THECLIMATOLOGYOFSPRINGTIMEFREEZEEVENTSINTHECENTRALAND
EASTERNUSA
By
TingWang
ATHESIS
Submittedto
MichiganStateUniversity
inpartialful˝llmentoftherequirements
forthedegreeof
GeographyMasterofScience
2021
ABSTRACT
THECLIMATOLOGYOFSPRINGTIMEFREEZEEVENTSINTHECENTRALAND
EASTERNUSA
By
TingWang
TheagriculturalproductionsinthecentralandeasternUnitedStatesaresensitivetospringtime
freezeevents.Asaresultofglobalwarming,increasingtemperatureshaveledtoearliershifted
springs,usuallycalledfalsesprings,whichhaveresultedindisastrousdamageonprematureplants
exposedtosubsequentfreezeevents.Thisstudyanalyzestheclimatologyofspringtimefreezesand
theirimpactsonagricultureintheMidwesternUnitedStatesfortheperiodof1981-2018.Thestudy
beganbyevaluatingtwopotentialdatasetsforthepurposeofthisanalysis:thePRISM(Parameter-
elevationRegressionsonIndependentSlopesModel,http://prism.oregonstate.edu)analysisandthe
ERA5(the˝fthmajorglobalreanalysisproducedbyEuropeanCentreforMedium-RangeWeather
Forecasts,Hersbachetal.,2018)reanalysis.ThePRISMdataarefoundtobeabetterrepresentation
oftheobservedfreezingeventsandthereforeusedforestablishingfreezeeventsclimatology,while
theERA5reanalysisisusedtounderstandtheweatherconditionsandclimatebackgroundofthe
freezeevents.FreezingdaysinMarchshowadecreasingtrendacrossourstudyregionfrom1981
to2018.EOFanalysisoffreezingdaysinMarchshowsarelativelylargervariationintheOhio
Valley,andthe˝rstEOFtimeseriesshowssubstantialinterannualvariability.Thepositivephase
ofNAO(NorthAtlanticOscillation)isusuallyassociatedwithlessfreezingriskinMarchacross
thestudyregion.Acropyieldsimulationmodelisusedtoinvestigatethehistoricalimpactsof
falsespringsandsubsequentfreezeeventsonfruitcropyieldsusingappleasanexample.Damage
tendstooccurattheearlygrowingstagesofappleswhentheyaremorevulnerable.Damageis
generallyoccurringonearlierandwarmerdays,whichcouldbeduetothemorefrequentfalse
springoccurrences.TheUpperMidwestandtheNortheastareregionsthatarelessvulnerableto
freezedamage.
ACKNOWLEDGEMENTS
Iwouldliketothankmyadvisors,Drs.Je˙AndresenandShiyuanZhong,andcommittee
memberandgraduatedirector,NathanMoore,fortheirhelpandguidanceinthisresearch.They
havebeenverysupportiveandinstructive,andIamsogratefultohavetheopportunitytowork
withthem.
IwouldliketoacknowledgetheDepartmentofGeography,EnvironmentandSpatialSciences
forgivingmetheopportunitytopursuemygraduatestudyandtothankallmyfellowgraduate
studentsandcollaboratorsfortheirhelpandencouragement.Itisawonderfulexperiencebeinga
geographerhere.Ihavelearnedalotandexpandedmyinterests.
IwouldliketothankresearchersintheForestServiceNorthernResearchStationLansing
O˚ce,particularlyXindiBian,WarrenHeilman,JosephCharney,andMikeKiefer,fortheir
scienti˝cdiscussionsandforprovidingcomputationalanddatastorageresourcesando˚cespace.
Lastbutnotleast,IwouldliketothankthebeautyofMichigan.Thegorgeousviewshere
attractmetochoosetostayherefortwoyears.IlovethesapphireshinesoftheGreatLakes;Ilove
thefantasticbikingtrailsaroundtheMackinacIsland;IlovethemelodioussoundoflakewavesI
recordedattheimpressiveAirbnbhouse;IlovethethickcarpetoffallenleavesinPictureRocks;I
lovethespectacularsunsetoftheLakeLansingNorth.
Solongasmencanbreatheoreyescansee,
Solonglivesthis,andthisgiveslifetothee.
iii
TABLEOFCONTENTS
LISTOFTABLES
.......................................
vi
LISTOFFIGURES
.......................................
viii
CHAPTER1INTRODUCTION
...............................
1
CHAPTER2EVALUATIONOFTHEERA5ANDPRISMDATASETS
..........
7
2.1Background......................................7
2.1.1Necessityofevaluatinggriddeddatasets...................7
2.1.2PRISManalysisandERA5reanalysisdatasets................8
2.1.3Studyobjective................................9
2.2DataandMethod...................................10
2.2.1Studyregion..................................10
2.2.2Griddeddatasets...............................10
2.2.3Observationaldatasets............................11
2.2.4EvaluationMethods..............................14
2.3Results.........................................15
2.3.1Assessmentofweathervariables.......................15
2.3.2Assessmentofderivedvariables.......................30
2.4Summary.......................................34
CHAPTER3FREQUENCYANDSEVERITYOFSPRINGTIMEFREEZEEVENTS
..
36
3.1Background......................................36
3.1.1FalseSpringsandSpringtimeFreezeEvents.................36
3.1.2SpatiotemporalVariabilityofSpringtimeFreezeEvents...........37
3.1.3StudyObjectives...............................38
3.2DataandMethod...................................39
3.3Results.........................................41
3.3.1Temporalandspatialvariabilityoffreezeevents...............41
3.3.2FreezingAreaPercentage...........................43
3.3.3ResultofEOFanalysisforfreezingdaysinspringtimemonths.......51
3.4Summary.......................................56
CHAPTER4SPRINGTIMEFREEZEEVENTSIMPACTSONAGRICULTURE
.....
58
4.1Background......................................58
4.1.1Springtimefreezetypesandtheirimpactsonperennials..........58
4.1.2Falsespringoccurrencesandimpactsoncrops................59
4.1.3In˛uenceofphenologyonfreezedamage..................60
4.1.4Studyobjectives................................61
4.2Datasetandmethods.................................62
4.2.1PRISMdataset................................62
4.2.2Appleyieldsimulationmodel........................62
iv
4.3Results.........................................64
4.3.1Characteristicsoftheyearlyresults......................64
4.3.2DailyResultsandsomederivedindices...................68
4.4Summary.......................................83
CHAPTER5CONCLUSION
.................................
85
APPENDIX
...........................................
88
BIBLIOGRAPHY
........................................
98
v
LISTOFTABLES
Table2.1:Percentofhourlyobservationsmissingatsixstations...............13
Table2.2:Hourlysummarystatistics,rootmeansquaredi˙erence(RMSD),andBias
betweenERA5productandobservationalnetworksatsixlocations.......23
Table2.3:Dailysummarystatistics,RMSDbetweenERA5productandobservational
networks,atsixlocations..............................24
Table2.4:Dailysummarystatistics,RMSDbetweenPRISMproductandobservational
networks,atsixlocations..............................25
Table2.5:Seasonalsummarystatistics,RMSDbetweenERA5productandobservational
networks,atsixlocations..............................27
Table2.6:Seasonalsummarystatistics,RMSDbetweenPRISMproductandobserva-
tionalnetworks,atsixlocations..........................28
Table2.7:Annualsummarystatistics,RMSDbetweenERA5productandobservational
networks,atsixlocations..............................29
Table2.8:Annualsummarystatistics,RMSDbetweenPRISMproductandobservational
networks,atsixlocations..............................29
Table2.9:Growingdegreeday(Base4
°
C)summarystatistics................30
Table2.10:FreezeeventscapturedratiosatEastLansingofERA5..............31
Table2.11:FreezeeventscapturedratiosatEastLansingofPRISM.............32
Table2.12:Freezeeventsbelow0
°
Ccapturedratiosatsixstationswithoutdetailedcate-
goriesofERA5...................................33
Table2.13:Freezeeventsbelow0
°
Ccapturedratiosatsixstationswithoutdetailedcate-
goriesofPRISM..................................33
Table3.1:39-yearmeanofregionalaveragedmonthlyFAP.................46
Table3.2:Theil-SenSlopeformonthlyFAP..........................46
Table3.3:Pearsoncorrelationvalues(r)comparingmonthlyFAPforthemonthsof
March,April,andMaytoNAOvalues0-3monthsearlier.............48
vi
Table3.4:Pearsoncorrelationvalues(r)comparingmonthlyFAPforthemonthsof
March,April,andMaytoPNAvalues0-3monthsearlier.............50
TableA.1.1:.RMSDsofairtemperaturebetweenWRFandobservationsatthreelocations
for48hoursfromMay12thtoMay13thwithdi˙erentverticalresolutions....91
TableA.1.2:.RMSDsofwindspeedbetweenWRFandobservationsatthreelocationsfor
48hoursfromMay12thtoMay13thwithdi˙erentverticalresolutions......93
vii
LISTOFFIGURES
Figure2.1:Studycomparisonsitelocations..........................12
Figure2.2:Di˙erencebetweenERA5andobservationsfordailymaximumtemperature
in2010,2012,and2014..............................16
Figure2.3:Di˙erencebetweenERA5andobservationsfordailyminimumtemperature
in2010,2012,and2014..............................17
Figure2.4:Di˙erencebetweenPRISMandobservationsfordailymaximumtemperature
in2010,2012,and2014..............................18
Figure2.5:Di˙erencebetweenPRISMandobservationsfordailyminimumtemperature
in2010,2012,and2014..............................19
Figure2.6:Timeseriesofthedi˙erencebetweenERA5monthlymeandailymeantem-
peratureandobservedvaluesatsixlocations...................20
Figure2.7:Timeseriesofthedi˙erencebetweenPRISMmonthlymeandailymean
temperatureandobservedvaluesatsixlocations.................21
Figure2.8:Scatterplotofdailyminimumandmaximumtemperaturebetweenobserva-
tionsandreanalysisdatasets............................22
Figure3.1:Average,standarddeviation,andtrendoffreezingdaysinspring(March,
April,May)from1981to2019..........................42
Figure3.2:Timeseriesofspringtimefreezeeventscountfrom1981to2019........43
Figure3.3:Subsetregionsde˝nedinKarlandKnight(1998)................44
Figure3.4:TimeseriesofmonthlyFAPvalues........................45
Figure3.5:TheleadingtwoEOFmodesandPCtimesseriesforfreezingdaysinMarch..52
Figure3.6:TheleadingtwoEOFmodesandPCtimesseriesforfreezingdaysinApril...53
Figure3.7:The˝rstPCtimesseriesforfreezingdaysinMarchrelatedtoatmospheric
variables......................................54
Figure3.8:ThesecondPCtimesseriesforfreezingdaysinMarchrelatedtoatmospheric
variables......................................55
viii
Figure3.9:TimesseriesofNAOinMarchrelatedtoatmosphericvariables.........55
Figure4.1:Average,absolutevariation,relativevariation,andtrendofthe˝rstgreendate
(Juliandate)from1981to2018..........................64
Figure4.2:Average,absolutevariation,relativevariation,andtrendofthebloomdate
(Juliandate)from1981to2018..........................66
Figure4.3:Average,absolutevariation,relativevariation,andtrendoftheyearlypoor
pollinationdaysfrom1981to2018........................67
Figure4.4:Average,absolutevariation,relativevariation,andtrendoftheyield(bud
survivalchance)from1981to2018........................68
Figure4.5:Average,absolutevariation,relativevariation,andtrendofdamagedaysfrom
1981to2018....................................69
Figure4.6:Averageofdamagedaysateachgrowthstagefrom1981to2018........70
Figure4.7:Thetrendofdamagedaysateachgrowthstagefrom1981to2018........71
Figure4.8:Theaverageoffreezedamageoccurringdateateachgrowthstagefrom1981
to2018.......................................72
Figure4.9:Thetrendoffreezedamageoccurringdateateachgrowthstagefrom1981to
2018........................................73
Figure4.10:Theaveragedailyminimumtemperatureduringdamageoccurringdateat
eachgrowthstagefrom1981to2018.......................74
Figure4.11:Thetrendofdailyminimumtemperatureduringdamageoccurringdateat
eachgrowthstagefrom1981to2018.......................75
Figure4.12:Thetimeseriesofarea-averageddamagedaysinsixsubregionsfrom1981to
2018........................................77
Figure4.13:Thesumofdamagedaysateachgrowthstageinsixsubregionsfrom1981to201878
Figure4.14:Theaverageofdamagevaluesateachgrowthstageinsixsubregionsfrom
1981to2018....................................79
Figure4.15:Theaveragedailyminimumtemperatureateachgrowthstageinsixsubregions
from1981to2018.................................80
ix
Figure4.16:Theaveragedailyminimumtemperatureduringdamagedaysateachgrowth
stageinsixsubregionsfrom1981to2018....................81
Figure4.17:Theaveragedamage-occurringdatesateachgrowthstageinsixsubregions
from1981to2018.................................82
FigureA.1.1:.Di˙erencesofsoilmoisturebetweenWRFoutputsandobservationsat
Williamsburg...................................91
FigureA.1.2:.Di˙erencesofairtemperaturebetweenWRFoutputsandobservationsat
threelocations...................................92
FigureA.1.3:.ComparisonofdailyminimumairtemperaturebetweenWRFandPRISM
onMay8th.....................................92
FigureA.1.4:.Dailyminimumtemperaturechangesinresponsetoincreasedsoilmoisture
byafactorof1.05onMay8thandMay9th...................94
FigureA.1.5:.Dailyminimumtemperaturechangesinresponsetoincreasedsoilmoisture
byafactorof1.05onMay12thandMay13th..................95
FigureA.1.6:.Majorsoilcategoriesinourstudyregion....................96
FigureA.1.7:.SoilmoisturechangebasedonsoiltypesonMay8thandMay9th.......96
FigureA.1.8:.SoilmoisturechangebasedonsoiltypesonMay12thandMay13th.....97
x
CHAPTER1
INTRODUCTION
TheMidwesternUS(includingthestatesofIllinois,Indiana,Iowa,Michigan,Minnesota,Mis-
souri,Ohio,andWisconsin)ranksamongthemostintensiveandextensiveagriculturalproduction
areasintheworldanda˙ectstheglobaleconomyconsistentlyanddynamically.Amongthevarious
weather-andclimate-relatedthreatstoagriculturalproductionsystemsintheregionareabnormally
mildwintersfollowedbywarm,earlysprings,usuallycalledfalsesprings,whichhaveresulted
indisastrousdamagetocropsphenologicallyadvancedbytheearlywarmthbutthenexposedto
subsequentfreezeevents(Guetal.,2008;O'Brienetal.,2019).Forexample,springtimefreeze
eventsin2002and2012causeddramaticlossesinregionalagriculturalproduction,includinga
yieldreductionofmorethan95%ofsourcherryyieldsinMichiganin2002andalossofmorethan
85%ofappleproductioninMichiganin2012(NASS,2002;Marinoetal.,2011;O'Brienetal.,
2019;Labeetal.,2015;Kistneretal.,2018;Guetal.,2008).Whiletheeconomicimpactsoffalse
springsandsubsequentfreezeeventsaresigni˝cant,thereisagenerallackofpreviousresearch
investigatingthenature,extent,andimpactsofsuchevents,whichistheprimaryobjectiveofthe
studypresentedinthisthesis.
Understandingthestatisticalcharacteristicsofspringtimefreezeeventsrequirescontinuous,
long-term(decades)climatedatafrombothstationobservationsandgriddedanalysis/reanalysis.
Griddeddatasetsmayprovideconsiderableadvantagesoversinglesitestationobservationswith
continuouscoverageoverspaceandtime,especiallyinareaswithsparsemeteorologicalstationsor
restricteddataaccessibilityandavailability(Ceglaretal.,2017).Griddeddatasetsincludeanalysis
orreanalysisproducts,bothofwhichinvolveperformingdataassimilation,aprocessrelyingon
bothobservationsfromavarietyofsourcesandforecastsfromnumericalweatherpredictionmodels
(Parkeretal.,2016).Initially,sometypeofobjectiveanalysisisappliedonamyriadofobservations
toanirregulargridtorepresenttheatmosphericstate(Deeetal.,2011).Asafollow-upprocess,
model-basedreanalysiscanbeperformedwithindividualdatalayersfromthedataassimilation
1
systemtoprovideamultivariate,spatiallycomplete,andcoherentrecordoftheglobalatmospheric
circulation(Deeetal.,2011).Areanalysisisaparticulartypeofanalysisdonewitha˝xed
softwaresystem.However,bothreanalysesandanalysesaresensitivetochangesinobservational
systems(Deeetal.,2011).Reanalysesarepotentiallymoreattractivesincetheyprovidemore
comprehensivegriddedestimatesofatmosphericconditionsatregularintervalsoverlongperiods
oftime(Parkeretal.,2016).Griddedreanalysesandanalyseshavebeenusedinawiderangeof
applications,includingunderstandingatmosphericdynamicsofjetstreams(Kidstonetal.,2010),
investigatingtheimpactofclimatevariabilityonagriculture,airpollutionapplications,andwind
energydevelopment(Toretietal.,2019;Essouetal.,2017;Gleixneretal.,2020;Cannonetal.,
2015),andevaluatingclimatemodels(Gleckleretal.,2008).
Therearealsolimitations.Despitetheirpopularity,griddeddatasetsrepresentarea-averaged,
andinsomecases,time-averagedestimatesofmeteorologicalvariablesandhencemaynotaccu-
ratelyrepresentmeteorologicalconditionsataparticularlocationortime(Tetzneretal.,2019;
Bosilovichetal.,2013).Itis,therefore,necessarybeforeaspeci˝capplicationtoassesshow
thegriddeddatasetsrepresentthelocalvariabilityinaregionofinterest.Thisnecessityisthe
motivationfortheworkpresentedinChapter2thatevaluatestwogriddeddatasetsusedforthe
studiesinthelaterchapters:TheanalysisproductPRISM(Parameter-elevationRegressionson
IndependentSlopesModel,http://prism.oregonstate.edu),agriddedanalysisdatasetprovidedby
thePRISMclimategroupfromOregonStateUniversity,andareanalysisproduct,ERA5,the˝fth
majorglobalreanalysisproducedbytheEuropeanCentreforMedium-RangeWeatherForecasts
(ECMWF;Hersbachetal.,2018).PRISMestimatestemperaturesoverspaceusingempirical
statisticalrelationshipswithobservedsinglesiteclimatedatafromnearbystationsadjustedfor
elevation(WaltonandHalletal.,2018).Stationobservationsaregivenhigherweightsifthey
aregeographicallyclosertothetargetgridcellandhavesimilarcoastalproximityortopographic
position,consideringallotherfactors(WaltonandHalletal.,2018).Previousresearchsuggests
thatconsiderationofthesefactorsresultedingreateraccuracyforPRISMincomplexterrainand
alongcoastlinescomparedtoothergriddedanalyses,includingDaymetandWorldClim(Walton
2
andHalletal.,2018).ERA5isaglobalreanalysisextendingbackto1979withahorizontal
resolutionof30kmandhasbeenevaluatedforanumberofregionsandhistoricalperiods.For
example,Tetzneretal.(2019)evaluatedERA5inthesouthernAntarcticPeninsulaandEllsworth
Land,Antarctica,andconcludedthatERA5ishighlyaccurateinrepresentingthemagnitudeand
variabilityofnear-surfaceairtemperature.Theyalsorevealedthatthehigherspatialandtempo-
ralresolutionofERA5comparedtoitspredecessor,ERA-Interim,signi˝cantlyreducedthecold
coastalbias.Gleixneretal.(2020)concludedthatERA5reducedstatisticalbiasesandimproved
therepresentationofinterannualvariabilityovermostofAfrica.However,fewstudieshavespeci˝-
callyevaluatedPRISMandERA5overtheMidwesternandEasternUnitedStates,necessitatingan
assessmentofhowbothdatasetsrepresentthelocalclimaticvariabilityoftemperaturepriortotheir
applicationasaproxyforobservationstoinvestigatethehistoricalfrequencyoffreezeevents.In
anevaluationofthetwogriddeddatasetsacrosstheregionfrom2001to2018,PRISMwasfound
tomoreaccuratelyrepresenthistoricalobservationscomparedtoERA5andwasusedasaproxy
forobservationsacrossthestudyregioninastudyofspringtimefreezeevents(Chapter3)andtheir
impactsonagriculture(Chapter4).
Toinvestigatethestatisticalcharacteristicsofspringtimefreezeevents,weneed˝rsttoconsider
thephysicalandcausalmechanismsassociatedwiththem.Therearetwomaintypesoffreezeevents,
advectionfreezesandradiationfreezes.Advectionfreezestendtooccurinawell-mixedboundary
layerwithstrongwindsassociatedwithcoldfrontsortheleadingedgesofpolar-originhigh-pressure
systems(Winkleretal.,2012).Radiationfreezestendtobeassociatedwithsurfaceanticyclones
locatedacrosstheregioncharacterizedbyrelativelycloud-freeandcalmconditionsinwhichthe
earth'ssurfacecoolsmorerapidlythantheatmosphereabove(Winkleretal.,2012).Radiation
freezesoccurmorefrequentlyandarecomparativelyeasiertopredictduetotheassociationwith
atemperatureinversion.Onthecontrary,advectionfreezesarelessfrequentandoccurwithout
inversiondevelopingandthusaredi˚culttopredict.
Therearealsotemporalchangesinspringtimefreezefrequencytoconsider.Dailyminimum
temperaturesacrosstheregionhavewarmedinrecentdecades(Andresenetal.2012),inmanyareas
3
morequicklythanmaximumtemperatures,resultinginadecreaseinthediurnaltemperaturerange
(Quetal.,2014).Therehavealsobeenchangesatbothendsofthetemperaturedistribution.Roweet
al.(2012)notedthatthedayswithrecordlowdailyminimumtemperatureshadbeensigni˝cantly
andsteadilydecreasingnearlyeverywhereacrosstheUnitedStates,whilethedayswithrecord
highdailyminimumtemperaturesincreasedconsiderably.Therehavealsobeendecreasesinthe
numberofdayswithfreezingtemperatures.DeGaetano(1996)detectedanoveralldecreasingtrend
infreezingdaysintheNortheasternUnitedStatesfrom1959to1993,whileEasterlingetal.(2002)
observedasigni˝cantreductioninspringtimefreezingdaysacrossthewesternandnorth-central
USfrom1948to1999.Inthesamestudy,˝rstfallfreezeoccurrenceswasfoundtobeoccurring
laterintheyearwithtime;theoccurrencesofthelastspringfreezeweretrendingearlier,resulting
inanoverallincreaseinthelengthofthefrost-freeseason(Easterlingetal.,2002).Yuetal.(2014)
alsorevealedthesametrendfor˝rstfallfreezedates,lastspringfreezedates,andthefrost-free
seasonlengthfrom1980to2010intheMidwest.Theirstudyalsoidenti˝edlesserinterannual
variabilityofthesedatesintheGreatLakesregionversusotherportionsoftheMidwest(Yuetal.,
2014).
