Вы находитесь на странице: 1из 27



P. D. Jones,
• M. New,• D. E. Parker,
3 andI. G. Rigor

Abstract. We review the surface air temperature and satellite temperature measurementsof the lower
recordof the past 150years,consideringthe homogene- troposphereand considerthe last 150 years in the con-
ity of the basicdata and the standarderrors of estima- text of the last millennium.We then provide a globally
tion of the averagehemisphericand global estimates. complete absolutesurfaceair temperature climatology
We presentglobal fieldsof surfacetemperaturechange on a 1ø x 1ø grid. This is primarily based on data for
over the two 20-year periods of greatestwarming this 1961-1990. Extensive interpolation had to be under-
century,1925-1944 and 1978-1997. Over theseperiods, taken over both polar regionsand in a few other regions
global temperaturesrose by 0.37ø and 0.32øC,respec- where basic data are scarce, but we believe the climatol-
tively. The twentieth-centurywarming has been accom- ogy is the most consistentand reliable of absolutesur-
paniedby a decreasein thoseareasof the world affected face air temperature conditions over the world. The
by exceptionallycool temperaturesand to a lesserextent climatologyindicatesthat the annual average surface
by increasesin areas affected by exceptionallywarm temperatureof the world is 14.0øC(14.6øCin the North-
temperatures.In recent decadesthere have been much ern Hemisphere (NH) and 13.4øCfor the Southern
greater increasesin night minimum temperaturesthan Hemisphere).The annualcycleof globalmean temper-
in day maximumtemperatures,so that over 1950-1993 atures follows that of the land-dominated NH, with a
the diurnal temperature range has decreasedby 0.08øC maximumin July of 15.9øCand a minimum in Januaryof
per decade.We discussthe recentdivergenceof surface 12.2øC.

1. INTRODUCTION areas and gradual changesin observingpracticesover

the ocean. Section 3 reviews several methods that have
The surfaceair temperaturedatabasehasbeen exten- been used to aggregatethe basicdata to a regular grid,
sivelyreviewed on severalearlier occasions,most nota- the basicaim of which is to reduce the effectsof spatial
bly by Wigleyet al. [1985,1986]andEllsaesser
et al. [1986] and temporal variationsin the spatial density of avail-
and by the IntergovernmentalPanel on Climate Change able data. Future improvementsto the network are
(IPCC) [Follandet al., 1990,1992;Nichollset al., 1996]. discussed.
This work extendstheseby usingslightimprovementsto Before consideration of what the data tell us about
the basicsurfacedatabaseand updatesthe recordsto the the post-1850 instrumental era, section 4 assessesthe
end of 1998. Most importantly,it also providesa com- accuracyof the estimatesof hemisphericand global
prehensiveanalysisof surfaceair temperaturesin abso- temperature anomalies, particularly during the nine-
lute rather than anomalyunits,by the developmentof a
teenth century, when coveragewas poorer. Section 5
1øby 1øgrid box climatologyfor all land, ocean,and sea
then analyzesthe surfacetemperature record. It com-
ice areasbasedon the 1961-1990 period.
The structure of this review is as follows. Section 2 pares patterns of warming over the 1978-1997 period
with thoseof the earlier 1925-1944 period, which expe-
discusses the quality of the raw land and marine tem-
peraturedata, addressingthe long-termhomogeneityof rienceda comparablerate of warming.It also considers
the basicdata and consideringthosesubtlechangesthat trends in the areas affected by extreme temperatures,
might influence the data, such as urbanization of land recent trends in Arctic temperatures and in maximum
and minimum temperatures, the recent divergence of
surface temperature trends and satellite retrievals of
lower tropospherictemperaturetrends,and the last 150
•ClimaticResearchUnit, Universityof EastAnglia,Nor-
years in a longer near-millennial context.
wich, England.
The discussion in section 5 is limited to what the
of Oceanography,
of Washington,Seattle. surfacerecord tells us about past temperaturechanges.
4AppliedPhysicsLaboratory,Universityof Washington, Extensivediscussion of the reasonsfor the changes(both
Seattle. variationsin atmosphericcirculationand past external

Copyright1999 by the AmericanGeophysicalUnion. Reviewsof Geophysics,

37, 2 / May 1999
pages 173-199
8755-1209/99/1999 RG900002 $15.00 Papernumber1999RG900002

forcing)may be foundin the IPCC assessments [Folland smallerthan individualdaily or evenhourlyoccurrences.

et al., 1990, 1992; Nicholls et al., 1996] and elsewhere Effectsup to 10øC,for individualdays,havebeenquoted
[e.g., van Loon and Williams, 1976; Karl et al., 1989, for some cities [see, e.g., Oke, 1974]. To assessurban-
1993a; Trenberth, 1990; Parker et al., 1994; and Hurtell, ization influenceson monthly and longer timescalesin
1995]. In section 6 the developmentof the absolute certain regions (Australia [Torok, 1996]; Sweden
climatology is described,and section 7 presents our [MobergandAlexandersson, 1997]), specificrural station
conclusions. data setshavebeen developedto allow estimationof the
magnitude of possible residual urban warming influ-
encesin the Jones[1994] data set. Resultsshowthat if
2. HOMOGENEITY OF THE BASIC DATA there are residualinfluences,they are an order of mag-
nitude smaller than the nearly 0.6øCwarming evident
However mathematically cunning or appealing an during the twentieth century, confirmingearlier work
analysistechniquemaybe, any resultsor conclusions are [Joneset al., 1989, 1990;Karl and Jones,1989].
dependentupon the quality of the basicdata. In clima-
tology, quality of the data is essential.A time seriesof 2.2. Marine Component
monthly temperature averagesis termed homogeneous Historical temperaturedata over marine regionsare
[Conradand Pollak, 1962] if the variationsexhibitedby largelyderivedfrom in situ measurementsof seasurface
the series are solely the result of the vagariesof the temperature(SST) and marine air temperature(MAT)
weather and climate. Many factorscan influencehomo- taken by shipsand buoys.Inhomogeneityproblemswith
geneity,which we next discussby divisionof the surface bothvariablesrelate principallyto changesin instrumen-
data into land and marine components. tation, exposure,and measurementtechnique.For SST
data the changein the samplingmethod from uninsu-
2.1. LandComponent lated buckets to engine intakes causesa rise in SST
The most importantcausesof inhomogeneityare (1) valuesof 0.3ø-0.7øC[Jamesand Fox, 1972]. Engine in-
changesin instrumentation,exposure,and measurement take measurementsdominate after the early 1940swith
technique;(2) changesin stationlocation(both position uninsulated bucket measurements the main method be-
and elevation);(3) changesin observationtimesand the fore. For MAT data, the averageheight of ships'decks
methods used to calculatemonthly averages;and (4) above the ocean surface has increased during the
changesin the environmentof the station, particularly present century, but the more seriousproblem is the
with referenceto urbanizationthat affectsthe represen- contaminationof daytimeMAT throughsolarheatingof
tativenessof the temperaturerecords.All of thesehave the ships'infrastructure,rendering only the nighttime
been discussedat length in the literature [see, e.g., MAT (NMAT) data of value [Parkeret al., 1994,1995a].
Mitchell, 1953;Bradleyand Jones,1985;Parker, 1994].A Follandand Parker [1995] developeda model to esti-
number of subjectiveand relatively objective criteria mate the amount of cooling of the seawaterthat occurs
[e.g.,Joneset al., 1985, 1986a,b, c; Easterlingand Peter- in bucketsof various types, dependingupon the time
son, 1995; Alexanderssonand Moberg, 1997] exist for between samplingand measurementand the ambient
testingmonthly data. Petersonet al. [1998a] provide a weather conditions.Adjustmentsto the bucket SST val-
comprehensivereview of these and numerous other ues for all yearsbefore 1941 are estimatedon the basis
techniques. of assumptions aboutthe samplingtime and the typesof
Here we use the land station data set developedby bucketin use at different times.The adjustments,which
Jones[1994]. All 2000+ station time seriesused have are greaterin winter, are tuned usingan assumptionthat
been assessed for homogeneityby subjectiveinterstation SST annual cycle amplitudeshave not experienceda
comparisonsperformed on a local basis.Many stations longer-term trend [Folland and Parker, 1995]. Canvas
were adjustedand someomitted becauseof anomalous buckets,which evaporativelycool more than wooden
warming trends and/or numerous nonclimatic jumps buckets, were in dominant use after the late nineteenth
(completedetailsare givenbyJoneset al. [1985,1986c]). century,havinggraduallyreplacedthe woodenbuckets
All homogeneityassessmenttechniqueswould per- that would havebeen usedinitially in the mid nineteenth
form poorly if all station time serieswere affected sim- century.Somebucketsare still in usetoday,but sincethe
ilarly by commonfactors(e.g. thoserelatedto urbaniza- early 1940s they have been much better insulated, al-
tion). Urban development around a meteorological though it is believedthat uninsulatedbucketswere still
station means that measuredtemperaturesare slightly used on some ships until the 1960s (C. K. Folland,
warmer than they would otherwisehavebeen. The effect personalcommunication,1995.).
is a real (rather than spuriousas in the other causesof In the combination of marine data with land-based
inhomogeneities)changein climate.It is consideredan surfacetemperatures,SST anomaliesare used in pref-
inhomogeneity,however,as the measuredtemperatures erence to NMAT because they are consideredmore
are no longer representativeof a larger area. The ur- reliable [Trenberthet al., 1992]. The basicassumptionis
banization influence we are interested in is the effect on that on monthlyand longer timescalesan anomalyvalue
the monthly time series. This effect is considerably of SST will be approximatelyequal to that of the air

