ContentslistsavailableatScienceDirect
Journal
of
Neuroscience
Methods
j ou rn a l h o m epa g e :w w w . e l s e v i e r . c o m / l o c a t e / j n e u m e t h
Research
article
Evaluation
of
the
RumiWatchSystem
for
measuring
grazing
behaviour
of
cows
J.
Werner
a,b,∗,
L.
Leso
a,c,
C.
Umstatter
d,
J.
Niederhauser
e,
E.
Kennedy
a,
A.
Geoghegan
a,
L.
Shalloo
a,
M.
Schick
d,
B.
O’Brien
aaTeagasc,Animal&GrasslandResearchandInnovationCentre,Moorepark,Fermoy,Co.Cork,Ireland bUniversityofHohenheim,InstituteforAgriculturalEngineering,70599Stuttgart,Germany cUniversityofFlorence,DepartmentofAgricultural,FoodandForestrySystems,50145Firenze,Italy dAgroscope,ResearchDivisionCompetitivenessandSystemEvaluation,8356Ettenhausen,Switzerland eInnoCleverGmbH,Tiergartenstrasse7,4410Liestal,Switzerland
h
i
g
h
l
i
g
h
t
s
•TheRumiWatchSystemcansuccessfullymonitorbehaviourofgrazingcows.
•HighcorrelationbetweenRumiWatchSystemandvisualobservationindeterminingtimebudgetfordifferentcowactivities. •HighcorrelationbetweenRumiWatchSystemandvisualobservationindetermininggrazingbitesandruminationchews. •TheRumiWatchSystemwasproventobeanaccurateresearchtoolformeasuringdetailedgrazingbehaviourofcows.
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received31January2017
Receivedinrevisedform15August2017 Accepted16August2017
Availableonline24August2017
Keywords: Validation Sensortechnology Monitoringbehavior Boutcriteria Grazingbites Grazingtime
a
b
s
t
r
a
c
t
Feedingbehaviourisanimportantparameterofanimalperformance,healthandwelfare,aswellas reflectinglevelsandqualityoffeedavailable.Previously,sensorswereonlyusedformeasuringanimal feedingbehaviourinindoorhousingsystems.However,sensorssuchastheRumiWatchSystemcanalso monitorsuchbehaviourcontinuouslyinpasture-basedenvironments.Therefore,theaimofthisstudy wastovalidatetheRumiWatchSystemtorecordcowactivityandfeedingbehaviourinapasture-based system.TheRumiWatchSystemwasevaluatedagainstvisualobservationacrosstwodifferent experi-ments.Thetimedurationperhouratgrazing,rumination,walking,standingandlyingrecordedbythe RumiWatchSystemwascomparedtothevisualobservationdatainExperiment1.Concordance Correla-tionCoefficient(CCC)valuesofCCC=0.96forgrazing,CCC=0.99forrumination,CCC=1.00forstanding andlyingandCCC=0.92forwalkingwereobtained.Thenumberofgrazingandruminationboutswithin onehourwerealsoanalysedresultinginCohen‘sKappa()=0.62and=0.86forgrazingand rumina-tionbouts,respectively.Experiment2focusedonthevalidationofgrazingbitesandruminationchews. TheaccordancebetweenvisualobservationandautomatedmeasurementbytheRumiWatchSystemwas highwithCCC=0.78andCCC=0.94forgrazingbitesandruminationchews,respectively.Theseresults indicatethattheRumiWatchSystemisareliablesensortechnologyforobservingcowactivityandfeeding behaviourinapasturebasedmilkproductionsystem,andmaybeusedforresearchpurposesinagrazing environment.
©2017TheAuthors.PublishedbyElsevierB.V.ThisisanopenaccessarticleundertheCCBYlicense (http://creativecommons.org/licenses/by/4.0/).
1. Introduction
Withincreasingscaleonfarms,anddecliningavailablelabour, thereisarequirementfortechnologiesthatassistfarmersintheir
∗ Correspondingauthorat:Animal&GrasslandResearchandInnovationCentre, Moorepark,Fermoy,Co.Cork,Ireland.
E-mailaddress:jessicaanna.werner@teagasc.ie(J.Werner).
daytodaymanagement.Animalmanagementinvolvesensuring thehealthandwelfareoftheanimals;reactingtocertaineventsin theanimalreproductivecycleandimprovingefficiencyinfeed pro-visionforconversionintoananimalproduct,suchasmilkormeat. Especiallyinapasturebasedsystemthebalancebetweenthefeed offeredandtheherddemandneedstobeoptimizedtomaximise grassutilisationwhilesimultaneouslyensuringthatanimalsare wellfedatpasture.Shortageinlabourandtimetoobserveanimals makes it difficult for farmers tomonitor all animals intensely. https://doi.org/10.1016/j.jneumeth.2017.08.022
Automated monitoring for quantifying physiological and behavioural parameters, e.g. oestrus, somatic cell count and feeding behaviour, can give an insight into overall health sta-tus, important animal events as well as helping with feeding management.Foracontinuousmonitoringofthesephysiological and behaviouralparameters sensor-based, easy-to-usetools for farmersneedtobedeveloped.
Oneofthebestindicatorsofhealthandwelfareofdairycowsis feedingbehaviour.AstudybyBareilleetal.(2003)showedthatfeed intakewasinfluencedbyanumberofdifferentdiseasessuchasmilk fever,ketosisorhooflesions.Thereisabenefittodetect emerg-ingdiseasesearlierbymonitoringthefeedingbehaviourofdairy cowsautomatically.Previousresearchhasshownthatadeclinein ruminationtimecanbeusedasareliablepredictorofbothhealth andfertilityeventsandisalsomentionedtobeanindicatorfor cowstress(Herskinetal.,2004).Feedingbehaviourcanalsobe usedtooptimise grasslandmanagement decisionswitha focus onincreasinganimalintakeandreducinggrassresiduals.Itisof keyimportancetomeasure,manageandallocateaccuratelythe feedavailableandofferedtothecows,irrespectiveofthefarming systeminordertooptimisefarmefficiencyandprofitability.The estimationoffeedintakebasedonbehaviouralparameters,such asfeedingtimeorbitefrequency,providesvaluableinformation thatcanbeusedtomanagecows.Pahletal.(2015)conducteda studyinanindoorfeedingsystemtocomparefeedingandchewing timewithmeasuredintakedataobtainedbyweighingofthe feed-ingtroughs.Theyconcludedthatitwasappropriatetousefeeding behaviourforestimatingintakeinbarnsystems.
Somemethodshavebeendevelopedtopredictintakein grass-basedsystems(Undietal.,2008), suchastheN-alkane-method (Dillon and Stakelum, 1988).This methoddetermines the feed intake by the usage of an orally applied bolus with synthetic faecesmarker.These measurementsare labour-intensive, time-consumingandinvasive,asthecowshavetobedosedorallytwice adayovera2-weekperiod.Analternativeapproachusedto deter-minefeedintakewastheIGERanimalrecordingsystem(Mezzalira etal.,2014).Thisconsistedofanosebandsensorthatmeasured jawmovementbyelectricalresistance(Rutteretal.,1997).Itcould identifyandmeasuregrazingandrumination.However,the max-imumrecordingperiodofthissystemwas24h,andtheanalysis ofthedataviathe“Graze software”wasverylaborious (Rutter, 2000).Furthermore,thedistributionandcommercialsupportfor thistechnologyhasceasedinrecentyears.Butanewtechnology, theRumiWatchSystemmayhavethepotentialtoimprovedata captureandreplacetheIGERanimalrecordingsystem.