Projectedfuturechangesinthefrequencyandseverityoffalsespringeventsandsubsequent
freezeeventsassociatedwithawarmingworldmaybecomplex(Allstadtetal.,2015).Both˝rst
greendatesand˝rstbloomdateswilllikelycontinuetoshiftearlierintheyearwithrisingglobal
temperatures,butthecomplexnatureofthisprocesscomplicatesmakingaprioriestimatesofhow
temperaturechangeswilla˙ectspringonset(Allstadtetal.,2015).Resultsfrompreviousstudies
suggestthattheintermountainWest,GreatPlains,andupperMidwestregionsoftheUnitedStates
mayexperienceincreasesinthefrequencyoffalsesprings(Allstadtetal.,2015;Petersonand
Abatzoglou2014;O'Brienetal.,2019).However,Winkleretal.(2012)concludedthatprojections
offreezeriskafterfalsespringsinMichiganareassociatedwithhighuncertainty.
FollowinganexaminationofhistoricalfreezeeventsacrosstheregioninChapter3,westudy
potentialassociatedagriculturalimpactsinChapter4.Springtimefreezeeventscancausesevere
damagetocrops,especiallyfollowingfalsesprings.Thepatternofabnormalearlywarmthwillnot
4
directlycausedamage,butthetimingofanysubsequentfreezeevents,particularlyinrelationtothe
cropphenologicalstage,iscritical(Augspurgeretal.,2007;Longstrothetal.,2012).Forexample,
intheearlyphenologicalstagesoftemperatetreefruitcropssuchasappleandcherry,coldinjury
anddamagebeginattemperaturesnear-6
°
C,butasthebloomstageapproaches,temperaturesas
warmas-2
°
Ccancauseconsiderabledamage(Longstroth,2005).Therefore,themagnitudeof
coldinjurymayvarygreatlydependingonthecropphenologicalphaseandthelevelofobserved
minimumtemperaturesatthetimeofthefreeze(Hufkensetal.,2012).Extensivedamagein
both2007and2012freezeeventswasassociatedwithrecord-breakingwarmtemperaturesduring
themonthofMarch,whichledtoabnormallyadvancedandvulnerablecropphenologicalstages,
followedbyaseriesofbothadvectiveandradiationalfreezesinApril(Augspurgeretal.,2007;
Kistneretal.,2017).Sincechangesinextremeweathereventssuchasfreezeeventsdependmore
onvariabilitythanoveralltemperaturetrends,studiesrevealthatthereisanincreasedriskoffreeze
damageoncropswithmoresigni˝canttemperaturevariabilityrelatedtoglobalclimatechange
(Rigbyetal.,2008;Augspurgeretal.,2013).
Deterministicorsemi-deterministiccropsimulationmodelsarefrequentlyappliedtobetter
understandthepotentialimpactsofclimatevariabilityandchangeonagriculturalproduction
systems.However,therelativelyslowgrowthandcomplexityofperennialstendtolimitthe
numberandtypesofavailablecropsimulationmodels(LobellandField,2011).Thegrowthand
developmentofvegetationisgenerallyafunctionofmanyenvironmentalvariables,withthemost
substantialcontributionistypicallyassociatedwithtemperature(Rigbyetal.,2008).Rigbyet
al.(2008)usedasimpleThermalTime(orDegreeDay)modeltoestimatethephenologyofbud
break,andtheirresultsrevealedthatfrostrisktovegetationisassensitivetoincreasesindaily
temperaturevarianceastoincreasesinthemeantemperature.Similarly,Marinoetal.(2011)used
amodeltoapproximatethetimingofleaf-outtoreconstructtheseasonalstartoftheseasonover
theperiod1901-2007acrossthesoutheasternUnitedStatesandnotedadecreasingtrendoffalse
springoccurrencesoverthestudyregion(Marinoetal.,2011).Variousstudieshavefocusedon
howthephenologicalstagesofplantsrespondtothetemperaturechangewithwarmingtrends,but
5
farfewerhaveinvestigatedpossibletreefruitproductionimpacts.InChapter4ofthisstudy,the
approachofZavallonietal.(2006)wasappliedovertheMidwesternandEasternUnitedStatesto
estimatetheyieldresponseofapple,withphenologicaldevelopmentbasedonthemethodologyof
Rijal(2017).
Giventheirimportanceindescribingyieldvariabilityoftreefruitproductionovertime,a
betterunderstandingoftheclimatologyofspringtimefreezesandtheirimpactsonagriculture
ishighlydesirable.ThisstudyconsidersproductionareasacrossthecentralandeasternUnited
States.InChapter2,ERA5reanalysisandPRISManalysisdatasetsarecomparedtosinglestation
observationsatdi˙erenttemporalscalesforseveralrelevantclimatevariablesandthenevaluatedto
determinehowthesedatasetscapturefreezeeventswithspeci˝edcriteria.Chapter3investigates
thefrequency,severity,andpotentialcausesofspringtimefreezeeventoccurrencesacrossthestudy
regionusingERA5reanalysistounderstandtheweatherconditionsandclimatebackgroundofthe
freezeeventsrepresentedbythePRISManalysisdataset.InChapter4,weinvestigatethehistorical
impactsofspringtimefreezeeventdamageonappleintheMidwesternUSfrom1981to2018with
acropsimulationmodelwithinputdatafromPRISM,includingtemporalchangesinassociated
agroclimaticvariables.Finally,inChapter5,wesummarizethemainresultsfromChapter2to
Chapter4andalsothelimitations.
6
CHAPTER2
EVALUATIONOFTHEERA5ANDPRISMDATASETS
2.1Background
2.1.1Necessityofevaluatinggriddeddatasets
Manyenvironmental-themedapplicationsincludingcropsimulationmodelsrequiredetailed
andcomprehensiveweatherandclimatedataasinputs(Ceglaretal.,2017).Suchinputdataare
usuallyobtainedfromvarioussources,includingdirectobservationsfromnetworksofweatheror
climatestations,griddedanalysesofstationobservations,orgriddedreanalysesblendingobserva-
tionswithnumericalweatherorclimatemodeloutputs.Griddedanalyses/reanalysesdatasetshave
advantagesoverstationobservationsastheycanprovidereasonableestimatesofmissingdataover
spaceandtime,allowingforcontinuouspatternanalysis(Ceglaretal.,2017).Thegriddedanaly-
sesofstationobservationsaretypicallygeneratedbymeansofspatialandtemporalinterpolation
techniques(Dalyetal.,2008;Waltonetal.,2018),whilegriddedreanalysesgenerategriddeddata
surfacesprimarilywithmodelsimulationsandtheiroutputs.Becauseofthecontinuousspatial
andtemporalcoverage,griddedanalysesorreanalysesdatasetshavebeenwidelyusedforunder-
standingweatherphenomena,climatevariabilityandtrends,andotherrelatedapplicationssuchas
agriculture,airpollution,windenergy(Toretietal.,2019;Essouetal.,2017;Gleixneretal.,2020;
Cannonetal.,2015).Therearealsochallengesintheirapplication.Inparticular,griddeddatasets
representarea-averaged,andinsomecases,time-averagedestimatesofmeteorologicalvariables
andhencemaynotaccuratelyrepresentmeteorologicalconditionsataparticularlocationortime
(Tetzneretal.,2019).Consequently,thesedatasetsshouldbeevaluatedagainstactualobservations
todocumentifbiasesorerrorsexistbeforebeingusedtounderstandweatherorclimateconditions
foraparticularapplication.
7
2.1.2PRISManalysisandERA5reanalysisdatasets
PRISM(Parameter-elevationRegressionsonIndependentSlopesModel;Dalyetal.,2008),a
griddedanalysisdatasetofsurfaceclimatevariablesprovidedbythePRISMclimategroupfrom
OregonStateUniversity,wasutilizedtocharacterizefreezeeventsinthestudydomain.Thedataset
was˝rstevaluatedbycomparingitwithobservationaldatasetsatdi˙erentlocationsintheregion.
ThePRISMclimategroupincorporatesclimateobservationsfromawiderangeofmonitoring
networksanddevelopscontinuousspatialclimatedatasetsbyapplyingsophisticatedqualitycontrol
measures.PRISMprovideslong-term(from1981topresent)dailyclimatevariables,including
dailyminimum/maximum/meantemperature,precipitation,andvaporpressurede˝cit(VPD)ata
uniform4kmgridspacing.ThecomparisonofthefreezecharacteristicsderivedfromthePRISM
datawiththosederivedfromstationobservationswilldeterminewhetheritisagoodproxyfor
observationsoffreezeoccurrences.
BecausethePRISMdatasetonlycontainsdailysurfacevariables,anotherdatasetthatprovides
informationabovethesurfaceisnecessarytoexplainthevariabilityandtrendsoffreezingevents
fromtheperspectiveoflarge-scaleatmosphericconditions.Therecentlyreleasedclimatereanalysis
dataset,ERA5,bytheEuropeanCentreforMedium-RangeWeatherForecasts(ECMWF)willbe
usedforthispurpose.ERArefersto`ECMWFRe-Analysis,'withERA5beingthe˝fthmajorglobal
reanalysisproducedbyECMWF.ERA5provideshourlydatafrom1979untilpresentat139pressure
levelsfrom1000hPato1hPawithahorizontalresolutionof30km.Asthelatestglobalreanalysis
dataset,ERA5bene˝tedfromadecadeofdevelopmentsinmodelphysics,coredynamics,anddata
assimilationrelativetoearlierreanalysisdatasetssuchasERA-Interim(Hersbachetal.,2019).The
ERA5datasetsigni˝cantlyenhanceshorizontalresolutionwiththe30kmgridspacingatdi˙erent
pressurelevelsand9kmgridspacingatthesurfaceonlandcomparedto79kmforERA-Interim.
AnumberofstudieshaveevaluatedtheaccuracyandapplicabilityofERA5andERA-Interim.For
example,Tareketal.(2019)revealthattheERA-InterimandERA5temperaturesaregenerally
similartositeobservationsoverNorth-America,andERA5demonstratesasubstantialreduction
inwarmerbiasescomparedtothoseintheERA-Interimdataset.Also,withagoodrepresentation
8
ofobservationalprecipitationdata,thedry/wetbiaspatternofERA-Interimissharplyreducedin
ERA5(Tareketal.,2019).Inthisstudy,wewillexaminetheaccuracyofERA5inrepresenting
freezeeventsinourstudyregion.
2.1.3Studyobjective
TheprimaryobjectiveofthisstudyistocompareERA5reanalysisandPRISManalysisdatasets
withsingle-siteobservationsatdi˙erenttemporalscalesforseveralrelevantclimatevariablesand
thenevaluatehowthesedatasetscapturefreezeeventswithspeci˝edcriteria.
9
2.2DataandMethod
2.2.1Studyregion
ThestudydomainisthecentralandeasternUnitedStatesencompassingtheregionfrom105
º
Wto75
º
Wlongitudeandfrom35
º
Nto50
º
Nlatitude.Thisregionencompassesoneofthelargest
andmostintensiveagriculturalproductionareasintheworld,includinglargeareasofspecialty-crop
agriculturesuchastreefruitwhicharehighlyvulnerabletospringfreezes.
2.2.2Griddeddatasets
ERA5isthe˝fthgenerationreanalysisfromECMWF.Thereanalysisisproducedata1-hourly
timestepusingasigni˝cantlymoreadvanced4D-varassimilationscheme(Tereketal.,2019).The
ERA5-Landreanalysisdatasetprovidesaconsistentviewoftheevolutionoflandvariablesover
severaldecadesatanenhancedresolution(9km)comparedtoERA5(MuñozSabateretal.,2019).
Hourlytemperature,relativehumidity,andprecipitationfrom2001to2018acrossourstudyregion
areobtainedfromERA5-Land.Datafromthecentersofgridboxesfromthereanalysisnearestto
theobservationalstationswerethenextractedforcomparison.Also,itisnoteworthythatvariable
outputsingridboxescontainingadjacentwaterareas(e.g.,alongtheshoresoftheGreatLakes)
onlyrepresentspatialaveragesacrossthelandareasanddonotincludethewater-coveredsurfaces
(MuñozSabateretal.,2019).
TheParameter-elevationRegressionsonIndependentSlopesModel(PRISM;Dalyetal.1994)
isusedtoderivegriddedsurfacevariablesfortheconterminousUnitedStates.Ateachgridcell,
anelevationregressionfunction,consideringmultiplephysicalfactorsthatre˛ecttheirsimilarity
tothetargetgridcell,is˝ttostationalobservationsusingamovingwindow(Waltonetal.,
2018).Thesefactorsincludedistance,cluster,elevation,coastalproximity,topographicfacet,
verticallayer,topographicposition,ande˙ectiveterrainheight(Waltonetal.,2018).PRISM
incorporatesdatafromawiderangeofnetworks,asmanyaspossible,includingCOOP,RAWS,the
CaliforniaDataExchangeCenter(CDEC),Agrimet,NRCS,theCaliforniaIrrigationManagement
10
InformationSystem(CIMIS),andmore(seehttp://prism.oregonstate.edufordetails).Weobtained
dailyminimumandmaximumtemperaturesfromPRISMfrom2001to2018acrossourstudy
regionforthiscomparison.
2.2.3Observationaldatasets
Threestationalobservationdatasetsareusedinthestudy.TheyarerespectivelytheMichi-
ganEnviroweather(EW)network,theNOAANationalWeatherServiceAutomatedSurfaceOb-
servingSystem(ASOS),andtheUSClimateReferenceNetwork(USCRN).TheMichiganEn-
viroweather(EW)networkisaninteractiveinformationsystemlinkingreal-timeweatherdata
fromastatewidemesonetwork,forecasts,andbiologicalandotherprocess-basedmodelsforas-
sistanceinoperationaldecision-makingandriskmanagementassociatedwithMichigan'sagri-
cultureandnaturalresourceindustries.Itaimstodevelopasustainableweather-basedinforma-
tionsystemthathelpsusersmakepests,plantproduction,andnaturalresourcemanagementde-
cisionsinMichigan(https://www.canr.msu.edu/resources/enviroweather_weather_data_and_pest-
_modeling).TheAutomatedSurfaceObservingSystem(ASOS)unitsareautomatedsensorsuites
designedtoserveasaprimarymeteorologicalobservingnetworkintheUnitedStates.There
arecurrentlymorethan900ASOSsiteswithone-minuteand˝ve-minutedataathourlyintervals
intheUnitedStates(https://www.ncdc.noaa.gov/data-access/land-based-station-data/land-based-
datasets/automated-surface-observing-system-asos).Five-minutedatawasobtainedforuseinthis
study.TheUSClimateReferenceNetwork(USCRN)isasystematicandsustainednetworkof
climatemonitoringstationswithsitesacrosstheconterminousUS,Alaska,andHawaii.Itincorpo-
rateshigh-qualityinstrumentstomeasuretemperature,precipitation,windspeed,soilconditions,
andmore.TheUSCRNprogramaimsatmaintainingasustainable,high-qualityclimateobserva-
tionnetworktoprovideacontinuousseriesofclimateobservationsformonitoringtrendsinthe
nation'sclimateandsupportingclimate-relatedresearch(https://www.ncdc.noaa.gov/crn/).
Forthepurposeofevaluatinggriddedweatherproducts,sixsinglesitestationsacrossMichigan
werechosenforreferenceobservations:EastLansing,Gaylord,EastLeland,SouthHaven,Grand
11
Figure2.1:
Studycomparisonsitelocations.
Therediconsinthe˝gurerefertothepointsofthe
stationalobservationdatasets,whiletheblueiconsrepresentthecentroidsoftherectanglesin
ERA5-landdatasetswhichcontainthestationalpoints.
Rapids,andTraverseCity(Figure2.1).EastLelandandSouthHavenwerebothchosentore˛ect
theareasofthestudydomainneartheGreatLakes,wherethelakesmaysigni˝cantlymodifylocal
climate(AndresenandWinkler,2009).Itisimportanttonotethatthecomparisonsofgridded
arealaveragestoindividualpointobservationsarenotatidenticalspatialandtemporalscalesbut
stillcloseenoughtodeterminetheirrepresentativenessandrelativeapplicability.
Weevaluatethegriddeddatasetswiththepointobservationsathourly,daily,monthly,annual,
anddecadalscales.Missinghourlytemperature,relativehumidity,andprecipitationobservations
forthesixchosenstationswereomittedfromthecomparisonandaresummarizedintable2.1.
ThefewestoverallmissingobservationswerefoundatEastLansing,EastLeland,andSouthHaven
sites.TheASOSnetworkmissesmorethanhalfoftheprecipitationrecord.Withafocusonfreeze
eventsthataremorerelatedtotemperature,wearenotgoingtoprovidemanyresultsofevaluating
precipitationdata.Also,wenoticethatmorethanaquarteroftherelativehumiditydataisnot
recordedatGaylord.
12
Network
Station
Lat
Lon
Date
Atmpmissing
Relhmissing
Prcpmissing
EW
EastLansing
42.67
°
-84.49
°
1/1/2001to12/31/2018
1.38%
0%
0.13%
EW
EastLeland
45.03
°
-85.67
°
5/1/2003to12/31/2018
0.37%
0.37%
0.96%
EW
SouthHaven
42.36
°
-86.29
°
4/6/2006to12/31/2018
0.94%
0.99%
0.8%
ASOS
GrandRapids
42.881
°
-85.523
°
1/1/2001to12/31/2018
1.97%
2.39%
54.91%
ASOS
TraverseCity
44.742
°
-85.582
°
1/1/2001to12/31/2018
3.06%
3.88%
54.11%
USCRN
Gaylord
44.91
°
-84.72
°
9/19/2007to12/31/2018
0.06%
26.63%
0.69%
Table2.1:
Percentofhourlyobservationsmissingatsixstations.
2.2.4EvaluationMethods
Toquantifythedi˙erencesbetweengriddeddatasetsandpointobservations,severalstatistics
werecomputed.RootMeanSquaredDi˙erence(RMSD),acommonmetricusedtomeasurethe
accuracyofvariables,isthesquarerootoftheaverageofsquareddi˙erencesbetweenestimations
andactualobservations.Inthisstudy,estimationsarethegriddeddatasets(ERA5andPRISM)
andobservationsarethestationaldatasets(EW,ASOS,andUSCRN).Thede˝nitionforRMSDis
asfollows.
H
9
and
b
H
9
areobservationsandestimations,respectively.
'"(ˇ
=
q
1
=
Í
=
9
=
1
¹
H
9

b
H
9
º
2
Andthede˝nitionforBiasis:
0B
=
1
=
Í
=
9
=
1
¹
H
9

b
H
9
º
14
2.3Results
2.3.1Assessmentofweathervariables
We˝rstprovidetimeseriesplotsbycomparingthereanalysisdatasetsandtheobservational
datasets(Figure2.2-Figure2.5).Weselectthreeyears,2010,2012,and2014(randomlyselected;
allthreeyearshavecompletedatafromobservations),andplotthedailydi˙erencesbetweenthe
reanalysisdatasetsandtheobservations.TheconclusionsfromthetimeseriesofDmaxT(daily
maximumtemperature)andDminT(dailyminimumtemperature)forbothERA5andPRISM
revealthatERA5showslargerperturbationsthanPRISMforbothDmaxTandDminT,indicating
thesmallerRMSDs(RootMeanSquaredDi˙erences)ofPRISM.Also,ERA5underestimates
DmaxTfornear-lakestations,especiallyforlatesprings,whichperiodwecareaboutmost.The
timeseriesofDminTforERA5showsthatitdramaticallyoverestimatesDminTforstationsnearthe
lake.Also,theDminTofERA5isconsistentlywarmerthantheobservationsnearthelakeduring
thesummertimeforourselectedthreeyears.Thedi˙erencebetweenERA5andobservationsof
DmaxTshowsoverallsmallerperturbationsthanDminT.
ThetimeseriesofDminTandDmaxTforPRISMalsorevealthatDminThasnoticeablylarger
errorsthanDmaxT.AndthetimeseriesofDmaxTshowsmuchsmootherpatternsforallsixstations
forthreeselectedyears.Thedi˙erencesbetweenPRISMandobservationsforDminTrevealthat
PRISMisconsistentlywarmerthanobservationsforthetwonear-lakestations.Also,sincethe
ASOSnetworkisincorporatedintothePRISMreanalysisdataset,GrandRapidsandTraversecity
shownoticeablysmallerperturbationsthanotherstations.
15
Figure2.2:
Di˙erencebetweenERA5andobservationsfordailymaximumtemperaturein2010,
2012,and2014.
16
Figure2.3:
Di˙erencebetweenERA5andobservationsfordailyminimumtemperaturein2010,
2012,and2014.
17
Figure2.4:
Di˙erencebetweenPRISMandobservationsfordailymaximumtemperaturein2010,
2012,and2014.
18
Figure2.5:
Di˙erencebetweenPRISMandobservationsfordailyminimumtemperaturein2010,
2012,and2014.
19
Figure2.6:
Timeseriesofthedi˙erencebetweenERA5monthlymeandailymeantemperature
andobservedvaluesatsixlocations.
Wealsoevaluatethereanalysisdatasetsonamonthlyscale.Thedi˙erencesbetweenERA5
reanalysisandobservationsofmonthlymeansofdailymeantemperatureshownoticeableannual
cycles(Figure2.6).ERA5reanalysistendstooverestimatedailymeantemperatureduringsummer
seasonsandunderestimateduringwinterseasons,whichisoutofourexpectation.Wethinkthat
reanalysisdatasetsshouldreducethecyclesincetheyarearea-averaged.Andthetwonear-lake
stationsstillshowlargeramplitudes.SincetheASOSnetworkisnotincorporatedintoERA5
reanalysisdatasets,wedon'tobservebetteragreementsbetweenthereanalysisandobservations.
