Ansell, T. J. et al. (2006) Daily mean sea level pressure reconstructions for the European

26 Pages·2006·6.03 MB·English

15JUNE2006 ANSELL ET AL. 2717
Daily Mean Sea Level Pressure Reconstructions for the European–North Atlantic
Region for the Period 1850–2003
T. J. ANSELL,a P. D. JONES,b R. J. ALLAN,a D. LISTER,b D. E. PARKER,a M. BRUNET,c A. MOBERG,d
J.JACOBEIT,e P. BROHAN,a N. A. RAYNER,a E. AGUILAR,c H. ALEXANDERSSON,f M. BARRIENDOS,g
T.BRANDSMA,h N. J. COX,i P. M. DELLA-MARTA,j A. DREBS,k D. FOUNDA,l F. GERSTENGARBE,m
K.HICKEY,n T. JÓNSSON,o J. LUTERBACHER,p Ø. NORDLI,q H. OESTERLE,m M. PETRAKIS,l A. PHILIPP,e
M.J.RODWELL,r O. SALADIE,c J. SIGRO,c V. SLONOSKY,s L. SRNEC,t V. SWAIL,u A. M. GARCÍA-SUÁREZ,v
H. TUOMENVIRTA,k X. WANG,u H. WANNER,p P. WERNER,m D. WHEELER,w AND E. XOPLAKIp
aHadleyCentre,MetOffice,Exeter,UnitedKingdom
bCRU,UniversityofEastAnglia,Norwich,UnitedKingdom
cUniversitatRoviraiVirgili,Tarragona,Spain
dStockholmUniversity,Stockholm,Sweden
eUniversityofAugsburg,Augsburg,Germany
fSwedishMeteorologicalandHydrologicalInstitute,Norrköping,Sweden
gUniversityofBarcelona,Barcelona,Spain
hKNMI,DeBilt,Netherlands
iDurhamUniversity,Durham,UnitedKingdom
jUniversityofBern,Bern,Switzerland
kFinnishMeteorologicalInstitute,Helsinki,Finland
lNationalObservatoryofAthens,Athens,Greece
mPotsdamInstituteforClimateImpactResearch,Potsdam,Germany
nNationalUniversityofIreland,Galway,Galway,Ireland
oIcelandicMeteorologicalOffice,Reykjavik,Iceland
pUniversityofBern,andNCCRClimate,Bern,Switzerland
qTheNorwegianMeteorologicalInstitute,Oslo,Norway
rECMWF,Reading,UnitedKingdom
sMcGillUniversity,Montreal,Quebec,Canada
tMeteorologicalandHydrologicalService,Zagreb,Croatia
uEnvironmentCanada,Ontario,Canada
vArmaghObservatory,Armagh,Ireland
wUniversityofSunderland,Sunderland,UnitedKingdom
(Manuscriptreceived5May2005,infinalform13October2005)
ABSTRACT
The development of a daily historical European–North Atlantic mean sea level pressure dataset
(EMSLP)for1850–2003ona5°latitudebylongitudegridisdescribed.Thisproductwasproducedusing
86continentalandislandstationsdistributedovertheregion25°–70°N,70°W–50°Eblendedwithmarine
datafromtheInternationalComprehensiveOcean–AtmosphereDataSet(ICOADS).TheEMSLPfields
for1850–80arebasedpurelyonthelandstationdataandshipobservations.From1881,theblendedland
andmarinefieldsarecombinedwithalreadyavailabledailyNorthernHemispherefields.Completecov-
erageisobtainedbyemployingreducedspaceoptimalinterpolation.Squaredcorrelations(r2)indicatethat
EMSLPgenerallycaptures80%–90%ofdailyvariabilityrepresentedinanexistinghistoricalmeansealevel
pressureproductandover90%inmodern40-yrEuropeanCentreforMedium-RangeWeatherForecasts
Re-Analyses(ERA-40)overmostoftheregion.AlackofsufficientobservationsoverGreenlandandthe
MiddleEast,however,hasresultedinpoorerreconstructionsthere.Errorestimates,producedaspartofthe
reconstructiontechnique,flagtheseasregionsoflowconfidence.ItisshownthattheEMSLPdailyfields
andassociatederrorestimatesprovideauniqueopportunitytoexaminethecirculationpatternsassociated
withextremeeventsacrosstheEuropean–NorthAtlanticregion,suchasthe2003heatwave,inthecontext
ofhistoricalevents.
Correspondingauthoraddress:Dr.T.J.Ansell,HadleyCentre,MetOffice,FitzRoyRoad,Exeter,DevonEX13PB,UnitedKingdom.
E-mail:[email protected]
©2006AmericanMeteorologicalSociety
JCLI3775
2718 JOURNAL OF CLIMATE VOLUME19
1. Introduction Worley et al. 2005), the marine observations are con-
strained to the major shipping routes of the time, pre-
The European Community (EC)-funded European dominantlybetweennorthwestEuropeandtheAmeri-
and North Atlantic Daily to Multidecadal Climate casandfromnorthwestEuropetoaroundtheCapeof
Variability Project (EMULATE) began in November Good Hope. Similarly, in remote terrestrial regions
2002.AninitialaimofEMULATEwastodefinechar- suchascentralGreenlandandnorthernAfrica,nosta-
acteristicatmosphericcirculationpatternsovertheEu- tion data are available. In these areas, our fields are
ropean and North Atlantic region. Changes in mean based purely on reconstruction. Because of this, it is
amplitudes, variability, persistence, and transition re- important to be able to constrain analyses in these re-
gimes of these dominant patterns over a 154-yr period gions of low confidence, and accordingly, our Europe-
would then be assessed with both traditional and new an–North Atlantic mean sea level pressure dataset
statistical techniques. These variations and trends (EMSLP)isavailablewitherrorestimatestoguidethe
wouldberelatedtoseasurfacetemperature(SST)pat- researcher.