immediately(typically 10 m) above the ocean surface The FDM is the newestapproachand works on the
(NMAT). The rationale is that SST observationsare first difference(FD) time series,the differencein tem-
more reliable [Parkeret al., 1995a]throughoutthe world. perature between successivevalues in a station time
Fewer individualSST than MAT readingsper unit area series, e.g., Jan(t) minus Jan(t - 1). FD values are
would be required to give averagesof the samereliabil- averaged together for all stationswithin each 5ø x 5ø
ity, owing to the strongerday-to-dayautocorrelationin grid box as with the CAM technique. To revert to a
SST comparedwith MAT values[seeParker,1984],even similar analysisas CAM, the cumulative sum of each
if the MAT observations had been as reliable as those of grid box seriesis calculated.The resultingseriesthere-
SST. In practice,the number of usable MAT observa- fore have similar potential changesin variance due to
tions is only about half that of SST. changingstation numbersbetween and within boxesas
The independenceof the land and marine compo- in CAM. This aspectwill alsobe presentin RSM series,
nents of the surface temperature record means that but to a lesser extent.
hemisphericand regionalseriescan be used to test the In all three techniques,large-scalehemisphericand
veracityof the other component.The componentshave global seriesare computedby a weighted averageof all
been shownto agreewith each other on the hemispheric the grid boxes with data. For CAM and FDM the
basisby Parkeret al. [1994] and, usingislandand coastal weightsare the cosineof the central latitude of the grid
regionsin the southwestern Pacific,byFollandand Salin- box, while for RSM the box size was chosen to be of
get [1995] and Folland et al. [1997]. approximatelyequal area. Petersonet al. [1998b] show
that the differences between the methodg for trends over
the 1880-1990 period are only a few hundredthsof a
3. • AGGREGATION OF THE RAW DATA degree Celsiusper 100 years,the differencesbeing due
more to differencesin techniquesfor calculationof a
3.1. LandComponent linear trend than to differences between the three meth-
Severaldifferent methodshave been usedto interpo- ods. The three methods also differ somewhat in their
late the station data to a regular grid. Petersonet al. interannual variances. CAM and FDM have similar vari-
[1998b] recognize three current and different tech- ances,but the RSM has a significantlysmallervariance
niques:(1) the referencestationmethod(RSM) usedby [Petersonet al., 1998b].
Hansen and Lebedeff [1987], (2) the climate anomaly A fourth approach, optimal interpolation, has been
method (CAM) usedby Jones[1994], and (3) the first proposed by l/innikov et al. [1990]. The approach is
difference method (FDM) used by Peterson et al. basedon pioneeringRussianwork on the averagingof
[199861. meteorologicalfields (recentlypublishedin Englishby
The CAM technique differs from the other two in Kagan [1997]). Regional and hemisphericaveragesand
requiringthat eachstationbe reducedto anomaliesfrom their errors are computeddirectlyand it is not necessary
monthly meanscalculatedfor a commonperiod. Jones to calculategrid box values.The most recent exampleof
[1994]uses1961-1990,requiringeachstationto have at its use is by Smith et al. [1994] in the calculation of
least 20 years of valuesfrom the base period, and esti- averageglobal sea surfacetemperature.
mates mean values,where possibleusingnearby series,
for stationsthat have long recordsbut that do not have 3.2. Marine Component
20 years data during 1961-1990. Station temperature For marine regionsthe raw data must be aggregated
anomaly values are then averagedtogether for all sta- in a different manner. Each SST or NMAT value is
tions within each 5ø x 5ø grid box. Resulting grid box accompaniedby location data. Climatologicalmonthly
time seriesdiffer in the number of stationsbetweengrid fields for each 1ø square and for each pentad (5-day
boxesand through time. Individual grid box time series period) and day of the year were calculatedon the basis
may therefore have differencesin varianceswith time of all in situ data from the 1961-1990 period. For SST a
due to changesin station numbers.The implicationsof backgroundclimatologybased partly on satellite data,
this are discussed in sections 4 and 5. adjustedto be unbiasedrelative to the in situ data, was
In the RSM technique,the world is divided into a used to guide interpolation of data voids in each indi-
number of areas (Hansenand Lebedeff[1987] use 80), vidual month in 1961-1990 [Parker et al., 1995a, b].
and a key long stationis selected,where possible,in each These complete monthly fields were then smoothed
area. Within each area, successivelyshorter station slightlyin an attempt to reduce errors in regionswhen
records are adjusted so that their averagesequal the there are few observationsbefore averagingover the 30
compositeof all stationsalready incorporatedover the years.Harmonic interpolationwasthen usedto generate
overlap with the composite.The method enables all the pentad climatology,whichwas then linearly interpo-
recordsto be usedprovidedthat they have a reasonable latedto the dailytimescale.Grid box(5ø x 5ø)anomalies
length of overlap (e.g., 10 years) with the composite. for eachmonth are now producedby trimmed averaging
Differences in variance of the area average serieswill of all 1ø anomalies(from their respectivedaily average)
also be present owing to changesin time of station for each of the pentads that make up the month. The
numbers. trimmingremovesthe effectsof outliers.A final checkis

then made for unreasonableoutliers,on a monthlybasis, would be slightlylarger using,for example,the period
by assessingwhether 5ø grid box anomaliesdeviatedby since t881.
more than 2.25øCfrom the averageof the eight neigh- Warming evident in Table 1 is slightly greater for
bors in space(of which four had to be nonmissingto both periodsfor the SouthernHemisphere(SH), but
make the test) or time (of whichboth the previousand there are greater seasonaldifferencesin the NH com-
next months had to exist to perform the test). If the pared with the SH. The global seriesis calculatedas the
value failed the test, it was replaced by the neighbor average of the two hemisphericseries.Nicholls et al.
average. The 2.25 threshold was chosen empirically: [1996]calculatethe globalseriesasthe weightedsumof
higher thresholdsadmitted bull's eyes;lower thresholds all grid boxes.The latter method can bias the global
causeddeletion of possiblycorrectvalues.Full detailsof averageto the NH value when there are marked differ-
the methodare givenby Parkeret al. [1995a,b]. The grid encesin hemisphericcoverage,but the effectis relatively
box anomaliesare based only on in situ measurements small [Wigleyet al., 1997].
from shipsand buoys;no satellite data are used. In the Northern Hemisphere, Figure 1 showsthat
year-to-yearvariabilityis clearlygreaterin winter than in
summer. The greater variability in all seasonsin both
3.3. Combinationof Landand Marine Components
hemispheresduring the yearsbefore 1881 is due princi-
The two data setsused are Jones[1994] for the land
pallyto the sparserdata availableduringtheseyearsand
areasand Parkeret al. [1995a]for the oceans.Both have
representsgreateruncertaintyin the estimates(see also
been producedon the same5ø x 5øgrid boxbasis.In the
section4). This feature is much more marked in the
combineddata set, the two data setsare merged in the
Northern Hemisphere, especiallyover the land regions
simplestfashion.Anomalyvaluesare taken, where avail-
[e.g.,Jones,1994].Both hemispheresclearlyexhibittwo
able, from each component.For grid boxeswhere both
periodsof warming:1920-1940, especiallyin the North-
components are available, the combined value is a
ern Hemisphere,and sincethe mid-1970sin both hemi-
weightedone, with the weightsdeterminedby the frac-
tions of land and ocean in the box. Because the land spheres.In the globalannualseries(Figure 2), 10 of the
12 warmestyearsoccurbetween 1987 and 1998. The two
componentfor oceanicislandsis potentiallymore reli-
yearsnot occurringin thisperiodare 1944and 1983.The
able than surroundingSST values, the land fraction is
assumed to be at least 25% for each of the island and
warmest year of the entire series on a global basis
occurredin 1998, 0.57øCabovethe 1961-1990 average.
coastalboxes.A correspondingcondition is applied to
the ocean fraction in boxes where this is less than 25%. The next warmest years were 1997 (0.43øC), 1995
(0.39øC), and 1990 (0.35øC). More discussionof the
The method is describedin more detail by Parkeret al. database will be undertaken in section 5.
[1994, 1995a].
3.5. FutureImprovementsto the Database
3.4. Hemisphericand Global AnomalyTime Series At present only --•1000 station records are used to
Figure 1 showsthe two hemispherictime seriesby monitor air temperatures over land regions [Jones,
seasonand year. The global annual series exhibits a 1994].This representsan apparentreductionby nearlya
warming (estimatedby a linear trend calculationusing factor of 2 from the number availableduringthe 1951-
least squares)of 0.57øCover the 1861-1997period. On 1980 period, but this reductionis principallyin areasof
these large spatial scales,the seriesshown agree with good or reasonable station coverage. The reduction
more recent analyses[e.g.,Petersonet al., 1998b],attest- stemsfrom the greater availabilityof past data in some
ingto someextentthat the homogeneity exerciseof the countries(e.g., United States, Canada, former Soviet
previoussectionhas been successful.Different station Union (fUSSR), China and Australia),data that are not
temperature compilationsnever use exactly the same available in real time.
basicland stationand marine data, but the vastmajority The reductionin land data availabilitydoesnot seri-
(>98%) of the basicdata is in common. ouslyimpair hemispherictemperature estimates,but it
Table 1 givesmonthly trend values for each hemi- doesreducethe quality of the grid box data set and the
sphereand the globecomputedover the 1861-1997and quality of subregionalseries,particularlyin the tropics
1901-1997 periods.Trends for all months,seasons,and [Jones,1995a]. Efforts are currentlyunder way to im-
years in both periods are significantat better than the prove the land componentof the data set. Nearly 1000
99% level. Estimation of the changesover the last 137 siteshave been designatedas the Global Climate Ob-
yearsusinglinear trendsis just one method that can be servingSystem(GCOS) SurfaceNetwork (GSN). Most
used.All monthly,seasonaland annualseriesare poorly World MeteorologicalOrganization(WMO) members
approximatedby linear trends,particularlyin the North- have agreed to attempt to maintain these stationsand,
ern Hemisphere (NH). Alternatives,using the differ- where possible,the sitesaroundthem well into the next
encesbetween the first and last 10- or 20-year periods, century.The selectionaimed to utilize the best stations
however, give similar results.The estimated trends are while achievingand improving the spatial coveragein
alsopartly dependentuponthe startingand endingyears recentyears,but accedingto memberswho havebudget
and the period used(see alsoKarl et al. [1994]). Trends constraints[Petersonet al., 1997]. Once the network is

' ' ' i i i , i , , ,

i i i • i i i , i , i i I I I I i I ! I i I , I i I i i , I ,



TABLE 1. Temperature ChangeExplained by the Linear the nineteenth century. Once available, the new data
Trend Over Two Periods
shouldsignificantlyimprove analysesover much of the
Pacific before World War II and in the Southern Hemi-
1861-1997 1901-1997

NH SH Globe NH SH Globe

Jan. 0.64 0.57 0.61 0.60 0.59 0.60

Feb. 0.70 0.55 0.63 0.75 0.60 0.68 4. ERRORS
March 0.66 0.56 0.61 0.75 0.64 0.69
April 0.52 0.58 0.55 0.65 0.62 0.64
May 0.50 0.66 0.58 0.61 0.72 0.66 This sectiondiscusses
the accuracyof the estimation
June 0.31 0.73 0.52 0.53 0.68 0.60 of hemisphericand global averageanomalies.The error
July 0.29 0.59 0.44 0.48 0.68 0.58 of estimation is due to two factors: measurement error
Aug. 0.40 0.58 0.49 0.49 0.67 0.58 and, most importantly, changesin the availability and
Sept. 0.41 0.56 0.48 0.49 0.66 0.57
Oct. 0.63 0.62 0.63 0.51 0.67 0.59
densityof the raw data, i.e., samplingerrors.
Nov. 0.74 0.60 0.67 0.59 0.68 0.63
Dec. 0.69 0.60 0.65 0.74 0.60 0.67 4.1. Measurement Errors
Year 0.54 0.60 0.57 0.60 0.65 0.62 Both the basic land and marine data are assessed for

All temperature changevaluesare in degreesCelsius. errors each month. Extreme land values exceeding3
standard deviations from the 1961-1990 statistics are all
checkedand spatialfields of normalizedvaluesare as-
fully operational,it will improvethe quality of the data, sessedfor consistency.Marine valuesmust passnear-
particularly for recent years,but it will succeedonly in neighbor checks, and 1ø area pentad anomalies are
partially restoringthe data coverageavailableduringthe trimmed values using winsorization [Afifi and Azen,
1951-1980 period. 1979]to temperthe effectof outliers.Our applicationof
Improvementsare alsobeingundertakenovermarine winsorizationwas to replace all valuesabovethe fourth
areas. Buoy data are becoming increasinglyavailable, quartile by the fourth quartile and all valuesbelow the
particularlyin the SouthernHemisphereand the tropi- first quartile by the first quartile, before averaging.Al-
cal Pacific.Efforts are underwayin the United States(at ternative thresholds such as the thirtieth and seventieth
the National Climatic Data Center (NCDC), through percentile would have been possible.Despite these
the Comprehensive Ocean-Atmosphere Data Set checks,it is likely that erroneousvalues enter both the
project)and in the United Kingdom(the Hadley Centre land and marine analyses,though any effect is consid-
of the MeteorologicalOffice) to digitize much of the ered to be essentiallyrandom [Trenberthet al., 1992].It
ships'log material that is knownto be availablein some is also probable, although less likely, that correct but
regions,principallyfor the two World War periodsand extremevaluesmay be discarded.




0.5 -

0.0 Figure 2. Hemisphericand global temperature

averageson the annual timescale(1856-1998),
relative to 1961-1990.

