,Krist,M.,…Procházka,P.(2019).Weakeffectsofgeolocatorsonsmallbirds:A

meta-analysiscontrolledforphylogenyandpublicationbias.Journal of Animal Ecology,1–14.https://doi.org/10.1111/1365-2656.12962

Dhanjal-Adams,K.L.,Bauer,S.,Emmenegger,T.,Hahn,S.,Lisovski,S.,&Liechti,F.(2018).Spatiotemporalgroupdynamicsinalong-distancemigratory bird. Current Biology, 28(17), 2824–2830.e3. https://doi.org/10.1016/j.cub.2018.06.054

Egevang,C.,Stenhouse, I. J.,Phillips,R.A.,Petersen,A.,Fox, J.W.,&Silk, J. R.D. (2010). TrackingofArctic terns Sternaparadisaea re-vealslongestanimalmigration.Proceedings of the National Academy of Sciences of the United States of America,107,2078–2081.https://doi.org/10.1073/pnas.0909493107

Finch, T., Butler, S. J., Franco, A.M. A., & Cresswell,W. (2017). Lowmigratory connectivity is common in long-distance migrant birds.Journal of Animal Ecology, 86, 662–673. https://doi.org/10.1111/ 1365-2656.12635

Fudickar,A.M.,Wikelski,M.,&Partecke,J. (2012).Trackingmigratorysongbirds: Accuracy of light-level loggers (geolocators) in foresthabitats. Methods in Ecology and Evolution, 3, 47–52. https://doi.org/10.1111/j.2041-210X.2011.00136.x

Hiemer,D.,Salewski,V.,Fiedler,W.,Hahn,S.,&Lisovski,S.(2018).Firsttracksof individual Blackcaps suggest a complexmigration pattern. Journal of Ornithology,159,205–210.https://doi.org/10.1007/s10336-017-1490-3

Hill,R.D.(1994).Theoryofgeolocationbylightlevels.InB.J.L.Boeuf,&R.M.Laws(Ed.),Elephant seals: Population ecology, behaviour and physiology(pp.227–236).Berkeley,CA:UniversityofCaliforniaPress.

Johnson,M.,Clarkson,P.,Goldstein,M.I.,Haig,S.M.,Lanctot,R.B.,Tessler,D.F.,&Zwiefelhofer,D.(2010).Seasonalmovements,winterrangeusemandmigratoryconnectivityoftheblackoystercatcher.Condor,112,731–743.

Knight, S. M., Bradley, D. W., Clark, R. G., Gow, E. A., Bélisle, M.,Berzins, L. L., … Norris, D. R. (2018). Constructing and evaluat-ing a continent-widemigratory songbird network across the an-nual cycle. Ecological Monographs, 88(3), 445–460. https://doi.org/10.1002/ecm.1298

Lam,C.H.,Nielsen,A.,&Sibert,J.R.(2010).Incorporatingsea-surfacetem-perature to the light-based geolocationmodel TrackIt.Marine Ecology Progress Series,419,71–84.https://doi.org/10.3354/meps08862

Lavers,J.L.,Lisovski,S.,&Bond,A.L. (2019).PreliminarysurvivalandmovementdataforadecliningpopulationofFlesh-footedShearwaterArdennacarneipes inWesternAustraliaprovides insights intoma-rinethreats.Bird Conservation International,29,327–337.https://doi.org/10.1017/S0959270918000084

Lisovski, S. (2018). Light-level geolocation in polar regions with 24-hour daylight.Wader Study, 125, 1–6. https://doi.org/10.18194/ws.00109

Lisovski, S., Bauer, S., Briedis,M., Davidson, S. C., Dhanjal-Adams, K.L.,Hallworth,M. T.,…Bridge, E. S. (2019). Light-LevelGeolocatorAnalyses: A user's guide – Supplementary online manual. Zenodo Digital Repository.https://doi.org/10.5281/zenodo.2670008

Lisovski,S.,Gosbell,K.,Hassell,C.,&Minton,C.(2016).Trackingthefullannual-cycleoftheGreatKnotCalidris tenuirostris,along-distancemi-gratoryshorebirdoftheEastAsian-AustralasianFlyway.Wader Study,123,177–189.