terns over the North Atlantic and worldwide, and to Quality control and gridding issues, central to this
natural and anthropogenic forcing factors, involving work, are described in sections 2 and 3, including the
various climate model integrations. A final aim was to useofinterpolationtechniquestoobtainspatiallycom-
relatethesetrendstoextremesintemperatureandpre- pletefields.Insection4,wecompareEMSLPtoexist-
cipitation over Europe. inganalysesandexamineitsabilitytoresolveextreme
Previousstudiesofthisnaturehavebeenlimitedbya events. In addition to improving the understanding of
lack of gridded mean sea level pressure (MSLP) prod- recent events [such as the 2003 summer heat wave in
uctsofsufficientlength.CentraltoEMULATE,there- westernEurope,theautumn2000floodsintheUnited
fore,hasbeenthedevelopmentofdailygriddedMSLP Kingdom(U.K.),andtheEuropeanfloodsinGermany,
fields over Europe and the extratropical North Atlan- Austria,andtheCzechRepublicin2002(Danubeand
tic, extending back to 1850. These fields will enable us Elberivers)],weexpectthatthisproductwillbeavalu-
to more fully examine whether relationships between able aid in the further understanding of historical
SSTandcirculationpatternsarestationary,andthento extremeeventsdatingbacktothemid-nineteenthcen-
morereliablyassesstherelativeimportanceofanthro- tury. Conclusions are given in section 5.
pogenic factors. They will also be used to study the
dynamicbackgroundsofextremeeventsandcirculation
extremes over a well-extended period. 2. Data sources and quality control
Here we present the development of this gridded
daily MSLP product on a 5° latitude by longitude grid OurstrategyhasbeentotakeadvantageofaNorth-
over the region 25°–70°N, 70°W–50°E (hereafter re- ernHemisphere(northof15°N)synopticdailygridded
ferred to as the EMULATE region). Daily gridded MSLPproductavailablefrom1881tothepresent(J86),
MSLP fields since 1881 for the Northern Hemisphere and to improve it while extending the analysis back to
north of 15°N are already available (Jackson 1986, 1850withtheinclusionofnewlandstationandmarine
hereafterreferredtoasJ86),butareonlyreliableover pressure observations. The J86 fields are an extremely
western and central Europe (e.g., Germany, Austria, valuable resource, as they are derived from synoptic
Switzerland, and northern Italy). These fields are im- operationalchartsandsoimplicitlycontainmanythou-
provedandextended,usingthelongstation–basedEu- sands of station observations. Because of this, we con-
ropean pressure series from earlier EC projects and centratedoureffortsoncollatinganddigitizingstation
recentlydigitizedlongstationrecords,particularlyover datafortheperiodbefore1881.TheEMSLPfieldsfor
EuropeandRussia.Overtheoceanwetakeadvantage 1850–80 are therefore based purely on a blending of
of the recently released International Comprehensive landstationdataandshipobservations;from1881,the
Ocean–AtmosphereDataSet(ICOADS;Worleyetal. blended land and marine fields are combined with the
2005;Diazetal.2002),usingallavailableobservations daily J86 fields.
overthe24-hperiod.Byblendingthesesourcesweare Because the number and times of observations per
able to produce daily fields from 1850. day varies markedly between stations, all the observa-
Thebiggestchallengetothisworkhasbeenthelack tions(includingthemarineandJ86fields)areadjusted
of daily observations, particularly before 1881. During to represent the 24-hourly mean. Accordingly, the
thisperiod,observationsareavailableoveronly15%of EMSLPdailyfieldsrepresenttheaveragepressureover
the EMULATE region. Despite the inclusion of the the 24-h period and so are different and indeed
recently digitized U.S. Maury collection (before 1863; smootherthansynopticand4(cid:1)dailyreanalysischarts.
15JUNE2006 ANSELL ET AL. 2719
the EMULATE Web site (www.cru.uea.ac.uk/cru/
projects/emulate/).
QUALITY CONTROL
Most of the 86 station series required some form of
qualitycontrolandhomogenization.Mostobservations
weremadewithmercurybarometers;anumberofcor-
rections are necessary for converting these measure-
FIG.1.Distributionofthe86continentalandislandstationsin mentsintoatruemeasureoftheatmosphericpressure.
EMSLP. Eighty-three station records begin between 1850 and
Thereadingfromamercurybarometer(usuallyinEn-
1880;Tasiilaq,Potsdam,andTenerifebeginafter1880.Theyears
glish inches or millimeters of mercury) is proportional
1850–80correspondtotheperiodwhennoJ86fieldsareavailable.
tothelengthofmercuryinacolumn,balancedagainst
the weight of the entire atmospheric column. The in-
struments were calibrated at “standard conditions,” so
a. Terrestrial data sources
correctionsmustbeappliedtoaccountforthethermal
The daily continental and island observations were expansionofmercuryandforthelocalgravityvalue.In
drawnfromanumberofsources.Dataalreadyinelec- most cases, the station data had been corrected at
tronic form were obtained for various Italian, Fenno- sourcetoastandardtemperatureof0°Candtoastan-
scandian, and U.K. stations compiled by earlier EC dardgravityequaltothatat45°N.SomedigitizedRus-
projects such as Improved understanding of past cli- sian series, however, had been calibrated at 13.33°C
maticvariabilityfromearlydailyEuropeaninstrumen- (e.g., Lugansk), so additional adjustments had to be
tal sources (IMPROVE; Camuffo and Jones 2002 and made. All pressures were converted to units of hecto
referencestherein)andWavesandStormsintheNorth Pascals (hPa).