Thedi˙erencesbetweenPRISMreanalysisandobservationsofmonthlymeansofdailymean
temperaturerevealthatthelakee˙ecttendstogetPRISMconsistentlyhigherthanobservations
(Figure2.7).Also,thetwoASOSstationsareshowingoverallsmallerperturbationsaround
zero.Weobserveabnormallylargedi˙erencesbetweenobservationsandbothERA5andPRISM
reanalysisdatasetsinSeptemberof2009andNovemberof2007atSouthHaven.Thereanalysis
datasetsbothgreatlyoverestimatethedailymeantemperatureforthesetwomonths.Wethencheck
theobservationaldata,andtherearenomissingdataforthesetwomonths.Theabnormalvalue
betweengriddeddatasetsandobservationsforEastLansinginDecemberof2009isduetothe
missingdatainthestationalobservations.
20
Figure2.7:
Timeseriesofthedi˙erencebetweenPRISMmonthlymeandailymeantemperature
andobservedvaluesatsixlocations.
Todemonstratetheagreementbetweenobservationsandreanalysisdatasetsinamorestraight-
forwardway,weprovidescatteredplotsatdailyscalesbetweenthemfortheselectedsixstations
(Figure2.8).Inthisscatterplot,ifthepointsareontheredline,itmeansthatgriddeddataagree
withtheobservations.Ifthepointsareabovetheredline,itmeansthatgriddeddataarelarger
thanobservations;ifbelow,smaller.ComparingERA5andPRISM,weobservethattheDmaxT
andDminTofPRISMareinoverallbetteragreementswithobservations.TheshapesofDminT
ofEAR5tellusthattheERA5reanalysisdatasettendstooverestimateDminTduringsummertime
whenDminTishigherandunderestimateDminTduringwintertimewhenDminTislower.The
agreementbetweenERA5andobservationsforDmaxTisbetterthanDminT,andthetwonear-lake
stationsareoverallunderestimated.ForPRISM,DmaxTalsoagreesbetterwithobservationsthan
DminT.ThesameasERA5,PRISMtendstooverestimateDminTofEastLelandandSouthHaven.
MostoftheDminTandDmaxTofPRISMarescatteredaroundtheredlinewithsmallvariations.
ThetwoASOSstations,GrandRapidsandTraverseCityshowgreatshapesoftheagreementsince
theyareincorporatedintothePRISMgroupmodel.
21
Figure2.8:
Scatterplotofdailyminimumandmaximumtemperaturebetweenobservationsand
reanalysisdatasets.
22
Station
Number
RMSD
RMSD
RMSD
Bias
Bias
Bias
Temp
RelHum
Precip
Temp
RelHum
Precip
°
C
%
mm
°
C
%
mm
EastLansing
157770
1.81
11.26
0.43
0.37
-3.75
0.1
EastLeland
137346
3.05
11.85
0.4
0.27
0.33
0.11
SouthHaven
111666
3.18
12.97
0.51
0.3
3.45
0.12
GrandRapids
157751
1.57
9.06
NaN
-0.32
-0.09
NaN
TraverseCity
157751
2.16
9.59
NaN
-0.69
2.76
NaN
Gaylord
98905
1.71
9.59
0.6
0.23
-0.15
0
Median
147548
1.99
10.43
0.51
0.25
0.12
0.1
Table2.2:
Hourlysummarystatistics,rootmeansquaredi˙erence(RMSD),andBiasbetween
ERA5productandobservationalnetworksatsixlocations.
SincetheERA5reanalysisdatasethashourlyproducts,weareabletoexaminetheperformance
oftheERA5onanhourlytimescale(Table2.2).Sincemorethanhalfoftheprecipitationdata
atGrandRapidsandTraverseCityarenotrecordedbythenetworksystem,wedonotassess
precipitationdataatthesetwostations.ERA5productsoveralloverestimatetemperature,relative
humidity,andprecipitationforoursixstations.Also,itisnoteworthythattheRootMeanSquared
Errors(RMSDs)arelargeforrelativehumidity,whichalsore˛ectedtheBiasintemperature.Our
resultsalsorevealthatthetwonear-lakestations,EastLelandandSouthHaven,havenoticeably
largerRMSDfortemperatureandrelativehumidity.SincePRISMonlyhasdailydata,wecould
notevaluateitonanhourlyscale.
WealsoevaluateERA5productsonadailyscale.TocomparewiththePRISMdataset,we
providetheassessmentfordailyminimumtemperature(DminT)anddailymaximumtemperature
(DmaxT)(Table2.3).PRISMde˝nesadayasthe24hoursendingatGreenwichMeanTime
(GMT,or7:00a.m.EasternStandardTime).ThismeansthatPRISMdata,likedailyminimum
temperature,forMay26th,referstothe24hoursendingat7:00a.m.onMay26th.Forconsistency,
weusethismethodtode˝neadayforthedataofERA5aswell.WeobservePRISMhas
overallsmallerRMSDsandBiasforboththeDminTandDmaxTcomparedtoERA5,showinga
betteragreementwithobservations.TheRMSDforDmaxTofERA5couldbeaslargeas4.49
degrees.ERA5overalloverestimatestheDminTandunderestimatestheDmaxT,whilePRISM
23
Station
Days
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
6573
2.49
1.14
1.23
-0.1
EastLeland
5722
4.27
3.15
2.76
-1.82
SouthHaven
4652
4.49
2.98
2.82
-1.83
GrandRapids
6572
1.88
1.33
0.05
-0.6
TraverseCity
6572
2.52
1.98
0.08
-1.23
Gaylord
4120
2.12
1.27
-0.05
0.34
Median
6147
2.51
1.66
0.66
-0.92
Table2.3:
Dailysummarystatistics,RMSDbetweenERA5productandobservationalnetworks,
atsixlocations.
overestimatesboththeDminTandDmaxT.TheunderestimationforDmaxTandoverestimation
forDminTofEAR5arereasonablesincethereanalysisdataarearea-averagedvaluesacrossthe
griddedboxes,whichcouldreducethediurnalcycleofobservations.Thelakee˙ectsarere˛ected
bythedramaticallylargeRMSDandBiasatEastLelandandSouthHavenfromtheEAR5dataset,
whichcouldbeasthreetimeslargeasthoseofotherstations.Thelakee˙ectsalsohavesigni˝cant
butsmallerin˛uencesonthePRISMdataset,whichmightbeduetoPRISMincorporatesmore
oftheobservationaldatasets.Also,itisnoteworthythatPRISMhasalargerRMSDandBiasof
DminTthanthatofDmaxT(Table2.4).ThismightbeduetothatDmaxTisusuallymeasured
atstableconditions,resultinginasmallerbiasinthearea-averagedreanalysisdatasets.Also,we
observedoverallbetterestimatesofPRISMattheASOSnetworkstationsthantheEWnetwork
sincestationsoftheASOSnetworkareincorporatedintothereanalysisdatasets.
24
Station
Days
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
6573
1.49
0.91
0.65
0.5
EastLeland
5722
2.98
0.96
2.13
0.08
SouthHaven
4652
2.84
1.27
1.51
-0.34
GrandRapids
6572
0.73
0.33
-0.01
-0.01
TraverseCity
6572
1.03
0.58
0.29
-0.04
Gaylord
4120
1.62
0.83
-0.74
0.6
Median
6147
1.56
0.87
0.47
0.04
Table2.4:
Dailysummarystatistics,RMSDbetweenPRISMproductandobservationalnetworks,
atsixlocations.
Theunitfortemperatureis
°
C.
25
WealsoprovideseasonalandannualsummarystatisticsforbothERA5andPRISMdatasets.
Wecalculatetheseasonalandannualaveragesforbothdailyminimumandmaximumtemperatures.
Theseasonsarede˝nedwithacontinuouswinterseasonfromDecembertoFebruary.Forexample,
thewinterseasonof2001isde˝nedasDecember2001,January2002,andFebruary2002.The
tableresults(Table2.5-2.8)tellusthatfromdailyscaletoannualscale,valuesofRMSDsare
overallreduced,andvaluesofBiasarenotsigni˝cantlya˙ected.Inaddition,PRISMalwaysshows
overallbetteragreementswithobservationsnomatterwhattimescales.Itisnoteworthythatfrom
dailytoannual,theRMSDsofDminTandDmaxT,especiallyforthetwonear-lakestations,forthe
ERA5areconsiderablydecreasing,thatis,atanannualscale,theRMSDsofERA5arecomparable
tothoseofPRISM.TheseasonalRMSDsandBiasarevariedbyseason.BothPRISMandERA5
tendtogenerallyoverestimateDminTandunderestimateDmaxTduringthespringandsummer
seasons.ERA5alsounderestimatesDmaxTduringthefallandwinterseasons.However,ERA5
overestimatesDminTduringfallandunderestimatesDminTduringwinter.ThemagnitudesofBias
ofERA5overallshowsmallervaluesduringthecoldseason,springandwinter,comparingtolarger
valuesduringthewarmseason,summerandfall.ThischaracteristicisespeciallyobservedforEast
LelandandSouthHaven.Forfallandwinter,PRISMoverestimatesbothDmaxTandDminT.The
magnitudesofBiasofPRISMaresimilarforallseasons.RMSDsofDmaxTareoverallsmaller
than(halfofthemagnitude)thoseofDminTforbothPRISMandERA5.However,theRMSDs
ofDminTaresmallerthanthoseofDmaxT,especiallyforthetwonear-lakestations.Inaddition,
ERA5generallyshowsthebestcomparisonresultsforDminTinthespringseasoncomparedto
otherseasonsconsideringbothRMSDsandBias.Incontrast,PRISMgenerallyshowsthebest
comparisonresultsforDminTinthewinterseasoncomparedtootherseasonsconsideringboth
RMSDsandBias.Still,PRISMshowsmorereliableresultsforbothDmaxTandDminTthroughout
alltheseasons.
26
Station
season
years
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
Spring
18
0.92
0.30
0.85
-0.11
EastLeland
Spring
16
1.32
3.97
1.18
-3.92
SouthHaven
Spring
13
1.49
3.71
1.34
-3.67
GrandRapids
Spring
18
0.44
0.63
-0.20
-0.51
TraverseCity
Spring
18
0.78
1.36
-0.55
-1.33
Gaylord
Spring
11
0.78
0.36
-0.67
0.20
Median
Spring
17
0.85
1.00
0.33
-0.92
EastLansing
Summer
18
2.39
0.35
2.36
-0.23
EastLeland
Summer
16
4.46
2.57
4.39
-2.51
SouthHaven
Summer
13
4.51
1.52
4.46
-1.40
GrandRapids
Summer
18
0.63
0.90
0.54
-0.83
TraverseCity
Summer
18
1.14
1.23
0.96
-1.15
Gaylord
Summer
11
1.12
0.54
1.10
0.47
Median
Summer
17
1.77
1.07
1.73
-0.99
EastLansing
Fall
18
2.05
0.76
1.95
-0.34
EastLeland
Fall
16
3.99
0.82
3.92
-0.69
SouthHaven
Fall
13
4.08
1.31
3.97
-1.26
GrandRapids
Fall
18
1.16
0.67
1.07
-0.58
TraverseCity
Fall
18
1.45
1.03
1.29
-0.94
Gaylord
Fall
12
0.97
0.52
0.91
0.47
Median
Fall
17
1.75
0.79
1.62
-0.64
EastLansing
Winter
19
0.76
0.31
-0.36
0.19
EastLeland
Winter
16
1.58
0.60
1.38
-0.25
SouthHaven
Winter
13
1.54
1.04
1.46
-0.98
GrandRapids
Winter
19
1.36
0.54
-1.19
-0.44
TraverseCity
Winter
19
1.62
1.57
-1.42
-1.53
Gaylord
Winter
12
1.71
0.33
-1.56
0.21
Median
Winter
17.5
1.56
0.57
-0.78
-0.35
Table2.5:
Seasonalsummarystatistics,RMSDbetweenERA5productandobservational
networks,atsixlocations.
Theunitfortemperatureis
°
C.
27
Station
season
years
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
Spring
18
0.89
0.62
0.82
0.50
EastLeland
Spring
16
2.34
0.23
2.32
-0.10
SouthHaven
Spring
13
1.45
0.49
1.41
-0.40
GrandRapids
Spring
18
0.20
0.16
0.14
-0.04
TraverseCity
Spring
18
0.38
0.21
0.31
-0.01
Gaylord
Spring
11
1.06
0.66
-1.02
0.63
Median
Spring
17
0.98
0.36
0.57
-0.03
EastLansing
Summer
18
1.03
0.61
0.94
0.43
EastLeland
Summer
16
2.76
0.36
2.74
-0.31
SouthHaven
Summer
13
2.30
1.04
2.25
-0.90
GrandRapids
Summer
18
0.11
0.15
-0.08
-0.02
TraverseCity
Summer
18
0.46
0.40
0.31
-0.07
Gaylord
Summer
11
0.73
0.93
-0.70
0.92
Median
Summer
17
0.88
0.51
0.63
-0.05
EastLansing
Fall
18
0.78
0.84
0.40
0.32
EastLeland
Fall
16
1.99
0.34
1.92
0.06
SouthHaven
Fall
13
1.81
0.36
1.61
-0.22
GrandRapids
Fall
18
0.09
0.14
0.07
0.01
TraverseCity
Fall
18
0.47
0.26
0.36
-0.10
Gaylord
Fall
12
0.56
0.51
-0.53
0.47
Median
Fall
17
0.67
0.35
0.38
0.04
EastLansing
Winter
19
0.54
0.72
0.36
0.65
EastLeland
Winter
16
1.45
0.60
1.39
0.58
SouthHaven
Winter
13
0.83
0.41
0.71
0.19
GrandRapids
Winter
19
0.22
0.11
-0.18
0.02
TraverseCity
Winter
19
0.31
0.27
0.17
0.02
Gaylord
Winter
12
0.83
0.42
-0.71
0.37
Median
Winter
17.5
0.69
0.42
0.27
0.28
Table2.6:
Seasonalsummarystatistics,RMSDbetweenPRISMproductandobservational
networks,atsixlocations.
Theunitfortemperatureis
°
C.
28
Station
years
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
18
1.23
0.57
1.11
-0.24
EastLeland
16
2.77
1.87
2.75
-1.86
SouthHaven
13
2.85
1.87
2.81
-1.85
GrandRapids
18
0.35
0.66
0.05
-0.6
TraverseCity
18
0.35
1.25
0.08
-1.23
Gaylord
12
0.29
0.41
0
0.36
Median
17
0.79
0.96
0.6
-0.92
Table2.7:
Annualsummarystatistics,RMSDbetweenERA5productandobservationalnetworks,
atsixlocations.
Theunitfortemperatureis
°
C.
Station
years
RMSD
RMSD
Bias
Bias
DminT
DmaxT
DminT
DmaxT
°
C
°
C
°
C
°
C
EastLansing
18
0.71
0.71
0.55
0.36
EastLeland
16
2.11
0.22
2.1
0.03
SouthHaven
13
1.52
0.46
1.49
-0.38
GrandRapids
18
0.07
0.12
-0.01
-0.01
TraverseCity
18
0.38
0.26
0.29
-0.04
Gaylord
12
0.75
0.61
-0.73
0.59
Median
17
0.73
0.36
0.42
0.01
Table2.8:
Annualsummarystatistics,RMSDbetweenPRISMproductandobservational
networks,atsixlocations.
Theunitfortemperatureis
°
C.
29
Station
Days
RMSD
RMSD
Bias
Bias
ERA5
PRISM
ERA5
PRISM
EastLansing
6573
0.99
0.77
0.49
0.38
EastLeland
5722
1.79
1.07
0.2
0.59
SouthHaven
4652
1.96
1
0.27
0.36
GrandRapids
6572
0.71
0.34
-0.07
0
TraverseCity
6572
0.99
0.47
-0.13
0.05
Gaylord
4120
0.8
0.49
0.3
0.02
Median
6147
0.99
0.63
0.24
0.21
Table2.9:
Growingdegreeday(Base4
°
C)summarystatistics.
2.3.2Assessmentofderivedvariables
Freezeeventscouldresultindi˙erentiatedrisksoncropswithdi˙erentphenologicalstages.
GrowingDegreeDay(GDD)isacommonlyusedagro-climaticindextosimulateplantphenologies.
WecalculatetheGDDwithabasetemperatureof4
°
C(Table2.9).Theresultsshowthatcalculations
basedonDmaxTandDminTfromPRISMhavearelativelysmallerRMSDandsmallerBiasthan
thosefromERA5.BothERA5andPRISMoverestimateGDDvalues.ERA5stillshowslarger
RMSDsatthesetwonear-lakestationsandthesamecaseforPRISM.
Sincewecaremostabouthowthereanalysisdatasetsrepresentfreezeoccurrences,weneedto
assesshowtheyarecapturedormissed.Forprecipitationvariableevaluation,someforecastscores
areutilizedtomeasuretheabilityofaproduct.Hereweusedi˙erentcriteriatode˝nefreeze
eventsandthencalculatehowreanalysisproductscapturetheseevents.Tables2.10and2.11show
anexampleofhowweexaminethis.We˝rstde˝ne˝velevelsoffreezeevents,rangingfrom0
°
C
to-25
°
Cwith5-degreesteps.Andthen,wecounttheseeventsforeachmonth(JulyandAugust
arenotshowingheresincenode˝nedfreezeeventsoccurred).Here107/131=0.82meansthat
throughoutthedataperiod,from1/1/2001to12/31/2018,thereare6573days,and131daysof
whichinJanuaryhaveDminTbetween-5degreesand0degrees.Andofthese131events,ERA5
captured107eventswitharatioof0.82.PRISMshowsbetterresultsbycapturing112eventswith
aratioof0.85.BycomparingERA5andPRISM,weobservethatPRISMoverallbehavesverywell
incapturingdi˙erenttypesofeventsthroughoutdi˙erentmonths.Theharderthefreezeevents,the
30
month
Type1
Type2
Type3
Type4
Type5
-5to0
-10to-5
-15to-10
-20to-15
-25to-20
°
C
°
C
°
C
°
C
°
C
1
107/131=0.82
91/141=0.65
67/101=0.66
49/84=0.58
17/28=0.61
2
88/122=0.72
80/141=0.57
66/108=0.61
43/70=0.61
11/23=0.48
3
151/173=0.87
93/170=0.55
40/68=0.59
10/17=0.59
5/6=0.83
4
120/193=0.62
7/32=0.22
0/0=nan
0/0=nan
0/0=nan
5
3/20=0.15
0/0=nan
0/0=nan
0/0=nan
0/0=nan
6
0/0=nan
0/0=nan
0/0=nan
0/0=nan
0/0=nan
9
0/0=nan
0/0=nan
0/0=nan
0/0=nan
0/0=nan
10
26/114=0.23
0/0=nan
0/0=nan
0/0=nan
0/0=nan
11
122/188=0.65
36/95=0.38
3/7=0.43
2/2=1.00
0/0=nan
12
171/192=0.89
98/159=0.62
26/59=0.44
13/21=0.62
3/4=0.75
Table2.10:
FreezeeventscapturedratiosatEastLansingofERA5.
Freezeeventsarecategorized
into˝vetypes,whichare,respectively,from-5
°
Cto0
°
C,from-10
°
Cto-5
°
C,from-15
°
Cto
-10
°
C,from-20
°
Cto-15
°
C,andfrom-25
°
Cto-20
°
C.
overalllowerthecapturingratioforEAR5,whichisnotthecaseforPRISM.PRISMstillshowshigh
capturingratiosforhardfreezeevents.Also,itisnoteworthythatbothERA5andPRISMshow
betterresultsforwintertimefreezeeventsthanspringtimefreezeevents.Forspringtimefreeze
events,wede˝nitelyseetheadvantagesofPRISM.Eventhoughnotasgoodaswintertimeratios,
thespringtimefreezeevents,especiallyforthelatespring,arecapturedprettywellbyPRISM.In
May,20daysarereportedtohavefreezeeventsthroughout18years.AndPRISMcaptures13of
20,whichismuchbetterthanERA5,whichonlycaptures3of20.Therefore,generally,PRISM
demonstratesanadvantageousabilitytocapturefreezeeventsfordi˙erenttypesandfordi˙erent
occurringtimes.Thisbene˝tofPRISMleadstoourlaterchoiceofusingPRISMasaproxyfor
observationstorepresentthesurfacefreezeeventsasanexcellentgriddeddataset.
Fortheother˝vestations,wedon'tpresentthedetailedtablesheresincetheresultsand
conclusionsaresimilar.PRISMshowsmuchbetterresultsthanERA5forallcategoriesoffreeze
events.Also,forthetwonear-lakestations,evenPRISMmissesalotofeventsforhardfreezes
andlatespringfreezes.AndthetwostationsfromtheASOSnetworkshowexcellentratiosof
PRISM,asweexpect.Weprovideasummarytablewithoutdetailedcategoriesforsixstations.In
tables2.12and2.13,wecountallthefreezeevents,de˝nedasDminTlessthan0degrees.For
31
month
Type1
Type2
Type3
Type4
Type5
-5to0
-10to-5
-15to-10
-20to-15
-25to-20
°
C
°
C
°
C
°
C
°
C
1
112/131=0.85
119/141=0.84
77/101=0.76
73/84=0.87
22/28=0.79
2
114/122=0.93
123/141=0.87
91/108=0.84
57/70=0.81
13/23=0.57
3
147/173=0.85
129/170=0.76
48/68=0.71
10/17=0.59
5/6=0.83
4
128/193=0.66
20/32=0.62
0/0=nan
0/0=nan
0/0=nan
5
13/20=0.65
0/0=nan
0/0=nan
0/0=nan
0/0=nan
6
0/0=nan
0/0=nan
0/0=nan
0/0=nan
0/0=nan
9
0/0=nan
0/0=nan
0/0=nan
0/0=nan
0/0=nan
10
86/114=0.75
0/0=nan
0/0=nan
0/0=nan
0/0=nan
11
159/188=0.85
62/95=0.65
6/7=0.86
0/2=0.00
0/0=nan
12
170/192=0.89
136/159=0.86
53/59=0.90
15/21=0.71
4/4=1.00
Table2.11:
FreezeeventscapturedratiosatEastLansingofPRISM.