0.15 -


-0.15 Global

' 18•80
' 19•00
' 19•20 19•40 19•60 19•80 2000


120E 180 120W 60W 0 60E 120E 120E 180 120W 60W 0 60E 120E


120E 180 120W 60W 0 60E 120E 120E 180 120W 60W 0 60E 120E







•20E •e0 •20W 60W 0 •0E •20E

Temperaturetrends (øCper 20 yr) over 192•44

-3 -2 -1 -0.5 0 0.5 I 2 3

Plate la. Trendof temperature ona seasonalandannualbasisfor the20-yearperiod1925-1944.Boxeswith

significantlineartrendsat the 95% level(allowingfor autocorrelation)
are outlinedby heavyblacklines.At
least2 (8) month'sdatawererequiredto definea season(year),andat leasthalf the seasons or yearswere
required to calculatea trend.


120E 180 120W 60W 0 60E 120E 120E 180 120W 60W 0 60E 120E


.•••••.••,•..•---•,•. _,•• -.

• o o ..:.. • ,


120E 160 120W 60W 0 60E 120E 120E 180 120W 60W 0 60E 120E





120E 18O 120W 60W 0 60E 120E

Temperaturetrends (øCper 20 yr) over 1978-97

-3 -2 -1 -0,5 0 0.5 I 2 3

Plate lb. Sameas Plate la but for the 20-yearperiod 1978-1997.








0 30E 60E 90E 120E 150E 180 150W 120W 90W 60W 30W 0

0 30E 60E 90E 120E 150E 180 150W 120W 90W 60W 30W 0

Figure3. Spatialpatterns
of (a) s/2and(b) ?,calculated
Figure 3b, dark shadingindicatesvaluesof <0.8.

4.2. SamplingErrors Samplingerrorsare now assessed in a more thorough

The availabilityand densityof data usedin the com- way,usingeitherbasicprinciples[Joneset al., 1997a]or
pilationof the surfacetemperaturefield are everchang- optimal averaging[Smithet al., 1994]. Standarderrors
ing. For mostof the post-1850period, apart from the (SEs) of regional and hemisphericestimatesdepend
two World War periodsand yearssinceabout1980,the upon two factors:the locationsand the standarderrors
changesshould have led to an improvementin the of the grid boxeswith data.Joneset al. [1997a]calculate
accuracyof the large-scaleaverageanomaliesthrough the SE of each grid box using
the useof morecompletedatafields.Mapsshowingdata
availabilityby decadefor the last 120yearsare givenby SE2 - s/27(1
- 7)/[1 + (n - 1)7] (])
Parker et al. [1994]. Samplingerrors were initially as-
sessedby using"frozengrids"[Jones,1994],i.e., compu- wheres•2istheaverage
of allstations
in thebox,
tation of the hemisphericanomaliesusingonly the grid 7 is the averageinterrecordcorrelationbetween these
boxesavailablein earlier decades.Although theseanal- sites,and n 'is the number of site recordsin the box. Over
ysesprovedthat samplingerrorswere relativelysmall marine boxesthe numberof stations(n) is taken to be
and that therewasno long-termbiasin the estimationof the number of individual SST measurementsdivided by
temperaturesusingthe sparsegridsof the nineteenth 5. The parameters
s•2 and7 maybe considered
century,they failed to take into accountthe effectsof temporalvarianceand a functionof the spatialvariance
regionsalwaysmissingand changesin the densityof the within each grid box, respectively.Standarderror esti-
networkwith time in someareas.The effectsof sampling mates are also made for grid boxeswithout data using
errors in the estimation of hemispherictemperature interpolation
of thes•2 and7 fields,withn setto zero.
trendshas alsobeen addressedby Karl et al. [1994]. The method takes into accountthe changingnumber

(a) Global mean decadal temperature anomaly







1860 1880 1900 1920 1940 1960 1980


(b) Northern hemisphere decadal temperature anomaly




Figure 4. One and two standarderrors
o.o (SE) on the hemisphericand globaltem-
peratureson the decadaltimescale.The
-0.2 shadedareas highlight +__
1 SE.



I , , • I , I , • • I , • • I • , • I • • • I • ,

1860 1880 1900 1920 1940 1960 1980


(c) Southern hemisphere decadal temperature anomaly

' ' i ' ' ' i ' ' ' I ' ' i I i



1860 1880 1900 1920 1940 1960 1980

of stationswith time in the individualgrid box seriesbut grid box size, and thus never approacheszero when (1)
doesnot correctfor it. Time seriesof individualgrid box would imply an SE value of zero. Joneset al. [1997a]
serieswill exhibit variance changesthat may be unreal- showthat an alternate assumptionabout the standard
isticand simplya resultof changingstationdensity.This error(SE = s/2,forn = 0, instead
of s•27)
factor is discussed more in section 5. largervalues(by ---0.01-0.02)than thosepresentedhere.
In midlatitude continental regions where there is The implicationsof the formula are that in any global
greatvariabilityof temperature with time,s/2will be SE assessment it is more important to havetemperature
higher than over lessvariable regionssuch as coastal estimatesfrom continentalregions,as opposedto oce-
areas, the oceansand tropics. In these continentalre- anic regions,of the same size, becauseof the former
gions, ?, which has a maximum value of 1, is lower regions' greater variability. If, for example, a limited
(---0.8-0.9) thanoveroceanicandtropicalregions(>0.9). network of siteswere to be deployedto measureglobal
Figure 3 illustratesthis with the fields for s/2and ?, temperature,a muchgreaterpercentageof themwouldbe
calculated for annual data on the interannual timescale. locatedoverlandthanthe land/oceanfractionwouldimply.
The value of ? is alwaysgreater than 0.7, for the 5ø x 5ø Figure3 shows thatthefieldsofs•2 and?arerelatively

smoothon the annualtimescale.This alsoapplieson the only and do not include additional uncertainties that
monthly timescale,where ? falls to minimum values of relate to the correction of SST bucket observations or to
about 0.7 in some regionsin the winter season.Joneset any residual urbanization effects over some land areas.
al. [1997a] used thesesmoothfeaturesto enablevalues
of these variables to be estimated for all grid boxes
without temperaturedata. SE valuesfor suchgrid boxes 5. ANALYSES OF THE SURFACE TEMPERATURE
can then be estimatedfrom Eq. (1)•with n = 0. The RECORD
reliability of these assessmentswas tested using 1000-
year controlrunsof generalcirculationmodels(GCMs), The record has been extensivelyanalyzed over the
for which completeglobalfieldsof surfacetemperature last 10-15 yearsin a seriesof reviewpapersor chapters
data were available[Joneset al., 1997a]. in nationaland internationalreviews,includingthosefor
With SE estimatesfor each grid box, which will vary the IPCC [e.g., Wigleyet al., 1985, 1986, 1997;Ellsaesser
in time due to data availability,we can calculate the et al., 1986;Folland et al., 1990, 1992;Jonesand Briffa,
averagelarge-scaleSE in a similar manner to the calcu- 1992; Jones, 1995a; Nicholls et al., 1996]. The hemi-
lation of the large-scaleaveragetemperatureanomalies, sphericand global seriesand their trends have already
beendiscussed in section3.4. In thissectionwe compare
the two periods of twentieth century warming, 1925-

Ng / Ng
SE2- • SE•
2cos(lati) • cos(lati)
i=1 i=1
(2) 1944 and 1978-1997, and discuss trends in the areas
affectedby extremewarmth, trends in Arctic tempera-
tures, trends in maximum and minimum temperatures,
where Ng is the total number of grid boxesin the region
the apparent disagreement between recent warming
under study.If all the grid boxeswere independentof
measuredat the surfaceand by satellitesfor the lower
eachotherthenthe large-scale SE2 wouldbe SEe di-
troposphere,and the last 150 yearsin the contextof the
vided by the number of grid boxes.However, the effec- millennium.
tive numberof independentpoints(Neff) is considerably
lessthan the total number of grid boxesand difficult to
estimate.Neff hasbeen discussed by earlier authors[e.g., 5.1. Comparison of the Two Twentieth-Century
Livezeyand Chen, 1983;Briffa and Jones,1993;Madden Warming Periods
Figures 1, 2, and 4 clearly indicate that most of the
et al., 1993 and Jonesand Briffa, 1996] and referred to
warming during this century occurred in two distinct
also as the number of spatial degreesof freedom. Neff
periods. The periods differ between the hemispheres
valuesin Joneset al. [1997a]were estimatedusingthe
and amongthe seasons,but the periodsfrom about 1920
approachsuggested by Maddenet al. [1993]. to 1945 and since 1975 stand out. Plate 1 shows the trend
The SE of the global, hemisphericor large-scaleav-
of temperature changeby seasonfor the two 20-year
erage is therefore
periods,1925-1944(Plate la) and 1978-1997 (Plate lb).
(3) The annual globaltemperatureincreasesduring the two
SmglobaI = sm2/Neff
periods were 0.37 and 0.32øC, respectively,with the
followingan analogyfrom earlier work by Smith et al. secondperiod on average0.27øCwarmer than the first.
[1994].Typicalvaluesof Neff dependon timescale,mak- In both periods,the regionswith the greatestwarming
ing comparisonsbetween different studies difficult. and coolingoccurover the northern continentsand,with
Joneset al. [1997a], for annual data on the interannual the exceptionof a few isolated grid boxes, during the
timescale, obtain values of ---20 for both surface obser- December-January-February (DJF), March-April-May
vations and similar fields from the control runs of (MAM), and September-October-November(SON)
GCMs. Values increase to --•40 for seasonal series on seasons.The warming is generallygreatestin magnitude
this timescale,with slightlyhighervaluesin the northern during the DJF and MAM seasonsover the high lati-
summer comparedto the winter season.The timescale tudesof the northerncontinents,althoughtheseregions
dependenceof all the formulaemeansthat the SE values are not alwaysthosewhere local statisticalsignificanceis
must be calculated on the timescale of interest. SE values achieved. Small trends over some oceanic areas are
on the decadaltimescaleare onlymarginallysmallerthan significantbecauseof low variabilityin year-to-yeartem-
those calculated on the interannual timescale. perature values. The 1978-1997 period has a much
Figure 4 showsthe global and hemispherictempera- larger area with statisticallysignificantwarming com-
ture values with appropriate standard errors on the pared with 1925-1944.The pattern of recentwarmingis
decadaltimescale.Given the greaterpercentageof miss- strongestover northern Asia, especiallyeasternSiberia.
ing regionsin the SouthernHemisphere, the SE values It is also evident over much of the Pacific basin, western
are highestthere. SE valuesdecreasewith time during partsof the United States,westernEurope, southeastern
the courseof the last 150 years,reachingtypicalvalues Brazil, and parts of southern Africa. The 1925-1944
for the globeduringthe last 50 yearsof 0.048øC(0.054) warming was strongest over northern North America
on the decadal(interannual)timescale.It must be re- and is clearlyevidentover the northernAtlantic, partsof
memberedthat these estimatesare for samplingerrors the westernPacificand centralAsia. The improvements

in coveragebetweenthe two periodsare apparent,and 5.3. Trendsin Arctic Temperatures