Atlantic(WASA;Schmithetal.1997),andforthecities Whensealevelpressureswerenotavailable,thesta-
of Montreal, Canada (Slonosky 2003); Gibraltar, tion level pressures were corrected to mean sea level.
United Kingdom; De Bilt, Netherlands; Paris, France; An expression for the reduction of station level pres-
Palermo, Italy; and Galway, Ireland (Hickey et al. sure to sea level can be obtained by combining the
2003),byindividualefforts(seetheacknowledgments). hypsometric equations with the ideal gas equation for
Wewereabletoobtainupdatesandadditionalhistoric air(seeSlonoskyetal.2001).Thisconversionrequires
dataforsomeWASAstations,extendingthembackto the temperature reading. Daily temperature records
1850andforwardto2003.Considerablematerialhadto were not always available, so in these few cases clima-
be digitized, however, from individual hard copy tological temperature values were employed.
records from Russian, British, French, and Spanish Adailypressurevaluewasobtainedforeachstation
daily weather records (DWRs), held in the U.K. Na- by taking the average of all available observations for
tional Meteorological Archives and Library. These each day. The number and times of observations per
were supplemented by scanned Algerian, French, and day varied markedly across all 86 stations, however,
U.S. observations on the National Oceanic and Atmo- causing biases because atmospheric pressure has
spheric Administration (NOAA) Library Web site marked semidiurnal and diurnal variations. This arises
(http://docs.lib.noaa.gov/rescue/data_rescue_home- from internal gravity waves in the atmosphere, gener-
.html). Old American Bulletin of International Meteo- ated by atmospheric solar heating through the absorp-
rological Observations volumes also provided valuable tionofsolarradiation,andupwardeddyconductionof
records for Nuuk (Godthåb), Greenland, and helped heat from the ground (Chapman and Lindzen 1970).
fill gaps in existing records. Data were also digitized Over the EMULATE region, the amplitude of both
fromcompilationsmadeundertheauspicesoftheU.K. diurnal and semidiurnal oscillation (also referred to as
Board of Trade, Royal Engineers, and Army Medical atmospheric tides) is generally (cid:2)1.0 hPa (Dai and
Department, and from Ottoman Empire records. Wang 1999). While this is small compared with daily
Inall,86continentalandislandstationsovertheEu- variability (especially when compared with the Trop-
ropean–North Atlantic region (see Fig. 1 for land sta- ics), it is important to account for it.
tion distribution) were selected. A detailed list of the Inorderforthedailyfieldstobetterapproximatethe
individualstationserieslengthsisprovidedinTable1; “true”dailymean,eachstationwascorrectedforthese
thecorrespondingdatasourcesaredetailedinTable2. atmospherictides.Duetoalackofsufficientsubdaily
Both the “uncorrected” and quality-controlled daily data,however,wewereunabletocalculatethediurnal
stationdataseriesusedintheprojectareavailablefrom and semidiurnal cycle at each station directly. Instead,
2720 JOURNAL OF CLIMATE VOLUME19
TABLE1.Listof86pressuresitesusedinEMSLP.Thestartandendyearoftherecordisgiven,aswellasthelatitudeand
longitudeindecimaldegrees(negativelongitudesare°W).AsourceIDisalsoprovided;seeTable2.
First Last First Last
StationandsourceID year year Lat Lon StationandsourceID year year Lat Lon
Aberdeen1,3 1861 1995 57.16 (cid:3)2.10 London3,4 1850 1881 51.46 0
Alexandria13 1876 1881 31.20 29.95 Lugansk6 1850 1880 48.60 39.30
Algiers4,15 1872 1881 36.76 3.10 Lund1 1864 2001 55.70 13.20
Tasiilaq1 1894 1995 65.60 (cid:3)37.63 Lyon4 1869 1881 45.72 4.95
Angra(deHeroismo)4 1871 1878 38.66 (cid:3)27.22 Madrid14,16 1853 1880 40.45 (cid:3)3.71
Archangel6,11 1866 2000 64.55 40.53 Malta4,10 1852 1880 35.83 14.00
Armagh14 1850 2001 54.35 (cid:3)6.65 Milan2,4 1763 1998 45.61 8.73