Freezeeventsare
categorizedinto˝vetypes,whichare,respectively,from-5
°
Cto0
°
C,from-10
°
Cto-5
°
C,from
-15
°
Cto-10
°
C,from-20
°
Cto-15
°
C,andfrom-25
°
Cto-20
°
C.
example,atEastLansing,from1/1/2001to12/31/2018with6573days,493daysinJanuaryhasa
DminTnohigherthan0degrees.The493daysaremorethanthesumof˝vede˝nedfreezeevent
daysbecausetherearedayslessthan-25degreesthatarenotcountedabove.Comparingratios
throughoutdi˙erentmonths,weconcludethatbothERA5andPRISMshowbetterfreezeevent
capturingbehaviorduringwintertimefreezesthanspringtimeandfallfreezes.ERA5onlycaptures
anoverall17percentofthetotalfreezeoccurrencesinMayand13percentofthetotalfreezesin
October.PRISMbehavesbetterwithanoverall63percentinMayand65percentinOctober.For
themonthsofJanuary,February,March,andDecember,bothERA5andPRISMshowexcellent
ratiosfrom0.97to0.99.Comparingratiosbetweendi˙erentstations,thetwonear-lakestations,
EastLelandandSouthHaven,showthelowestratiosforbothERA5andPRISM.AndGaylord,
whichisintheUSCRNnetwork,showsthebestratioforbothEAR5andPRISM.Theseveral
freezeoccurrencesinJuneandSeptemberarenotcapturedbyeitherERA5orPRISM.Therefore,
wehopetounderstandhowthespringtimefreezeeventsaredeterminedinthelaterchapters.
32
month
EastLansing
EastLeland
SouthHaven
GrandRapids
TraverseCity
Gaylord
Median
1
490/493=0.99
416/435=0.96
313/330=0.95
502/508=0.99
516/519=0.99
333/334=1.00
0.99
2
471/475=0.99
406/410=0.99
295/301=0.98
472/474=1.00
483/486=0.99
305/305=1.00
0.99
3
427/435=0.98
388/396=0.98
244/269=0.91
427/433=0.99
453/459=0.99
296/296=1.00
0.98
4
157/225=0.70
174/276=0.63
53/127=0.42
143/164=0.87
230/292=0.79
186/197=0.94
0.74
5
3/20=0.15
0/71=0.00
0/15=0.00
3/12=0.25
8/41=0.20
13/32=0.41
0.17
6
0/0=nan
0/6=0.00
0/0=nan
0/0=nan
0/1=0.00
0/1=0.00
0
9
0/0=nan
0/4=0.00
0/4=0.00
0/0=nan
0/3=0.00
0/1=0.00
0
10
26/114=0.23
0/83=0.00
0/34=0.00
12/70=0.17
9/96=0.09
48/98=0.49
0.13
11
230/292=0.79
104/253=0.41
50/169=0.30
207/265=0.78
198/281=0.70
229/252=0.91
0.74
12
429/437=0.98
370/436=0.85
239/304=0.79
449/461=0.97
458/472=0.97
347/351=0.99
0.97
Table2.12:
Freezeeventsbelow0
°
CcapturedratiosatsixstationswithoutdetailedcategoriesofERA5.
month
EastLansing
EastLeland
SouthHaven
GrandRapids
TraverseCity
Gaylord
Median
1
484/493=0.98
426/435=0.98
320/330=0.97
503/508=0.99
518/519=1.00
333/334=1.00
0.99
2
472/475=0.99
403/410=0.98
291/301=0.97
474/474=1.00
484/486=1.00
305/305=1.00
0.99
3
414/435=0.95
362/396=0.91
241/269=0.90
424/433=0.98
451/459=0.98
295/296=1.00
0.97
4
162/225=0.72
156/276=0.57
65/127=0.51
144/164=0.88
242/292=0.83
188/197=0.95
0.77
5
13/20=0.65
6/71=0.08
1/15=0.07
10/12=0.83
25/41=0.61
31/32=0.97
0.63
6
0/0=nan
0/6=0.00
0/0=nan
0/0=nan
0/1=0.00
0/1=0.00
0
9
0/0=nan
0/4=0.00
0/4=0.00
0/0=nan
0/3=0.00
1/1=1.00
0
10
86/114=0.75
13/83=0.16
8/34=0.24
57/70=0.81
52/96=0.54
91/98=0.93
0.65
11
266/292=0.91
174/253=0.69
108/169=0.64
255/265=0.96
259/281=0.92
246/252=0.98
0.92
12
423/437=0.97
416/436=0.95
275/304=0.90
456/461=0.99
466/472=0.99
348/351=0.99
0.98
Table2.13:
Freezeeventsbelow0
°
CcapturedratiosatsixstationswithoutdetailedcategoriesofPRISM.
2.4Summary
ThisstudycomparesERA5andPRISMreanalysisproductstostationalobservations.We
concludethatthePRISMdataset,withahigherhorizontalresolutionof4km,overalldemonstrates
abetteragreementoftemperatureswithobservations.However,ERA5outperformsPRISMinits
availabilityofvariousweathervariables.Themajor˝ndingsareasfollows.
1)BothreanalysisdatasetsshowbetterresultsforDmaxTthanDminT,asshowninthelower
DmaxTperturbationsofthetimeseriesandthesmallervaluesofDmaxTRMSDs.
2)Griddedreanalysisdatasetstendtoreducetheobserveddailycycles.ERA5underestimates
DmaxTwhileoverestimatesDminT,resultinginreduceddailycycles.Incontrast,PRISMtends
tooverestimatebothDminTandDmaxT.However,onamonthlyscale,ERA5reanalysistendsto
overestimatedailymeantemperatureduringsummerseasonsandunderestimateitduringwinter
seasonsinthetimeseries,whichisconsistentwiththescatterplotresults.
3)TheASOSnetworkstations,whichareincorporatedintothePRISManalysis,showexcellent
comparingresultsforbothdirectweathervariablesandderivedvariables.Thisbehaviorisnot
observedintheevaluationoftheERA5dataset.
4)Thetwonear-lakestationsshowlargeRMSDsandBiasandalsolargeperturbationsoftime
seriesforbothreanalysisproducts.ButPRISMstillshowsabetteragreement.
5)PRISMalsoshowsabetterbehaviorincapturingthederivedvariables.Withabase
temperatureof4
°
C,bothERA5andPRISMoverestimateGDDvalues.PRISMshowsexcellent
abilityincapturingdi˙erenttypesoffreezeeventsthroughoutdi˙erentmonths.BothPRISMand
ERA5showbetterfreezeeventcaptureduringwintertimethanotherseasons.
Inconclusion,wewillusePRISMasaproxyofobservationstorepresentfreezeeventsinlater
chapters.Sincethewholethesisisfocusedonfreezeevents,wejustevaluatetemperaturesand
temperature-derivedvariables.However,toassessbothERA5andPRISMdatasets,morework
needstobedoneinthefuture.Forexample,withtheavailabilityandaccessibilityofprecipitation
dataintheERA5andPRISM,wecouldevaluatehowprecipitationeventsarecapturedbycomputing
analyticalscoreslikethebiasscore(BS),thethreatscore(TS),andtheHeidkeskillscore(HSS).In
34
addition,sincewindconditionsalsohaveaprofoundimpactonfreezeformation,itisanecessity
toassessthewindspeedanddirectionoftheERA5(PRISMdoesnotprovidewinddata).Also,
ERA5providesland-onlydatasetswithabetterresolutionof9kmthanthesingle-leveldatasetswith
30km.Itissigni˝canttoevaluatehowthisresolutionimprovementa˙ectstheoverallagreement
withobservations.
35
CHAPTER3
FREQUENCYANDSEVERITYOFSPRINGTIMEFREEZEEVENTS
3.1Background
3.1.1FalseSpringsandSpringtimeFreezeEvents
Asaconsequenceoftheglobalwarmingtrend,earlierspringsinduceprematureplantdevelop-
ment,resultinginvulnerableandsusceptiblecropsexposedtosubsequentspringtimefreezeevents
(Guetal.,2008;Allstadtetal.,2015).Intheearlyphenologicalstages,temperaturesnear-6
°
C
begintocausedamagetothecrops,butasthebloomstageapproaches,temperaturesaswarmas
-2
°
Ccouldcauseconsiderabledamage(Longstroth,2005).Springtimefreezeevents,especially
thoseafteranuntimelyextendedperiodofwarmertemperaturesknownas"falsespring",couldhave
devastatingconsequencestoagricultureproduction.Prominentexamplesincludefreezeeventsthat
occurredinlatespringof2002,2007,and2012followinganextendedperiodofwarmweather
andwipedoutmorethan90%offruitproductionandothercropsacrosstheUSMidwest,which
resultedinanagriculturallossofmorethan2billiondollars(NASS,2002;Marinoetal.,2011;
O'Brienetal.,2019;Labeetal.,2015;Kistneretal.,2018;Guetal.,2008).Theseeventsshare
twocharacteristics:theyoccurredfollowinganextendedperiodofunusuallywarmconditions,and
theycauseddisastrousconsequencestoprematuredevelopingvegetationandcrops.
Earlierspringscouldresultintheearlierdevelopmentofcropsandmightincreasethevulner-
abilityofcropstofreezeeventdamage(Winkleretal.,2013).Allstadtetal.(2015)projected
thatthewidespreadhistoricaladvancesinspringplantphenologywouldextendintothefuture,
withanincreasedriskoffalsespringsintheMidwesternUS.Besides,KatzandBrown(1992)
alsoemphasizedthatchangesinvariabilityweremoreimportanttoextremeeventfrequenciesthan
changesinaverageswithstatisticalmodels.
Findingsfrompreviousresearchalsorevealedthatfalsespringsandsubsequentfreezeevents
36
couldberelatedtolarge-scaleatmosphericcirculationpatterns.Whenarelativelow-pressure
systememergesovertheNorthPaci˝cOcean,creatingaripplee˙ectdownstream,warmair
massintrudes,andthenearlyspringoccursacrossmuchofwesternNorthAmerica(Aultetal.,
2013).Kunkeletal.(2004)speculatedthattheshortergrowingseasonsintheearly1900swere
characterizedbydrierairmassesandmoreglobalsolarradiationwhilethelongergrowingseasons
inrecentyearsmaybeassociatedwithairmassesthataremoisterincomparisonandresultinless
globalsolarradiation.
3.1.2SpatiotemporalVariabilityofSpringtimeFreezeEvents
Manypreviousstudieshaveconsideredthespatiotemporalvariabilityandtrendofthespring-
timefreezeevents.DeGaetano(1996)detectedadecreasingtrendoffreezingdaysintheNorth-
easternUnitedStatesfrom1959to1993,withdi˙erentthresholdsatanindividualstationlevel.
Easterlingetal.(2002)suggestedthatthespringtimefreezingdayssigni˝cantlydecreasedforthe
period1948-1999inwesternandnorth-centralUS.Theyalsonotedthattheregionaltrendsinfrost
daysaresubstantiallyrelatedtothemeanannualtemperaturechanges.Yuetal.(2014)revealed
earlieroccurrencesoftheLastSpringFrost,delayedeventsoftheFirstFallFrost,andlengthening
periodsoftheFrost-FreeSeasonfrom1980to2010.Theirstudyalsoshowedweakerinterannual
variabilityofthesedatesintheGreatLakesregionthanotherregionsintheMidwest(Yuetal.,
2014).
Mostofthesefreezeeventstudieswerecarriedoutattheindividualstationlevel.Nevertheless,
thegridded4-kmdatasetreleasedbythePRISM(Parameter-elevationRegressionsonIndependent
SlopesModel)climategroupfromOregonStateUniversity(PRISMClimateGroup,Oregon
StateUniversity,http://prism.oregonstate.edu,createdFeb2020)allowscontinuousspatiotemporal
analysisofspringtimefreezeeventsacrosstheMidwesternUS.Besides,littleresearchhasfocused
onfreezeeventsintheindividualspringtimemonths.
37
3.1.3StudyObjectives
Withthenecessityofunderstandingtheclimatologyofspringtimefreezeevents,wetryto
investigatethefrequency,severity,andpotentialcausesofspringtimefreezeeventoccurrences.
ThisstudyaimstouseERA5reanalysis(the˝fthmajorglobalreanalysisproducedbytheEuropean
CentreforMedium-RangeWeatherForecasts)tounderstandtheweatherconditionsandclimate
backgroundofthefreezeeventsrepresentedbythePRISManalysisdataset.
38
3.2DataandMethod
ThestudydomainisthecentralandeasternUnitedStatesencompassingtheregionfrom
105
º
Wto75
º
Wlongitudeandfrom35
º
Nto50
º
Nlatitude.Thisregioncoversoneofthe
mostintensiveandextensiveareasofagriculturalproductionintheworld.Sincetheagricultural
productioninthisregionissubstantiallysensitivetoclimatevariability,especiallyfreezeevents,
itisworthwhiletostudythefrequencyandseverityoffreezeeventsrepresentedbythegridded
datasetsoverthisregion.
PRISM,agriddedanalysisdatasetprovidedbythePRISMclimategroupfromOregonState
University,hasabetteragreementwiththeobservationalfreezingeventsandprovidesestimatesof
sixbasicclimateelements:precipitation(ppt),minimumtemperature(tmin),maximumtempera-
ture(tmax),meandewpoint(tdmean),minimumvaporpressurede˝cit(vpdmin),andmaximum
vaporpressurede˝cit(vpdmax).PRISMhasanexcellentspatialresolutionof4km.Butitjust
providesclimatevariablesatdailyscaleswithtimecoveragefrom1981untilthepresent.We
obtaindailyminimumandmeanairtemperaturefromPRISMdatasetstorepresentthefrequency
andseverityoffreezeeventsonthesurface.ERA5isaclimatereanalysisdatasetreleasedbythe
EuropeanCentreforMedium-RangeWeatherForecasts(ECMWF),coveringtheperiodfrom1950
tothepresent.ThenameERArefersto'ECMWFReAnalysis',withERA5beingthe˝fthmajor
globalreanalysisproducedbyECMWF(afterFGGE,ERA-15,ERA-40,ERA-Interim)(Henner-
mannetal.,2020).TheglobalERA5reanalysisdatasetincludesgriddeddatasetswithahorizontal
resolutionof30kmat37pressurelevels.WeuseERA5datasetsforunderstandingtheweather
conditionsandclimatebackgroundintheupperlevelsthatoccursimultaneouslywiththefreeze
eventsinthesurface.
Wede˝nethefreezedaysasdaysbydi˙erentcriteriawhenthedailyminimumairtemperature
isbelowdi˙erentdegreesCelsius,includingthecommonlyusedone,0
º
C.Thefreezeeventsare
identi˝edusingthePRISMdatasetthatiscomprisedof357
Ö
722horizontalgridpointswitha4
kmgridspacingacrossthestudyregion.
WeobtaintheteleconnectionindicesfromtheClimatePredictionCenter(CPC).Niño-3.4isan
39
indexindicatingtheSSTanomaliesacrosstheregionspans5
°
N-5
°
S,120
°
-170
°
Winthetropical
Paci˝cOcean.BoththeNAO(NorthAtlanticOscillation)andPNA(Paci˝cNorthAmerican)
indicesarebasedontheleadingrotatedprincipalcomponentanalysesofmean500-mbheightsin
theNorthernHemisphere(20
°

°
Nlatitude)(Dixonetal.,2007).
WealsousetheEmpiricalOrthogonalFunction(EOF)techniquetoidentifythespatialpatterns
ofthefreezingdaysindi˙erentmonths.TheEOFanalysisproducesasetofmutuallyorthogonal
modesthatconsistofspatialstructures(EOFs)withcorrespondingtimeseries(principalcompo-
nentsorPCs)(Yuetal.,2014).Thecorrespondingeigenvalueofeachmodedescribesthevariance
explainedbythemode(Yuetal.,2014).Inthisstudy,weanalysethe˝rsttwomodesthattogether
explainmorethan60%ofthevariance.
40
3.3Results
3.3.1Temporalandspatialvariabilityoffreezeevents
We˝rstexaminethespatialvariabilityoffreezeeventsacrossthecentralandMidwestern
UnitedStates.Themean,standarddeviation,andtrendoffreezingdaysinMarch,April,and
Mayacrossourstudyregionfrom1981to2019areshowninFigure3.1.Theaveragenumbersof
freezingdaysforthethreemonthsallshowalatitude-dependentpattern,andalsodecreasesasthe
monthproceeds.InMay,someareasaremovingtowardthefrost-freeseason,resultinginnear-zero
freezeeventsthere.Thestandarddeviationhereisdividedbytheaveragenumbersoffreezingdays
ateachgridpointtorepresenttherelativevariationsofthefreezingdays.Ourresultsrevealthat
freezingdaysinMarchhavetheleastvariationacrosstheregionthroughoutthestudyperiod,and
thoseinAprilshowarelativelyhighervariationinthesouthernregion.FreezingdaysinMayshow
ahighvariationinthecentral-southernregion,whichisdominatedbythesmallmeanvaluesinthese
areas.Wealsoconducttrendanalysesbylinearleast-squaresregression.Theresultsrevealthat
freezingdaysinMarchshowadecreasingtrendacrossthecentralandMidwesternUnitedStates,
whichisconsistentwiththedecreasingtrendinspringtimefreezeeventsinthepreviousstudies
(DeGaetano,1996;Easterlingetal.,2002).However,nearlyhalfoftheregionshowsanupward
trendinApril,whichisespeciallysigni˝cantinNorthernMichigan,partofwesternWisconsin,
andcentralOklahoma.FreezingdaysinMayshowasigni˝cantdownwardtrendinmostofthe
Northeast,incontrasttoanupwardtrendinthewesternregion.Wetrytounderstandwhythese
di˙erentiatedresultsareobservedinAprilandMayinthefollowingpart.
Figure3.2showsthetimeseriesofSpatiallyAveragedFreezingDays(de˝nedasthedays
withdailyminimumtemperaturesbelow0
º
C)Counts(SAFDC)andSpatiallyAveragedDaily
MinimumTemperature(SADMT)inspring(March,April,May)from1981to2019.Wecalculate
theTheil-Senslope,whichisthemedianofallthepairedvalues,toavoidhavingoutliers'e˙ects
onthetrendofthetimeseries.Anegativeslopeof-0.043forSAFDCisobserved,indicatinga
generaldecreasingtrendofspringtimefreezeeventsacrossthecentralandMidwesternUS,which
41
Figure3.1:
Average,standarddeviation,andtrendoffreezingdaysinspring(March,April,May)
from1981to2019.
agreeswithpreviousstudies(DeGaetano,1996;Easterlingetal.,2002).Theyear1996,2013and
2018allshowabruptlyhighvaluesofSAFDC,whiletheyear2012showsthelowestSAFDC.
TheresultsofSADMTwereexactlyonthecontrary,asexpected.Itisnoteworthythat,although
boththeyears2007and2012showalowfrequencyoffreezeevents,considerablecropdamage
wasobservedinthesetwoyears,whichwerelikelyaconsequenceoffalsespringsexposingthe
earlierdevelopedcropstothesubsequentfreezeevents.Asweexpect,theSADMTshowsan
exactoppositephasetotheSAFDCwithapositiveTheil-Senslopeof0.005.Wealsorelate
theSAFDCtothespatiallyaverageddailyminimumandmeantemperature.Theresultreveals
bothsigni˝cantnegativerelationshipswithR-squaredvaluesof0.7894and0.7414,indicatinga
deterministicrelationshipbetweenthemonthlyaverageddailyminimum/meantemperatureand
freezingdaysinthatcorrespondentmonth.Thesigni˝canttestsfortheserelationshipsshowthat
42
Figure3.2:
Timeseriesofspringtimefreezeeventscountfrom1981to2019.
The˝ttedlinewas
˝ttedwithouttheyear2018.
allthesetrendsarenotsigni˝cant.
3.3.2FreezingAreaPercentage
Tofurtherunderstandandexplainvariationsinsomeregionsofinterest,weintroduceavariable
calledFreezingAreaPercentage(FAP),whichisde˝nedasthepercentageofgridpointswhose
dailyminimumtemperatureisbelow0
º
Coneachdayinapre-de˝nedsubsetofthestudyregion.
Hereweusetheregionsde˝nedinKarlandKnight(1998)andalsousedasclimateregions
byNOAA(NationalOceanicandAtmosphericAdministration)asthesubsets(Figure3.3).We
calculatetheFAPineachsubregionrespectivelyoneachdayandsumupthedailyvaluesforthree
springtimemonths(Table3.1).ThelargertheFAP,themoreareasofthatregionarelikelyto
43
Figure3.3:
Subsetregionsde˝nedinKarlandKnight(1998).
su˙erfromfreezingrisks.ThemonthlyFAPvaluescouldbeexplainedasthenumberoffreezing
daysoccurringinthede˝nedregionatsomespringtimemonth.HighervaluesofmonthlyFAPare
observedinthenorthernsubregionsandtheearliermonths.
Figure3.4showsthetimeseriesofmonthlyaggregatedFAPvalues.Largervaluesareshownin
thenorthernregions,theNorthernGreatPlains,theUpperMidwest,andtheNortheast.Throughout
thewholestudyregion,anoticeablylowFAPvalueisintheyear2012issandwichedbetween
twolargevalues.Weobservedsomeinterannualandinterdecadalperiodicityofthetimeseriesof
monthlyFAPvaluesbyspectralanalysis(notshownhere).Sincethecausesofspringtimefreezes
couldberelatedtolargecirculations,welaterinvestigatetherelationshipbetweenFAPvaluesand
someteleconnectionindices.
WeconducttrendanalysesformonthlyFAPvalues(Table3.2).Theresultsrevealdecreasing
trendsofFAPinalmostallregionsforthethreemonths,exceptforincreasingtrendsintheNorthern
GreatPlainsandalsotheUpperMidwestinMarchandintheNortheastandSoutheastinMay.
Thetrendsfortheaccumulatedspringtimearedownwardwithasigni˝canceintheOhioValley,
whichextendsthe˝ndingsinEasterlingetal.(2002)fromtheperiod1948-1999into1981-2019.
44
Figure3.4:
TimeseriesofmonthlyFAPvalues.
Easterlingetal.(2002)revealedanon-signi˝canttrendof-0.1daysperdecadeintheOhioValley
forspringtimefrom1948to1999,whileourresultsshowasigni˝canttrend(p-valuelessthan0.1)
of-0.113daysperyearforspringtimefrom1981to2019,furtheringindicatingfewerfreezingdays
occurringintheOhioValleyduringrecentdecades.