TheRumiWatchSystem wasinitially developed byNydegger andBollhalder(2010)attheSwissFederalResearchInstitute (Agro-scope, Tänikon, Switzerland) for behavioural measurements on cowsfedindoorsandiscommerciallydistributedbythecompany (Itin+HochGmbH,Liestal,Switzerland)since2010.Itiswell estab-lishedasasensortechnologyinindoorhousingsystems(Ruuska etal.,2016)andhasundergoneanumberofmodificationsin devel-opmentasaresearchandadvisorytool(Zehneretal.,2012).Most ofthemodificationswereconductedtooptimizetheanalysis soft-ware,theRumiWatchConverter,basedondevelopmentofdifferent algorithmstoanalysetherawdata.InastudyofZehneretal.(2017) differentversionsoftheRumiWatchConverterwerevalidatedin indoorhousingsystemsagainstvisualobservation.Further modi-ficationsofappliedalgorithmsintheRumiWatchConverter,such asthedefinitionofeachgrazingbite,wereconductedtorecord grazingbehaviourasfeedingbehaviourdiffersinindoorhousing systemsandpasture-basedsystems.Asitisabsolutelycriticalthat anyanimalbehavioursensoroperatescorrectlyinmonitoringthe appropriateparameters, theobjectiveofthis studywasto vali-dateanupdated andadapted versionof theRumiWatchSystem forthemeasurementofgrazingbehaviourinapasture-basedmilk
productionsystem.Twoseparateexperimentswereconductedto validateparameterssuchasgrazing,rumination,walking,standing andlyingtime,aswellasgrazingbitesandruminationchews. 2. Materialandmethods
ValidationoftheRumiWatchSystemwasconductedintwo sep-arateexperimentswithindividualcowherdsatTeagasc,Animal andGrasslandResearchandInnovationCentre(Fermoy,Co.Cork, Ireland,50◦07N;8◦16W).Experiments1and2tookplaceinthe periodsof10thto19thofMay2016and31stofMayto2ndofJune 2016,respectively. The experimentalgrazing areas represented permanentgrasslandwith70%perennialryegrassand30%annual meadowgrass.Thisstudywaspartofalargerstudywhere differ-entlevelsoffeedallowancewereallocatedtodairycowsacross differentperiodsofthelactationandfordifferentdurations. 2.1. Sensortechnology
TheRumiWatchSystemconsistsoftwoseparatedeviceswith associated software packages for managing the sensors (Rumi-WatchManager)andanalysingdata(RumiWatchConverter).
TheRumiWatchhalterincorporatesanosebandpressure sen-sor,a3-axisaccelerometeraswellasadatalogger.Thenoseband sensorcomprisesanoil-filledtubewithabuilt-inpressuresensor. Thepressureinsidethetubechangeswithalternationdueto chew-ingactivitiesandisrecordedina10Hzresolution.Thoserawdata aresavedonanintegrated4GBSD-cardforupto4months,which isimplementedtogetherwiththedataloggerinaprotectiveboxon therightsideofthehalter.Ontheleftside,thereisapowersupply withtwo3.6Vbatteriesintegratedinasimilarlyconstructedbox. Basedondifferentpressuresignaturesofthenosebandpressure sensor,theRumiWatchConverterisabletodetectjawmovements andclassifiesthembasedonfrequencyandrhythmtogetherwith accelerationpatternsintograzingbites/chews,ruminationchews oranyotheractivity.Additionally,thetimedurationofthose differ-entclassificationsisrecorded.Furtherinformationabouttechnical componentscanbefoundinWerneretal.(2016),Zehneretal. (2012)andZehneretal.(2017).
TheRumiWatchpedometerconsistsofaprotectiveplasticbox witha3-axisaccelerometerintegrated.Similartothehalter,there isanSD-Cardandadataloggerconnectedwiththe23.6Vbatteries builtintothebox.Accelerationrawdataforthex-,y-andz-axisare recordedina10Hzresolution.TheRumiWatchConverterisableto classifytheaccelerationrawdatabyapplyingspecificalgorithms intostanding,walking,lyingaswellasamountofstrides.Further informationaboutthedevelopmentandvalidationcanbefoundin Alsaaodetal.(2015).
TheRumiWatchManager2(V.2.1.0.0)andtheRumiWatch Con-verter(V.0.7.3.36)wereusedforExperiments1and2ofthecurrent study.Thereweretwodifferentapproachesfortimeresolutions. The 1-min summary data were categorized into differentfocal behaviourclassificationsbytheRumiWatchConverter.The Con-verteralsocreatednumericalvaluesbasedonacalculationofthe timeduration(minperperiod)offocalbehavioursineachofthe definedtimeresolutions,e.g.5-min,10min,1hsummaries.
The RumiWatch Converter V.0.7.3.36 used three different parameterstomonitorandcalculategrazingtime.Twoparameters consideredinthisstudywereusedtocalculategrazingtime.EAT1 determinedgrazingwithheadpositiondown,EAT2determined grazingwithheadpositionup.Furthermore,therewere parame-tersforgrazingandruminationboutbehaviourintegratedinthe RumiWatchConverterV.0.7.3.36withgrazingboutsand rumina-tionbouts.Agrazingboutwasdefinedasanevent,wheregrazing wasdetectedforaminimumdurationof7minandtheinter-bout
Table1
Experimentalprotocolfordatacollectionofcowbehaviourbyvisualobservation.
Day Time Observer1 Observer2
Cownumbers Cownumbers
1 09:00–11:00 1,2,3 4,5,6 12:00–14:00 4,5,6 1,2,3 17:00–19:00 1,2,3 4,5,6 2 09:00–11:00 7,8,9 10,11,12 12:00–14:00 10,11,12 7,8,9 17:00–19:00 7,8,9 10,11,12 3 05:00–07:00 4,5,6 1,2,3 11:00–13:00 1,2,3 4,5,6 19:00–21:00 4,5,6 1,2,3 4 05:00–07:00 10,11,12 7,8,9 11:00–13:00 7,8,9 10,11,12 19:00–21:00 10,11,12 7,8,9
interval wasdefined witha minimumthreshold of 7min.That meant,thatifacowwasnotshowinganygrazingbehaviourfor ≤7minorshecommencedtoruminate,thedetectionofthe graz-ingboutwasstopped.Thesedefinitionsweresimilartothoseused inastudybyPérez-Ramírezetal.(2009)afterBrunetal.(1984).A secondboutparameter,aruminationboutwasdefinedashavinga minimumdurationof3minandaninter-boutintervalwitha mini-mumthresholdof1min.ThestudyofWolfgeretal.(2015)applied similarcriteria.