Astrakhan6,11 1850 2000 46.35 48.03 Montreal14 1850 1873 45.53 (cid:3)73.60
Athens14 1850 1880 37.90 23.73 Moscow6,11 1850 2000 55.76 37.66
Baghdad7,4 1869 1876 33.23 44.23 Nikolayev6 1850 1880 46.58 31.95
Barcelona14,16 1850 2002 41.50 2.01 Nordby1 1874 2002 55.43 8.40
Beirut7,13 1874 1881 33.82 35.48 Oksøyfyr1 1870 2002 58.07 8.05
Bergen1 1868 2002 60.38 5.33 Orenburg6 1850 1876 51.75 55.10
Bermuda10 1852 1880 32.28 (cid:3)64.50 Padua2 1766 1997 45.40 11.85
Biarritz3,4 1860 1880 43.46 (cid:3)1.53 Palermo4,14 1851 1880 38.13 13.33
Biskra4,15 1878 1881 34.80 5.73 Paris14,4 1851 1880 48.81 2.33
Bodø1,14 1868 1994 67.26 14.43 Plymouth3 1861 1881 50.35 (cid:3)4.15
Brest3,4 1861 1881 48.45 (cid:3)4.16 Potsdam1 1893 1993 52.38 13.06
Cadiz2,14 1786 2002 37.46 (cid:3)6.28 Prague14 1850 1880 50.08 14.42
Corfu10,13 1852 1880 39.61 19.91 Providence15 1850 1860 41.68 (cid:3)71.25
DeBilt14 1850 2001 52.10 5.18 Reykjavík14 1820 2001 64.13 (cid:3)21.90
Diyarbakir4,7 1869 1876 37.88 40.18 Riga6,11 1850 1990 56.81 23.89
Durham14 1850 1881 54.76 (cid:3)1.58 Rochefort3,4 1862 1881 45.93 (cid:3)0.93
Fao4,7 1869 1876 29.98 48.50 Rome4 1869 1881 41.95 12.50
Funchal4 1871 1881 32.63 (cid:3)16.90 Scutari10 1866 1880 41.00 29.05
Galway3,14 1850 1880 53.28 (cid:3)9.02 Sevastopol6,11 1850 1990 44.61 33.55
Gibraltar4,14 1850 2002 36.10 (cid:3)5.35 Sibiu4 1878 1881 45.80 24.15
Nuuk(Godthåb)8 1875 1880 64.16 (cid:3)51.75 St.Johns10,12 1852 1876 47.56 (cid:3)52.70
Gothenburg1 1860 2002 55.70 11.98 Stockholm1 1756 1998 59.33 18.05
Halifax9,12 1850 1875 44.63 (cid:3)63.50 Stornoway3,4 1872 1881 58.22 (cid:3)6.32
Hammerodde1 1874 1995 55.30 14.78 St.Petersburg6,11 1850 2000 59.93 27.96
Haparanda1 1860 2002 65.82 24.13 Stykkisholmur1 1874 2003 65.08 (cid:3)22.73
Harnosand1 1860 1995 62.61 17.93 Tenerife16 1901 2002 28.47 (cid:3)16.32
Helsinki1 1844 2001 60.17 24.95 Tbilisi6,11 1850 1990 41.68 44.95
Hohenpeissenberg14 1850 2002 47.80 11.02 Tórshavn1 1874 2002 62.02 (cid:3)6.77
Jena14 1850 2000 50.93 11.58 Toulon3,4 1868 1881 43.10 5.93
Kazan6,11 1850 2000 55.78 49.13 Uppsala1 1722 1998 59.86 17.63
Kem6 1866 1880 64.95 34.65 Valentia1,3,4 1861 1995 51.93 (cid:3)10.25
Kiev6,11 1850 1990 50.40 30.45 Vardø14 1861 2003 70.36 31.10
Kostroma6 1850 1880 57.73 40.78 Vestervig1 1874 1995 56.77 8.32
LaCoruna3,16,5 1865 2002 43.16 (cid:3)8.50 Visby1 1860 2002 57.63 18.28
Lesina(Split)4,13 1869 1881 43.53 16.30 Wilna6,11 1850 1990 54.68 25.30
Lisbon4 1869 1881 38.77 (cid:3)9.13 Zagreb14 1862 2000 45.82 15.98
we used the seasonal phase and amplitude gridded 1987) and also includes several million new observa-
fields calculated by Dai and Wang (1999) and interpo- tions from the U.S. Maury collection, amongst others.
latedfromthenearestgridpoint.Observationhoursfor Daily marine gridded MSLP fields were generated us-
eachdayofthestationserieswerecollatedandusedto ing these data for 1850–1997, supplemented with the
determine the appropriate adjustment required. National Centers for Environmental Prediction
(NCEP) Global Telecommunication System (GTS)
b. Marine data sources
data for 1998–2003.
Marine pressure observations were obtained from
SUMMARY OF MARINE GRIDDING PROCEDURE
ICOADS (Worley et al. 2005; Diaz et al. 2002). This
dataset combines the Met Office’s Marine Data Bank Toqualitycontrolandgridthemarineobservations,
with the previous version of COADS (Woodruff et al. we have modified the procedure used in the develop-
15JUNE2006 ANSELL ET AL. 2721
TABLE 2.Summary of sources corresponding to source IDs provided in Table 1. A number of the stations used in EMSLP were
availableasaresultofindividualefforts;thesestations(andtheresearcher’sname)arelistedhereundermiscellaneoussources.Full
detailsofthesourcesareprovidedontheEMULATEWebsite(www.cru.uea.ac.uk/cru/projects/emulate/).