45
NorthernGreatPlains
SouthernGreatPlains
UpperMidwest
OhioValley
Northeast
Southeast
March
22.641
12.173
26.135
16.387
26.724
14.154
April
10.06
2.699
14.416
4.6
14.884
3.92
May
1.312
0.135
2.58
0.163
2.639
0.175
Springtime
34.013
15.007
43.131
21.15
44.248
18.248
Table3.1:
39-yearmeanofregionalaveragedmonthlyFAP.
Relationshipswithpvalueslessthan0.10arecharacterizedwithonestar
andthoselessthan0.05arewithtwostars.
NorthernGreatPlains
SouthernGreatPlains
UpperMidwest
OhioValley
Northeast
Southeast
March
0.027
-0.001
0.004
-0.031
-0.006
-0.004
April
-0.012
-0.037*
-0.03
-0.06
-0.041
-0.013
May
-0.031*
-0.002
-0.016
-0.001
0.005
0
Springtime
-0.065
-0.065
-0.016
-0.113*
-0.019
-0.003
Table3.2:
Theil-SenSlopeformonthlyFAP.
Relationshipswithpvalueslessthan0.10arecharacterizedwithonestarandthoseless
than0.05arewithtwostars.
TofurtherinvestigatetheperiodicityofmonthlyFAPvalues,werelatethemtosometeleconnec-
tionindices,Niño-3.4,NAO,PNA,andPDO.ThepositivephaseoftheNAOre˛ectsbelow-normal
heightsandpressureacrossthehighlatitudesoftheNorthAtlanticandabove-normalheightsand
pressureovertheeasternUnitedStates.Thepatternisreversedforthenegativephase.Thestrong
positivephaseofNAOtendstobeassociatedwithabove-normaltemperaturesintheeasternUnited
States(CPC,2012).Table3.3showsPearsoncorrelationvaluescomparingmonthlyFAPvaluesfor
thespringtimemonthstoNAOvalueswith0-3monthsearlier.Itisnoteworthythatweobservea
signi˝cantlystrongnegativecorrelationbetweenFAPvaluesandNAOinMarchfrom1981to2019.
ThisresultleadstotheconclusionthatthepositivephaseofNAOisusuallyassociatedwithless
freezingriskinMarchacrossthestudyregion.ThePearsoncorrelationvaluecouldbeashighas
-0.469intheOhioValley,indicatingahighlysigni˝cantcorrelationwiththepositivephaseofNAO
occurringduringMarchforrecentdecades.Eventhoughnotsigni˝cant,somepositivecorrelations
areoccurringformonthlyFAPvaluesinAprilandMayafterthesigni˝cantnegativecorrelationin
March,especiallyintheSouthernGreatPlainsandOhioValley.Thisislikelybecausethee˙ects
ofthepolarjetstreambecomeweakerinthelatespringtime.Withlagginglengthsincreasing
from1to3months,morepositivecorrelationsareobserved.ThemonthlyFAPvaluesinApril
arepositivelycorrelatedwithNAOinFebruary,indicatingapositivephaseofNAOinFebruaryis
statisticallyassociatedwithmorefreezingrisksinAprilacrossthestudyregion,especiallyinthe
NorthernGreatPlains,UpperMidwest,OhioValley,andNortheast.Also,whenthepositivephase
ofNAOinDecemberoccurs,fewerfreezingrisksinMarchareobservedacrossourstudyregion,
especiallyintheUpperMidwest,Northeast,andSoutheast.
47
NorthernGreatPlains
SouthernGreatPlains
UpperMidwest
OhioValley
Northeast
Southeast
Lag0
March
-0.368**
-0.576**
-0.357**
-0.469**
-0.466**
-0.376**
April
-0.105
0.108
0.039
0.093
-0.096
-0.069
May
0.008
0.145
-0.029
0.047
-0.193
-0.08
Springtime
-0.016
-0.045
0.036
-0.05
-0.114
-0.206
Lag1
March
0.142
-0.159
0.129
0.028
0.055
0.08
April
0.226
0.294*
0.159
0.237
0.067
0.023
May
-0.116
-0.175
0.07
-0.089
0.005
0.02
Springtime
0.039
-0.172
0.134
-0.06
-0.021
-0.188
Lag2
March
0.178
0.075
0.138
0.239
-0.07
0.081
April
0.494**
0.251
0.416**
0.317**
0.289*
0.089
May
-0.176
0.041
-0.015
-0.128
-0.102
-0.343**
Springtime
0.166
-0.244
0.19
0.042
0.071
-0.064
Lag3
March
-0.125
-0.085
-0.346**
-0.179
-0.443**
-0.418**
April
0.07
-0.158
0.093
-0.087
0.176
-0.011
May
-0.216
0.009
0.137
-0.098
0.053
0.005
Springtime
0.233
-0.055
0.125
0.082
0.042
-0.067
Table3.3:
Pearsoncorrelationvalues(r)comparingmonthlyFAPforthemonthsofMarch,April,andMaytoNAOvalues0-3months
earlier.
Relationshipswithpvalueslessthan0.10arecharacterizedwithonestarandthoselessthan0.05arewithtwostars.
Table3.4shows0-3monthlaggedPearsoncorrelationvaluesbetweenthemonthlyFAPvalues
foreachofthethreemonthstoPNAvalues.ThepositivephaseofthePNApatternisassociated
withabove-averagetemperaturesoverwesternCanadaandtheextremewesternUnitedStatesand
below-averagetemperaturesacrossthesouth-centralandsoutheasternUnitedStates(Dixonetal.,
2007).ThecorrelationsbetweenPNAandFAPareoverallnegative.Nevertheless,FAPvalues
inMayaresigni˝cantlypositivelyassociatedwithPNAinMay,especiallyintheNorthernGreat
PlainsandtheUpperMidwest.ThisindicatesthatthepositivephaseofPNAinMaycouldlead
tomorefreezingrisksacrossmostofourstudyregionsinMay.Also,wenoticethatthetotal
springtimeFAPvaluesarealwayssigni˝cantlynegativelycorrelatedwithPNAvalues1-3months
earlier,exceptintheSouthernGreatPlainsregion.Thisrelationshipprovidesforecasttoolsfor
predictingspringtimefreezingrisksacrossrelevantareasforuptothreemonthsahead.
49
NorthernGreatPlains
SouthernGreatPlains
UpperMidwest
OhioValley
Northeast
Southeast
Lag0
March
-0.17
0.121
-0.228
0.01
-0.197
-0.124
April
-0.205
-0.08
-0.299*
-0.185
-0.266
-0.12
May
0.356**
0.267
0.343**
0.304*
0.148
0.272*
Springtime
-0.153
0.073
-0.198
-0.102
-0.158
-0.163
Lag1
March
-0.350**
-0.217
-0.338**
-0.26
-0.315*
-0.185
April
-0.465**
-0.249
-0.362**
-0.397**
-0.17
-0.243
May
0.058
-0.105
-0.174
0.017
-0.175
-0.223
Springtime
-0.407**
-0.184
-0.429**
-0.358**
-0.324**
-0.271*
Lag2
March
-0.067
0.125
-0.133
-0.041
-0.086
-0.138
April
-0.191
-0.229
-0.238
-0.286*
-0.134
-0.223
May
0.166
-0.144
-0.041
0.061
-0.136
-0.035
Springtime
-0.397**
-0.111
-0.430**
-0.362**
-0.358**
-0.322**
Lag3
March
-0.142
-0.095
-0.269*
-0.085
-0.245
-0.16
April
-0.232
-0.097
-0.357**
-0.267
-0.436**
-0.302*
May
0.157
-0.003
-0.064
0.015
-0.044
-0.058
Springtime
-0.285*
-0.147
-0.429**
-0.299*
-0.401**
-0.320**
Table3.4:
Pearsoncorrelationvalues(r)comparingmonthlyFAPforthemonthsofMarch,April,andMaytoPNAvalues0-3months
earlier.
Relationshipswithpvalueslessthan0.10arecharacterizedwithonestarandthoselessthan0.05arewithtwostars.
ElNiñoeventsusuallyoccurwithabove-normaltemperaturesinthenorthernregionsofthe
UnitedStates(CPC,2008).SincethepositivePNAisoftenassociatedwithanElNiñoevent
(CPC2007),theoverallrelationshipsbetweenFAPvaluesandNiño-3.4arenegativeaswell.Itis
noteworthythattheFAPvaluesacrosstheNorthernGreatPlainsinMarchshowsigni˝cantnegative
correlationswithNiño-3.40-3monthsearlier.Ourresultsalsorevealgeneralpositiverelationships
betweenmonthlyFAPandPDOvalues.InMay,theFAPsarepositivelycorrelatedwithPDO,
especiallyintheNorthernGreatPlains,theSouth,theUpperMidwest,andtheOhioValley.
3.3.3ResultofEOFanalysisforfreezingdaysinspringtimemonths
EmpiricalOrthogonalFunction(EOF)analysisisperformedtoidentifythedominantspatial
patternsofinterannualvariabilityoffreezingdaysinMarch,April,andMay,respectively.We
discussheretheresultsoftheleadingtwomodesforeachmonth.
ForfreezingdaysinMarch(Figure3.5),the˝rstEOFmodeexplains61.97%ofthetotal
variancewithin-phase˛uctuationacrossourstudyregion.The˝rstEOFspatialpatternshows
arelativelylargevariationintheOhioValleyandthe˝rstEOFtimeseriesshowssubstantial
interannualvariability.Weconductthespectralanalysistoidentifymajorperiods,whichreveals
prominentperiodsof2.5to4.5year,indicatingthatthein-phase˛uctuationpatternoffreezing
daysacrosstheregioninMarchoccursevery2.5to4.5years.Thesecondmodeonlycontributes
to8.3%ofthetotalvariance.ThesecondEOFspatialpatternshowsoppositephasesbetweenthe
regionencompassingtheOhioValley,theSoutheast,andtheNortheast,andtheregioncovering
theNorthernGreatPlainsandtheUpperMidwest.ThespectralanalysisofthesecondEOFtime
seriesrevealsamajorperiodof6.5years.
Tofurtherunderstandhowthefrequencyofthesepatternsassociatedwithlargecirculations,we
alsorelatetheEOFtimeseriestosometeleconnectionindices.Theresultrevealsthatthe˝rstEOF
timeseriesissigni˝cantlycorrelatedwithpositivephaseNAOinMarchaswellasDecemberof
thepreviousyear.ThisindicatesthatifweobserveapositivephaseofNAOintheaforementioned
months,thein-phase˛uctuationacrossourstudyregionwithahigherintensityovertheOhioValley
51
Figure3.5:
TheleadingtwoEOFmodesandPCtimesseriesforfreezingdaysinMarch.
islikelytooccurinMarch.
ForfreezingdaysinApril(Figure3.6),the˝rstmodeofin-phase˛uctuationsoverthedomain
explains56.242%.The˝rstspatialpatternalsoshowsalargervariationacrosstheNorthernGreat
Plains.The˝rstEOFtimeseriesshowsinter-decadalvariabilitywithmajorperiodsof5to7years.
Thesecondmodeaccountsfor12.33%ofthetotalvariance.Thespatialpatternofthesecond
modeshowsout-of-phasevariabilitybetweentheNorthernGreatPlainsandtherestofourstudy
region,andthetimeseriesofitshowsinterannualvariabilitywithmajorperiodsof2.5to3.5years.
Thecorrelationofcoe˚cientsbetweenfreezingdaysinAprilandthethreeteleconnectionindices
revealthatthe˝rstEOFtimeseriesissigni˝cantlycorrelatedtopositivephaseNAOinFebruary
andnegativephasePNAinMarchandDecemberofthepreviousyear.
ForfreezingdaysinMay(notshownhere),the˝rstmodeexplains57.8%ofthetotalvariance.
SinceinMay,somesouthernareasareinfrost-freeseason,weonlyobservein-phase˛uctuation
52
Figure3.6:
TheleadingtwoEOFmodesandPCtimesseriesforfreezingdaysinApril.
inthenorthernpartofthestudyregion.Andthetimeseriesforitshowsmajorperiodsof3to7
years.Thesecondmodeexplains10.71%ofthetotalvarianceandshowsoppositephasesbetween
theNorthernRockiesandPlainsandtherestofourstudyregion.Thetimeseriesforitreveals
interannualvariability.Therelationtoteleconnectionindicesshowsthatthesecondmodetime
seriesissigni˝cantlyrelatedtobothpositivephasesofNAO,PNA,andPDOinMay.
TounderstandthespatialpatternsofthetwoleadingEOFmodesoffreezingdaysindi˙erent
monthsacrossourstudyregioninthecontextofatmosphericcirculationanomalies,weregressthe
timeseriesofthe˝rsttwoEOFmodesoffreezingdaysinMarchtoseveralmonthlyanomalous
atmosphericvariables,includingthe500hPageopotentialheight(H500),1000hPageopotential
height(H1000),the500-1000hPathickness,andthe1000hPawind.
Theanomalous500-hPageopotentialheightregressedonnormalizedPC1inMarchshows
thatthisvariableisnegativelycorrelatedwithnormalizedPC1throughoutourstudyregion,while
53
Figure3.7:
The˝rstPCtimesseriesforfreezingdaysinMarchrelatedtoatmosphericvariables.
anomalous1000-hPageopotentialheightispositivelycorrelated(Figure3.7).Thisresultindicates
thattheoverallnegativefreezingdayanomaliesovertheentirestudydomain(Figure3.5)are
associatedwithlower-(higher-)than-normal500-hPa(1000hPa)heightandshallower-than-normal
thicknessbetweenthetwolevels.TheresultsforPC2tellsadi˙erentstory(Figure3.8)thatwhen
PC2offreezingdaysinMarchispositive,whichisassociatedwiththeout-of-phasevariability
EOF2spatialpatterninFigure3.5,thegeopotentialheightislowerthanaverageattheupperlevel
andbelownormalatthesurface,resultinginathicknessthickerthannormal.Theregressionresults
betweentheanomalousatmosphericvariablesandNAOinMarchrevealsigni˝cantRossbywave
patterns.ThethicknessregressionpatternofPC1issimilartothatofNAO,whichisconsistentwith
theaboveconclusionthatfreezingdaysinMarcharesigni˝cantlycorrelatedtoNAO(Figure3.9).
The1000hPawindanomaliesregressedonnormalizedPC1andPC2showsimilarresults,which
arebothsimilartotheresultofNAO.Thisanomalouswind˝eldischaracterizedbyananomalous
cycloneovertheSouthernGreatPlainsandananomalouseasterlywindfromtheAtlantictothe
OhioValley.
54
Figure3.8:
ThesecondPCtimesseriesforfreezingdaysinMarchrelatedtoatmospheric
variables.
Figure3.9:
TimesseriesofNAOinMarchrelatedtoatmosphericvariables.
55
3.4Summary
Thisstudyinvestigatesthefrequency,severity,andclimatebackgroundofspringtimefreeze
eventoccurrences.Weconducttrendanalysisforfreezingdaysateachgridpointandalsoat
theregion-averagedlevel.WealsousetheEOFtechniquetoidentifythespatialpatternand
correspondingtrendoffreezingdayseachspringtimemonth.Andwealsouseclimatevariablesto
explaintheseresults.Themajor˝ndingsareasfollows.
1)TrendanalysisateachgridpointrevealsthatfreezingdaysinMarchshowadecreasingtrend
acrossthecentralandMidwesternUS.Trendanalysisoverthespatiallyaveragedspringtimefreeze
events,whichissigni˝cantlycorrelatedtospatiallyaverageddailyminimumtemperature,also
revealsageneraldecreasingtrendacrossthestudyregion.Althoughtheyears2007and2012show
alowfrequencyoffreezeevents,considerablecropdamagewasobservedinthesetwoyears,asa
consequenceoffalsespringsexposingtheearlierdevelopedcropstothesubsequentfreezeevents.
2)Wethende˝neavariable,freezingareapercentage(FAP),overtheregionsofinterest.The
trendsfortheaccumulatedspringtimearedownwardwithasigni˝canceintheOhioValley,which
extendsthe˝ndingsinEasterlingetal.(2002)fromtheperiod1948-1999into1981-2019.To
furtherinvestigatetheperiodicityofmonthlyFAPvalues,werelatethemtosometeleconnection
indices,Niño-3.4,NAO,PNA,andPDO.WeconcludethatthepositivephaseofNAOisusually
associatedwithlessfreezingriskinMarchacrossthestudyregionandthetotalspringtimeFAP
valuesarealwayssigni˝cantlyassociatedwiththenegativephaseofPNAvalues1-3monthsearlier,
exceptintheSouthernGreatPlainsregion.
3)EOFanalysisoffreezingdaysinMarchshowsarelativelylargervariationintheOhioValley,
andthe˝rstEOFtimeseriesshowssubstantialinterannualvariability.Andwhenapositivephase
ofNAOinMarchandDecemberofthepreviousyearoccurs,thein-phase˛uctuationacrossour
studyregionwithahigherintensityovertheOhioValleyisexpectedtooccurinMarch.Andthis
patternislikelyassociatedwiththegeopotentialheightlowerthanaverageattheupperleveland
higherthanaverageatthesurfaceandathicknessshallowerthannormal.
Theanalysisworkdoneinthisstudyhassomelimitations.First,togainbetterresultsatthe
56
subregionlevel,thestudyregionshouldbeextendedtothecontiguousUnitedStates.Weonly
getlimitedpartsofthede˝nedsubregionsoftheNortheast,theSoutheast,andtheSouthernGreat
Plains.Therefore,thefreezingdaysaveragedacrossthesede˝nedregionsinthisstudycouldnot
representthegeneraltrendofthewholeregionde˝nedinKarlandKnight(1998).Second,to
understandtheweatherconditionsoffreezeeventformation,speci˝ccasestudiescouldbedone.
Forexample,wecouldselectacoupleoffreezeeventsthatprofoundlyimpactcrops'growthto
investigatetheparticularweatherconditionsofwind,temperature,andhumidity.
Inspiteofthelimitationsofthisstudy,theimplicationsoftrendsandcausesofspringtime
freezeeventsareimportantforfruitgrowersinthecentralandmidwesternUnitedStates.NAOis
anexcellentindextoindicatethefreezeeventoccurrencesinMarch.Althoughthefreezeevent
occurrencestendtodecreaseinrecentdecades,cropsareexposedtohigherfalsespringrisks,
whichwillbeinvestigatedinthenextchapter.
57
CHAPTER4
SPRINGTIMEFREEZEEVENTSIMPACTSONAGRICULTURE
4.1Background
4.1.1Springtimefreezetypesandtheirimpactsonperennials
Di˙erenttypesoffreezeeventscouldcausedi˙erentlevelsofdamagetoperennials.Conse-
quently,itisimportanttounderstandwhenandhowthesefreezeeventsoccurandthus˝ndways
tominimizetheimpactsoftheseadverseweathereventsoncrops.Althoughthetermsfrostand
freezeareoftenusedinterchangeably,theyrepresenttwokindsofweatherevents.Thetermfreeze
istypicallyusedtodescribeaninvasionofalarge,frigidairmassfromtheArcticorCanadian
regions,whichiscommonlycalledtheadvectivefreeze(Powelletal.,2000).Theadvectivefreezes
tendtooccurinawell-mixedboundarylayerwithstrongwindsassociatedwithcoldfrontsor
withtheleadingedgesofpolar-originhigh-pressuresystems(Winkleretal.,2012).Frosts,also
knownastheradiativefreeze,usuallyoccurundercalm,clearconditionsoncoldnightswhere
thegroundorplantcanopysurfacecoolsradiativelymorequicklythantheairaboveit,whichis
usuallyassociatedwithhigh-pressuresystems.(Winkleretal.,2012).Therearetwotypesoffrost:
hoarfrostandblackfrost.Hoarfrostsarevisiblewithiceformation,andblackfrostsareinvisible
duetoinsu˚cientwatervapor(Powelletal.,2000).Empirically,radiativefreezeshappenmore
frequently,andadvectivefreezesresultincolderandmoreextendedperiods.Bothadvectivefreezes
(intrusionofchilledair)andradiativefreezes(negativeheatenergybalance)couldsigni˝cantly
decreaseagriculturalproduction.Radiativefreezesoccurmorefrequentlyandareassociatedwith
temperatureinversionandarecomparativelyeasiertopredict.Onthecontrary,advectivefreezes
arelessfrequentandoccurwithoutinversiondevelopingandthusaredi˚culttopredict.
Knowingthetypeoffreezecouldhelpsuggestwaysofmanagingcropplantationdistributions
sincethetwotypesoffreezeeventsproducedi˙erentdamagepatternsacrossthegrowingregion.
58
Researchhasshownthatfreezeeventthatdamageplantsaremoreoftenradiativethanadvective
(Charrieretal.,2015).Radiativefreezestypicallyoccuraftersunsetwhentheweatherconditions
arecalm.Sincethereisnoincomingsolarradiationaddingheattothesystem,theoutgoinglong-
waveinfraredradiationcoolsthegroundandthusmakestheheatenergybalancenegative(Charrier
etal.,2015).Therefore,radiativefreezesoccurmorefrequentlyinnarrowvalleys,inconcaveor
˛atlocations,andlessregularlyinelevatedandconvexareasmoreexposedtowind(Lindkvistet
al.,2000).Consequently,propercropsiteselectionisane˙ectivewaytoavoidfreezinginjury
causedbyradiativefreezes.Hilltopsandslopesareconsideredasgoodorchardsitesfromwhich
relativelycoldaircan˛owdownslopeawayfromtheorchardsite,whilelandscapedepressionsare
regardedaspoorsiteswherecoldairaccumulatesatnightduetoairdrainage(Winkleretal.,2012).
Incontrast,advectivefreezesmaycausemoresigni˝cantdamagebyenhancingtheheatlossand
coolingofplanttissue(Perry1998)atthosegoodsitesunderradiativefreezes(elevatedandconvex
areasmoreexposedtowind),whicharemoreexposedtowindwithsubfreezingtemperatures
(Winkleretal.,2012).Therefore,thecombinationofbothtypesoffreezescouldresultincrop
siteselectionfailures.Inaddition,localtopographycomplexityandvariationsplayanimportant
roleinleadingtofreezeseventsformationbyin˛uencingtemperatureandwindinvalleyorlake
systems(Lindkvistetal.,2000).