Lisovski, S., & Hahn, S. (2012). GeoLight – Processing and analysinglight-based geolocator data in R.Methods in Ecology and Evolution,3,1055–1059.

Lisovski, S., Hewson, C. M., Klaassen, R. H. G., Korner-Nievergelt,F., Kristensen, M. W., & Hahn, S. (2012). Geolocation bylight: Accuracy and precision affected by environmental fac-tors. Methods in Ecology and Evolution, 3, 603–612. https://doi.org/10.1111/j.2041-210X.2012.00185.x

Lisovski,S.,Sumner,M.D.,&Wotherspoon,S.J. (2015).TwGeos: Basic data processing for light based geolocation archival tags. GithubRepository,Retrievedfromhttps://github.com/slisovski/TwGeos

MacPherson,M.P.,Jahn,A.E.,Murphy,M.T.,Kim,D.H.,Cueto,V.R.,Tuero, D. T., & Hill, E. D. (2018). Follow the rain? Environmental

16  |    Journal of Animal Ecology LISOVSKI et aL.

driversofTyrannusmigrationacrosstheNewWorld.The Auk,135,881–894.

McKinnon,E.A.,&Love,O.P.(2018).Tenyearstrackingthemigrationsofsmalllandbirds:Lessonslearnedinthegoldenageofbio-logging.The Auk,135,834–856.https://doi.org/10.1642/AUK-17-202.1

Meier,C.M.,Karaardıç,H.,Aymí,R.,Peev,S.G.,Bächler,E.,Weber,R.,…Liechti,F.(2018).WhatmakesAlpineswiftascendattwilight?Novelgeolocatorsrevealyear-roundflightbehaviour.Behavioral Ecology and Sociobiology,72,45.https://doi.org/10.1007/s00265-017-2438-6

Merkel, B., Phillips, R. A., Descamps, S., Yoccoz, N. G., Moe, B., &Strøm, H. (2016). A probabilistic algorithm to process geolo-cation data. Movement Ecology, 4, 26. https://doi.org/10.1186/s40462-016-0091-8

Nielsen, A., Bigelow, K. A., Musyl, M. K., & Sibert, J. R. (2006).Improving light-based geolocation by including sea surface tem-perature. Fisheries Oceanography, 15, 314–325. https://doi.org/10.1111/j.1365-2419.2005.00401.x

Nielsen, A., & Sibert, J. R. (2007). State-space model for light-basedtrackingofmarineanimals.Canadian Journal of Fisheries and Aquatic Sciences,64,1055–1068.https://doi.org/10.1139/f07-064

Phillips,R.,Silk,J.,Croxall,J.,Afanasyev,V.,&Briggs,D.(2004).Accuracyofgeolocationestimatesforflyingseabirds.Marine Ecology Progress Series,266,265–272.https://doi.org/10.3354/meps266265

Rakhimberdiev, E., Saveliev, A., Piersma, T., & Karagicheva, J. (2017).FLightR: An r package for reconstructing animal paths from solargeolocationloggers.Methods in Ecology and Evolution,8,1482–1487.

Rakhimberdiev, E., Senner, N. R., Verhoeven, M. A., Winkler, D. W.,Bouten,W.,&Piersma,T.(2016).Comparinginferencesofsolargeo-location data against high-precisionGPS data: Annualmovementsofadouble-taggedblack-tailedgodwit.Journal of Avian Biology,47,589–596.https://doi.org/10.1111/jav.00891

Rakhimberdiev, E., Winkler, D. W., Bridge, E., Seavy, N. E., Sheldon,D., Piersma, T., & Saveliev, A. (2015). A hiddenMarkovmodel forreconstructing animal paths from solar geolocation loggers usingtemplatesforlightintensity.Movement Ecology,3,1–15.https://doi.org/10.1186/s40462-015-0062-5