2.2. Experiment1
2.2.1. Animalsandtreatments
Agroupoftwelvespringcalvingdairycowsfromaherdof15 inapasture-based milkproductionsystemwasused.Thecows wereonaverage91±12daysinmilkatthestart ofthe experi-ment.ThegroupconsistedofanequalnumberofHolstein-Friesian (HF)andJerseycrossbred(JEX)cowswithfourprimiparous1and eightmultiparouscows2,ranginginlactationnumberfrom2to 6.Themeanbodyweightwas477±65kgandtheaveragebody conditionscore(BCS)was2.8±0.2rangingfrom1to5,measured undertheEdmonsonetal.(1989)scoringsystem.Theaveragemilk yieldovertheexperimentalperiodwas22.5±4.5kg/cow/day.All cowsfollowedasimilarmilkingroutine.Cowsweremilkedtwice aday(6:30hand14:30h)andspentapproximately1.5−2.0hper milkingawayfromthepaddock.
Cowswereonagrassonlydietandreceivedagrassallocation twicedailyaftermilking.Pre-andpost-grazingswardheightwas measureddailywitharisingplatemeter(diameter355mmand 3.2kg/m2;Jenquip,Fielding,NewZealand).Pre-grazingheightof grassaveraged11.9±2.5cm,whilepost-grazingheightaveraged 4.5±0.8cmduringtheexperimentalperiod.Theaverage experi-mentalpaddockareaovertheexperimentalperiodwas0.15ha.All cowswereidentifiablebynumberspaintedontheirsides(1–12). 2.2.2. Experimentaldesign
Cowbehaviouraldatawascollectedbyvisualobservationand byautomatedrecordingoftheRumiWatchSystem.Twopreviously trainedobserverswereusedtomonitorthecows.Thecowgroup
wasdividedinto4subgroupsof3cowseach.Eachsubgroupwas observedbyeach observeron3occasionsperdayover a4day period(Table1).Observationstookplaceover2-hperiodsbetween dawn(05:00) anddusk (21:00)excluding milkingtimes,which extendedfrom7:00to9:00and14:00to17:00h.
Visualobservationwasperformedby1-minscansamplingas usedinthestudyofBüchelandSundrum(2014).Behaviouraldata wascategorizedasfeedingbehaviour(FB)andactivitybehaviour (AB).Feedingbehaviourwasfurtherclassifiedasgrazing, ruminat-ingandotheractivities,whileactivitybehaviourwasclassifiedas standing,walkingandlying(Beaucheminetal.,1989).Thedifferent behaviourclassificationsaredescribedinTable2.Thedatawere recordedonamanualspreadsheet andweretransferred manu-allytoanelectronicspreadsheet(Microsoft ExcelVersion2010; MicrosoftCorporation,Redmond,USA)foranalysis.
Withregardtotheautomateddatacollection,the RumiWatch-Systemcomprisedoftwodevices,ahalterplacedontheheadand a pedometerplacedonthelefthind leg.TheRumiWatch Man-ager2wasusedtosynchronizebothRumiWatchdevicestoUTC (UniversalTimeCoordinated)atthebeginningoftheexperiment. 2.2.3. Datapreparation
Therewere144h(8640min)ofvalidobservations.Thisvisually recordeddatawereassembledinthreeways.Firstlythedatawas assignedtotheappropriateclassificationwithinthecategoriesof FBandABat1-minintervals.Secondly,thetimedurationofthe specificclassificationsofFBandABweretotalledfor1-hintervals. Finally,thenumberofstartedgrazingboutsandfinishedgrazing boutsandthenumberofstartedruminationboutsandfinished ruminationboutswerecalculatedforeach1-hperiod.
Theautomaticallycaptureddataofthehalterandpedometer weredownloadedusingtheRumiWatchManager2.Forthis,the microUSB-portintegratedinthesensorswasconnectedviaa USB-cabletothecomputerandtherawdataweretransferred.Further, theRumiWatchConverter(V.0.7.3.36)wasusedtoconverttheraw datainto1-minand1-hsummaries.Thosesummariescomprised comparableparameterstothoseoutlinedaboveforthevisualdata (classificationat 1-minintervals,duration of specificbehaviour classification perhour and numbers ofgrazing and rumination bouts).Fortheanalysisofgrazingtime,thetotalofEAT1TIMEand EAT2TIMEwascalculated.
2.3. Experiment2
2.3.1. Animalsandtreatments
TheobjectiveofthistrialwastovalidatetheRumiWatch nose-bandsensorforthenumberofgrazingbitesandruminationchews. Agroupoftwelvespringcalvingdairycowswerefittedwith Rumi-Watchnosebandsensors.Allcowswereidentifiablebynumbers (1–12)paintedonbothsidesofeachcow.Theexperimentextended overa2-day-periodwithanadjustmentperiodtothenoseband sensorof8dayspriortothestartingoftheexperiment.Twocow breedswereincludedwith7Holstein-Friesian(HF)and5Jersey crossbreds(JEX).Thegroupconsistedof3primiparousand9 mul-tiparouscows,rangingfrom2to6parities.Averagecowmilkyield
Table2
Definitionofdifferentbehaviourcategoriesforobservers,adaptedfromBikkeretal.(2014)andAlsaaodetal.(2015). Behaviour Definition
Feedingbehaviour
Grazing Cows’muzzleislocatednearorabovethegrassandmakesbitingmotiontoingestgrass,orcow’s headpositionupandmakingchewingmotion
Ruminating Regurgitation,chewing,salivationandswallowingofingestedgrass
Otheractivities Anyothermovementsofthemuzzle,whicharenotassociatedwithgrassintake Activitybehaviour
Standing Cowisinanuprightpositionbutisnotwalking
Walking Cowtakesatleast3consecutivestridesinthesamedirection(forwardorbackward) Lying Cowisrestingontheground(notstanding)
Table3
Experimentalprotocolfordatacollectiononcowbehaviour(ruminationchewsandgrazingbites),byvisualobservation.Observedcowsweremonitoredin5-minperiods inthestatedorderascendingordescendingbyeachobserver.
Day Time Observer1 Observer2 Observer3 Observer4
CowNo. CowNo. CowNo. CowNo.
1 09:30–11:30 17→→612 71→→612 126→→17 612→→17 12.00–14.00 17→→612 71→→612 126→→17 612→→17 17:00–19:00 1→6 7→12 6→1 12→7 7→12 1→6 12→7 6→1 2 04:30–06:30 17→→612 71→→612 126→→17 612→→17 11:00–13:00 17→→612 71→→612 126→→17 612→→17 19:00–21:00 1→6 7→12 6→1 12→7 7→12 1→6 12→7 6→1
overtheexperimentalperiodwas20.8±4.7kg/cow/day.Cowshad anaveragebodyweightof496±69kgandaBCSof2.9±0.2.The cowswereonagrassonlydietandweremaintainedinthesame paddockthroughouttheexperimentalperiod.Theyhadadlibitum accesstowater,andfreshgrasswasprovidedoncedailydirectly after morning milking. Milking times were twice daily (6:30h and 14:30h) during which thecows spend 1.5-2.0hper milk-ingawayfromthepaddock.Pre-andpost-grazingswardheight wasmeasuredwithanautomatedrisingplatemeterwhich isa grass measuringdevice(Grasshopper, TrueNorth Technologies, Shannon,Co.Clare,Ireland;(McSweeneyetal.,2015))onone occa-sionduringtheexperiment.Averagepre-grazingandpost-grazing heightwere14.7cmand6.1cm,respectively,onanexperimental areaof0.36ha.