some shippingroutes are surroundedby data voids in There has been recent controversyregardingArctic
1925-1944 (see alsoParkeret al. [1994]). temperaturetrends.We discussin this sectionthe three
principalanalyses.Chapmanand Walsh[1993]examined
5.2. Trendsin the AreasAffectedby Extreme the overall seasonal behavior of Arctic land station tem-
Warmth peratures between 1961 and 1990. Their results show
The previous section showsthat 20-year warmings that in winter and spring,warming dominatedwith val-
were greatestover continentaland high latitude regions uesof about0.25 and 0.5øC/decaderespectively.In sum-
where temperatures vary most on interannual time- mer, the trend was near zero, and in autumn the trend
scales:extreme colorsin Plate 1 are not apparent over either was neutral or showeda slightcooling.Consider-
the tropicsand the oceans.The 95% significancelevels ing the longerrecordback to the mid nineteenthcentury
on Plate 1, however,showthat warming,particularlyfor [Jones,1995c], only the spring seasonhad its warmest
the mostrecent 20 years,wasjust assignificantin the less yearsof the entire period recently,i.e., in the 1980sand
interannually and decadallyvarying regions compared 1990s.
with the highly variable regions.Recently,Joneset al. For the temperature trends over the pack ice, there
[1999] has transformedthe surfacetemperature data- are two investigations[Kahl et al., 1993; Martin et al.,
basefrom one in anomalieswith respectto 1961-1990 to 1997], which yield contradictoryresults. We refer to
percentilesfrom the sametime period. This enablesthe these as KCZ and MMD, respectively.For the western
rarity of extreme monthly temperaturesto be directly Arctic (70ø-80øN, 90ø-180øW) and the time period
comparedbetween grid boxes.The transformationwas 1950-1990, KCZ used a combination of surface air
performed usinggamma distributions,which treat each temperature data determined from dropsonde data
grid box impartially, accountingfor the effects of both taken by the U.S. Ptarmigan flights for 1950-1961 and
differences in interannual variability acrossthe world, radiosonde data from the fUSSR-manned North Pole
and differences in skewness. (NP) drifting ice stationsfor 1954-1990 (data are dis-
Plate 2 compares the anomaly and the percentile cussedin more detail in section6.4). From the combi-
method for displayingannual temperaturesfor the year nation of these two data sets, KCZ find a significant
1998(the year 1997is shownbyJoneset al. [1999,Figure cooling trend of 1.2øC/decadein autumn and 1.1øC/
7]). The percentilemap indicatesunusualtemperatures decade in winter, which contradicts the land-based stud-
over many tropical and oceanicregionsthat would not ies. Walsh [1993, p. 300], however,observes"that only
warrant a secondglance in the anomaly form. Similar one [NP] track extends as far south as 75øN in the
maps could be constructedusing the normal distribu- Western Arctic domain..." and that many of the drop-
tion, but seasonaland annual temperatures are often sondeprofileswere obtained from the climatologically
negativelyskewedover many continentaland high lati- warmer marginal ice zone, "... where all these (drop-
tude regions.The gammadistributionalsotakesaccount sonde)measurements were made in the first 12 years...
of this. of the study period." This implies that their observed
Figure 5 showsthe percentageof the analyzedarea of coolingtrends may be generatedby the geographically
the globe with annual temperatures greater than the warm bias of the dropsonde measurements.Another
ninetieth and lessthan the tenth percentile since 1900. sourceof potential biasis the instrumenttime lag as the
An increasein the percentageof the analyzedareaswith dropsondefalls through the Arctic inversion,and the
warm extremesis evident,but by far the greatestchange fact that the NP radiosondesurface temperaturesare
is a reductionin the percentageof the analyzedareawith always taken from multiyear ice or ice islands,while
cold extremes. Some caution should, however, be exer- some of the dropsondesurface temperaturescould be
cisedwhen interpretingthese results.Large changesin from open leads or thin ice.
coveragehave occurredover the twentieth century,so Given the impact that the KCZ paper had on the
some of the changesmay be due to areas entering the sciencecommunity,MMD searchedfor trendsin the NP
analysisduringthistime. Coveragechangesare minimal, 2-m air temperature data for 1961-1990, where these
however, since 1951. Second,grid box values based on temperatureswere taken at 3-hour intervals inside a
fewer stations or few individual SST observations, a Stevensonscreen, then resampled to 6-hour intervals.
more commonsituationat the beginningof the century, Their area of interestis an irregularregionin the central
are more likely to be extreme in this percentile sense Arctic, excludingthe climatologicalwarmer regions of
(see earlier discussionin section4.2 and Joneset al. the Beaufort Sea southof 77øN and the vicinityof Fram
[1997a]). Again, changesin the densityof observations Strait. In their trend analysis, MMD used both the
per grid box are minimal since1951.In future analysesit measuredand anomalyfields,where the anomalieswere
shouldbe possibleto correct for this secondproblem, the difference between the daily NP temperaturesand
producinggrid box time seriesthat do not havechanges the griddedmean daily field definedfrom optimal inter-
in variance due to changesin station or SST numbers, polation of the buoy, NP, and coastalstation tempera-
using some of the formulae developedby Joneset al. tures for 1979-1993 [Martin and Munoz, 1997]. The
[1997a]. purpose in using the anomaly fields was to remove






180 120W 60W 0 60E 120E 180

-5 -3 -1 -0.5 -0.2 0 0.2 0.5 1 3 5

Ocean: Parker et al (Climatic Change199•) Land: Jones(J.Climate 1994)

Surface Temperature Anomaly Percentiles(w.r.t. 1961-90)

• (Anomalies fitted to Gamma Distributions)

90N' •_.•• • _•_,, .___ __ I

60N --•'-"X• '-'• '-"•'•x._/--
30N "

30S x..


•0 120W 60W 0 60E 120E 180

0 2 10 20 30 40 50 60 70 80 90 98 100
Plate 2. Surfacetemperaturesfor 1998,relativeto the 1961-1990average,expressed (a) as anomaliesand
(b) as percentiles.The percentileswere definedby fitting gammadistributionsto the 1961-1990 annual
deviationsrelativeto the 1961-1990average,for all 5ø x 5øboxeswith at least21 yearsof annualdata in this

1910 1920 1930 1940 1950 1960 1970 1980 1990

Figure 5. Percentageof the monitored area

40 ' of the globe, for eachyear from 1900 to 1998,
with annual surface temperaturesabove the
3O ninetieth percentileand below the tenth per-
centile.The percentileswere definedby fitting
gamma distributions,based on the 1961-1990
period, usingall 5ø x 5ø boxeswith at least 21
years of data.

1910 1920 1930 1940 1950 1960 1970 1980 1990

geographicalinhomogeneitiesfrom the temperatures. by includingmore data (54% of the globalland areas)

For both fields, MMD found statisticallysignificant and extendingthe time seriesto 1993. (Until very re-
warming to occur in May (0.8øC/decade)and June cently,mean monthlymaximumand minimumtemper-
(0.4øC/decade),and on a seasonalbasis, they found atureswere not routinely exchangedbetweencountries.
significantwarmingin the summeranomalyfield (0.2øC/ This is one data set it is hoped will be dramatically
decade).Although the other seasonaltrendswere not improvedthroughGCOS (see section3.5).) This study
significantat the 95% level, MMD alsofound warming confirmedthe earlier study,showingthat for nonurban
in winter (0.35øC/decade)and spring (0.2øC/decade), stationsover 1950-1993,minimumtemperaturesroseby
and a coolingin autumn (-0.2øC/decade).Their num- 0.18øC/decade, maximaroseby 0.08 per decade,and the
bers are consistentwith land stations(Chapman and DTR decreasedby 0.08øCper decade.The new study
Walsh[1993] and the data used in 5ø grid box analyses loweredthe trend of minimumtemperatures(0.18 com-
here). paredwith 0.21) somewhatby incorporatingmuchmore
tropicalstationdata. It alsodecreasedthe possibilityof
5.4. Trends in Maximum and Minimum urbanizationeffectsbeing the causeby usingonly non-
Temperatures urban stationsin the analysis.
Up to now all the surfacetemperatureanalysespre- Plate 3 showsthe trendsfor each5øby 5øgrid box on
sentedin this paper have usedmonthlymean tempera- an annual basis for the 1950-1993 period. Minimum
tures. This situation arisesprincipally from the wide- temperatures decreasein only a few areas, while the
spread availability of this variable. Over the last few DTR decreasesin most regionsexceptArctic Canada
decades,greaterunderstandingof the causesof someof and some Pacific Islands.The new studymade use of
the changeshas been gained by consideringrelated manyregionalassessments reportedin a specialissueof
changesin other variables,suchas circulationand pre- the journalAtmospheric Research[seeKukla et al., 1995].
cipitation [e.g., Parker et al., 1994; Jonesand Hulme,
1996]. Other temperature variables to receive similar 5.5. Comparisons of RecentSurfaceWarmingWith
widespread interest are the monthly mean maximum Satellite Estimates
and minimum temperaturesand their difference, the Figures l, 2, and 3, and Plates lb and 3 all indicate
diurnal temperaturerange (DTR). These temperature that surfacetemperatureshave risen this century and
extremeserieshavemeaningonly over land regions,and particularlyover the last 20 years or so. The increaseis
there are only enoughdata spatiallyto warrant analyses seen in most areas and is evident over land and marine
sinceabout 1951. Longer data setsshouldexistback to regions(see especiallyPlate lb). Recently,somedoubt
the late nineteenthcenturybut are availabledigitallyfor hasbeen caston the surfacerecordby comparisons with
only a few countries(e.g., United Statesand fUSSR). satellite measuresof the average temperature of the
Karl et al. [1993b]were the first to extensivelyanalyze lower troposphere.The satelliterecord usesmicrowave
trends in these variables,using gridded data covering soundingunit (MSU) instrumentson the NOAA series
37% of the global land area. They found that for the of polar orbiters (see Christyet al. [1998] for the most
1951-1990 period, minimum temperatureswarmed at 3 recentreviewof the lower tropospheresatelliterecord).
times the rate of maximum temperatures.DTR de- Comparisonsare usuallymade with the MSU2R record
creasedover this period by 0.14øC/decade,while mean (whichmeasuresthe weightedverticaltemperaturepro-
temperaturesincreasedby 0.14øC/decade.Their study file of the tropospherefrom the surfaceup to 250 hPa,
confirmedearlier studiesfor the United States[Karl et centeredon -750 hPa) but the original MSU2 record
al., 1984].Easterlinget al. [1997]haveextendedthisstudy (surfaceup'to 200 hPa, centeredon -500 hPa) is also

Plate 3. Trends(in degreesCelsiusper 100years)for each5ø x 5øgridboxfor (a) maximum,(b) minimum,

and (c) diurnal temperaturerange for nonurban stations.Redrawn with permissionfrom Easterlinget al.
[1997]. Copyright 1997 American Associationfor the Advancementof Science.



, , ..,. -


Plate 3. (continued)

relevant [see Huh'ell and Trenberth,1996, 1997, 1998]. a trend close to -0.04øC/decade. However, over the
The latter is known to have partial lower stratospheric longer period from 1965 to 1996, the radiosonderecord
influences[see Christyet al., 1998]. agreeswith the surface (0.15øC/decadewarming). Ra-
The doubt comes from the fact that over the 1979- diosonderecordsare known to contain many problems
1996 period the surfacewarmed relative to MSU2R by of long-term homogeneity[Parkeret al., 1997, and ref-
0.19øCper decade[Joneset al., 1997b].Althoughneither erencestherein],but it doesseemstrangeto concentrate
the surfacewarming,the slightMSU2.R cooling,nor the on the 1979-1996 agreementsbetweenradiosondesand
difference is statisticallysignificant,doubts have been MSU2R and forgel the longer 32-year agreementbe-
caston the veracityof the surfacerecord since1979 and, tween the radiosondes and the surface. Fourth, and
by inference,the pre-1979period aswell. The doubtsare often ignoredin the comparisons, is the long-termagree-
often expressedin the non-peer-reviewed,"grey litera- ment, both over this centuryand since1979,betweenthe
ture," in newspaperarticles,and on Web pages. !and and marine componentsof the surfacerecord (see
Several intercomparisonshave been undertaken to also section2.2).
unravel reasonsfor the differences[see, e.g. Christyet Several reasonshave been postulatedfor the differ-
al., 1998;Hurrell and Trenberth,1996, 1997, 1998;Joneset ences.The MSU2R record is an amalgamationof many
al., 1997b].Severalissuesare undisputedin thesestud- different satellite series, each requiring calibration
ies. First, 20 years is too short a period over which to againsteachother aswell ascorrectionsfor orbital decay
calculate trends and draw meaningful conclusions.Sec- of the satellite and possibledrift in the satellitecrossing
ond, the two records are not the same, and their trends times. Changesrelative to the surfacerecordwere most
need not be equivalent.The recordsmeasuretempera- noticeable during 1981-1982 and again during 1991
tures at different levels.At the surface,recent warming (both times of satellite changes);these may be at least
is greater in minimum temperatures,and the shallow partly instrumentaland havebeen highlightedby Hurrell
nocturnal boundary layer over land areas is often de- and Trenberth[1996, 1997, 1998]andJoneset al. [1997b].
coupled from the lower troposphere. More realistic There are no evident reasonsfor heterogeneityin the
comparisons with lower tropospheric temperatures land or marine surface record at these dates. The
might be made with maximum temperatures,but these MSU2R-radiosonde agreement is generally good, al-
are less spatially extensive.Also, becausethe MSU2R though a similar comprehensivecomparisonsimilar to
record is short, it may be possiblefor its trend to differ the surface-MSU2R assessment of Joneset al. [1997b]
substantiallyfrom that for the surfacewithout the two has yet to be carried out. Comparisonsare usuallycar-
being physicallyincompatible. ried out as zonal and hemisphericmeans.Hurrell and
Third, the MSU2R trend agreesalmost exactlywith Trenberth[1998] showthat local colocatedcomparisons
radiosondedata [Parkere! al., 1997]aggregatedtogether showpoor agreementover tropicalcontinents.They also
to mimic the equivalentverticalprofile of MSU2R, with highlight the original MSU2 record, which shows a