4.1.2Falsespringoccurrencesandimpactsoncrops
Extremetemperature˛uctuationsduringthespringtimecanhaveenormousimpactsonthe
growthofvegetation,andmanyagriculturalandhorticulturalcrops.Anextendedperiodofwarmer
temperatures,identi˝edasa"falsespring",couldcauseperennialcropstobreakdormancyandstart
theirgrowingseasonearlierthanusual(Kistneretal.,2018).Asaresult,whenthecoldaircomes
back,thesecropsareespeciallyvulnerableandsusceptibleduringtheirbudstages,and,hence,
springfreezeshaveprofoundimpactsoncrops.In2002,sourcherryproductioninMichiganwas
reducedbyatleast95%fromthepreviousannualyieldsduetospringtimefreezeeventsfollowing
theunexpectedextendedwarmspells(NASS,2002).InApril2007,afteranextendedwarmth
59
promptingearlygrowthofvegetationandcrops,coldarcticairin˝ltratedsouthwardintothecentral
andeasternUS,causingtemperaturestodroptofreezingandthusleadingtoanestimated$2billion
ofyieldreductionsforcropssuchaswinterwheat,corn,andforagelegumesduetofrostburnand
plantmortality(Marinoetal.,2011;Kral-O'Brienetal.,2019).DuringMarch2012,apersistent
upperairridgingfeatureresultedinthewarmesttemperatureanomaliesatmanylocationsacross
theMidwesternUSsince1900forthatmonth,whichbroughtmanycropsoutofdormancy(Labe
etal.,2015).NeartheendofMarchandthroughoutApril2012,averagetemperaturesreturnedto
theseregions,resultinginaseriesofcropyieldlossduetofreezedamages,includingthelossof
approximately85%ofapplecropyieldsand90%oftartcherrycropyieldsfortheyearinMichigan
(Kistneretal.,2018).
Globalclimatechangehassparkedconsiderableresearchonextremeweathereventsandtheir
impactsonvegetationgrowth.Nevertheless,researchfocusedontheoccurrences,immediate
e˙ects,andlong-termconsequencesoffalsespringeventshavebeenlimited,despitetheirpotential
forextremeenvironmentalandeconomicconsequences(Kral-O'Brienetal.,2019).Falsesprings
arerelatedtoglobalwarminginthewaythattheonsetofspringhasgenerallyshiftedearlierinthe
yearoverthepastseveraldecadesduetorisingaveragetemperatures(Allstadtetal.,2015).False
springs,whichusuallyhappeninlatewinterorearlyspring,aresu˚cientlymildandlongtobring
vegetationoutofdormancyearlier,renderingthevegetationvulnerabletolaterfreezeevents(Ault
etal.,2013).Thelaterthefreezeoccursaftergrowthbegins,themoredamageislikelytobecaused
sinceplantsareinamorevulnerablephenologicalstage(Marinoetal.,2011).Additionally,itis
projectedthatfalsespringrisksmightincreaseinportionsoftheMidwesternUS(Allstadtetal.,
2015).
4.1.3In˛uenceofphenologyonfreezedamage
Freezeeventsthatoccuratadi˙erenttimecouldresultindi˙erentinjuriestocrops.Usually,
freezeeventsafterfalsespringscouldcauseseveredamagetogrowingcrops.Also,phenological
processesareparticularlycrucialforfreezeeventsavoidanceinspringandautumn(Charrieret
60
al.,2015).Microclimatefactorsthata˙ectcropdevelopmentincludechillunits,growingdegree
hours,andminimumtemperatures,varyingindiversi˝edtopography(Loganetal.,2000).Since
phenologicalprocessesarestronglydependentontemperature,microclimatecoulda˙ectcrop
stagesbydelayingdevelopmentwithcoolingandacceleratinggrowthwithwarming.During
winter,fruittreesinthestageofdormancyareveryresistanttocoldtemperatures.Duringthe
annualcycleofgrowthanddormancy,thetransitionperiodsinautumnandspringaretheriskiest
sincethereisamoderateprobabilityoffreezingwhenplantsaremostvulnerable(Charrieretal.,
2015).Thetemperaturesthatcausedamageinspringaremuchlowerthanthetemperaturesthat
treesencounterdamageinautumn(Larcheretal.,2010).Therefore,moderatelylowtemperatures
inspringcouldresultindeadlyimpactsoncrops.Asfruittreesdevelopinthespringandbudsstart
toswell,theylosetheabilitytowithstandcoldwintertemperatures(Longstrothetal.,2012).The
stageofbuddevelopmentdetermineshowsusceptibleanygivenfruitcropiswhenfreezeevents
occur(Longstrothetal.,2012).Whenaspellofunexpectedwarmweatherinthespringoccurs,the
plantswilldevelopquickly.Andwhenthetemperaturesreturntonormal,thesedevelopingcrops
thatareveryvulnerablewouldsu˙ersigni˝cantdamages.
4.1.4Studyobjectives
Thisstudyaimstoinvestigatethehistoricalimpactsofspringtimefreezeeventdamageoncrops
intheMidwesternUSfrom1981to2018.Inparticular,thestudyfocusesonthe˝rstgreendates,
bloomdates,poorpollinationdays,andthebudsurvivalchanceoftheyearforapplesusingcrop
simulationmodelswithinputdatasetsfromPRISM.Also,damagedaysoftheyear,dailyminimum
temperature,anddamageoccurringdatesforeachphenologygrowingstageareincludedtoexplore
theimpactsoffalsesprings.
61
4.2Datasetandmethods
4.2.1PRISMdataset
PRISM,agriddedanalysisdatasetprovidedbythePRISMclimategroupfromOregonState
University,o˙ersestimatesforsixessentialclimateelements:precipitation(ppt),minimumtem-
perature(tmin),maximumtemperature(tmax),meandewpoint(tdmean),minimumvaporpressure
de˝cit(vpdmin),andmaximumvaporpressurede˝cit(vpdmax).PRISMhasanexcellentspatial
resolutionof4km.Evaluationshavebeendoneinthesecondchapter,andconclusionsrevealthat
thePRISManalysisdatasetgenerallyagreeswellwiththeobservationsandcouldrepresentthe
freezeevents.Asinputsforourappleyieldsimulationmodel,weobtainedthreevariablesfrom
PRISM,dailyminimumtemperature,dailymaximumtemperature,anddailyprecipitation.
4.2.2Appleyieldsimulationmodel
Aprimarycauseofyear-to-yearyield˛uctuationsintreefruitproductionacrosstheUnited
Statesandinternationalproductionareasisbudlossassociatedwiththecoldinjurythattypically
occurseitherduringthespringorwinterseasons(AndresenandBaule,2018).Budsensitivityto
coldenvironmentaltemperaturesisstronglydependentonthephenologicalstageofdevelopment
andgenerallyincreasesfromfulldormancythroughvegetativeintoreproductivestages.Inthis
study,potentialseasonalyieldlossesforappleproductionassociatedwithcoldtemperatureswere
estimatedfollowingtheapproachofZavallonietal.(2006),inwhichphenologicaldevelopment
andcoldsensitivityaresimulatedonadailybasisusingobservedmaximumandminimumair
temperatures.Estimationofphenologicaldevelopmentofapplewasbasedonthemethodologyof
Rijal(2017)usingsummedseasonaltotalsofbase4
°
CgrowingdegreedaysfromMarch1st.Ten
andninetypercentdamagethresholdtemperaturesforapplesatanumberofphenologicalstages
wereobtainedfromBallardetal.(1998).Ingeneral,thedamagethresholdtemperaturesvaryfrom
-34.4
º
Cduringdormancyto-2.2
º
Cduringlatevegetativestagesthroughfullbloom.Assuminga
linearrelationshipbetweenthedamageseveritylevelandthecriticalfreezingtemperatures(Dennis
62
andHowell,1972),dailypercentdamageestimatesbasedonobservedminimumtemperatureswere
developedforeachphenologicalstagedormancythroughfullbloom.Potentialyieldlossesdueto
coldtemperatureswerethengeneratedseasonallyateachlocationasthecumulativesumofdaily
colddamagesimulatedduringtheseason.
63
Figure4.1:
Average,absolutevariation,relativevariation,andtrendofthe˝rstgreendate(Julian
date)from1981to2018.
4.3Results
4.3.1Characteristicsoftheyearlyresults
Thecropsimulationmodeloutputstwo˝lesateachgridpoint,dailyresultsandyearlyresults.
Theyearlyresultscontainthe˝rstgreendate('fgreen'inFigure4.1,referredtoasleafoutinsome
literature),thebloomdates,thepoordays(intermsofpollination),andtheyield(i.e.,budsurvival
chance).AndthedailyresultsprovideinformationabouttheGrowingDegreeDay(GDD),the
phenologicalstageofapples,andthefreezedamage.
Figure4.1showscharacteristicsofthe˝rstgreendate,includingannualmean,absoluteand
relativevariation,andthetrend.ThevaluesaretheJuliandatesoftheyear.Forexample,60means
thedateofMarch1st,whichistheearlyspringtime.Theaveragevaluesarelatitude-dependent
64
withlaterdatesinthenorthernregionandearlierdatesinthesouthernregion,similartothe
averagedtemperaturepattern.ForMichigan,weobserveagradualincreasefromthesouthto
thenorth.The˝rstgreendateinnorthernMichigancouldbearoundearlyApril,butthatinthe
upperpeninsulacouldbeaslateasthemiddleofMay.Theinterannualstandarddeviationshows
theabsolutevariation,whilethecoe˚cientofvariationistherelativevariationcalculatedasthe
absolutevariationdividedbythemeanvalues.Therefore,theabsolutevariationofthe˝rstgreen
dateindicatesthatthenorthernregionshavegenerallysmallervariationsthanthesouthernregions.
ThehighestabsolutevariationisobservedintheSoutheastofthestudyregionandthelowestin
theNortheast.Sothe˝rstgreendateismorevariableinwarmerregionscomparedtocoldregions.
Sincetheabsolutevariationvaluesaregenerallysmallcomparedtothemeanvaluesofthe˝rst
greendate,therelativevariationsaredominatedbytheaverages,resultinginanoppositepattern
tothemeanvalues.Thevaluesinthetrendgrapharetheslopesoftheregressionofthe˝rstgreen
dateonthe18years,withnegativevaluesindicatingearlierdatesandpositivevaluesindicating
laterdates.Also,thesigni˝canttrendisindicatedbythegreyshades.Theresultsrevealthatthe
˝rstgreendatesarebecomingearlierinmostareasexceptforsomenorthernmostregionsofthe
northernplainsandMichigan'sUpperPeninsula.
Similarpatternsarefoundforthebloomdate(Figure4.2)thatoccurslaterintheyearthanthe
˝rstgreendates.Theabsolutevariationsofthebloomdateareoverallsmallerthanthoseofthe
˝rstgreendate,especiallyacrosstheentireNortheast,wherethevariationsaresmall.Themean
valuesstilldominatetherelativevariations.Thegeneraltrendofthebloomdateisstillbecoming
earlier.However,moreareasoftheNorthernGreatPlainsandtheUpperMidwestshowlaterbloom
dates.Also,moresigni˝cantpointsareshownintheSouthernGreatPlains,theNortheast,andthe
Southeast.
Thepatternsforthepoorpollinationdays,however,aredi˙erent(Figure4.3).Themeanvalue
patternisnotlatitude-dependent.Mostareasofourstudyregionshowvaluesfrom2to3days.And
largevaluesareobservedacrosstheOhioValleyandpartoftheNorthernGreatPlains.Mostof
theareasshowaninterannualstandarddeviationoffewerthantwodays.However,thelargemean
65
Figure4.2:
Average,absolutevariation,relativevariation,andtrendofthebloomdate(Julian
date)from1981to2018.
valueareasalsoshowlargerabsolutevariationsofaroundthreedays.Therelativevariationshows
anexactlydi˙erentpattern,whichrevealslargevaluesintheNorthernandtheSouthernGreat
Plains.Also,sincetheabsolutevariationvaluesareinthesamemagnitudeasthemeanvalues,the
relativevariationsarenotdominatedbythemeanvalues.Regressionofpoorpollinationdaysover
theyearsrevealsthatthepoordaysincreaseacrossmostofourstudyregionsfrom1981to2018,
withnoticeablesigni˝canceintheNorthernGreatPlains,theUpperMidwest,andtheNortheast.
Onthecontrary,thepoordaysinthenorthernpartoftheOhioValleyandthesouthernpartofthe
SouthernGreatPlainssigni˝cantlydecreasethroughoutourstudyperiod.
Similartopoorpollinationdays,thepatternforthemeanbudsurvivalchanceisalsonot
latitude-dependent(Figure4.4).ThevaluesarelargerintheUpperMidwestandtheNortheastthan
intherestoftheregions.Itisnoteworthythatthemeanvaluesaroundlakeshoresareconsistently
66
Figure4.3:
Average,absolutevariation,relativevariation,andtrendoftheyearlypoor
pollinationdaysfrom1981to2018.
highthroughoutthestudyperiod,whichcouldbearesultofthelargebiasintheinputPRISM
datainlake-modi˝edareas,asdiscussedinChapter2.Thelargebiasintheinputdatamayalso
explainthesmallvaluesintheinterannualstandarddeviationinareasalongtheshorelines.The
largestabsolutevariationsareobservedintheNorthernGreatPlains.SouthernMichiganandthe
westoftheSouthernGreatPlainsshowsmallerabsolutevariationscomparedtootherareas.Itis
alsonoticeablethatthere'sahighlightregionintheupperpartofMichigan,whichshowssmaller
budsurvivalchancesandlargerstandarddeviationscomparedtootherpartsofMichigan.Since
theabsolutevariationpatternisconsistentwiththemeanpattern,therelativevariationpatternis
stillconsistent,whichistheoppositeofthemeanpattern.Regressionofbudsurvivalchanceover
theyearsrevealsthatmostareasofourstudyregionincreasethroughoutourstudyperiod.And
WisconsinandthesouthernpartoftheOhioValleyshowsigni˝cantincreases.
67
Figure4.4:
Average,absolutevariation,relativevariation,andtrendoftheyield(budsurvival
chance)from1981to2018.
4.3.2DailyResultsandsomederivedindices
Freezedamageoneachdayischaracterizedbyafractionnumberlessthan1.0inthedaily
output˝les.Therefore,wecouldknowthedayswhendamageoccursandalsotheintensityofthe
damage.Also,wecouldknowthedatewhenthedamageoccurs,thephenologicalstageofthe
plantwhendamageoccurs,inadditiontothedailyminimumtemperaturewhendamageoccurs.
Weconductsomeanalysesbasedontheseresults.
Figure4.5showstheresultsforyearlyaccumulateddamagedays.Sincethebudsurvivalchance
andthedamagefractionnumbersaddupto1.0,themeanpatternofdamagedaysistheopposite
ofthatofthebudsurvivalchance.Incontrarytotheyieldvalues,damagedaysaroundtheGreat
LakesRegionsarefewercomparedtootherareas.Theinterannualstandarddeviationgraphtells
usthattheareaswheremoredamagedaysoccuralsoshowhigheramplitudesofperturbations.
68
Figure4.5:
Average,absolutevariation,relativevariation,andtrendofdamagedaysfrom1981to
2018.
TheresultsarecharacterizedbylowervariabilityintheUpperMidwestandtheNortheastversus
highervariabilityintheSouthernGreatPlains,thesouthernOhioValley,andtheSoutheast.The
relativevariationpatterntellsadi˙erentstory.Dividedbytheaverages,theamplitudesaround
theGreatLakesRegionsandtheNortheastarelargerthanotherareas.Thetrendofdamagedays
acrossourstudyregionisoveralldownward,whichistheoppositeofthebudsurvivalchance.And
signi˝cantincreasesareobservedintheSouthernGreatPlains.Itisalsonoteworthythatmost
areasofMichiganshowaslightupwardtrendofdamagedays.
Figure4.6showstheaveragedamagedaysateachgrowthstage.Stagesarede˝nedaccording
totherunningtotalofbase4oCgrowingdegreedayseachyear,calculatedbasedontemperature.
Byde˝nition,wehaveStage0andfrom2till9withoutStage1.InStage0,damageonlyoccurs
inthenorthernmostregionsneartheUSandCanadaborder,whichisexpectedbecauseofthe
69
Figure4.6:
Averageofdamagedaysateachgrowthstagefrom1981to2018.
temperature-dependence.InStage2,damagesoccuroverlargeareasoftheSouthernGreatPlains
andinpartsofthesouthernOhioValleyandtheSoutheast.Damageareasstayintheseregions
throughtherestofthestages,withthenumberofdamagedaysandtheareasin˛uencedbecoming
smallerinStages3-7,andlargeragaininStages8and9.
Thetrendsofthedamagedaysforeachstagecanbedeterminedbyregressingthetimeseriesof
averagedamagedaysduringeachgrowthstageonthetimeseriesoftheyearsoverthestudyperiod
(Figure4.7).ThedamagedaysinStage0thatoccuroverthenorthernPlainexhibitadownward
trend.Onthecontrary,thetrendsaregenerallyupwardinregionsofthesouthernPlains,especially
forStage2andStage9.
ExceptforStage0,thedateswhendamageoccursduringeachgrowthstageshowlatitude-
dependencewithprogressivelylaterdatestowardsnorth.(Figure4.8).InStage0,damageonly
occursinthenorthernpartsoftheGreatPlainsandtheMidwestaswellaspartsoftheOhioValley.
70
Figure4.7:
Thetrendofdamagedaysateachgrowthstagefrom1981to2018.
Thedamageacrosstheseregionshappensaroundsimilardates,withaslightdelayinnorthern
MichiganandWisconsin,possiblyduetothelakee˙ect.InStage2,damagehappensfromlate
FebruaryinthesouthernedgeofthestudyregiontolateMarchandearlyAprilinthesouthern
PlainsandtheOhioValley,andfromlateApriltoearlyMayinthenorthernPlainsandupper
Midwest.Similardelayedoccurrencetowardsnorthernlatitudesalsooccursduringotherstages.
Thetrendanalysis(Figure4.9)revealsthat,ingeneral,theoccurringdamagedatesarebecoming
earlieracrossthestudyregionoverthe38-yearstudyperiod,especiallyforStage9whendamage
occursmuchearlieracrosssouthernMichiganandnorthernOhio.TheonlyexceptionisStage0,
duringwhichdamageinMichiganandWisconsinhappenssigni˝cantlylater.
Thespatialpatternsoftheaveragedamagedaysandthedateswhendamageoccursareconsistent
withthepatternsofdailyminimumtemperatureduringthedamagedays(Figure4.10).InStage
0,thedailyminimumtemperaturesduringdamagedaysarebelow-10
°
C.TheninStage2,
71
Figure4.8:
Theaverageoffreezedamageoccurringdateateachgrowthstagefrom1981to2018.
mostdamageoccurringintheSouthernGreatPlainswithdailyminimumtemperaturefrom-8
°
Cto-6
°
C,whileintheUpperMidwestwithdailyminimumtemperaturefrom-6
°
Cto-4
°
C.
ThenthedailyminimumtemperaturesforStage3andStage4aresimilar,from-5
°
Cto-3
°
C
acrossourstudyregion.Thenasthephenologymarchesintolaterstages,thedailyminimum
temperaturesarehigher.Generally,thedamageoccurringinthenorthernareaswithhigherdaily
minimumtemperaturescomparedtothesouthernareas.Theregressionanalysis(Figure4.11)
revealsthatthedailyminimumtemperaturesareoverallbecomingwarmerduringdamagedays.
AndlargertrendvaluesareonlyshowingaroundtheGreatLakesRegioncomparedtootherareas.
ItisnoteworthythattheareasshowingearlierdamageoccurringdatesinFigure4.9showdaily
minimumtemperaturetrendsupward.Therefore,weconcludethatdamageisgenerallyoccurring
onearlierandwarmerdays.Thisconclusioncouldberelatedtothe'FalseSpring'occurrences.
Falsespringsarerelatedtotheglobalwarmingtrendinthewaythattheonsetofspringhasgenerally
72
Figure4.9:
Thetrendoffreezedamageoccurringdateateachgrowthstagefrom1981to2018.
shiftedearlierintheyearwithrisingaveragetemperatures.Asaconsequenceofearlierspringsand
increasingtemperatures,thedamageisoccurringearlierwithwarmertemperatures.
73
Figure4.10:
Theaveragedailyminimumtemperatureduringdamageoccurringdateateach
growthstagefrom1981to2018.
74
Figure4.11:
Thetrendofdailyminimumtemperatureduringdamageoccurringdateateach
growthstagefrom1981to2018.
75
FollowingthedivisionofthesubregionsinChapter3,wecalculatethedamagedaysaveraged
acrossdi˙erentsubregions(Figure4.12).We˝rstcalculatetheannualdamagedaysateachgrid
pointandthencomputethespatialaverageineachsubregion.Therefore,eachsubregionhasa
timeseriesofdamagedays.We˝nddamagedaysshowanabrupthighvaluein2012intheUpper
MidwestandtheNortheast,whichisconsistentwiththeseveredamagetocropsreportedduring
thatyear.Theannualmeandamagedaysandstandarddeviationsrangefrom1.56daysyr-1and
0.99yr-1overtheUpperMidwestto7.88dayyr-1and4.85daysyr-1overtheSouthernGreat
Plains.Thetimeseriesofdamagedaysacrossdi˙erentsubregionssuggeststhattheSouthernGreat
PlainsandtheSoutheastarenotfavorableforapplegrowth,whereastheUpperMidwestismost
suitableforapplegrowthasfarastheriskoffreezedamageisconcerned.Althoughonlyasmall
partoftheNortheastiscoveredinourstudy,theresultshowsthattheNortheastisalsofavorable
forappleplanting.
Tofurtherexaminetheregionaldi˙erencesofthedamagedaysforeachgrowthstage,wecount
theannualdamagedaysovereachgrowthstageateachgridpointandaverageoverallgridpoints
ineachsubregion.Wethensumtheannualmeanoverthe38yearsforeachsubregion,andthe
resultsareshowninFigure4.13.First,theresultsrevealthattherearemoredamagedaysoccurring
inthe˝rsttwophenologicalstagessinceweknowthatcropsaremorevulnerableatearlygrowing
stages.TheUpperMidwestandtheNortheastshowlowervaluesforallninestagescomparedto
otherregions.Also,therearenodamagedaysduringStage0fortheSouthernGreatPlainsand
theSoutheastandalmostnodamagedaysduringStage0fortheOhioValleyandtheNortheast.