2.3.2. Experimentaldesign
Theaccordancebetweenallfourobserverswasmeasuredover twodaysin24 5-minperiods.Fourpreviously trainedobservers thenmonitoredonecowperobserverfor5-minperiodstovalidate the number of grazing bitesand ruminationchews. The num-berofgrazingbitesandruminationchewswasrecordedusinga handheldcomputerwithaspeciallyprogrammedapplication.This applicationwasprogrammedtorecordeachgrazingbiteor rumi-nationchewbypressingabuttononthehandheldcomputer.Each bite/chewinthe5-minmeasurementperiodwasthensummarized intheoutputtothetotalamountofbites/chewsper5min.Agrazing bitewasdefinedasacombinationofjaw,tongueandneck move-menttoripgrassaccompaniedbyagrassbitingsound(Baileyetal., 1996).Ruminationchewswerecountedafterregurgitation took placeandabolustravelledthroughtheoesophagustoreachthe mouth(Schirmannetal.,2009).AsinExperiment1,the observa-tionalperiodswereextendedto2-hperiodsand occurredthree timesaday(Table3).Allobservationstookplacebetween04:30 and21:00h.Every5-minobservationperiodwasalternatedwitha 5-minbreakperiod.Cowswererotatedacrossobserversforevery 5-minobservationperiod,suchthattheobservermonitoredeach ofthe12cowsduringeach2-hourobservationperiod.
2.3.3. Datapreparation
Thedatarecordedonthehandheldcomputerincludedthe num-berofvisuallyobservedruminationchewsandgrazingbites.This datawassubsequentlytotalledwithinaspreadsheetapplication. Intotal249observationperiodswereanalysed,whichhadbeen collectedoverthe2-dayperiod.Theobservationperiodsconsisted of181periodsofgrazingbiteobservationsand71periodsof rumi-nationchewobservations.Datawereexcluded,whentwodifferent behaviourtypes(grazingbitesandruminationchews)were moni-toredwithinone5-minperiod.Periodswhereneithergrazingbites
norruminationchewswereobserved,werealsoexcludedfromthe dataset.
Withregardtothedataautomaticallyrecordedbythe Rumi-Watch noseband sensor, these data were classified into the categoriesofgrazingbitesandruminationchews.Thedatawere thenconvertedbytheRumiWatchConverter(V.0.7.3.36)into 5-minintervals,summarizingthenumbersofruminationchewsor grazingbites.Foranalysingruminationchews,thealgorithmswith theplausibilitycheckfunctionforminimumdurationof3minwere turnedoff.
2.4. Statisticalanalysis
StatisticalanalysiswasperformedusingRversion3.3.1(R Foun-dationforStatisticalComputing,Vienna,Austria).Thefollowing analyseswerecarriedouttoassessagreementbetweenthe Rumi-WatchSystemandvisualobservations,dependingonthetypeof datarecordedatdifferenttimeperiods.
InExperiment1, Cohen’sKappa()wascalculatedtoassess agreementbetween RumiWatchSystemand visualobservations whenFBandABwererecordedat1-minresolution(Cohen,1960). The–valueswereinterpretedinasimilarmannerasbyLandis andKoch(1977),where:poor:<0.00,slight:=0.00–0.20,fair: =0.21–0.40,moderate:=0.41–0.60,substantial:=0.61–0.80, andalmostperfect:=0.81–1.00.
Thepercentageagreement(PA)ofthe1-minresolutiondatafor bothcategoriesofFBandABandspecificclassifications,e.g.grazing, rumination,standing,walking,etc.recordedbyvisualobservation andRumiWatchSystemwascomputedusingthefollowingformula, usedbyMartinetal.(1993):
PA[%]= total numbers of agreement
total numbers of agreement+total numbers ofdisagreement×100
Anumberoftestswereconductedonthevariableswithnumeric valuestoassessagreementbetweendataoftheRumiWatchSystem andvisualobservations.ThebehaviouraldatafromtheRumiWatch noseband sensor and from thepedometer were subjectedto a graphicalanalysisinaBland-Altman-PlotandaSpearman’sRank correlation(rs)andaconcordancecorrelationcoefficient(CCC)was calculated,usingtheU-statistics(Carrascoetal.,2007). Interpreta-tionofrs-valuesandCCCwerebasedoncriteriadefinedbyHinkle etal.(2003)asfollows:Negligible=0.0–0.3,low=0.3–0.5, moder-ate=0.5–0.7,high=0.7–0.9andveryhigh=0.9–1.00.
TheBland-Altman-Plotsdemonstratedtheagreementbetween bothmeasurementmethods.Thiswasconductedbyplottingthe differences(automatedmeasurement−visualobservation)against the means of visual observationand automated measurement. TheBland-Altman-analysisindicatedthemeandifferences(bias), betweenthepairedautomaticallyrecordedandvisuallyobserved
values,with95%-confidenceintervals (CI). It alsodisplayedthe lower and upper limits of agreement along with their relative 95%-confidenceintervals.Thelimitsofagreementwerecalculated as±1.96*standarddeviationfrommeandifference.Althoughthe parameterswerenotnormallydistributed,theBland-Altman-plots wereused,asthedifferencesbetweenthepairedvaluesfollowed anormaldistribution.Inaddition,amethodtodetermine signifi-cantdifferencesbetweentwomeasurementsbyGiavarina(2015) wasalsoappliedontheBland-Altman-analysis.Thebias(ormean difference)wasconsideredtobesignificantwhenthelineof equal-itywasnotwithinthe95%CIofthemeandifference.Therefore,a significantunder-oroverestimationwasdeclaredwhenthelineof equalitywasnotincludedinthe95%CIofthemeandifference.
Forthevalidationofbouts(grazingandrumination),therewere valuesbetween0and2forboutsstartedorfinishedwithineach 1-hperiodmeasured.Therefore,thesevaluesweretreatedas ordi-nalvariables.Theagreementforgrazingandruminatingboutswas assessedusingweighedkappastatisticsandpercentageagreement, asexplainedabove.
In Experiment 2 grazing bites and rumination chews were countedbyhumanobservers during 5-minperiods. Agreement betweeneverypairedobserverwasevaluatedusingCCCandamong allobserversviaoverallCCC.Numberofgrazingbitesand rumi-nation chews per 5-min period were analysed using the same methods for numeric variables as described in Experiment 1, includingSpearman’s Rankcorrelation, CCC and Bland-Altman-Analysis.