warmingof 0.05øC/decade
over the 1979-1993 period. If 1200 1400 1600 1800 2000

the MSU2R is in error, the implication is that the

radiosonde record must also contain errors. The reason
favoredby Christyet al. [1998] is that the differencesare
real and representdifferencesin warming rates at dif- 0.0

ferent levelsin the lower troposphere.

Recently, another potential inhomogeneity in the
MSU2R recordhasbeenidentifiedby Wentzand Schabel
[1998]. It relates to the orbital decay of each satellite,
which is driven by variations in solar activity causing
variationsin orbital drag.They calculatethat on a global
basis, the effect should result in a correction of the
1200 1400 1600 1800 2000
MSU2R data by +0.12øC/decade.The effect does not
changeother MSU channels,includingthe raw MSU2
Figure 6. Northern Hemisphere temperature reconstruc-
data. J. R. Christy (personal communication,1998) tions from paleoclimaticsources.The three seriesare Mann et
claimsthat the effect is alreadyimplicitly accountedfor al. [1998,1999](thick),Briffaet al. [1998](medium)andJones
duringthe numerousintersatellitecalibrationsnecessary et al. [1998] (thin). All three annuallyresolvedreconstructions
to producethe continuousseriessince1979.Santeret al. havebeen smoothedwith a 50-yearGaussianfilter. The fourth
[1999], in an extensivestudyof troposphericand strato- (thickest) line is the short annual instrumentalrecord also
spheric temperature trends from radiosondes, MSU smoothedin a similar manner.All seriesare plotted as depar-
data, and model-basedreanalysessince 1958, indicate tures from the 1961-1990 average.
that all have potential biasesand inhomogeneitiessuch
that the wholeissueis unlikelyto be resolvedin the near
future. Only comprehensivecolocated comparisonsof [Mann et al., 1998, 1999] and two definitionsof summer
radiosondesand the MSU2R and greaterunderstanding [Briffaet al., 1998;Joneset al., 1998]). The shortinstru-
and correction of the biases in radiosondes will resolve mental record on an annual basis is superimposed.
the issue.In manyregionsthe necessarymetadata,giving Agreement with the annual instrumentalrecord is poor-
information on which sondes have been flown, are not est during the nineteenth century,partly becauseof the
alwaysavailable. different seasons(summer in two of the series)used.
The instrumental record also rises considerablyin the
5.6. The Last 150 Years in the Context of the Last last 2 decades,and this cannotbe seenin the multiproxy
Millennium seriesbecausethey end before the early 1980s,as some
The globalsurfacetemperaturehas clearlyrisen over of the proxy recordswere collectedduring theseyears.
the last 150 years (Figure 2 and Table 1), but what The most strikingfeature of the multiproxyaverages
significancedoesthis havewhen comparedwith changes is the warming over the twentieth century, for both its
over longer periodsof time? The last millennium is the magnitude and duration. The twentieth century is the
period for whichwe know mostaboutthe preinstrumen- warmestof the millenniumand the warmingduringit is
tal past, although it must be remembered that such unprecedented(see alsodiscussion by Mann et al. [1998,
knowledgeis considerablypoorer than for the post-1850 1999]andJonesetal. [1998]).The four recentyears1990,
period. Information aboutthe pastcomesfrom a variety 1995, 1997, and 1998, the warmest in the instrumental
of proxy climatic sources:early instrumental records series,are the warmest since 1400 and probably since
back to the late seventeenthcentury,historicalrecords, 1000. The end of the recent E1 Nifio event (suchevents
tree ring densitiesandwidths,ice core and coral records tend to warmer temperaturesglobally)and the greater
(see Joneset al. [1998], Mann et al. [1998], and many likelihood of La Nifia (which tends to lead to cooler
referencestherein for more details). Uncertaintiesare temperatures)as opposedto E1 Nifio conditionsduring
considerablygreater because information is available 1999and 2000 meansthat 1998will likely be the warmest
only from limited areas where these written or natural year of the millennium.The coolestyear of the last 1000
archives survive and, more importantly, because the years, based on these proxy records,was 1601. If stan-
proxyindicatorsare only imperfectrecordsof past tem- dard errors could be assignedto this assessment,they
perature change(see, e.g., Bradley[1985], Bradleyand would be considerablygreater than errors given during
Jones[1993],and the papersin volumessuchas thoseof the instrumentalperiod [Joneset al., 1997a]. The stan-
Bradleyand Jones[1995] andJoneset al. [1996]). dard errors givenby Mann et al. [1998, 1999] may seem
Over the last few years,a number of compilationsof relatively small, but they only relate to their regressions
proxy evidencehave been assembledfollowingthe pio- and include neither the large additional uncertaintiesin
neeringwork of BradleyandJones[1993].In Figure 6 we someof the proxyrecordsbefore the nineteenthcentury,
show three recent reconstructions of Northern Hemi- nor the standard errors of the instrumental record.
spheretemperaturefor part of the last millennium.The Numerous paleoclimaticstudieshave consideredthe
reconstructionsare all of different seasons(annual last millennium and two periods, the Little Ice Age

[Grove,1988;Bradleyand Jones,1993] and the Medieval north of 60øN. The four componentsare discussedsep-
Warm Epoch [Hughesand Diaz, 1994], are often dis- arately.
cussed.The latter two studiesquestionperceivedwis-
dom that these periods were universallycolder and 6.1. LandAreasExcludingAntarctica
warmer, respectively,than the present.The questioning For these regions we make use of the 1961-1990
arisesfrom the considerableadvancesin paleoclimatol- climatologydevelopedfrom station data by New et al.
ogythat havebeen made over the past 15-20 years.The [1999].In this analysis,12,092stationestimatesof mean
series used in Figure 6 contain from ten to several temperature for 1961-1990 are used to constructa 0.5ø
hundred Northern Hemisphere proxy climatic series, latitude by 0.5ø longitude climatology for global land
considerablymore seriesthan previouslyanalyzed.The areas excludingAntarctica. The temperature averages
coldest century of the period since 1400, the seven- were givenreliabilitycodes,expressedasa percentageof
teenth, was on averageonly 0.5ø-0.8øCbelow the 1961- the number of years available during the 1961-1990
1990 base. The differences between these studies and period. In regions of dense coverage, stations with
someearlier work is that they are basedon considerable higher percentageswere used preferentially.For most
amountsof proxy data rather than individual seriesor stations,mean temperaturewas calculatedas the aver-
schematicrepresentations of perceivedwisdom[Folland age of maximumand minimum temperature,thusavoid-
et al., 1990, Figure 7.1]. ing potential differencesdue to the method of calculat-
ing mean temperature(see discussion in section2.1).
The station data were interpolated as a function of
6. ANOMALY VERSUS ABSOLUTE TEMPERATURES latitude, longitude,and elevationusingthin plate splines
[e.g., Hutchinson,1995]. The use of elevation enabled
Throughout this review, temperature serieshave al- the splinesto define spatiallyvaryingtemperaturelapse
waysbeen quotedin degreesCelsiuswith respectto the rates in different mountainousareas,providedthat the
"normal" period 1961-1990.Earlier analysesused1951- basic data were adequate (see New et al. [1999] for
1970 [Joneset al., 1986a,b] or 1951-1980 [Parkeret al., further discussionof this point). Station data were in-
1994]. The only systematicdifferencesbetween these sufficientto achievespatiallyvaryinglapseratesin Ant-
analysesare simplychangesin referencelevel,dueto the arcticaand Greenland(seelater in sections6.3 and 6.4).
different normal periods. There will, however,be ran- The elevationdata usedoriginatedfrom a 5-min digital
dom differencesdependingon the normal period, due to elevationmodel (DEM), availablefrom the National
missingdata values affecting the computation of the GeophysicalData Center (NGDC) (see Acknowledg-
referenceperiod means(see also Wigleyet al. [1997]). ments) which had been degradedto 0.5ø cellsby aver-
Anomaly valuesovercomemost of the problemswith agingthe thirty-six5-min pixelsin each cell. The accu-
absolutetemperaturessuchas differencesbetween sta- racy of the interpolationshasbeen assessed usingcross-
tions in elevation, in observation times, the methods validationand by comparisonwith earlier climatologies
used to calculate monthly mean temperatures, and (see referencesof New et al. [1999]). We believe this
screentypes.Any changeswith time in these aspects climatologyrepresentsan important advanceover pre-
should be corrected in the initial homogeneityassess- viouswork in that it is strictlyconstrainedto the period
ment of a station temperature time series. Anomaly 1961-1990, elevationis explicitlyincorporated,and er-
valuesshouldbe all that are required in any analysisof rors are evaluated on a regional basis.
the basic5ø x 5ø grid box data set. For combinationwith the other three components,
The first absolutetemperature climatologywas pro- the 0.5ø x 0.5øresolutionwasdegradedto 1ø x 1ø.In this
ducedby Crutcherand Meserve[1970]for the NH and by degradation,four 0.5ø x 0.5ø cellswere averagedto give
Taljaardet al. [1969]for the SH. The 5øgrid point values one 1øcell. The DEM was also degradedto providea 1ø
from theseearlyworksare availabledigitally.Data avaiI- elevation field for all land areas.
abilitymusthavebeen poorerthen than now.Also, these
earlier analysesmust be basedon few yearsof record in 6.2. Oceanic Areas From 60øN to 60øS
data sparseregionssuchas Antarctica and parts of the In the anomaly data set we used SST anomaliesover
tropics,makingit difficultfor eachregionto be basedon marine regionsbecauseof their superioraccuracycom-
a consistentcommonperiod. paredwith MAT data. For the climatology,however,it is
Recent work, however,has enabled absolutetemper- preferable to use MAT, as MAT shouldmatch better
atures for the 1961-1990 period to be much better with land temperature on islandsand from coastlines
defined, using considerablyenhanceddata sets.In this than an SST climatology.
sectionwe produce a globallycompleteanalysisof the MAT climatologies,however,will be affectedby solar
absolute value of surface temperature for a common heating on board ship (see section2.2). To overcome
period, 1961-1990.Our absoluteclimatologyis made up this problem, we combineda new SST climatologyfor
of four components:land areas excludingAntarctica, 1961-1990 [Parkeret al., 1995b]with an air-seatemper-
oceanicareasfrom 60øNto 60øS,Antarcticapolewardof ature difference climatology,also for 1961-1990. The
60øS,and northern oceanic areas including the Arctic, SST climatologymakesuse of all availablein situ and