Salewski,V.,Flade,M.,Poluda,A.,Kiljan,G.,Liechti,F.,Lisovski,S.,&Hahn,S.(2013).Anunknownmigrationrouteofthe“globallythreatened”AquaticWarblerrevealedbygeolocators.Journal of Ornithology,154,549–552.https://doi.org/10.1007/s10336-012-0912-5

Schmaljohann, H., Lisovski, S., & Bairlein, F. (2017). Flexible reactionnorms to environmental variables along the migration route and

the significance of stopover duration for total speed of migra-tion in a songbirdmigrant. Frontiers in Zoology,14, 17. https://doi.org/10.1186/s12983-017-0203-3

Shaffer, S. A., Tremblay, Y., Awkerman, J. A., Henry, R.W., Teo, S. L.H.,Anderson,D.J.,…Costa,D.P.(2005).Comparisonoflight-andSST-based geolocation with satellite telemetry in free-ranging al-batrosses.Marine Biology, 147, 833–843. https://doi.org/10.1007/s00227-005-1631-8

Sjöberg, S., Pedersen, L., Malmiga, G., Alerstam, T., Hansson, B.,Hasselquist,D.,…Bäckman,J.(2018).Barometerloggingrevealsnewdimensionsofindividualsongbirdmigration.Journal of Avian Biology,49,e01821.https://doi.org/10.1111/jav.01821

Stutchbury,B.J.,Tarof,S.A.,Done,T.,Gow,E.,Kramer,P.M.,Tautin,J.,…Afanasyev,V.(2009).Trackinglong-distancesongbirdmigrationbyusing geolocators.Science,323, 896. https://doi.org/10.1126/science.1166664

Sumner,M.D.,Wotherspoon,S.J.,&Hindell,M.A.(2009).Bayesianes-timationofanimalmovementfromarchivalandsatellitetags.PLoS ONE,4,e7324.https://doi.org/10.1371/journal.pone.0007324

Wilson, R. P., Ducamp, J.-J., Rees,W. G., Culik, B.M., &Niekamp, K.(1992).Estimationoflocation:Globalcoverageusinglightintensity.InI.G.Priede,&S.M.Swift(Eds.),Remote monitoring and tracking of animals(pp.131–134).England,UK:EllisHorwood.

Wotherspoon,S.J.,Sumner,D.A.,&Lisovski,S.(2013a).R Package SGAT: Solar/Satellite Geolocation for Animal Tracking. GitHub Repository,Retrievedfromhttps://github.com/SWotherspoon/SGAT

Wotherspoon, S. J., Sumner, D. A., & Lisovski, S. (2013b). R Package BAStag: Basic data processing for light based geolocation archival tags. GitHubRepository,Retrievedfromhttps://github.com/SWotherspoon/BAStag

Yamaura, Y., Schmaljohann, H., Lisovski, S., Senzaki, M., Kawamura,K., Fujimaki, Y., & Nakamura, F. (2017). Tracking the Stejneger'sstonechatSaxicola stejnegerialongtheEastAsian-AustralianFlywayfromJapanviaChinatosoutheastAsia.Journal of Avian Biology,48,197–202.

How to cite this article:LisovskiS,BauerS,BriedisM,etal.Light-levelgeolocatoranalyses:Auser'sguide.J Anim Ecol. 2019;00:1–16. https://doi.org/10.1111/1365-2656.13036

Graphical AbstractThecontentsofthispagewillbeusedaspartofthegraphicalabstractofhtmlonly.Itwillnotbepublishedaspartofmainarticle.

Overthepastdecade,light-levelgeolocatortagshaveemergedasanimportanttoolfortrackinganimals.However,duetovariouscaveats,thetranslationoflightrecordingsintolocationestimatescanbedaunting.Inthis'Howto...'paper,weprovideapracticalguidefortheanalysisofgeolocatordatabydescribinggoodpracticesanddiscussingcommonpitfalls.