3. Results 3.1. Experiment1
Thecomparisonofcategoricaldataofthenosebandsensorand ofthepedometerisshowninTable4.TheCohen’sKappavalue was=0.84forthevisualfeedingbehaviourmeasurements com-paredwiththenosebandsensorand=0.89forthevisualactivity measurementscomparedtothepedometer.Usingan interpreta-tionofLandisandKoch(1977),theseresultsindicateanalmost perfectagreementofvisualandautomaticallyrecordeddataona 1-minresolution.Thisresultissupportedbytheoverallagreement of91.1%forthenosebandsensorand95%forthepedometer.
Theevaluationof1-hsummariesoffeedingbehaviourmeasured bythevisualobservationandbythenosebandsensorispresented inFig.1andTable5.Grazingwasdetectedbyvisualobservation ashavingoccurredfor40.5min/h(median), whilegrazing time detectedbytheautomated sensorsystemwasrecorded witha medianof47min/hour.Aslightoverestimationoftheautomated systemingrazingminperhourisdisplayedinFig.1(a).According toBland-Altman-Statisticsthemeandifferencewas4.41min/hour, andthisoverestimationisshownasthesolidlineinFig.1(a).The correlationofrs=0.96andaCCC=0.96isclassifiedasveryhighfor determinegrazingtime.
Thecomparisonofruminationtimemeasuredbyvisual obser-vationandby theautomated methodisshown inFig.1(b) and Table5.Thecorrelationofrs=0.98andaCCC=0.99isveryhigh. Theautomatedsystemrecordedarangeofmeasuredminof rumi-nationbetween0and59,withamedianof0.Alternatively,the observersrecordedarangefrom0to57minwithamedianof2min ruminationperhour.InTable5theanalysisofthe Bland-Altman-Plotispresented. Thebiasof0.03min/hour alongwiththe95% CIof−3.04and3.10demonstratedaperfectagreementbetween theautomatedsystemandtheobservers.Themeanofallvalues wascompletelyaccurateand95%ofallrecordedvaluesdistributed themselvesinadifferencerangeof±3min/hour.
Fig.1.AgreementofautomatedRumiWatchnosebandsensormeasurementsand visualobservationsoffeedingbehaviour(a)grazingand(b)ruminationtimein1-h periods,displayedinBland-Altman-plots(solidlineindicatesthemeandifference; dashedlinesindicateupperandlower95%limitsofagreement).
Thetotalagreementformeasuringstartedandfinishedgrazing boutswithin1-hperiodswasPA=84.7% andPA=85.4%, respec-tively,whereastheagreementforruminationboutsstartedand finishedwasPA=93.1%and PA=93.8%.Theweighedkappa val-uesshowedamoderateagreementbetweenvisualandautomated measurement with values of =0.62 and =0.66 for grazing bouts started and finished, respectively. The rumination bouts showedimprovedperformance,withanalmostperfectagreement of=0.86forboutsstartedandboutsfinished.
Thevalidationofthepedometerintermsofmeasuringactivity behaviourinapasture-basedsystemisshowninFig.2.Standing timeandlyingtimeweredeterminedaccuratelybythesensor sys-temwitha correlationofrs=0.99,while walkingtimewasless accuratelydeterminedbythesensorwithacorrelationofrs=0.77. Therewerelessminutesperhourofwalkingdetectedin compari-sontostandingandlying.Themedianofwalkingtimewas1minfor visualobservationandrangedfrom0to18minperhourwhereas theautomatedsystemrecorded2minperhour,rangingfrom0to 17minperhour.Grazingtimewassignificantlyoverestimatedand timeatotheractivitieswassignificantlyunderestimated. Signifi-cantdifferencesofbiaswerealsoobservedbetweenthepedometer andvisualrecordingsinstandingtimeandwalkingtime(Table5). 3.2. Experiment2
Inthis experimenttheaccuracyof theRumiWatchSystemin measuringgrazing bites and rumination chews was examined. Thedegreeofagreementbetweenobserverswasanalysedinitially
Table4
Cohen’sKappa()andpercentageagreementbetweenvisualobservationsandautomatedmeasurementsbyRumiWatchforfeedingandactivitybehaviourona1-min resolution.
Category Cohen’s Agreementbetweenvisualand automatedmeasurement(%)
Classification Agreementbetweenvisualand automatedmeasurement(%) Feedingbehaviour(Grazing,
Ruminating,OtherActivities) 0.84 91.1
Grazing 91.5
Ruminating 94.3 OtherActivities 81.4 Activitybehaviour(Standing,
Walking,Lying) 0.89 95.0
Standing 96.3
Walking 95.4
Lying 98.7
Table5
Spearman’srho(rs),ConcordanceCorrelationCoefficient(CCC),andBland-Altman-analysis(Bias,upperandlower95%limitsofagreementwith95%CI)ofautomated
measurementsversusvisualobservationsina1-hresolutionfordifferentbehaviourclassifications.
Behaviourtimeinmin/h rs CCC Bias(95%CI) Lower(95%CI) Upper(95%CI)
Grazing 0.96 0.96 4.41 −4.69 13.51 (3.64;5.17) (−6.02;−3.37) (12.19;14.84) Rumination 0.98 0.99 −0.03 * −3.10 3.10 (−0.29;0.22) (−3.55;−2.66) (2.59;3.48) Otheractivities 0.91 0.90 −4.38(−5.12;−3.63) −13.20(−14.49;−11.92) 4.45(3.17;5.74) Standing 0.97 1.00 −0.69(−0.92;−0.47) −3.35(−3.74;−2.96) 1.96 (1.57;2.35) Lying 0.99 1.00 0.05*(−0.07; −1.35 1.45 0.17) (−1.55;−1.15) (1.24;1.65) Walking 0.78 0.92 0.85(0.63;1.06) −1.71(−2.08;−1.33) 3.40(3.02;3.77) *=nosignificantover-estimationorunder-estimationbetweenautomatedsystemandvisualobservation
Fig.2. AgreementofautomatedRumiWatchpedometermeasurementsandvisualobservationsforactivitybehaviour(a)standing;(b)walking;(c)lyingin1-hperiods displayedinBland-Altman-plots(solidlineindicatesthemeandifference;dashedlinesindicateupperandlower95%limitsofagreement).
(Table6).TheCCC-valuesdeterminedforgrazingbitesand rumi-nationchewswereCCC=0.98andCCC=1.00respectively,which demonstratedaveryhighagreementbetweenallfourobservers.
The comparison between the automated system and visual observationsinmeasuringgrazingbitesarepresentedinFig.3a) andTable7.Thevisuallycountedgrazingbitesrangedfrom0to
387per5-minperiod,withamedianof232bites.However,the RumiWatchSystemrecordedgrazingbitesbetween0and419,with a median of280bites.The Bland-Altman-Plot showedthat the automatedmeasurement slightlyoverestimated thenumbersof grazingbites.Abiasof36grazingbites/5min,withalower95% limitofagreementof−66grazingbites/5minandanupper95%
Table6
Concordance Correlation Coefficient (CCC) for each observer paired with all observersandoverallCCCbetweenallfourobserversinmeasuringgrazingbites andruminationchews.