satellite SST data during the period, new marginal sea resultsin a linear increasein temperaturebetweenthe
ice zone SST estimates,and new Laplacian-basedadjust- ice sheetmargin and 60øS.In reality, the field is likely to
ment schemes[Reynolds,1988] in areas where there exhibita sharpincreasein air temperatureat the margin
were insufficientdatato producea true climatology.The of the continental ice sheet, although this step will be
air-seatemperaturedifferencedata usedsimilarinfilling smoothedout to someextentby using30-year averages.
techniquesbut made useof only NMAT and in situ SST Plate 5 illustratesthe Antarctic climatologyfor the
values from 1 hour before local sunset to 1 hour after four midseasonmonths. Even at the relatively coarse
local sunrise,to minimize the effect of solar heating on resolution, the influence of the sea ice can be seen,
ships'MAT. In the developmentof the air-seatemper- principallyby its absenceduring January. The strong
ature difference climatology, we assume that values gradientsacrossthe Antarctic Peninsuladuring summer
based on nighttime-onlyvalues are the same as those and autumn are evident as well as the coastal-sea ice
basedon all 24 hoursof the day. Both climatologiesare extent effectsin the Weddell and Ross Seasand Prydz
availableon a 1ø x 1ø resolution.Our air temperature Bay.
version is the sum of the air-sea and the SST climatol-
ogy. 6.4. Arctic Areas North of 60øN
Plate 4 illustratesthis climatologycombinedwith the In this region, data for land areascome from the New
land data for the four midseasonmonthsfor the region et al. [1999] climatology.There are considerablymore
60øN to 60øS. At this contour resolution, most of the stationsin this region than in the Antarctic, although
large spatialchangesare due to latitude, elevation,and there is only one high-elevationstationin the interior of
land-oceancontrasts.The latter are markedly stronger Greenland(and eventhisis for a relativelyshortperiod
in the winter hemispherebut are still evident in the of record).As with Antarctica,elevationover Greenland
transition seasons in some areas. couldbe incorporatedonly as a covariate.Elsewherein
the Arctic, elevationwasas an independentvariableand
6.3. Antarctica Poleward of 60øS followedNew et al. [1999].
For this region, averagetemperaturesfor 35 stations The most difficult part of this region is the Arctic
were collectedfrom similar sourcesto those given by Ocean and the extensivesea ice regionsin the northern
New et al. [1999]. Only 15 were near completefor the Atlantic and PacificOceans.The systematiccollectionof
1961-1990 period, the remaining 20 being either from temperaturedata within the packice of the Arctic basin
siteswith recordsfor part of 1961-1990 or from loca- beganwith Nansen'sdrift acrossthe Arctic in the Fram
tions with data only in the 1950s. Additional sources during 1893-1896. During this drift, temperatureswere
includeda number of Antarctic data publicationsfrom recordedat 6-hour intervals.Sverdrup[1933] describesa
countriesoperatingthe stations(e.g.,Schwerdtfeger et al. similar time seriesthat was taken during the 1919-1926
[1959],JapanMeteorological Agency[1995],and sources Maud drift, where the temperaturerecordextendsfrom
listedbyJones[1995b]).In a few years'time it shouldbe August 1922 to September1924. Sverdrup[1933] com-
possibleto increasethe number of sites available by bined these two sets of drift observations with land
using the then extended information from automatic stationobservationsand publishedthe first monthlycli-
weatherstations(AWSs). A numberof new AWS sites matologyfor the basin.
are located in regions considerable distances from In 1938, the fUSSR began their seriesof North Pole
manned stations[Steamset al., 1993]. As yet, though, stationice camps,whichwere oceanographicand mete-
record lengthsare too short. orologicalstationsestablishedby aircraft on either mul-
The monthlystationtemperaturenormalswere inter- tiyearice floesor ice islandsin the centralArctic [Colony
polatedusingthe samesplinesoftwareas the other land et al., 1992].The seriesbeganwith NP-1, whichgathered
areas. Only four of the 35 stationsare located at eleva- data from May 1938 to March 1939. Following World
tions in excessof 1500 m, the remainderbeing at eleva- War II, NP-2 gatheredduringApril 1950 to April 1951;
tions below 300 m. This was too sparsea network for a then in 1954 the establishment of NP-3 marked the
full three-dimensional interpolation. Elevation was beginning of the continuousoccupationof the basin,
therefore used as a covariaterather than a truly inde- where the fUSSR maintained 1-4 stationsuntil April
pendent variable, enabling the definition of monthly 1991, when NP-31 was recovered. The NP 2-m air tem-
continent-widerelationshipswith elevation.Theselapse peratureswere taken at 3-hour intervalsinsidea Steven-
ratesvariedmonthlyfrom 8 (summer)to 11 (winter) øC son screen.They have recentlybeen resampledto every
km-•, largerthanwouldbeexpected.
Thisismostlikely 6 hours and are available on CD-ROM from the Na-
due to the strongcovarianceof elevationand position, tional Snowand Ice Data Center (NSIDC). During the
i.e., elevationincreasesand temperaturedecreases
pole- NP period the United States and Canada also carried
wards.The interpolationassigned moreweightto eleva- out a seriesof shorter-termoccupationsof ice floesand
tion than position.The interpolationover land was ex- ice islands; these records are available from the NCDC.
tendedoveradjacentoceanareas(seaice areasfor part The next major set of temperatureobservationsbe-
of the year) to 60øS,where the MAT climatologypro- gan on January 19, 1979, as the part of the U.S. contri-
vided the northern boundary.The use of sucha control bution to the First Global AtmosphericResearchPro-

gram (GARP) Global Experiment (FGGE), and correlationdecay length of 1300 km, to produce a 12-
consistedof an array of air-dropped,satellite-tracked hour, 200-km griddedtemperaturefield for the basinfor
buoys [Thomdike and Colony, 1980]. The purpose of the 1979-1994 period.
thesebuoyswas to record both the large-scaleice defor- Buildinguponthiswork,Rigoretal. [1999]studiedthe
mation by determiningthe buoy position,and the pres- monthly mean dependence of the correlation length
sure field driving this deformation.A temperature sen- scales.Although they find that during autumn, winter,
sor was included only to aid in the pressure sensor and spring,their correlationlengthscalesapproximately
calibration. The instruments were mounted inside a ven- equal the Martin and Munoz value, during summerthe
tilated 0.6-m-diametersphericalhull. This programwas correlationlength scalesare smaller than during other
known as the Arctic Ocean Buoy Program until 1991, seasons,and much smaller between the coastal land and
when to reflect the growing contributionsfrom other ice temperaturesthan betweenthe differentice temper-
nations,the program name was changedto the Interna- atures, i.e., the buoy or NP temperatures.They used
tional Arctic Buoy Program(IABP). Since 1979 a large these length scaleestimatesto derive an improvedop-
number and variety of buoyshave been deployedby timal interpolationdata set calledIABP/POLES, which
parachute, aircraftlandings,ship,andsubmarine. Begin- provides12-hourtemperatureson a 100-km rectangular
ning in 1985the programbeganthe installationof buoys grid, for 1979-1997. In this analysisthe summerbuoy
having ventilated and shielded thermistors at 1-2 m data were alsoadjustedto matchthe NP statistics.Rigor
(R. L. Colony,private communication,1998). As illus- et al. find that the IABP/POLES data set has higher
trated on the IABP Web site (see acknowledgments), a correlationsand lower biasescomparedwith the NP and
large variety of designswere deployed,with both inter- land station observations than the POLES data, and
nal and external thermistors.Today buoyswith external providesbetter temperatureestimatesespeciallyduring
and radiation shielded thermistors mounted at 2-m summer in the marginal ice zones. The IABP/POLES
height are deployedas much as possible. data set is available from the IABP Web site.
As Martin and Munoz [1997] describe, there is a We use this new analysisfor the Arctic Ocean for the
general problem with the buoy thermistorsduring sum- years1979-1997 and combineit with the land data north
mer, which became apparent when the buoy summer of 60øN from New et al. [1999] and the marine air
meanswere comparedwith the NP means.For the air temperaturesnorth of 60øNin the northernPacificand
temperaturemeasurementsabovethick sea ice, a natu- Atlantic Oceans.The Arctic Ocean analysesare there-
ral calibrationperiod occursin summer.The reasonfor fore based on a slightly more recent period, but the
this is asfollows:at the beginningof summer,becauseof qualityof the 1979-1997 data for the regionjustifiesthis.
the natural sea ice desalinationcycle, the upper ice Any temperature change between these two periods
surfaceis virtually saltfree. Then as summerprogresses, over this region is thoughtto be minimal (see section
freshwatermelt ponds form on the ice surface,so that 5.3). Plate 6 illustratesthe derived climatologyfor the
the air temperature measurementsabove the ice are Arctic for the four midseasonmonthsof January,April,
provided with a natural calibration point that is very July, and October. The coolestparts of the Arctic are
close to 0øC. From the Fram and Maud data, Sverdrup over the land regionsof northern Canada and eastern
[1933] first describedthis isothermalperiod. From anal- Siberia. Over the central Arctic in winter, temperatures
ysis of those NP stationsnorth of 82øN, Martin et al. are coldest acrossthe region joining the coldest land
[1997] found for the period June 23 to July 31 a mean regions.In the summerseason,the Arctic Ocean is near
temperatureof -0.2øC with a standarddeviationof 0.9øC. 0øC almost everywhere,as may be expectedfrom the
Martin and Munoz [1997] found that most of the previous discussion.The influence of the warm North
individualbuoysfor the sameregionand for the summer Atlantic watersis evidentin all the seasonalplotsexcept
thatwere 1ø-3øCabove summer.

freezing, or much warmer than the means of the NP

temperatures.We attribute this temperatureincreaseto 6.5. Combinationof the FourComponents Into the
solar heating of the thermistors.Out of 261 buoysde- 1961-1990 Climatologyof SurfaceAir Temperatures
ployedduringthe period 1979-1993, only29 had a mean Combination of the analysesfor the variousdomains
temperature that fell within one standarddeviation of was achieved in the most straightforwardmanner. For
the NP summermean, and only four buoyslay within the zone 60øN to 60øS,the climatologicalvalue for each
half a standarddeviation.Of the remainder, only four 1ø squareis selectedfrom either the land or the marine
had a cold bias, so that 229 buoys had a warm bias. analysis.If both analyseswere available,the averageof
Informal investigationsby two of us (S.M. and I.R.) the two was taken. The two separateanalysesfor the
showthat the problemwith summeroverheatingcontin- Arctic and Antarctic regionswere then combined.The
ues throughthe 1997 summer.In spite of this problem, climatologyavailable on a 1ø x 1ø basis,together with
Martin and Munoz [1997] used a variety of filtering the surface DEM used, can be found at the Climatic
techniquesto remove these summeroffsets,then used ResearchUnit Web site (see acknowledgments).
an optimal interpolationschemefor the NP, buoysand It is now a trivial task to calculate the mean hemi-
coastalland station temperatures,assuminga constant sphericand global temperaturesfor the 1961-1990pe-


30N _

0 c


60S ß t f ! f • f

180W 135W 90W 45W 0 45E 90E 135E 180E


30N _


60S I ! I I I ! I

180W 135w 90w 45w 0 45E 90E 135E 180E




180W 135w 90w 45w 0 45E 90E 135E 180E


" t


180W 135W 90W 45W 0 45E 90E 135E 180E

-40 -30 -20 -10 -5 0 5 10 15 20 25 30

... :......

Plate 4. Climatologicalvalues of averageair temperature for the 1961-1990 period for the midseason
monthsof January,April, July, and October for the region 60øNto 60øS.