Formostregions,thelargestnumberofdamagedaysoccurduringStage2,whilefortheUpper
Midwest,mostdamagedaysoccurduringStage0.Damagedaysforotherstagesvaryfromregion
toregion.
Asimilarcalculationisdoneforannualdamagevaluestounderstandhowdamageintensity
varieswithstageandregion(Figure4.14).ItturnsoutthattheUpperMidwestandtheNortheast
notonlyhavefewerdamagedaysbutalsoshowlowerdamageintensity,whichiscontrarytothe
SouthernGreatPlainsandtheSoutheast.Also,theresultsrevealthatStage2andStage3showthe
76
Figure4.12:
Thetimeseriesofarea-averageddamagedaysinsixsubregionsfrom1981to2018.
largestdamageintensityofallninestagesforalloursubregions.Therefore,forthesixsubregions
exceptfortheUpperMidwest,Stage2notonlyshowsthemostdamagedaysbutalsoshowsthe
mostintensedamage.AlthoughStage0showsthemostdamagedaysintheUpperMidwest,the
damagevaluesforStage0arenotlarge.Hence,Stage2isthephenologicalstagethatconsiderable
damagemightbemostlikelytooccur.
Theregion-andstage-dependenceofdailyminimumtemperature(Figure4.15)showsageneral
increasingtrendfromtheearlierstagestothelaterstages.However,fortheNorthernGreatPlainand
theUpperMidwest,thedailyminimumtemperaturesduringStage8arehigherthanthoseduring
Stage9.Also,forthetwoconsiderabledamageoccurringregions,theSouthernGreatPlainsand
theSoutheast,thedailyminimumtemperaturesduringStage9increasealmosttwicefromStage8,
indicatingasigni˝canttemperaturechange.CombinedwithFigure4.14,weconcludethatnearly
77
Figure4.13:
Thesumofdamagedaysateachgrowthstageinsixsubregionsfrom1981to2018.
nosigni˝cantdamageoccursduringStage9asthetemperatureincreases.
Thespatialandtime-averageddailyminimumtemperatureondamagedaysonly(Figure4.16)
revealthatthedailyminimumtemperaturesduringdamagedaysoftheUpperMidwestandthe
Northeastaregenerallyhigherthaninotherregions,consistentwiththeconclusionthatthese
tworegionshavetheleastdamage.AtStage2,thephenologicalstagewhenthemostsevere
damageoccurs,theUpperMidwestandtheNortheastshowalmost1DegreeCelsiushigherthan
theSouthernGreatPlain,wherethemostseveredamageoccurs.Thedailyminimumtemperatures
duringdamagedaysatStage2forthetwoleastfreeze-damagedregionsareabove-6
°
C,while
thoseforotherregionsareallbelow-6
°
C.Therearesharpdailyminimumtemperatureincreases
duringdamagedaysfromStage2toStage3andfromStage4toStage5.
Finally,thespatialandtime-averageddamage-occurringdatesforeachgrowthstage(Figure
78
Figure4.14:
Theaverageofdamagevaluesateachgrowthstageinsixsubregionsfrom1981to
2018.
4.17)revealalargeincreasefromanaverageJuliandateof20forStage0toanaverageJuliandate
ofmorethan60forStage2.ThenfromStage2toStage9,thedatesincreasegradually.Thetwo
mostseveredamage-occurringregions,theSouthernGreatPlainsandtheSoutheastshowmuch
earlierdamage-occurringdatescomparedtootherregions.
79
Figure4.15:
Theaveragedailyminimumtemperatureateachgrowthstageinsixsubregionsfrom
1981to2018.
80
Figure4.16:
Theaveragedailyminimumtemperatureduringdamagedaysateachgrowthstagein
sixsubregionsfrom1981to2018.
81
Figure4.17:
Theaveragedamage-occurringdatesateachgrowthstageinsixsubregionsfrom
1981to2018.
82
4.4Summary
Thisstudyinvestigatesthehistoricalimpactsofspringtimefreezeeventdamageoncropsin
thecentralandmidwesternUSfrom1981to2018usingcropsimulationmodelswithinputdatasets
fromPRISM.Inparticular,thevariablesexaminedinthisstudyincludethe˝rstgreendates,bloom
dates,poorpollinationdays,andthebudsurvivalchanceoftheyearforapples.Also,damagedays
oftheyear,dailyminimumtemperature,anddamageoccurringdatesforeachphenologygrowing
stageareincludedtoexploretheimpactsoffalsesprings.Themajor˝ndingsareasfollows.
1)Trendanalysisfortheonsetdatesofspringsrevealsatendencytowardsearlier˝rstgreen
datesandbloomdatesduringourstudyperiod,indicatingearlierspringsinrecentdecades.This
earliershiftofspringscouldresultinmorefrequentfalsespringoccurrences.Also,theonsetof
springdatesismorevariableinthewarmersouthernregionsthanthecoldernorthernregions.
2)Higherbudsurvivalchancesandfewerpoorpollinationdayswithlowervariationaroundthe
GreatLakesRegionthanotherregionsintheMidwesternandCentralUSareshowninourresults,
indicatingthefavorablegrowingconditionsforapplesintheseareas.Thisresultisconsistent
withthefewerfreezedamagerisksintheUpperMidwestandtheNortheastwithlowervariability
comparedtotheSouthernGreatPlainsandtheSoutheast.
3)Damageisgenerallyoccurringonearlierandwarmerdays,whichcouldbeaconsequenceof
morefrequent'FalseSpring'occurrences.Asaresultoftheglobalwarmingtrend,earliersprings
andincreasingtemperaturesleadtodamageoccurringearlierwithwarmertemperatures.
4)Wealsoobservedamagedaysshowanabrupthighvaluein2012intheUpperMidwestand
theNortheast,whichisconsistentwithobservedseveredamageduringthen.Thisabruptchangeis
notobservedinotherde˝nedregions.
5)Therearemoredamagedaysinstage2withhigherseveritysinceapplesarevulnerable
duringtheearlygrowingstages.Thisinformationsuggeststhatifapplesbegintheirgrowing
seasonsearlier,theycouldsu˙erfromsigni˝cantdamageduetosubsequentfreezeevents.
6)TheMidwestandtheNortheastaretworegionsfavorableforappleplantingsincethese
tworegionsshowfewerdamagedayswithlowerdamageintensityandalsohigherdailyminimum
83
temperatureduringdamageoccurringdays,whiletheSouthernGreatPlainsandtheSoutheastare
intheopposite.
Thisstudysimulatesthehistoricalyieldsofapplestoshowtheimpactsoffalsespringswithsome
limitations.First,thecropsimulationmodelonlytakesinputsoftemperaturesandprecipitation,
whichmightbeproblematicsincecropgrowthcouldberelatedtomorecomplicatedfactors.
Second,howthissimulationmodelrepresentstheobservationsneedstobeevaluated.Some
observationalrecordscouldbeusedtotestifythemodel'sbehavior.Also,ifthismodelisreliable,
somefutureprojectionscouldbemadetoprovidesomevaluablepredictionsforappleplanters.
Despitethelimitations,thisstudyprovideshelpfulinformationforappleplantersaboutthedamage
occurringearlierwithwarmertemperaturesduetofalsesprings.
84
CHAPTER5
CONCLUSION
Thisresearchfocusesontheclimatologyofspringtimefreezesandtheirimpactsonagriculture,
especiallyforthecentralandeasternUnitedStates.Chapter2assessesERA5reanalysisand
PRISManalysisdatasetsbycomparingthemtostationobservationsatselectedlocations.Chapter
3investigatesthefrequency,severity,andpotentialcausesofspringtimefreezeeventoccurrences.
TheImpactsoffalsespringsandsubsequentfreezeeventsareexaminedinChapter4.
InChapter2,ERA5andPRISMreanalysisproductsarecomparedtostationobservations.
TheconclusionisthatthePRISMdataset,demonstratingabetteragreementoftemperatureswith
observations,couldbeusedasaproxytorepresentobservedfreezeevents.Rootmeansquared
errors(RMSDs)andbiasarecalculatedtoevaluatehowthegriddeddatasetsagreewiththe
stationobservations.Griddedreanalysisdatasetstendtoreducetheobserveddailycycles.Ata
dailylevel,ERA5underestimatesdailymaximumtemperaturewhileoverestimatesdailyminimum
temperature,resultinginreduceddailycycles.Incontrast,PRISMtendstooverestimateboth
dailyminimumandmaximumtemperature.Sincedailymaximumtemperaturesaremeasuredin
moremixedconditions,theobservationsshouldagreewiththearea-averagedgriddeddatasetsto
alargerextentthandailyminimumtemperatures.Bothgriddeddatasetsshowbetterresultsfor
dailymaximumtemperaturethandailyminimumtemperature,asshownintheloweramplitudesof
perturbationsofthetimeseriesandthesmallervaluesofRMSDsofdailymaximumtemperature.
Lakee˙ectsareshownbylargeRMSDsandBiasandalsolargeperturbationsoftimeseriesof
near-lakestationsforbothgriddeddatasets.PRISMshowsexcellentabilityincapturingdi˙erent
typesoffreezeeventsforbothwintertimeandspringtime.Generally,freezeeventsarecaughtina
betterwayduringwintertimethanotherseasons.
InChapter3,weinvestigatethefrequency,severity,aswellasclimatebackgroundofspringtime
freezeeventoccurrences.TrendanalysisandEOFanalysisareconductedtorevealthecharacteris-
ticsoffreezingdaysduringthespringtime.Ourresultsindicateageneraldecreasingtrendinthe
85
frequencyofspringtimefreezeevents.Considerablydamageyearsof2007and2012showalow
frequencyoffreezeevents,indicatingtheprofoundimpactsoffalsespringsthatexposetheearlier
developedcropstothesubsequentfreezeevents.Bydividingourstudyregionintosixsubregions,
weobservesigni˝cantdownwardtrendsforthetotalfreezingdaysduringthespringtimeinthe
OhioValley,whichextendsthe˝ndingsinEasterlingetal.(2002)fromtheperiod1948-1999
into1981-2019.Byrelatingthearea-averagedfreezingdaystosometeleconnectionindices,we
concludethatthepositivephaseofNAOisusuallyassociatedwithlessfreezingriskinMarch
acrossthestudyregion.EOFanalysisoffreezingdaysinMarchshowsarelativelylargervariation
intheOhioValley,andthe˝rstEOFtimeseriesshowssubstantialinterannualvariability.Whena
positivephaseofNAOshowsinMarchandDecemberofthepreviousyear,thein-phase˛uctuation
acrossourstudyregionwithahigherintensityovertheOhioValleytendstooccurinMarch.And
thispatternisusuallyassociatedwiththegeopotentialheightlowerthanaverageattheupperlevel
andhigherthanaverageatthesurfaceandathicknessshallowerthanaverage.
Chapter4investigatesthehistoricalimpactsofspringtimefreezeeventdamageoncropsinthe
centralandmidwesternUSfrom1981to2018usingcropsimulationmodelswithinputdatasets
fromPRISM.Ourresultsrevealthattheonsetofspringsisshiftedearlierinrecentdecadeswith
theindicatorsof˝rstgreendatesandbloomdatesthataremorevariableinthewarmersouthern
regionsthanthecoldernorthernregionsduringourstudyperiod.Thisearliershiftofspringsis
associatedwithmorefrequentfalsespringoccurrences.ResultsalsorevealthattheUpperMidwest
andtheNortheastarelessvulnerabletofreezedamageforappleplantingthantheSouthernGreat
PlainsandtheSoutheastindicatedbyhigherbudsurvivalchances,andfewerfreezedamagerisks
withlowervariability.Damageisgenerallyoccurringonearlierandwarmerdaysasaconsequence
ofmorefrequent'FalseSpring'occurrences.TherearemoredamagedaysinStage2withhigher
severitysinceappleisvulnerableduringtheearlygrowingstages.
Therearesomelimitationsinthisresearch,asdiscussedinmoredetailintheindividualchapters.
Windspeedanddirectionthathaveprofoundimpactsonfreezeeventformationarenotevaluatedin
Chapter2.Also,precipitationdataneedstobeassessedbetweenthegriddeddatasetsandstational
86
observations.InChapter3,limitedbyourstudyregion,oursubregionscouldnotexactlyrepresent
theregionsde˝nedinKarlandKnight(1998).Speci˝cally,theNortheast,theSoutheast,andthe
SouthernGreatPlainsareonlypartsoftheregionsde˝nedinKarlandKnight(1998),leadingto
amisrepresentationoftheconclusionsoffreezeeventsovertheseregions.Inaddition,speci˝c
weatherconditionsneedtobeinvestigatedtounderstandthecausesoffreezeeventformation.In
Chapter4,thecropyieldsimulationmodelshouldbenotonlyevaluatedbutalsoimproved.Also,
somefuturetrendsofdamageriskscouldbeprojectedtoprovideguidelinesforfruitgrowers.
Inconclusion,themostvaluableinformationthatcouldprovidefruitgrowersisthatassociated
withglobalwarming,fewerspringtimefreezeeventsbutmorefrequentfalsespringeventsoccur
inrecentdecades.Thedamagetocropsoccursonearlierandwarmerdaysinrecentdecades.
87
APPENDIX
88
A.1ImpactsofSoilMoistureonSpringtimeFreezeEvents
Thethermalconductivityofthesoildependsonitsmineralogicalcomposition,texture,as
wellasitswatercontent(DeVriesetal.,1986).Sincesoilmoisturechangehasaprofounde˙ecton
soiltemperature,soilre˛ectance,andsoilheatstorage(AL-KAYSSletal.,1990),itisquestionable
howthesoilmoisturechangea˙ectstheapparenttemperature,whichiscloselyrelatedtotheheat
balancebetweenthegroundandaboveatmosphere.Whensoilsholdmorewater,theabilitytostore
andtransferheatduringthenightisenhanced.TheresultsfromAL-KAYSSletal.(1990)reveal
thatanincreaseinsoilmoisturecontentdecreasesthesoiltemperaturedi˙erencebetweenday-time
andnight-time.Andthisdecreasedtemperaturechangecouldprotectplantrootsystemsfromsharp
andsuddenchangesinsoiltemperature(AL-KAYSSletal.,1990).Finally,plantgrowthrateand
yieldmightincreaseduetothemodi˝cationofplantclimateathighersoilmoisturecontent.This
valuableinformationcouldprovidecropgrowerswithinstructionsonwhenandhowtoprotecttheir
cropsconfrontingapredictedsuddenandextremefreezeevent.
Thisstudyinvestigatestheimpactsofincreasingsoilmoisturecontenton2mairtemperature
change.Withtheapplicationofadeterministicregional-scalenumericalforecastmodel,thisstudy
aimstoinvestigatetheextentofhowthesoilmoisturechangea˙ectsthedailyminimumtemperature
changeintheGreatLakesRegion.
WeuseNAM(theNorthAmericanMesoscaleForecastSystem)analysestoinitiateour
modelcon˝guration.NAMisoneofthemajorweathermodelsrunbytheNationalCenters
forEnvironmentalPrediction(NCEP)forproducingdozensofweathervariables,fromtemper-
atureandprecipitationtoturbulentkineticenergy.TheNAMgeneratesmultiplegrids(ordo-
mains)ofweatherforecastsovertheNorthAmericancontinentatvarioushorizontalresolutions
(https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/north-american-mesoscale-forecast-
system-nam).WetakeinputdatafromNAManalysesandthenuseWRF(WeatherResearchand
Forecasting)modeltoforecasthowthetemperatureissensitivetosoilmoisturechange.
Speci˝cally,weconductWRFsensitivitystudiestoinvestigatethesensitivitytochanging
verticalresolutionandchanginginitialsoilmoisturesettings.Weobservetwonoticeablefreeze
89
eventsonthenightsofMay8thandMay12th.Therefore,basedonthesetwoevents,weconduct
aseriesofstudies.Beforewemodifyanyparameters,we˝rstcomparetheWRFmodeloutputs
toobservationstoknowtherepresentationoftheWRFoutputs.AsshowninFigureA.1.1,the
WRFoutputsmeasuresoilmoisturevaluesatdi˙erentlayerscomparedtotheobservationaldataat
Williamsburg.TheresultsrevealthattheWRFsimulationsconsistentlyoverestimatethemoisture
valuesattheuppersoillevelsandunderestimatethematthebottomlayerforboththesetwofreeze
events(May8thandMay12th)atthislocation.Thenwecompareotherweathervariablesatthree
stations,EastLansing,Gaylord,andWilliamsburg.Theresultsrevealthatforbothtwofreeze
events,WRFoutputscapturethedailycycleswellforallthreevariables,airtemperature,relative
humidity,andwindspeed,andthedi˙erencevaluesvarybylocation.WecompareWRFair
temperaturesat2mtoobservationalairtemperaturesat1.5m(FigureA.1.2).Thedi˙erencesrange
from-5degreesto5degrees.Generally,WRFoutputsunderestimatemaximumairtemperatures
intheafternoonandoverestimateminimumairtemperaturesduringthenight.Thisbiasisjust
whatweexpectsinceWRFoutputsarearea-averagedresults,whichshouldreducethedailycycle
amplitudes.Itisnoteworthythatinthegraph,wenotethetimefromWRFoutputs,whichisthe
standardUTCthatisfourhoursearlierthanthelocaleasterntime.WRFdoesnotshowexcellent
behaviorinrepresentingtherelativehumiditysincerelativehumidityissensitivetotemperature
values.Forwindspeedcomparison,wealsoshowthatWRFoutputstendtooverestimatethe
minimumvelocityduringthenightwhiletendtounderestimatethemaximumvelocityinthe
afternoon.Also,itisnoteworthythatwearecomparingWRFwindspeedat10mtoobservational
windspeedat3m.
SincePRISMhasbeenevaluatedinChapter2,wechoosetousePRISMasaproxytorepresent
theobservationaldailyminimumtemperaturevaluestoassessWRFoutputs'quality.Asshown
inFigureA.1.3,WRFcapturesthespatialvariationonMay8th.Weusealinearmodel,OLS
(ordinaryleastsquares)regressionmodel,toquantitativelyinvestigatehowtheWRFoutputs˝t
theobservationvaluesfromPRISM.OurresultsrevealthatWRFoutputsrepresent68.3%,27.1%,
31.8%,and1.1%,respectively,forMay8th,May9th,May12th,andMay13th,asshowninthe
90
FigureA.1.1:
Di˙erencesofsoilmoisturebetweenWRFoutputsandobservationsat
Williamsburg.
EastLansing
Gaylord
Williamsburg
30levels
2.172
2.1566
1.898
38levels
2.3653
2.0065
1.9866
40levels
2.3213
2.068
1.3364
45levels
2.3725
2.0733
1.3687
TableA.1.1:
RMSDsofairtemperaturebetweenWRFandobservationsatthreelocationsfor48
hoursfromMay12thtoMay13thwithdi˙erentverticalresolutions.
RsquaredvaluesfromtheOLSmodel.ExceptforMay13th,WRFgenerallycapturesthespatial
characteristicsovertheGreatLakesRegion.
Thenwetrytounderstandhowthehigherverticalresolutiona˙ectstherepresentationofWRF
outputs.Bymodifyingtheinitialverticallayersetting,werunWRFwith30levels,38levels,
91
FigureA.1.2:
Di˙erencesofairtemperaturebetweenWRFoutputsandobservationsatthree
locations.
FigureA.1.3:
ComparisonofdailyminimumairtemperaturebetweenWRFandPRISMonMay
8th.
92
EastLansing
Gaylord
Williamsburg
30levels
1.6087
1.6753
1.5199
38levels
1.5629
1.6919
1.8894
40levels
1.267
1.1656
1.6803
45levels
1.3944
1.3109
1.8251
TableA.1.2:
RMSDsofwindspeedbetweenWRFandobservationsatthreelocationsfor48hours
fromMay12thtoMay13thwithdi˙erentverticalresolutions.
40levels,and45levels.Weincreasetheverticalresolutionatlowerlevels.Wedonotobserve
ageneralimprovementinthecomparisonbetweenmodeloutputsandobservationsatthethree
selectedlocationsforairtemperatureandwindspeedfor48hoursfromMay12thtoMay13th.
Varyingbylocationandvariables,increasingtheverticalresolvedresolutionimprovestemperature
representationinGaylordandWilliamsburgandwindspeedrepresentationinEastLansingand
Gaylord(TableA.1.1andTableA.1.2).Itisinterestingtoshowthatwhenincreasingthelayers
from40to45,usuallytheRMSDsarenotbecomingsmaller;instead,largerRMSDsareobserved
inoursimulations.FromtheresultsoftheOLSmodel,wealsorevealthatincreasingthevertical
resolutiondoesnotnecessarilyimprovetherepresentationofthespatialcharacteristicsofWRF.
Next,weexaminehowtheairtemperatureisa˙ectedbyanincreaseinsoilmoisture.We˝rst
runtheWRFsimulationwithoutanychangesintheinitialinputdata.Thenwegettheoriginal
dailyminimumairtemperatureforfourdays,asshownintheleftcolumnofFigureA.1.4.Then
wechangethesoilmoisturevaluesintheinputdatabymultiplyingtheoriginalvaluesby1.05,
increasingtheinitialsoilmoistureby5percent.Theresultsarejustasweexpect;thedaily
minimumairtemperaturesonfourdi˙erentdaysareallincreasedlessthan1degree.However,
itisnotalwaysphysicallyreasonabletoincreasethesoilmoisturebymultiplyingafactor.As
showninFigureA.1.5,themaximumandminimumsoilmoisturevaluesactuallydependonthe
soilcategories.Forexample,sandyloamisthetypeofsoilthatshowsthebestabilitytostorewater.
Sowealsochangethesoilmoisturevaluestothemaximumandminimumvaluesbasedonthe
soiltypes(FigureA.1.6andFigureA.1.7).Theresultsrevealthatdecreasingsoilmoisturetothe
minimumvaluesconsiderablydecreasesdailyminimumtemperatureswithamaximumamplitude
93
FigureA.1.4:
Dailyminimumtemperaturechangesinresponsetoincreasedsoilmoisturebya
factorof1.05onMay8thandMay9th.
of8degrees,whichislargerthanthemagnitudeofchangingsoilmoisturetomaximumvalues.