ID Sourcedescription
1 WASAProject(Schmithetal.1997)
2 IMPROVE(CamuffoandJones2002)
3 U.K.dailyweatherrecords(MetOffice,1860–81)
4 Frenchdailyweatherrecords(BureauCentralMeteorologique,1869–81)
5 Spanishdailyweatherrecords(BoletinMeteorologicoDiario1875)
6 St.Petersburgyearbooks(NicolasCentralPhysicalObservatory,1850–87)
7 OttomanEmpirerecords(ConstantinopleObservatoireImperial,1870–74:Bulletin,1869–74)
8 Simultaneousinternationalmeteorologicalobservations(WashingtonSignalOffice,1875–81)
9 MeteorologicalobservationsatBermuda,1853–54,Halifax,1854–55,andmiscellaneouspapers(BoardofTrade1863).
10 RoyalEngineersandtheArmyMedicalDepartmentobservations(MetOffice1890)
11 Six-andthree-hourlymeteorologicalobservationsfrom223U.S.S.R.stations(Razuvaevetal.1998)
12 EnvironmentCanada
13 Austrianyearbooks(ZentralanstaltfürMeteorologieundGeodynamik,1854–1984)
14 Miscellaneousdatasources:
Armagh:ArmaghObservatoryrecords(A.García-Suárez)
Athens:AthensObservatoryrecords(D.Founda,M.Petrakis,E.Xoplaki)
Barcelona:ADVICEProject(M.Barriendos)
Bodø:Ø.Nordli
Cadiz:RealObservatoriodelaArmadaenSanFernando(RoyalObservatoryoftheSpanishNavyatSanFernando)
Debilt/Utrecht:KoninklijkNederlandsMeteorologischInstituut(KNMI)yearbooks(T.Brandsma)
Durham:DurhamUniversityObservatory(D.ListerandN.J.Cox)
Galway:Hickeyetal.(2003)
Gibraltar:GibraltarChronicleandRoyalEngineers(M.RodwellandD.Wheeler)
Hohenpeissenberg:GermanWeatherService(DWD)
Jena:F.Gerstengarbe
Madrid:ElNoticiosoNewspaper,RicoSinobasPaper,RicoSinobasManuscript,LaGacetaNewspaper,RealObservatorio
AstronomicodeMadrid
Montreal:ObservationsbyDr.Smallwood,Dr.Sunderland,Dr.HallnearMcGillObservatory,Montreal(V.Slonosky)
Palermo:AstronomicalObservatoryPalermo(M.Barriendos)
Paris:Journaldesobservationsmétéorologiquesetmagnétiquesfaitesàl’ObservatoiredeParis:1823–62,1861–72,1873–80
(M.Barriendos)
Prague:R.Brazdil
Reykjavík:P.Jones(http://www.cru.uea.ac.uk/cru/data/nao.htm)*
Vardø:Ø.Nordli
Zagreb:MeteorologicalandHydrologicalService,Gric(L.Srnec)
15 NOAAlibraryscannedimages(http://docs.lib.noaa.gov/rescue/data_rescue_home.html)
Algiers:Bulletinmétéorologiedugouvernmentgénéraledel’Algérie(1877–81)
Providence:Caswell(1859)
16 InstitutoNacionaldeMeteorología(SpanishNationalMeteorologicalOffice)
*The1854–80valuesforReykjavíkareclimatologicallyderived(seeJonesetal.1997).
ment of the First Hadley Centre Sea Level Pressure all 1° residuals over a 7° area centered on the 1° (cid:1) 1°
dataset (HadSLP1). This historical gridded global targetbox.Thisprocedureservestoinfilldata-sparseor
monthly MSLP product is described in Basnett and -missing areas and smooth over data-rich areas. For
Parker(1997).Thequalitycontrolandgriddingisbased example, if the target box value is missing, but one of
on residuals (anomalies) formed by removing a thesurroundingforty-eight1°boxesinthe7°areacon-
monthlybackgroundfield(describedbelow)fromeach tains an observation, then the target value is replaced
ship observation. The residuals were compared to a with this value.1 If all grid box values in the 7° degree
measure of intramonthly variability (see below); the area contain data, including the target, then the target
median residual of only those observations that were valuewillbereplacedwiththemedianofall49gridbox
equal to or less than 3 times this intramonthly value
wereassignedto1°latitudebylongitudegridboxes.A
1In the case of an isolated observation, this procedure could
smoother daily value for each 1° latitude by longitude resultinablockofanomalousvaluesbeingspreadovera7°(cid:1)7°
gridboxwasthenformedbytakingthemedianvalueof cell.
2722 JOURNAL OF CLIMATE VOLUME19
TABLE3.ListofsourcesfortheMetOffice’soperationalNorthernHemisphereJ86fieldsupto2003(Jackson1986).
Period Source
1Dec1880–31Dec1898 DeutscheWetterdienst“morning”charts(derivedfromthe“TaglicheSynoptischeWetterKarten”
coveringanareafrom80°Eto100°W)
1Jan1899–31Dec1939 ExtendedForecastDivisionoftheU.S.WeatherBureau1200UTCcharts
1Jan1940–31Dec1948 Offenbach0001UTCcharts*
1Jan1949–31Dec1965 ExtendedForecastDivisionoftheU.S.WeatherBureau1200UTCcharts
1Jan1966–20Aug1975 U.K.MetOffice0001UTCcharts
21Aug1975–31Dec2002 U.K.MetOffice0001UTCmodelcharts
1Jan2003–present NCEP–NCARreanalysis6-hourlySLPfields(dailyaverageof0600and1200UTCcharts)
*Recently,G.Compocomparedthe1948NCEP–NCARreanalysisandOffenbachchartsandshowedthattheOffenbachchartswere
actually1200UTC,not0001UTC.WedonotknowwhetherthisincorrecttimestampistrueforalloftheOffenbachcharts.
values.The7°areawaschoseninitiallytohelpcombat their monthly means agreed with HadSLP1 [described
the sparseness of the available observed data, but it in section 2c(2) below]. The maximum adjustments
results in considerable smoothing, particularly in well- were5–6hPa.Theyears1999–2002remainuncorrected
sampledmidlatituderegions.Thisisconsideredfurther (HadSLP1extendsonlyto1998),2butpotentialhetero-
in section 4. geneities are more likely before 1975 (see Table 3).
We also corrected each observation for the diurnal Fields were also regridded onto a 5° (cid:1) 5° grid. We
and semidiurnal oscillation, using the Dai and Wang applied a correction for the diurnal and semidiurnal
(1999)fields,aswiththelanddata.Considerableeffort oscillation using the Dai and Wang (1999) fields and
wasalsospentexploringanddocumentingtheextentof basedonanalysistimesgiveninTable3tomakethem
previously undetected duplicates and a low (anoma- consistent with the 24-h “daily” station and marine
lously negative) MSLP bias in the early 1850s; these data.
issuesareelaborateduponinappendixesAandB.The Intramonthlystandarddeviationfields,introducedin
background fields were added back to the screened the subsection of 2b, were calculated from the daily
gridded residuals to yield gridded actual values of ma- four6-hourlyNCEP–NationalCenterforAtmospheric
rine MSLP. Research (NCAR) reanalysis fields (Kalnay et al.