Behaviour [n/5min]
CCC Overall
CCC Observer1 Observer2 Observer3 Observer4 Grazing bites 1 0.97 0.97 0.99 0.98 2 0.99 1.00 3 0.99 Rumination chews 1 1.00 1.00 0.99 1.00 2 0.99 0.99 3 0.98
Fig.3.AgreementofautomatedRumiWatchnosebandsensormeasurementsand visualobservationsfor(a)grazingbitesand(b)ruminationchews,in5-minperiod, displayedinBland-Altman-plots(solidlineindicatesthemeandifference;dashed linesindicateupperandlower95%limitsofagreement).
limitofagreementof138grazingbites/5minconfirmedthe sig-nificantoverestimationofgrazingbitesbytheRumiWatchSystem. Overall,theagreementforgrazingbitesbetweenthetwo measure-mentapproachesforgrazingbiteswashigh,withrs=0.81anda CCC=0.78(Table7).
Theevaluationofthemeasuredruminationchewsbyvisualand automatedrecordingsdemonstratedpositiveresultsasshownin Fig.3(b).Thevisuallycountedruminationchewsrangedfrom2 to386chews/5-minperiodwithamedianof323chews. Alterna-tively,theRumiWatchSystemrecordedamedianof330rumination chews/5min.The agreement between observer and automated systemishighercomparedtothatforgrazingbites,witha correla-tionofr=0.81andaCCC=0.94.TheBland-Altman-plotofFig.3(b) alsoillustratedgraphically,thatthemeandifferencebetweenboth measurementmethodsisverysmallwithavalueof7.24 rumina-tionchewsper5-minperiod.Limits ofagreement(dashedlines inFig.3(b)indicatethatthemeandifferencesbetweenautomated measurementandvisualobservationliebetween−51.44and65.92 chewsper5-minperiod.
4. Discussion 4.1. Experiment1
The automated measurement of grazing time by the Rumi-WatchSystemwascomparedwithvisualobservationandahigh level of accuracy with a very high correlation of rs=0.96 was observed.Thecorrelationwasslightlyhigherthanthatofother sys-temsusedtodetectfeedingbehaviourinthestudyofBorchersetal. (2016).Inthatstudy,anr=0.88andCCC=0.82wasestablishedfor theCowManager‘SensOor’system(Agis,Harmelen,Netherlands) andr=0.93 andCCC=0.79for the‘TrackACow’system(ENGS, RoshPina,Israel).Theresultsinthecurrentstudyalsoindicated thattheRumiWatchsensorslightlyoverestimatedgrazingtimein comparisontovisualobservation.Whentheparameter‘EAT1TIME’ (headpositiondown)wasconsideredinisolation,thecorrelation wasincreased tors=0.97 and CCC=0.99. Thisshowed thatthe parameter‘EAT2TIME’(headpositionup)mayhaveincludedsome behaviourswhichshouldnotbeconsideredasfeedingbehaviours (e.g.licking).Thus,itmaybebeneficialtodefineingreaterdetail, thebehaviours that should beincluded in thedifferentoutput parametersoftheRumiWatchConverter.
Ruminationtime wasnot significantlydifferentbetweenthe RumiWatchnosebandsensorandvisualobservation.Thisresult iscomparabletoastudyofKrögeretal.(2016),inwhichthe nose-bandsensorwasvalidatedfordifferentfeedingregimesinanindoor situation.Furthermore,astudyofBorchersetal.(2016)showed theRumiWatchSystemtobeslightlymore accurate(CCC=0.99) thantheSmartbowsystem(SmartbowGmbH,Jutogasse,Austria) (CCC=0.96)indetectingruminationtime.
AlthoughtheappliedalgorithmsoftheRumiWatchConverter differeddependingonthechosenoutputresolution,theaccuracy in1-minsummariesand1-hsummarieswasveryhigh.For1-min summariesthechallengeforthealgorithmswastoevaluateeach minuteseparatelywithoutanymajorplausibilitychecksontime periodsbeforeorafterthemeasuredminute.Additionally,error
Table7
Spearman’srho(rs),ConcordanceCorrelationCoefficient(CCC),andBland-Altman-analysis(Bias,upperandlower95%limitsofagreementwith95%CI)ofautomated
measurementsagainstvisualobservationsofgrazingbitesandruminationchewsin5-minperiods.
Behaviour(n./5min) rs CCC Bias(95%CI) Lower(95%CI) Upper(95%CI)
Grazingbites 0.81 0.78 36.01 −66.16 138.19
(28.36;43.66) (−79.41;−52.93) (124.95;151.44) Ruminationchews 0.81 0.94 7.24(−0.15;−14.33) −51.44(−63.72;−39.17) 65.92(53.64;78.19)
detectedbetweenautomatedandvisualobservationshaspotential tobemoreobviouswith1-minsummariesthan1-hsummaries,as someoftheerrorsmaybecompensatedfor,inthetotallingofthe 60min(1-h)value.Whilethehighresolutionof1-minrecording mightnotbeasimportantforacommercialapplication,the bene-fitforresearchpurposesissignificant.Particularly,inbehavioural research,thereactiontotreatmentsonaveryhightimeresolution isvaluable.Visualobservations,ontheotherhand,aremore dif-ficultduetothepersonsandtimecommitmentrequired,aswell asthereducedpracticabilityofconductinghighresolution1-min scansamplingduringnighthours.Therefore,anautomated mea-surementsystemwouldallowincreasedmeasurementandadds functionalitytomanyoftheexperimentsconductedonanon-going basis.
Duetotheexperimentalsetupusedforvalidatingbout param-eters againstvisual observationsthere were limitations forthe automatedsysteminusingtheplausibilityalgorithms.The plausi-bilitycheckscanbeperformedbestoncontinuousdata.However, theshortperiodsof1-hobservationsasusedinthisstudy accumu-latedtoomanycut-offpoints.Therefore,theresultsofamoderate agreementbetweenvisualandautomatedmeasurementmightbe explainedinthisoverallcontext.Itislikelythatwithcontinuous observationsoverextendedperiodsoftime,thoseerrorsoffalsely allocatedstartedorfinishedboutswithin1-hperiodswouldbe reduced.
TheresultsoftheRumiWatchpedometerwereveryaccurate indetectingstandingandlyingbehaviours.Alsaaodetal.(2015) describedandvalidatedthealgorithmfortheRumiWatch pedome-terintheirpublicationandachievedsimilaraccurateresultsfor standingandlyingtimes.Theaccuracyofdetectingwalking cor-rectlyinthecurrentstudywasweakerwithrs=0.78thantheother classificationsofstandingwithrs=0.97andlyingwithrs=0.99.This maybeduetothefactthatbehavioursoflyingandstandingare moreeasilydetected,andthereforemorecorrectlydetectedthan walkingbehaviour.