O3 I"/1 O3 m

08 I. 08 I.

-80 -50 -30 -20 -15 -10 -5 -2 0 2 5 10 15 20

Plate 5. Climatologicalvaluesof averageair temperaturefor the 1961-1990period for midseasonmonths

of January,April, July, and Octoberfor the Antarctic regionsouthof 60øS.


10 1:0

Jb.,... '•




o•o C
[' ,

! !

-80 -50 -30 -20 -15 -10 -5 -2 0 2 5 10 15 20

Plate 6. Climatologicalvaluesof averageair temperaturefor the 1961-1990 period for midseasonmonths

of January,April, July, and October for the Arctic region north of 60øN.

25 i i i i


Figure 7. Seasonalcycleof hemispheric

and globalmean temperaturesin absolute
degreesCelsiusbased on the 1961-1990
NH period.

_ NH _



_ _





riod. The averageglobal annual temperatureis 14.0øC, from the 1961-1990 period. Our main conclusions
are as
with the NH warmerthan the SH (14.6øCcomparedwith follows:
13.4øC).In the CrutcherandMeserve[1970]and Taljaard 1. Annual global surfacetemperatureswarmed by
et al. [1969] atlas climatologiesthe global average is 0.57øCover the period 1861-1997 and by 0.62øCover
14.1øC(NH, 14.9øC;SH, 13.3øC).Given the warmth of 1901-1997.Over both periodsthe warmingwas slightly
the 1961-1990 period compared with earlier periods greater in the SH than in the NH.
(--•1931-1965)usedby theseworkers,the differencesare 2. The measured temperature trends on a hemi-
lessthan 0.5øC.We would expectour climatologyto be sphericbasisare not very dependentupon the method
about 0.2øC warmer than the earlier ones if climatic
usedto combinethe basicland and oceanicdata, pro-
changeswere the only causesof the differences. vided that all the basicdata have been adequatelyad-
New et al. [1999] have estimatedcross-validationer- justed for variousinhomogeneities.
rors for their climatologyover large regionalland areas. 3. The warmestyearsof the record all occur in the
Typical errors are in the range 0.5ø-1.0øC,the larger 1990s.The four warmestyears,in descendingorder, are
valuesoccurringover either data sparseregionsor areas 1998 (0.57øC above the 1961-1990 average), 1997
with complex terrain. Because there are potentially (0.43øCaboveaverage),1995 (0.39øCaboveaverage)
greater errors in our climatology over the Southern and 1990 (0.35øCaboveaverage).
Ocean and Antarctica, due to sparseand incomplete 4. The errors of estimation,due to sampling,are
(with respectto 1961-1990) data and the slightlydiffer- dependentupon the timescaleof interest.On the inter-
ent Arctic Oceanbaseperiod,we havenot attemptedto annualtimescale,recentindividualglobalmean, annual
estimatestandarderrors for the climatology.Compari- mean temperature anomalies have standard errors of.
sonwith the earlier climatologiessuggests that the value _0.05øC.
of 14øC is within 0.5øC of the true value.
5. Patternsof warmingover the two majorwarming
Figure 7 showsthe annual cycleof temperaturesin periodsin this century(1925-1944 and 1978-1997) in-
the two hemispheresand the globe. The magnitudeof dicate that the warming has been greatest over the
the hemisphericseasonalcycleis 13.1øCin the NH and northern continents and in the DJF and MAM seasons.
5.7øC in the SH. In the NH the warmest and coldest
The warmingover the earlier periodwasslightlygreater
months are the expectedJuly and January.In the SH, (0.37øCcomparedwith 0.32øC)on a globalbasis.The
Januaryis warmest,but August rather than July is the recent warming has been accompaniedby increasesin
areas affected by significantlywarm temperaturesand
reductionsin the areas affected by significantlycool
7. SUMMARY 6. A recent reanalysisof Arctic temperaturedata
from the packice areasindicatesa slightwarmingon an
We have revieweda surfaceair temperaturedata set annual basiswith statisticallysignificantwarmingover
and produceda climatologyof surfacetemperatureson the 1961-1990 period in May and June.
a 1ø x 1øbasisbased, as best as can be achieved,on data 7. The recent increasein mean temperatureis the

result of strongerwarming in nighttime comparedwith REFERENCES

daytime.Over the 1950-1993 period, nighttime (mini-
mum) temperatureshavewarmedby 0.18øCper decade Afifi, A. A., and S. P. Azen, StatisticalAnalysis:A Computer
and daytime(maximum)temperatureshavewarmedby OrientatedApproach, 2nd ed., 442 pp., Academic, San
Diego, Calif., 1979.
0.08øCper decade. Alexandersson, H., andA. Moberg,Homogenizationof Swed-
8. Recent surfacetemperatureincreasessince1979 ish temperature data, I, A homogeneitytest for linear
have not been observedby satellite retrievalsof lower trends, Int. J. Climatol., 17, 25-34, 1997.
tropospherictemperatures.Much has been written and Bradley,R. S., QuaternaryPaleoclimatology:
Methodsof Paleo-
climaticReconstruction, 472 pp., Allen and Unwin, Win-
said on this issue,highlightingthe agreementbetween chester, Mass., 1985.
satellite and radiosondemeasurementsof lower tropo- Bradley, R. S., and P. D. Jones,Data basesfor isolatingthe
spherictemperaturesover the same period. However, effects of the increasingcarbon dioxide concentration,in
surfaceand lower tropospherictemperaturechangescan Detectingthe ClimaticEffectsof IncreasingCarbonDioxide,
editedby M. C. MacCrackenand F. M. Luther, pp. 29-53,
and do differ, and the available databasesare not yet U.S. Dep. of Energy, Washington,D.C., 1985.
good enoughto estimateconfidentlythe magnitudeof Bradley, R. S., and P. D. Jones, "Little Ice Age" summer
possibledifferencesover this near-20-yearperiod. The temperaturevariations:Their nature and relevanceto re-
agreementbetweensurfaceand lowertropospherictem- cent globalwarmingtrends,Holocene,3, 367-376, 1993.
perature trendsfrom radiosondesover the longer 1965- Bradley,R. S., andP. D. Jones(Eds.), ClimateSinceA.D. 1500,
706 pp., Routledge,London, 1995.
1997 period is generallyconvenientlyforgottenin many Briffa, K. R., and P. D. Jones,Global surfaceair temperature
commentaries. variations over the twentieth century, 2, Implications for
9. Proxy evidence for the preinstrumentalperiod large-scalepalaeoclimaticstudiesof the Holocene,Holo-
providesour only estimatesof temperature capable of cene, 3, 165-179, 1993.
Briffa, K. R., P. D. Jones, F. H. Schweingruber,and T. J.
puttingthe periodsince1851into a longer-termcontext. Osborn,Influenceof volcaniceruptionson Northern Hemi-
The most recent proxy compilationsindicate that the sphere summer temperaturesover 600 years,Nature, 393,
twentieth century was the warmest of the millennium 450-455, 1998.
and that the warmingduringit wasunprecedentedsince Chapman,W. L., and J. E. Walsh, Recent variationsof seaice
1400. The coldest century of the millennium was the and air temperaturein high latitudes,Bull. Am. Meteorol.
Soc., 74, 33-47, 1993.
seventeenth, with a temperatureaverageonly0.5ø-0.8øC Christy, J. R., R. W. Spencer,and E. Lobl, Analysisof the
below the 1961-1990average. mergingprocedurefor the MSU daily temperaturetime
10. The 1ø x 1ø climatology,developedfor surface series,J. Clim., 11, 2016-2041, 1998.
air temperature,indicatesan averageglobal tempera- Colony, R., I. Appel, and I. Rigor, Surface air temperature
observations in the Arctic Basin, Tech. Memo. TM 1-92,
ture of 14.0øC(14.6øCin the NH and 13.4øCin the SH). Appl. Phys.Lab., Univ. of Wash., Seattle, 1992.
Conrad,V., and L. D. Pollak,Methodsin Climatology,459 pp.,
Harvard Univ. Press,Cambridge,Mass., 1962.
ACKNOWLEDGMENIS. The authors thank Alan Rob- Crutcher, H. L., and J. M. Meserve, Selected-levelheights,
ock for suggestingthis review,and ChrisFolland, Mike Hulme, temperatures,
for theNorthern
Tim Osborn,the technicalreviewersof Reviewsof Geophysics Hemisphere,NAVAIR 50-1C-52 rev., U.S. Navy, Chief Na-
for suggestionsthat improvedit. BrionyHorton is thankedfor val Operations,Washington,D.C., 17 pp. plus charts,11
producingFigure6 and assistance with Figure7. This work has pp. + 18 figs., 144 charts,1970.
Easterling, D. R., and T. C. Peterson, A new method for
been supportedby a number of researchprograms.P.D.J.
detectingand adjustingfor undocumented discontinuities
acknowledges the supportof the U.S. Department of Energy climatological time series,Int. J. Climatol.,15, 269-377, 1995.
(grant DE-FG02-98ER62601). M.N. has been supportedby Easterling,D. R., B. Horton, P. D. Jones,T. C. Peterson,T. R.
the U.K. Natural Environment Research Council (grant Karl, D. E. Parker, M. J. Salinger,V. Razuvayev,N. Plum-
GR3/09721). Much of the data collectionof climatological ruer, P. Jamason,and C. K. Folland, Maximum and minimum
(1961-1990)data hasbeensupportedby the U.K. Department temperaturetrendsfor the globe,Science,
277, 364-367, 1997.
of Environment,Transport, and the Regions(contractEPG Ellsaesser, H. W., M. C. MacCracken, J. J. Walton, and S. L.
1/1/48).The developmentof the GISST2.2seasurfacetemper- Grotch, Global climatictrendsas revealedby the recorded
data, Rev. Geophys.,24, 745-792, 1986.
ature and seaice climatologyhasbeen supportedby the U.K. Folland, C. K., and D. E. Parker, Corrections of instrumental
Public Meteorological Service Contract. S.M. and I.R. both biasesin historicalsea surfacetemperature data, Q. J. R.
gratefullyacknowledgethe supportof NASA grant NAGS- Meteorol. Soc., 121,319-367, 1995.
4375, Polar Exchangeat the Sea Surface.I.R. also acknowl- Folland, C. K., and M. J. Salinger,Surfacetemperaturetrends
edges the support of the United States Inter-Agency Buoy in New Zealand and the surroundingocean,1871-1993,Int.
Program(USIABP) throughgrant N00014-96-C-0185.Many J. Climatol., 15, 1195-1216, 1995.
of the data sets upon which some of the figures depend, Folland, C. K., T. R. Karl, and K. Ya. Vinnikov, Observed
together with the climatology,are available on the Climatic climate variations and change, in Climate Change: The
IPCC ScientificAssessment,
edited by J. T. Houghton, G. J.
ResearchUnit, University of East Anglia, Web site (http://
Jenkins,andJ. J. Ephraums,pp. 195-238,CambridgeUniv.
www.cru.uea.ac.uk/). The Arctic buoy data are also available Press, New York, 1990.
from the IABP Web site (http://iabp.apl.washington.edu).The Folland, C. K., T. R. Karl, N. Nicholls, B. S. Nyenzi, D. E.
DEM is availablefrom the NGDC Web site (http://www.ngdc. Parker, and K. Ya. Vinnikov, ObservedClimate Variability
noaa.gov/seg/topo). and Change,ClimateChange1992: The Supplementary Re-

port to the IPCC ScientificAssessment,edited by J. T. Data Setfor the SouthernHemisphere,Tech.Rep. TR027, 73