1996).
c. Gridded data sources
2) MONTHLY
1) DAILY
As described in the subsection of 2b, a background
WehavetakenadvantageoftheMetOffice’shistori-
fieldisrequiredforquality-controllingandgriddingthe
calJ86dataset,aNorthernHemisphere(northof15°N)
marine observations. We have principally relied on
synoptic daily MSLP product extending from 1881 to
HadSLP1 [an updated version of the global mean sea
thepresentona5°latitudeby10°longitudegrid.Until
levelpressuredataset(GMSLP2.1f;BasnettandParker
the 1970s these J86 fields were derived from digitized
1997)],whichisaglobalmonthlygriddedproduct,ona
hand-drawnsynopticcharts(seeTable3),includingthe
5° latitude by longitude grid for 1871–1998. For 1854–
GermanMorningchartsandU.S.forecastcharts.Inthe
70, we used Kaplan et al.’s (2000) optimally interpo-
1970s, they were replaced with model analysis charts.
lated(marineonly)fields.Aclimatologicalaveragewas
The J86 product is an extremely valuable resource, as
used for the period 1850–54. A land and marine back-
each daily field implicitly contains many thousands of
ground field is also required when blending the land
station observations that went into the original opera-
andmarinefields(section3a).AsKaplanetal.’s(2000)
tional charts each day.
optimally interpolated fields are marine only, a clima-
The many changes in sources, detailed in Table 3,
tological monthly average for 1871–1900 from Had-
mean that the J86 fields are, however, subject to het-
SLP1 was used for the period 1850–70.
erogeneities,elevationcorrections,andincreasingdata
For homogeneity checks and validation, a historical
availability(seealsoJones1987;Jonesetal.1999).Po-
gridded product developed as part of the Annual to
tential problems are important over southeastern Eu-
rope,theMiddleEast,andpartsoftheNorthAtlantic
Ocean, particularly before 1940. To account for these 2HadSLP2(AllanandAnsell2006)wasnotyetavailableatthis
heterogeneities, the J86 fields were adjusted such that time.
15JUNE2006 ANSELL ET AL. 2723
Decadal Variability in Climate in Europe Project FortheCanadianstationsandthoseinthefareastof
(ADVICE;Jonesetal.1999)wasused.TheADVICE theEMULATEregion,afinaladjustmentwasrequired
monthly data are available either as 51 individual sta- so that their daily average represented the same 24-h
tionseriesorona5°latitudeby10°longitudegridfor periodastheotherseries.TheCanadianstationsare5
1780–1995.NotethatADVICEusedmonthlyaverages h behind central Europe; the easternmost Russian sta-
ofJ86datasince1881andadditionalmonthlyfieldsfor tionsare4hahead.Asonlydailyaverageswereavail-
1873–80; accordingly, ADVICE is not independent of able for these stations, we were forced to interpolate
EMSLP. between the preceding (following) day and the actual
day for the Canadian (far Russian) stations.
d. Homogenization of land station data
Toobtainalongandhomogenizedseries,issuessuch
3. Gridding and reconstruction
as change of station location, instrument, and instru-
ment height need to be identified and accounted for. a. Blending land and marine fields
These can be identified in metadata records, but such
The quality-controlled land station data were
records are often unavailable.
blended with the gridded marine fields, using a proce-
Potential heterogeneities can be identified with a
duresimilartothatemployedtogridthemarineobser-
standard normal homogeneity test (Alexandersson
vations(subsectionof2b).Foreachdayineachmonth
1986). This and similar techniques, described in
ineachyearfor1850–2003andineach1°(cid:1)1°gridbox,
Slonosky et al. (1999), rely on comparisons with “ref-
the marine grid box value and all terrestrial individual
erence”seriesthatareknowntobehomogeneous.We
MSLP station observations (if present) were collated.
made near-neighbor comparisons, for example, be-
Residuals were formed by subtracting a monthly (land
tweenDurhamandAberdeenintheUnitedKingdom,
andmarine)backgroundfieldfromeachterrestrialob-
ToulonandBrestinFrance,Galway,Ireland,Armagh,
servationandmarinegridboxvalue,andthentheme-
NorthernIreland,andGibraltarandCadiz,Spain.Not
dianvalue(bothlandandmarine)wasselected.Allof
allpotentialheterogeneitiescouldbeidentifiedbythis
the1°(cid:1)1°medianvalueswerethenaveragedto5°(cid:1)
method, however, owing to a lack of suitable nearby
5°gridpointvalues,takingaccountoftheirspatialdis-
referenceseries.Occasionallythereferenceseriesitself
tribution.
wasfoundtocontaininhomogeneities(viz.Cadiz).Itis
The J86 fields (from 1881 to 2003) were then com-
also preferable to use records from at least two differ-
binedwiththeseblended(landandmarine)1850–2003
entobservingcountries(Slonoskyetal.1999),asmeth-
fields using Poisson blending (Reynolds 1988). Using
odological changes may have taken place at the same
the blended land and marine observations (anomalous
time within individual countries.
values) as the ground truth, the Laplacian (second de-
To address these homogenization issues more fully,
rivative) of the “less reliable” J86 anomalies was used
weappliedadjustmentfactors,similartothoseapplied
to interpolate between the blended observations and
to the J86 fields [section 2c(1)]. Specifically, the
J86anomalies.Forgridboxeswithnoobservations,the
monthly means were calculated for each daily series
J86 value was taken.