4.2. Experiment2
Theinter-observerreliabilitywassuccessfullytestedtoensure thevalidityoftheresults.Theoverallresultamongallobservers for grazing and rumination was indicated by a CCC=0.98 and a CCC=1.00, respectively. Similar results were considered by Schirmannetal.(2009)tobeasufficientlyaccuratereferenceto beusedasagroundtruthinplaceofvisualobservationtovalidate anothersensorsystem.Furthermore,theobservationalperiodsin thecurrentstudywereadjustedfrom10minperiodsasoutlinedin Werneretal.(2016)to5minperiods.Thiswastoensuremaximum concentrationbytheobserverwithpotentiallygreateraccuracyin countingallgrazingbites.Additionally, thefocuswasplaced on grazingbitesandruminationchewstoimprovereliability. Record-ingeveryindividualjawmovement(e.g.licking,chewing,etc.)was identifiedaschallengingfortheobservers.
TheRumiWatchSystemslightlyoverestimatedthenumberof grazingbites.Thismaybeduetothenosebandsensorbeingvery sensitiveindetectingpressuredifferences.However,anerrorin visualobservationofgrazingbiteswasalsodetected.Thiswaslikely duetothehighfrequencyofgrazingbites.Thereforethetrueor cor-rectnumberofgrazingbitesmayliebetweenthevaluesrecorded byautomated measurementand visualobservation. Astudyby Championetal.(1997)didconsiderthatdetectionofgrazingbites bytheautomated systemmaybemore accuratethandetection byvisualobservation.AfurtherstudybyDelagardeetal.(1999) alsoshowed,theRumiWatchSystemtobeasstronglycorrelated tovisualobservationinrecordingnumberofbitesperminasthe audiorecordingsystemtheyusedintheirstudywiththeR2values rangingfrom0.72to0.98.
InarecentlypublishedstudybyKrögeretal.(2016),the Rumi-WatchSystem was compared to visual observationin terms of accuracyincapturingruminationchewsat10-minresolution,and aCCCof0.92wasobserved.Inthecurrentstudy,aCCCvalueof 0.94 wasobserved fora similarcomparison.Thisrepresenteda slightincrease in accuracy,which maybe aconsequence ofan updatedversionoftheRumiWatchConverterusedinthecurrent study.However,itisalsotruethatthedietdifferedinboth stud-ies,withafeedingregimeincludingroughage,concentrateanda mixeddietbeingusedinthestudyofKrögeretal.(2016)whilea grassdietwaspredominantinthisstudy.Theobservationalperiods alsodifferedbetweenthestudies,withtheKrögeretal.(2016)and currentstudiesusing10-minand5-minresolutions,respectively. Astheplausibilitychecksforruminationarebasedoncontinuous datathethresholdofaminimumof3mintodetectruminationhad tobeturnedoff.Therefore,theresultsofthetwostudiesarenot entirelycomparable.Forrecordingofcontinuousdataover long-termperiods,thisplausibilitycheckfunctionshouldnotinfluence theaccuracyindetectingruminationchews.
In contrast to most commercially available systems in use on-farm, thecost of the RumiWatchSystemand itsapplication involvinga halter mightnot befeasible forwidespread useon farms.However,asaresearchtoolitcouldbeveryvaluable.With thewide ability ofrecording everyjawmovement,thespecific differentiation in grazing bites and rumination chews, as well asthequantification oftime durationsofruminationand graz-ing,theRumiWatchSystemisauniquetoolformeasuringgrazing behaviourlong-term.Withfurtherdevelopment,itmaybesuitable toestimatefeedintakeofdairycowsonpasture,basedon mea-suringbitesandchews.AsBerckmans(2006)mentioned, visual observationcanrepresentalimitingfactorinbehaviouralresearch, duetothenecessityforrestrictedobservationperiodsandahigh demandinlabour.Therefore,theavailability ofanaccurateand reliableautomatedtechnologymaybeveryusefulandappropriate. 5. Conclusions
Overall, the results indicated that the RumiWatchSystem showedahighlevelofaccuracyinmeasuringfeedingbehaviourof cowsinagrazingenvironment,eventhoughitwasoriginally devel-opedforcowsinindoorfeedingsystems.Theaccuracyof1-min summariesand1-hsummarieswasveryhighwithkappa-values between0.85and0.89andCCC-valuesrangingfrom0.92to0.99, respectively.Thenosebandsensortendedtooverestimategrazing timeandgrazingbitesslightly,whereasthedetectionof rumina-tiontimewasveryaccurate.Theaccuracybetweenobserverand automatedmeasurementwasmoderatefordetectingrumination chewsduetoissuesregardingplausibilitychecksforthespecific experimentalperiods.Thevalidationof thenewlyimplemented parametersintheRumiWatchConverterV.0.7.3.36forgrazingand ruminationboutsdelivered moderateor highconcordance.The pedometershowedverypromisingresultswithCCC-valuesranging from0.92to1.00.Itsperformanceindetectingwalkingwasslightly weakerthanindetectingstandingandlying,butstillappropriate forameasurementsystem.Basedonthoseresults,itmaybe con-cludedthattheRumiWatchSystemisareliablesensortechnology tomonitorgrazingbehaviourandthus,isausefultoolforresearch purposesinagrazingenvironment.
Acknowledgements
We gratefully acknowledgeCassandre Matras and Diarmuid McSweeneyforconductingfieldobservationsinthepaddocks. Fur-thermore,wewouldliketothankScienceFoundationIrelandfor fundingtheproject“PrecisionDairy”.Wewouldalsoliketothank
Itin+HochGmbHforcontributinghardwaresolutions.Forfunding thefirstauthor,wewouldliketoacknowledgetheWalsh Fellow-shipProgram.
References
Alsaaod,M.,Niederhauser,J.J.,Beer,G.,Zehner,N.,Schuepbach-Regula,G.,Steiner, A.,2015.Developmentandvalidationofanovelpedometeralgorithmto quantifyextendedcharacteristicsofthelocomotorbehaviorofdairycows.J. DairySci.98,6236–6242.
Büchel,S.,Sundrum,A.,2014.Technicalnote:evaluationofanewsystemfor measuringfeedingbehaviorofdairycows.Comput.Electron.Agric.108,12–16.
Bailey,D.W.,Gross,J.E.,Laca,E.A.,Rittenhouse,L.R.,Coughenour,M.B.,Swift,D.M., Sims,P.L.,1996.Mechanismsthatresultinlargeherbivoregrazingdistribution patterns.J.RangeManage.49,386–400.
Bareille,N.,Beaudeau,F.,Billon,S.,Robert,A.,Faverdin,P.,2003.Effectsofhealth disordersonfeedintakeandmilkproductionindairycows.Livest.Product. Sci.83,53–62.
Beauchemin,K.A.,Zelin,S.,Genner,D.,Buchanan-Smith,J.G.,1989.Anautomatic systemforquantificationofeatingandruminatingactivitiesofdairycattle housedinstalls.J.DairySci.72,2746–2759.
Berckmans,D.,2006.Automaticon-linemonitoringofanimalsbyprecision livestockfarming.In:GEERS,R.,Madec,F.(Eds.),LivestockProductionand Society.WageningenAcademicPublishers.
Bikker,J.P.,VanLaar,H.,Rump,P.,Doorenbos,J.,VanMeurs,K.,Griffioen,G.M., Dijkstra,J.,2014.Technicalnote:evaluationofanear-attachedmovement sensortorecordcowfeedingbehaviorandactivity.J.DairySci.97,2974–2979.