Inconclusion,thiscasestudyprovidesvaluableinformationforcropgrowersthatirrigation
beforepredictedfreezeeventscouldhelptoprotectcropsfromseveredamageduetothepositive
responseoftemperaturetoincreasedsoilmoisture.
94
FigureA.1.5:
Dailyminimumtemperaturechangesinresponsetoincreasedsoilmoisturebya
factorof1.05onMay12thandMay13th.
95
FigureA.1.6:
Majorsoilcategoriesinourstudyregion.
FigureA.1.7:
SoilmoisturechangebasedonsoiltypesonMay8thandMay9th.
96
FigureA.1.8:
SoilmoisturechangebasedonsoiltypesonMay12thandMay13th.
97
BIBLIOGRAPHY
98
BIBLIOGRAPHY
Andresen,J.A.andW.J.Baule,2018.TemperateFruitTreesandGrapes.In:AgronomyMono-
graphs,60:Agroclimatology:LinkingAgriculturetoClimate.Eds.:J.Hat˝eld,M.Sivakumar,J.
Prueger,.ASA,CSSA,andSSSA,Madison,WI.doi:10.2134/agronmonogr60.2014.0035.
AboutUs-MSUEnviroweather-WeatherforIPMAccessedMarch6,2020.
https://enviroweather.msu.edu/about.php.
Allstadt,AndrewJ,StephenJVavrus,PatriciaJHeglund,AnnaMPidgeon,WayneEThog-
martin,andVolkerCRadelo˙.ingPlantPhenologyandFalseSpringsintheConterminousUS
duringthe21stCenturyEnvironmentalResearchLetters10,no.10(October1,2015):104008.
https://doi.org/10.1088/1748-9326/10/10/104008.
AL-KAYSSl,AW,AAAl-Karaghouli,andAMHasson.ofSoilMoistureContent
onSoilTemperatureandHeatStorageunderGreenhouseConditions,n.d.,12.
Augspurger,CarolK.econstructingPatternsofTemperature,Phenology,andFrostDam-
ageover124Years:SpringDamageRiskIsEcology94,no.1(2013):
https://doi.org/10.1890/12-0200.1.
AutomatedSurfaceObservingSystem(ASOS)|NationalCentersforEnvironmentalInforma-
tion(NCEI)FormerlyKnownasNationalClimaticDataCenter(NAccessedApril5,2021.
https://www.ncdc.noaa.gov/data-access/land-based-station-data/land-based-datasets/automated-surface-
observing-system-asos.
ing2007WarmthandFrost:Phenology,DamageandRefoliationinaTemperate
DeciduousForesFunctionalEcology23,no.6(2009):https://doi.org/10.1111/j.1365-
2435.2009.01587.x.
Ault,T.R.,G.M.Henebry,K.M.deBeurs,M.D.Schwartz,J.L.Betancourt,andD.Moore.
TheFalseSpringof2012,EarliestinNorthAmericanREos,TransactionsAmerican
GeophysicalUnion94,no.20(2013):https://doi.org/10.1002/2013EO200001.
Ballard,J.,E.Proebsting,andR.Tukey,1998.CriticalSpringTemperaturesForTreeFruitBud
DevelopmentStages.WashingtonStateUniversityExtensionBulletinEB1128,WashingtonState
University,Pullman,WA.
Bosilovich,MichaelG.,JohnKennedy,DickDee,RobAllan,andAlanO'Neill.theRe-
processingandReanalysisofObservationsforInClimateScienceforServingSociety:
Research,ModelingandPredictionPriorities,editedbyGhassemR.AsrarandJamesW.Hurrell,
Dordrecht:SpringerNetherlands,2013.https://doi.org/10.1007/978-94-007-6692-1-3.
99
DennisFGandHowellGS,1974.Coldhardinessoftartcherrybarkand˛owerbuds.Res.
Rep.220,MichiganStateUniversity,Ag.Exp.Stn.,EastLansing,MI,USA.
Menzel,Annette.TrendsinPhenologicalPhasesinEuropebetween1951andInterna-
tionalJournalofBiometeorology44,no.2(August1,2000):https://doi.org/10.1007/s00484-
0000054.
MuñozSabater,J.,(2019):ERA5-Landhourlydatafrom1981topresent.CopernicusClimate
ChangeService(C3S)ClimateDataStore(CDS).
Cannon,D.J.,D.J.Brayshaw,J.Methven,P.J.Coker,andD.Lenaghan.ReanalysisData
toQuantifyExtremeWindPowerGenerationStatistics:A33YearCaseStudyinGreatBr
RenewableEnergy75(March1,2015):https://doi.org/10.1016/j.renene.2014.10.024.
Cayan,DanielR.,SusanA.Kammerdiener,MichaelD.Dettinger,JosephM.Caprio,andDavid
H.Peterson.esintheOnsetofSpringintheWesternUnitedSBulletinoftheAmer-
icanMeteorologicalSociety82,no.3(March1,2001):https://doi.org/10.1175/1520-
0477(2001)082<0399:CITOOS>2.3.CO;2.
Chai,T.,andR.R.Draxler.ootMeanSquareError(RMSD)orMeanAbsoluteError(MAE)?
ArgumentsagainstAvoidingRMSDintheGeoscienti˝cModelDevelopment7,no.
3(June30,2014):https://doi.org/10.5194/gmd-7-1247-2014.
Charrier,Guillaume,JérômeNgao,MarcSaudreau,andThierryAméglio.ofEn-
vironmentalFactorsandManagementPracticesonMicroclimate,WinterPhysiology,andFrost
ResistanceinTFrontiersinPlantScience6(2015).https://doi.org/10.3389/fpls.2015.00259.
Daly,Christopher.forAssessingtheSuitabilityofSpatialClimateData
InternationalJournalofClimatology26,no.6(2006):https://doi.org/10.1002/joc.1322.
Daly,Christopher,MichaelHalbleib,JosephI.Smith,WayneP.Gibson,MatthewK.Doggett,
GeorgeH.Taylor,JanCurtis,andPhillipP.Pasteris.ysiographicallySensitiveMappingofCli-
matologicalTemperatureandPrecipitationacrosstheConterminousUnitedSInternational
JournalofClimatology28,no.15(2008):https://doi.org/10.1002/joc.1688.
Dee,D.P.,S.M.Uppala,A.J.Simmons,P.Berrisford,P.Poli,S.Kobayashi,U.Andrae,
etal.TheERA-InterimReanalysis:Con˝gurationandPerformanceoftheDataAssimilation
SysQuarterlyJournaloftheRoyalMeteorologicalSociety137,no.656(2011):
https://doi.org/10.1002/qj.828.
DeGaetano,ArthurT.ecentTrendsinMaximumandMinimumTemperatureThresholdEx-
ceedencesintheNortheasternUnitedSJournalofClimate9,no.7(July1,1996):
https://doi.org/10.1175/1520-0442(1996)009<1646:RTIMAM>2.0.CO;2.
100
DeVriesD.A.,Thermalpropertiesofsoils.WijkW.R.(Ed.),PhysicsofPlantEnvironment,
North-HollandPublishingCo.,Amsterdam(1963),pp.210-235
Dulière,Valérie,YongxinZhang,andEricP.Salathé.PrecipitationandTemperature
overtheU.S.Paci˝cNorthwest:AComparisonbetweenObservations,ReanalysisData,andRe-
gionalJournalofClimate24,no.7(January7,2011):64.https://doi.org/10.1175/-
2010JCLI3224.1.
Easterling,DavidR.CHANGESINFROSTDAYSANDTHEFROST-FREESEA-
SONINTHEUNITEDSTATES,n.d.,6.
Essou,GillesR.C.,FrançoisBrissette,andPhilippeLucas-Picher.TheUseofReanaly-
sesandGriddedObservationsasWeatherInputDataforaHydrologicalModel:Comparisonof
PerformancesofSimulatedRiverFlowsBasedontheDensityofWeatherSJournalof
Hydrometeorology18,no.2(February1,2017):https://doi.org/10.1175/JHM-D-16-
0088.1.
Fitchett,JenniferM.,StefanW.Grab,DaveI.Thompson,andGholamrezaRoshan.
creasingFrostRiskAssociatedwithAdvancedCitrusFloweringDatesinKermanandShiraz,
Iran:InternationalJournalofBiometeorology58,no.8(October2014):
https://doi.org/10.1007/s00484-013-0778-0.
MSUExtension.DamageDependsonTreeFruitStageofDevAccessed
March17,2021.https://www.canr.msu.edu/news/freeze_damage_depends_on_tree_fruit_stage_of-
_development.
ProtectionAccessedFebruary11,2020.https://ssl.acesag.auburn.edu/-
department/peaches/freeze.html.
Gleixner,Stephanie,TeferiDemissie,andGulilatTeferaDiro.ERA5ImproveTempera-
tureandPrecipitationReanalysisoverEastAfrAtmosphere11,no.9(September17,2020):
996.https://doi.org/10.3390/atmos11090996.
Gu,Lianhong,PaulJ.Hanson,W.MacPost,DaleP.Kaiser,BaiYang,RamakrishnaNe-
mani,StephenG.Pallardy,andTildenMeyers.The2007EasternUSSpringFreeze:In-
creasedColdDamageinaWarmingWorBioScience58,no.3(March1,2008):
https://doi.org/10.1641/B580311.
Hat˝eld,Jerry,andMariaJanowiak.AMidwestandNorthernForestsRegionalClimate
Hub:AssessmentofClimateChangeVulnerabilityandAdaptationandMitigationStrategies,n.d.,
55.
Hufkens,Koen,MarkA.Friedl,TrevorF.Keenan,OliverSonnentag,AmeyBailey,John
O'Keefe,andAndrewD.Richardson.ImpactsofaWidespreadFrostEventFollowing
EarlySpringGlobalChangeBiology18,no.7(2012):https://doi.org/10.1111-
101
/j.1365-2486.2012.02712.x.
IntegratedPestManagement.viroweatherWeatherDataandPestAccessed
April5,2021.https://www.canr.msu.edu/resources/enviroweather_weather_data_and_pest_modeling.
Janowiak,MariaK,DanielNDostie,MichaelAWilson,MichaelJKucera,RHowardSkinner,
JerryLHat˝eld,DavidHollinger,andChristopherWSwanston.ADAPTATIONRESOURCES
FORAGRICULTURE,n.d.,72.
JJ.andRMSDWhichMetricIsMedium,March23,2016.https://medium.com-
/human-in-a-machine-world/mae-and-RMSD-which-metric-is-better-e60ac3bde13d.
Karl,ThomasR.,andRichardW.Knight.TrendsofPrecipitationAmount,Frequency,
andIntensityintheUnitedSBulletinoftheAmericanMeteorologicalSociety79,no.2
(February1,1998):https://doi.org/10.1175/1520-0477(1998)079<0231:STOPAF>2.0.CO;2.
Katz,RichardW.,andBarbaraG.Brown.EventsinaChangingClimate:Vari-
abilityIsMoreImportantthanAveragClimaticChange21,no.3(July1992):
https://doi.org/10.1007/BF00139728.
Kidston,J.,D.M.W.Frierson,J.A.Renwick,andG.K.Vallis.vations,Simulations,
andDynamicsofJetStreamVariabilityandAnnularJournalofClimate23,no.23(2010):

Kistner,Erica,OliviaKellner,Je˙reyAndresen,DennisTodey,andLoisWrightMorton.ul-
nerabilityofSpecialtyCropstoShort-TermClimaticVariabilityandAdaptationStrategiesinthe
MidwesternUSClimaticChange146,no.(January2018):https://doi.org/10.1007-
/s10584-017-2066-1.
Kral-O'Brien,KatherineC.,PeterL.O'Brien,andJasonP.Harmon.eedforFalseSpring
ResearchintheNorthernGreatPlains,USAgricultural&EnvironmentalLetters4,no.1
(January2019):https://doi.org/10.2134/ael2019.07.0025.
Kunkel,KennethE.,DavidR.Easterling,KennethHubbard,andKellyRedmond.Temporal
VariationsinFrost-FreeSeasonintheUnitedStates:GeophysicalResearchLetters
31,no.3(2004).https://doi.org/10.1029/2003GL018624.
Kurlus,Robert,LukaszMalarzewski,andTadeuszNiedzwiedz.TemperatureCharacteristics
andSpringFrostRiskintheSourCherry(PrunusCerasusL.)BloomingPeriodin
BulletinofGeography.PhysicalGeographySeries6,no.1(December1,2013):
https://doi.org/10.2478/bgeo-2013-0006.
Labe,Zachary,TobyAult,andRaulZurita-Milla.AnomalouslyEarlySpring
OnsetsintheCESMLargeEnsembleClimateDynamics48,no.(June2017):
102
https://doi.org/10.1007/s00382-016-3313-2.
Larcher,Walter,ChristineKainmüller,andJohannaWagner.vivalTypesofHighMoun-
tainPlantsunderExtremeTFlora-Morphology,Distribution,FunctionalEcology
ofPlants205,no.1(January1,2010):https://doi.org/10.1016/j.˛ora.2008.12.005.
Lawhorn,Brandon,andRobertC.Balling.TemporalandSpatialVariationsinFreezeWarn-
ingsintheConterminousUnitedStates:TheoreticalandAppliedClimatology139,
no.3(February1,2020):https://doi.org/10.1007/s00704-019-03056-8.
León,JoséAndrésPérez.Reanalysis:GoodbyeERA-Interim,HelloText.
ECMWF,April15,2019.https://www.ecmwf.int/en/newsletter/159/meteorology/global-reanalysis-
goodbye-era-interim-hello-era5.
Lindkvist,L.,Gustavsson,T.,Bogren,J.,2000.Afrostassessmentmethodformountainous
areas.AgriculturalandForestMeteorology102,doi:10.1016/S0168-1923(99)00087-8
Lobell,DavidB.,andChristopherB.Field.orniaPerennialCropsinaChanging
ClimaticChange109(December2011):http://dx.doi.org.proxy1.cl.msu.edu/10.1007/s10584-
011-0303-6.
Logan,Joanne,MarcellaA.Mueller,andMichaelJ.Searcy.PeachBud
Phenology,andFreezeRisksinaTopographicallyDiverseOrcHortTechnology10,no.2
(January2000):https://doi.org/10.21273/HORTTECH.10.2.337.
Longstroth,Mark.DamageDependsonTreeFruitStageofDevMSUExten-
sion,March2012.https://www.canr.msu.edu/news/freeze_damage_depends_on_tree_fruit_stage-
_of_development.
Marino,GarrettP,DalePKaiser,LianhongGu,andDanielMRicciuto.econstructionof
FalseSpringOccurrencesovertheSoutheasternUnitedStates,AnIncreasingRisk
ofSpringFreezeDamagEnvironmentalResearchLetters6,no.2(April1,2011):024015.
https://doi.org/10.1088/1748-9326/6/2/024015.
Maurer,E.P.,H.G.Hidalgo,T.Das,M.D.Dettinger,andD.R.Cayan.TheUtilityof
DailyLarge-ScaleClimateDataintheAssessmentofClimateChangeImpactsonDailyStream-
˛owinCaliforHydrologyandEarthSystemSciences14,no.6(June1,2010):
https://doi.org/10.5194/hess-14-1125-2010.
NASS,2015.CherryProduction.NationalAgriculturalStatisticsService,AgriculturalStatis-
ticsBoard,U.S.DepartmentofAgriculture.
NASS,2012.CherryProduction.NationalAgriculturalStatisticsService,AgriculturalStatis-
ticsBoard,U.S.DepartmentofAgriculture.
103
NASS,2002.CherryProduction.NationalAgriculturalStatisticsService,AgriculturalStatis-
ticsBoard,U.S.DepartmentofAgriculture.
orthAmericanMesoscaleForecastSystem(NAM)|NationalCentersforEnvironmentalIn-
formation(NCEI)FormerlyKnownasNationalClimaticDataCenter(NAccessedApril4,
2021.https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/north-american-mesoscale-
forecast-system-nam.
CSNationalWaterandClimateCenter-ClimateProducts-AccessedMarch
6,2020.https://www.wcc.nrcs.usda.gov/climate/prism.html.Parker,WendyS.eanalysesand
Observations:What'stheBulletinoftheAmericanMeteorologicalSociety97,no.9
(September1,2016):https://doi.org/10.1175/BAMS-D-14-00226.1.
erformanceMetricsforClimateModels-Gleckler-2008-JournalofGeophysicalResearch:
Atmospheres-WileyOnlineLibraryAccessedMarch17,2021.https://agupubs.onlinelibrary.wile-
y.com/doi/10.1029/2007JD008972.
Perry,K.B.,1998.Basicsoffrostandfreezeprotectionforhorticulturalcrops.HortTechnology
8,
Peterson,AlexanderG.,andJohnT.Abatzoglou.vedChangesinFalseSprings
overtheContiguousUnitedSGeophysicalResearchLetters41,no.6(2014):
https://doi.org/10.1002/2014GL059266.
Powell,ArlieA.andHimelrick,DavidProtectionAccessedApril5,
2021.https://ssl.acesag.auburn.edu/department/peaches/freeze.html.
overMonsoonAsia:AComparisonofReanalysesandObservationsin:Journal
ofClimateVolume30Issue2AccessedMarch12,2021.https://journals.ametsoc.org/view/-
journals/clim/30/2/jcli-d-16-0227.1.xml.
ClimateGroup,OregonStateUAccessedMarch30,2020.http://www.prism.oregon-
state.edu/.
Qu,Michael,JoeWan,andXianjunHao.AnalysisofDiurnalAirTemperatureRangeChange
intheContinentalUnitedSWeatherandClimateExtremes4(August1,2014):
https://doi.org/10.1016/j.wace.2014.05.002.
eanalyses.OrgHomePage|AdvancingReanalyAccessedMarch30,2020.https://reanaly-
ses.org/reanalysesorg-home-page.
Rigby,J.R.,andAmilcarePorporato.ingFrostRiskinaChangingGeophysical
ResearchLetters35,no.12(2008).https://doi.org/10.1029/2008GL033955.
104
Rijal,I.,2017.UseOfWaterMistToReduceTheRiskOfFrostDamageInTreeFruits.Ph.D.
Dissertation,Dept.ofGeography,MichiganStateUniversity,EastLansing,MI.
Rodrigo,J.ingFrostsinDeciduousFruitTreesMorphologicalDamageand˛owerHar-
ScientiaHorticulturae,2000,19.
Rowe,ClintonM.,andLoganE.Derry.TrendsinRecord-BreakingTemperaturesfortheCon-
terminousUnitedSGeophysicalResearchLetters39,no.16(2012).https://doi.org/10.1029/-
2012GL052775.
Skaggs,KariE.,andSuatIrmak.TermTrendsinAirTemperatureDistributionand
Extremes,GrowingDegree-Days,andSpringandFallFrostsforClimateImpactAssessmentson
AgriculturalPracticesinNJournalofAppliedMeteorologyandClimatology51,no.11
(November1,2012):https://doi.org/10.1175/JAMC-D-11-0146.1.
Tarek,Mostafa,FrançoisP.Brissette,andRichardArsenault.aluationoftheERA5Reanal-
ysisasaPotentialReferenceDatasetforHydrologicalModelingoverNorth-AmerHydrology
andEarthSystemSciencesDiscussions,July10,2019,https://doi.org/10.5194/hess-2019-
316.
Tetzner,Dieter,ElizabethThomas,andClaireAllen.AValidationofERA5ReanalysisData
intheSouthernAntarcticPorthLandRegion,andItsImplicationsforIceCore
SGeosciences9,no.7(June29,2019):289.https://doi.org/10.3390/geosciences9070289.
Toreti,A.,A.Maiorano,G.DeSanctis,H.Webber,A.C.Ruane,D.Fumagalli,A.Ceglar,S.
Niemeyer,andM.Zampieri.ReanalysisinCropMonitoringandForecastingSys
AgriculturalSystems168(January1,2019):https://doi.org/10.1016/j.agsy.2018.07.001.
.S.SurfaceClimateObservingReferenceNetwAccessedApril5,2021.https://www.-
ncdc.noaa.gov/crn/.
Walton,Daniel,andAlexHall.AnAssessmentofHigh-ResolutionGriddedTempera-
tureDatasetsoverCaliforJournalofClimate31,no.10(May15,2018):
https://doi.org/10.1175/JCLI-D-17-0410.1.
AretheDi˙erentMethodsofStatisticalDownscalingandWhatAretheStrengths/Weak-
nesses?|TheClimateWAccessedMarch4,2020.http://www.glisaclimate.org/faqs/what-
are-the-di˙erent-methods-of-statistical-downscaling-and-what-are-the-strengthsweaknesses.
IsERA5-CopernicusKnowledgeBase-ECMWFCon˛uenceWAccessedMarch
6,2020.https://con˛uence.ecmwf.int/display/CKB/What+is+ERA5.
Winkler,J.,Andresen,J.A.,Bisanz,J.M.,Guentchev,G.,Nugent,J.,Piromsopa,K.,Rothwell,
N.,Zavalloni,C.,Clark,J.,Min,H.K.,Pollyea,A.,Prawiranata,H.,2012.Ch.8Michigan'sSour
CherryIndustry:VulnerabilitytoClimateVariabilityandChange,in:Pryor,S.C.(Ed.),Climate
105
ChangeintheMidwest:Impacts,Risks,Vulnerability,andAdaptation.IndianaUniversityPress,
Bloomington,IN,pp.
Winkler,J.A.,Cinderich,A.B.,Ddumba,S.D.,Doubler,D.,Nikolic,J.,Perdinan,Pollyea,
A.M.,143Young,D.R.,Zavalloni,C.,2013.2.04-UnderstandingtheImpactsofClimateon
PerennialCrops,in:ClimateVulnerability:UnderstandingandAddressingThreatstoEssential
Resources.Elsevier,pp.
Yu,Lejiang,ShiyuanZhong,XindiBian,WarrenE.Heilman,andJe˙reyA.Andresen.Tem-
poralandSpatialVariabilityofFrost-FreeSeasonsintheGreatLakesRegionoftheUnitedS
InternationalJournalofClimatology34,no.13(November2014):https://doi.org/10.-
1002/joc.3923.
Zavalloni,C.,J.A.Andresen,andJ.A.FloreJ.A.WinklerJ.R.BlackandT.L.Beedy,2006.
ThePileusProject:ClimaticImpactsonSourCherryProductionintheGreatLakesRegioninPast
andProjectedFutureTimeFrames.ActaHort707:101-108.
106