and compared with a reference value (the correspond-
ingADVICEmonthlystationseriesoraninterpolation
b. Interpolation
from the nearest ADVICE or HadSLP1 grid point;
preference was given to the ADVICE station series To create spatially complete fields, Reduced Space
where possible). The difference in monthly means was OptimalInterpolation(RSOI)wasused[seeKaplanet
then used to adjust the daily SLP values. To avoid al.(1997)andreferencestherein].TheRSOItechnique
jumps in adjustments at the end of each month and reconstructs fields by least squares fitting incomplete
year, a binomial filter with seven terms was applied to observed data to yield the amplitudes of a predeter-
thewholemonthlyadjustmentseries.Thisprocessgives mined subset of the empirical orthogonal functions
a smooth daily adjustment series, but almost the full (EOFs)ofthespatialcovariancematrix.Theanalysisis
daily variability of the station data is still preserved. constrained to give the greatest weight to data with
This homogenization stage is very important, as some smallerestimatederrorvariance,sothatnoisyorsparse
of the inhomogeneities found were very large. For ex- data are prevented from producing noisy or spurious
ample, the London series required adjustments of (cid:4)7 fields (Kaplan et al. 1997). The series of EOFs is trun-
hPa during 1850. Without these adjustments, EMSLP catedtoremoveasmuchofthenoiseaspossiblewhile
would be flawed, as the station observations are retainingtruesignals.Byconstruction,therefore,large-
weighted heavily (see section 3a). scale features of the variable are recovered, which are
2724 JOURNAL OF CLIMATE VOLUME19
presumed to be those of the greatest climatic impor- believe, the smooth NCEP–NCAR fields from which
tance(Kaplanetal.2000).Amajorassumptionofthis the covariance matrix was estimated.
methodisthattheEOFsdescribeasetofpatternsthat Anestimateofthesamplingerrorisalsorequiredfor
occur throughout the reconstruction period. RSOI.Weusedanaverage(1961–90)ofthecombined
Duringtheperiod1850–80,theblendedlandandma- marine, land, and J86 sampling error. For the marine
rinefieldshaveverypoordatacoverage;85%–90%of observations,thesamplingerrorforeachmonth(after
thegridboxeshavemissingvalues.Toassesshowwell Parker1984)wascalculatedaspartofthemarinegrid-
RSOI would perform with such sparse coverage, we dingprocedure.Thiswasmultipliedbythesquareroot
performed verification experiments by subsampling ofthenumberofdayswithdatatoyieldthedailyerror.
daily NCEP–NCAR reanalysis fields for 1990–2000 to In addition, we took account of the errors inherent in
represent the historical sampling of 1850–60. These the ship observations. A value of 0.25 hPa for geo-
fieldswerethenreconstructedusingRSOI,usingEOFs graphicallyrandomone-sigmabiaswasestimatedfrom
calculated over the period 1951–89 (a relatively well- thedifferencesbetweensynopticchartsandoperational
sampled yet independent period). The root-mean- model analyses and added vectorially to the sampling
square(rms)errorsforthereconstructedand“original” error.
dailyfieldsweregenerally(cid:2)2.5hPa,indicatingthatthe Over land, estimated errors were based on the alti-
technique generally performs very well, even with tude of the station. An estimate of h/1500 as the bias
(cid:4)85%ofthedatawithheld.OverGreenlandandinthe associatedwiththereductiontomeansealevel,where
far northeast and west of the EMULATE region, rms h is the altitude of the stations (in meters), was identi-
errors were as large as 5 hPa, highlighting regions fiedbycomparingthepressurereducedtosealeveland
where the technique performs less well. modelanalysesatanumberofhigh-altitudegridpoints.
Again, 0.25 hPa was added (vectorially) to the eleva-
tion-related bias to reflect the random bias error. For
PROCEDURE
gridpointswheretheJ86datawereused,webasedthe
We now describe how RSOI was applied, detailing error on the intrabox standard deviation (calculated
first the calculation of the covariance matrix and error from1°(cid:1)1°resolutiondatawithinthe5°(cid:1)5°gridbox
fields. from the NCEP–NCAR reanalysis) divided by the
The most crucial element of RSOI is to obtain a re- numberofobservationsatthatgridpoint.Asampleof
liableestimateofthespatialcovariancematrix(Kaplan historicalsynopticchartswasusedtoestimatethenum-
etal.2000).Itisdesirabletousearelativelylongtime berofobservationsateachgridpoint.Inregionsofzero
period of well-sampled fields, so here, NCEP–NCAR observations (central Greenland and northeast Euro-
reanalysis data from 1951–2002 were used. We em- peanRussia),theerrorwassettotheintraboxstandard
ployed anomalies relative to smoothed daily averages deviation at that grid point.
(“normals”) for 1951–2000 on 5° latitude by longitude Complete MSLP anomaly fields were reconstructed
resolution. The unsmoothed normals were created by using the leading 20 EOFs and the error field. Follow-
taking the climatological 50-yr average for each calen- ing Rayner et al. (2003), the available “observations”
dardayoftheyear.A50-yraverageof29Februarywas (as anomalies) were then superimposed on the recon-
formedbysimulating29Februaryinnon–leapyearsby struction.Thengridpointswereflaggedwherethegrid
averaging 28 February and 1 March. After the un- point anomaly minus the average of its neighbors was
smoothed climatology was formed, a binomial (time greater than a maximum permitted difference. This
wise) filter with 21 terms (removing noise under 15 maximum permitted value was calculated as the mean
days) was applied at each grid point to yield the anomalous value plus 3 times the standard deviation,
smoothed (daily) climatology. based on 1961–90 daily averages and standard devia-
EOFswerecalculatedforeachcalendardayoverjust tions derived from the NCEP–NCAR reanalysis.