Borchers,M.,Chang,Y.,Tsai,I.,Wadsworth,B.,Bewley,J.,2016.Avalidationof technologiesmonitoringdairycowfeeding,ruminating,andlyingbehaviors.J. DairySci.99,7458–7466.
Brun,J.,Prache,S.,Béchet,G.,1984.Aportabledeviceforeatingbehaviourstudies. In:5thGrazingWorkshop,Midlothian,UK.
Carrasco,J.L.,Jover,L.,King,T.S.,Chinchilli,V.M.,2007.ComparisonofConcordance correlationcoefficientestimatingapproacheswithskeweddata.J.Biopharm. Stat.17,673–684.
Champion,R.,Rutter,S.,Orr,R.,1997.Distinguishingbitesandchewsinrecordings ofthegrazingjawmovementsofcattle.Proceedingsofthe5thResearch ConferenceoftheBritishGrasslandSociety,8–10.
Cohen,J.,1960.Acoefficientofagreementfornominalscales.Educ.Psychol.Meas. 20,37–46.
Delagarde,R.,Caudal,J.-P.,Peyraud,J.-L.,1999.Developmentofanautomatic bitemeterforgrazingcattle.Annalesdezootechnie48,329–339.
Dillon,P.,Stakelum,G.,1988.Theuseofn-alkanesandchromicoxideasmarkers fordeterminingfeedintake,faecaloutputanddigestibilityindairycows.In: Proceedingsofthe12thGeneralMeetingoftheEuropeanGrassland Federation,Dublin,pp.154–158.
Edmonson,A.,Lean,I.,Weaver,L.,Farver,T.,Webster,G.,1989.Abodycondition scoringchartforHolsteindairycows.J.DairySci.72,68–78.
Giavarina,D.,2015.Understandingblandaltmananalysis.Biochem.Med.25, 141–151.
Herskin,M.S.,Munksgaard,L.,Ladewig,J.,2004.Effectsofacutestressorson nociception:adrenocorticalresponsesandbehaviorofdairycows.Physiol. Behav.83,411–420.
Hinkle,D.E.,Wiersma,W.,Jurs,S.G.,2003.AppliedStatisticsfortheBehavioral Sciences.CengageLearning,Wadsworth.
Kröger,I.,Humer,E.,Neubauer,V.,Kraft,N.,Ertl,P.,Zebeli,Q.,2016.Validationofa nosebandsensorsystemformonitoringruminatingactivityincowsunder differentfeedingregimens.Livest.Sci.193,118–122.
Landis,J.R.,Koch,G.G.,1977.Themeasurementofobserveragreementfor categoricaldata.Biometrics,159–174.
Martin,P.,Bateson,P.P.G.,Bateson,P.,1993.MeasuringBehaviour:AnIntroductory Guide,2nded.CambridgeUniversityPress.
McSweeney,D.,Foley,C.,Halton,P.,O’Brien,B.,Pol-VanDasselaar,A.,Aarts,H., Vliegher,A.D.,Elgersma,A.,Reheul,D.,Reijneveld,J.,2015.Calibrationofan automatedgrassheightmeasurementtoolequippedwithglobalpositioning systemtoenhancetheprecisionofgrassmeasurementinpasture-based farmingsystems.Grasslandandforagesinhighoutputdairyfarmingsystems. In:Proceedingsofthe18thSymposiumoftheEuropeanGrasslandFederation, Wageningen,TheNetherlands,15–17June2015.WageningenAcademic Publishers,pp.265–267.
Mezzalira,J.C.,DeFaccioCarvalho,P.C.,Fonseca,L.,Bremm,C.,Cangiano,C.,Gonda, H.L.,Laca,E.A.,2014.Behaviouralmechanismsofintakeratebyheifersgrazing swardsofcontrastingstructures.Appl.Anim.Behav.Sci.153,1–9.
Nydegger,F.,Bollhalder,H.,2010.VorrichtungzumErfassenderKauaktivität. Switzerlandpatentapplication.September15,2010.
Pérez-Ramírez,E.,Peyraud,J.,Delagarde,R.,2009.Restrictingdailytimeatpasture atlowandhighpastureallowance:effectsonpastureintakeandbehavioral adaptationoflactatingdairycows.J.DairySci.92,3331–3340.
Pahl,C.,Hartung,E.,Grothmann,A.,Mahlkow-Nerge,K.,Haeussermann,A.,2015.
Suitabilityoffeedingandchewingtimeforestimationoffeedintakeindairy cows.Animal10,1–6.
Rutter,S.,Champion,R.,Penning,P.,1997.Anautomaticsystemtorecordforaging behaviourinfree-rangingruminants.Appl.Anim.Behav.Sci.54,185–195.
Rutter,S.M.,2000.Graze:aprogramtoanalyzerecordingsofthejawmovements ofruminants.Behaviorresearchmethods.Instrum.Comput.32,86–92.
Ruuska,S.,Kajava,S.,Mughal,M.,Zehner,N.,Mononen,J.,2016.Validationofa pressuresensor-basedsystemformeasuringeating:ruminationanddrinking behaviourofdairycattle.Appl.Anim.Behav.Sci.174,19–23.
Schirmann,K.,VonKeyserlingk,M.A.,Weary,D.M.,Veira,D.M.,Heuwieser,W., 2009.Technicalnote:validationofasystemformonitoringruminationin dairycows.J.DairySci.92,6052–6055.
Undi,M.,Wilson,C.,Ominski,K.,Wittenberg,K.,2008.Comparisonoftechniques forestimationofforagedrymatterintakebygrazingbeefcattle.Can.J.Anim. Sci.88,693–701.
Werner,J.,Umstatter,C.,Zehner,N.,Niederhauser,J.,Schick,M.,2016.Validation ofasensor-basedautomaticmeasurementsystemformonitoringchewing activityinhorses.Livest.Sci.186,53–58.
Wolfger,B.,Timsit,E.,Pajor,E.A.,Cook,N.,Barkema,H.W.,Orsel,K.,2015.Technical note:accuracyofaneartag-attachedaccelerometertomonitorrumination andfeedingbehaviorinfeedlotcattle1.J.Anim.Sci.93,3164–3168.
Zehner,N.,Niederhauser,J.J.,Nydegger,F.,Grothmann,A.,Keller,M.,Hoch,M., Haeussermann,A.,Schick,M.,2012.Validationofanewhealthmonitoring system(RumiWatch)forcombinedautomaticmeasurementofrumination, feedintake,waterintakeandlocomotionindairycows.In:International ConferenceofAgriculturalEngineering-CIGR-AgEng2012:Agricultureand EngineeringforaHealthierLife,8–12July2012Valencia,Spain,C-0438.
Zehner,N.,Umstätter,C.,Niederhauser,J.J.,Schick,M.,2017.Systemspecification andvalidationofanosebandpressuresensorformeasurementofruminating andeatingbehaviorinstable-fedcows.Comput.Electron.Agric.136,31–41.