the EMULATE region using a covariance matrix of Flagged anomalies and their neighbors were then
MSLP anomalies and applying a fourth-order Shapiro weightedbasedonwhethertheywereanobservationor
filter (Shapiro 1971), following Kaplan et al. (1997). reconstruction and on the numbers of constituent ob-
Kaplan et al. (2000) found that it was necessary to re- servations. This gave greater weight to well-observed
estimatethesignalcovariancetoobtainamorerealistic areasbecausereconstructedvaluesweretreatedasbe-
estimate of the signal covariance and, by association, ingbasedononeobservation.Theflaggedanomalywas
morerealistictheoreticalerrorestimates(Kaplanetal. then replaced by the average of the weighted anoma-
2000, their appendix). However, it was found in this lies, that is, the flagged point and its eight nearest
study that this step was not required because of, we neighbors.Overthe154-yrperiod,anaverageof1.5%
15JUNE2006 ANSELL ET AL. 2725
ofgridpointsmonth(cid:3)1wereinitiallyflagged,with0.2% ICOADSobservationsinEMSLPandtheirsubsequent
remainingafterthefirstweightedaverage.Thisproce- impact on the diurnal cycle. While both products are
durewasreiterateduntildatarejectionsceased;90%of based on J86 fields, which are derived from synoptic
these flagged grid points were “corrected” after two charts(Table3),thesefieldsinEMSLPwerecorrected
iterations. Finally, the climatology was added back to forthesemidiurnalanddiurnalcycle.Eventhoughour
yield absolute values. correction is coarse, the ICOADS observations over
theoceanarewellsampledthroughouttheday,andso
4. Assessment of results
thedailymarineMSLPfieldsinEMSLPareclosertoa
TodiagnoseanypotentialproblemswithEMSLP,we 24-h mean than in the original J86 fields. This is sup-
havecomparedthedatasetwiththemonthlyADVICE portedbytheagreementbetweenEMSLPandERA-40
product (Jones et al. 1999) and with the daily 6-hourly over the oceans in Fig. 3 (see below).
40-Yr European Centre for Medium-Range Weather ComparisonswithERA-40areshowninFig.3;both
Forecasts (ECMWF) Re-Analyses (ERA-40) MSLP productsareaveragedfor1959–2001.Differencesover
fields. Both long-term averages and individual events theoceanareverysmall((cid:2)0.5hPa),exceptintheeast-
havebeenexamined.Herewehighlightissuesthatmay ern Mediterranean and off the east Greenland coast,
be most relevant to the potential user. north of and extending into Iceland.
ThebiasesinGreenlandandIcelandaremostpromi-
a. Climatologies
nent during the winter months and are believed to be
We compare the EMSLP monthly climatology with relatedtoseasonalvariationsintheestimatedairtem-
the ADVICE gridded product over 1850–1995 in peraturesusedinthereductiontomeansealevelpres-
Fig. 2. The ADVICE region is only a subset of the sure. This results in fictitious high pressure areas over
EMULATE region, but there is generally excellent Greenland and central Iceland during the winter in
agreement between the two products, with differences ERA-40. The lower pressures over Greenland in
of the order (cid:5)0.5 hPa (only those values (cid:4)0.5 and EMSLP compared with ERA-40 are also evident in
(cid:2)(cid:3)0.5areplotted).Theseresultsaresimilartoarecent comparisons with NCEP–NCAR reanalysis fields (not
griddedSLPdatasetofLuterbacheretal.(2002)going shown). Models estimate the MSLP there by extrapo-
back to A.D. 1500. One region where differences are latingthesurfacepressureover2000mthroughtheice
prominent is central Spain. Pressures near Madrid in to sea level, so the result is sensitive to the assumed
EMSLP are around 1.5 hPa higher during November– temperature profile.
JanuarythanintheADVICEgrid,whereasfromJune Over Europe, the differences are larger than that
to August (JJA), EMSLP pressures are 1.5 hPa lower seenwithADVICE(Fig.2),andagainthesearelargest
than in ADVICE. The EMULATE station series are in November–February with differences up to 1.5 hPa
adjusted so that their monthly means are equal to the in eastern Europe. As in Greenland and Iceland, the
ADVICE station series (section 2d); hence the differ- reduction to sea level during the winter season prob-
encebetweenthetwoproductscannotbeexplainedby ablyaccountsforthedifferencesherebetweenEMSLP
adiscrepancyinthetwoinputseries.ThecoarserAD- and ERA-40. We suspect that differences in the num-
VICE grid may, however, account for the differences; ber of observations available to each product, particu-
theADVICEgridpointforMadridincludesdatafrom larly before the late 1970s,3 may also have had some
Oporto,Portugal,thatarenotincludedinEMSLP,be- influence. In North Africa and the Middle East, the
cause only monthly data are available. Differences be- differences may be associated with the diurnal cycle
tweencoastalandinlandsitesaremarkedinthisregion, correction (see section 4b).
due to thermal pressure systems that develop over the
interior. b. Correlations
In the Middle East, ADVICE MSLP is lower than
Spatial correlations and temporal squared correla-
EMSLP during most of the year, though particularly
tions were performed to compare EMSLP with the
during summer. The inclusion of the monthly Cairo,
ADVICE and ERA-40 products. Generally, the ex-
Egypt;andJerusalem,Israel;recordsinADVICEmay
plained variances are larger in the winter months, be-
accountfortheseregionaldifferencesinthetwoprod-
cause of the greater meteorological signal in this sea-
ucts; EMSLP has only limited records for Alexandria,
Egypt; and Beirut, Lebanon (see Table 1); and for
1881–2003reliessolelyonreconstructionandJ86fields
3TheinputarchiveforERA-40wassmallerintheearlyyears
here. Differences between the gridded products over
(Simmonsetal.2004)thanwhatwouldhavebeenavailabletothe
the ocean, we believe, are a result of the inclusion of J86fieldsatthetime.
2726 JOURNAL OF CLIMATE VOLUME19
FIG.2.(a)(toptobottom)January–Junemonthlynormalsfor(left)EMSLP,(middle)ADVICE,and(right)monthlydifferences
(cid:2)(cid:3)0.5and(cid:4)0.5inhPa.Normalsarecalculatedover1850–1995.ToaidthecomparisonwehaveinterpolatedEMSLPtotheADVICE
5°latitudeby10°longitudegrid.(b)Sameasin(a),exceptforJuly–December.
Fig 2alive4/C