Houghton, B. A. Callander, and S. K. Varney, pp. 135-170, pp., U.S. Dep. of Energy,Washington,D.C., 1986c.
Cambridge Univ. Press,New York, 1992. Jones,P. D., P.M. Kelly, C. M. Goodess,and T. R. Karl, The
Folland, C. K., M. J. Salinger,and N. Rayner,A comparisonof effect of urban warmingon the Northern Hemispheretem-
annual South Pacific island and ocean surface tempera- perature average,J. Clim., 2, 285-290, 1989.
tures, Weather Clim., 17, 23-42, 1997. Jones, P. D., P. Ya. Groisman, M. Coughlan, N. Plummet,
Grove, J. M., TheLittle Ice Age, 498 pp., Methuen, New York, W.-C. Wang, and T. R. Karl, Assessmentof urbanization
1988. effectsin time seriesof surfaceair temperaturesover land,
Hansen, J. E., and S. Lebedeff, Global trendsof measuredsurface Nature, 347, 169-172, 1990.
air temperature,J. Geophys.Res.,92, 13,345-13,372,1987. Jones,P. D., R. S. Bradley, and J. Jouzel (Eds.), Climatic
Hughes, M. K., and H. F. Diaz, Eds., The Medieval Warm Variationsand ForcingMechanismsof the Last 2000 Years,
Period, Clim. Change,26, 109-342, 1994. 649 pp., Springer-Verlag,New York, 1996.
Hurrell, J. W., Decadal trends in the North Atlantic Oscillation Jones,P. D., T. J. Osborn, and K. R. Briffa, Estimating sam-
and relationshipsto regional temperature and precipita- pling errors in large-scaletemperatureaverages,J. Clim.,
tion, Science,269, 676-679, 1995. 10, 2548-2568, 1997a.
Hurrell, J. W., and K. E. Trenberth, Satellite versus surface Jones,P. D., T. J. Osborn,T. M. L. Wigley, P.M. Kelly, and
estimatesof air temperaturesince1979,J. Clim., 9, 2222- B. D. Santer,Comparisonsbetweenthe microwavesound-
2232, 1996. ing unit temperature record and the surfacetemperature
Hurrell, J. W., and K. E. Trenberth,Spurioustrendsin satellite record from 1979 to 1996: Real differencesor potential
MSU temperatures from merging different satellite discontinuities?,J. Geophys.
Res.,102, 30,135-30,145,1997b.
records, Nature, 386, 164-167, 1997. Jones, P. D., K. R. Briffa, T. P. Barnett, and S. F. B. Tett,
Hurrell, J. W., and K. E. Trenberth, Difficultiesin obtaining High-resolutionpalaeoclimaticrecordsfor the last millen-
reliable temperature trends: Reconcilingthe surfaceand nium: Interpretation,integrationand comparisonwith gen-
satellite Microwave SoundingUnit records,J. Clim., 11, eral circulationmodel controlrun temperatures,Holocene,
945-967, 1998. 8, 455-471, 1998.
Hutchinson,M. F., Interpolatingmean rainfall usingthin plate Jones, P. D., E. B. Horton, C. K. Folland, M. Hulme, D. E.
smoothingsplines,Int. J. Geogr.Inf Syst.,9, 385-403, 1995. Parker, and T. A. Basnett,The use of indicesto identify
James,R. W., and P. T. Fox, Comparativesea-surfacetemper- changesin climaticextremes,Clim. Change,in press,1999.
ature measurements,Mar. Sci.Affairs Rep. 5, WMO 336, 27 Kagan,R. L., Averagingof Meteorological
Fields,379 pp., Klu-
pp., World Meteorol. Organ., Geneva, 1972. wet Acad., Norwell, Mass., 1997.
Japan MeteorologicalAgency,AntarcticMeteorological Data, Kahl, J. D., D. J. Charlevoix, N. A. Zaitseva, R. C. Schnell, and
Spec.VoL VI, 364 pp., Tokyo, 1995. M. C. Serreze,Absenceof evidencefor greenhousewarm-
Jones,P. D., Hemisphericsurfaceair temperaturevariations:A ing over the Arctic Ocean in the past40 years,Nature,361,
reanalysis andan updateto 1993,J. Clim.,7, 1794-1802,1994. 335-337, 1993.
Jones,P. D., Land surfacetemperatures•Is the networkgood Karl, T. R., and P. D. Jones, Urban bias in area-averaged
enough?,Clim. Change,31,545-558, 1995a. surfaceair temperaturetrends,Bull.Am. Meteorol.Soc.,70,
Jones,P. D., Recent variationsin mean temperature and the 265-267, 1989.
diurnal temperature range in the Antarctic, Geophys.Res. Karl, T. R., G. Kukla, and J. Gavin, Decreasingdiurnal tem-
Lett., 22, 1345-1348, 1995b. perature range in the United States and Canada from
Jones,P. D., Temperature changein the Arctic and the Ant- 1941-80, J. Clim. Appl. Meteorol.,23, 1489-1504, 1984.
arctic from instrumentaland high-frequencypaleoclimatic Karl, T. R., J. D. Tarpley, R. G. Quayle, H. F. Diaz, D. A.
(since 1400) reconstructions,in Wadati Conferenceon Robinson,andR. S. Bradley,The recentclimaterecord:What
Global Changeand the Polar Climate, pp. 5-8, Inst. of it can and cannottell us,Rev. Geophys.,27, 405-430, 1989.
Geosci.,Univ. of Tsukuba,Tsukuba,Japan, 1995c. Karl, T. R., P. Ya. Groisman, R. W. Knight, and R. R. Heim
Jones,P. D., and K. R. Briffa, Global surfaceair temperature Jr., Recent variations of snow cover and snowfall in North
variationsduringthe twentiethcentury,1, Spatial,temporal America and their relation to precipitation and tempera-
and seasonaldetails, Holocene, 2, 165-179, 1992. ture variations, J. Clim., 6, 1327-1344, 1993a.
Jones, P. D., and K. R. Briffa, What can the instrumental Karl, T. R., P. D. Jones,R. W. Knight, G. Kukla, N. Plummet,
record tell us about longer timescalepaleoclimaticrecon- V. Razuvayev,K. P. Gallo, J. Lindesay,R. J. Charlson,and
structions?,in Climatic Variationsand ForcingMechanisms T. C. Peterson,A new perspectiveon recent global warm-
of the Last 2000 Years,edited by P. D. Jones,R. S. Bradley, ing--Asymmetric trends of daily maximum and minimum
andJ. Jouzel,pp. 625-644, Springer-Verlag, New York, 1996. temperatures, Bull.Am. Meteorol.Soc.,74, 1007-1023,1993b.
Jones,P. D., and M. Hulme, Calculatingregionalclimatictime Karl, T. R., R. W. Knight, and J. R. Christy, Global and
seriesfor temperature and precipitation:Methods and il- hemispherictemperaturetrends:Uncertaintiesrelated to
lustrations,Int. J. Climatol., 16, 361-377, 1996. inadequatespatialsampling,J. Clim., 7, 1144-1163, 1994.
Jones,P. D., S.C. B. Raper, B. D. Santer, B. S. G. Cherry, Kukla, G., T. R. Karl, and M. Riches(Eds.), Minimax Work-
C. M. Goodess,P.M. Kelly, T. M. L. Wigley, R. S. Bradley, shop,Atmos. Res.,37, 1-275, 1995.
and H. F. Diaz, A Grid Point SurfaceAir TemperatureData Livezey,R. E., and W. Y. Chen, Statisticalfield significance
Setfor the NorthernHemisphere,Tech.Rep. TR022, 251 pp., and its determinationby Monte Carlo techniques,Mort.
U.S. Dep. of Energy, Washington,D.C., 1985. WeatherRev., 111, 46-59, 1983.
Jones,P. D., S.C. B. Raper, R. S. Bradley, H. F. Diaz, P.M. Madden, R. A., D. J. Shea, G. W. Brantstator, J. J. Tribbia, and
Kelly, and T. M. L. Wigley, Northern Hemispheresurface R. Webber, The effects of imperfect spatial and temporal
air temperaturevariations:1851-1984,J. Clim. Appl. Me- sampling on estimates of the global mean temperature:
teorol., 25, 161-179, 1986a. Experimentswith model and satellite data, J. Clim., 6,
Jones,P. D., S.C. B. Raper, and T. M. L. Wigley, Southern 1057-1086, 1993.
HemiSpheresurfaceair temperaturevariations:1851-1984, Mann, M. E., R. S. Bradley, and M. K. Hughes,Global-scale
J. Clim. Appl. Meteorol.,25, 1213-1230,1986b. temperaturepatternsand climate forcingover the past six
Jones,P. D., S.C. B. Raper, C. M. Goodess,B. S. G. Cherry, centuries, Nature, 392, 779-787, 1998.
and T. M. L. Wigley,A Grid Point SurfaceAir Temperature Mann, M. E., R. S. Bradley, and M. K. Hughes, Northern

Hemispheretemperaturesduring the past millennium:In- Smith, T. M., R. W. Reynolds,and C. F. Ropelewski,Optimal

ferences,uncertaintiesand limitations,Geophys.Res.Lett., averagingof seasonalsea surfacetemperature and associ-
26, 759-762, 1999. ated confidenceintervals(1860-1989),J. Clim., 7, 949-964,
Martin, S., and E. A. Munoz, Propertiesof the Arctic 2-Meter 1994.
Air temperaturefield for 1979 to the presentderivedfrom Stearns, C. R., L. M. Keller, G. A. Weidner, and M. Sievers,
a new gridded dataset,J. Clim., 10, 1428-1440, 1997. Monthly mean climatic data for Antarctic automatic
Martin, S., E. A. Munoz, and R. Drucker, Recent observations weather stations,in AntarcticMeteorologyand Climatology:
of a spring-summer surfacewarmingoverthe Arctic Ocean, StudiesBased on Automatic Weather Stations,Antarct. Res.
Geophys.Res.Lett., 24, 1259-1262, 1997. Set., 61, edited by D. H. Bromwichand C. R. Stearns,pp.
Mitchell, J. M., Jr., On the causesof instrumentallyobserved 1-21, AGU, Washington,D.C., 1993.
seculartemperaturetrends,J. Meteorol.,10, 244-261, 1953. Sverdrup, H. U., The NorwegianNorth Polar ExpeditionWith
Moberg, A., and H. Alexandersson,Homogenization of a theRS Maud 1918-1925,Scientific Results,vol. II, Meteorology,
Swedishtemperature data, II, Homogenized gridded air part 1, Discussion,331 pp., GeofysikInst., Bergen, 1933.
temperaturecomparedwith a subsetof global gridded air Taljaard, J. J., H. van Loon, H. C. Crutcher, and R. L. Jenne,
temperaturesince 1861,Int. J. Climatol.,17, 35-54, 1997. Climate of the UpperAir, SouthernHemisphere,vol. I, Tem-
New, M., M. Hulme, and P. D. Jones,Representingtwentieth peratures,Dew Pointsand Heightsat SelectedPressure Levels,
centuryspace-timeclimatevariability,1, Developmentof a NAk'AIR 50-1C-55, 135 pp., Chief Nav. Operations,Wash-
1961-90 mean monthlyterrestrialclimatology,J. Clim., 12, ington, D.C., 1969.
829-856, 1999. Thorndike, A. S., and R. L. Colony, Arctic Ocean Buoy Pro-
Nicholls,N., G. V. Gruza, J. Jouzel,T. R. Karl, L. A. Ogallo, gram data report, 19 January 1979-31 December 1979,
and D. E. Parker, Observedclimatevariability and change, Polar ScienceCenter Data Rep., 131 pp., Univ. of Wash.,
in The IPCC SecondScientificAssessment, edited by J. T. Seattle, 1980.
Houghton et al., pp. 133-192, CambridgeUniv. Press,New Torok, S., The development of a high-quality temperature
York, 1996. database for Australia, Ph.D. thesis, Univ. of Melbourne,
Oke, T. R., Reviewof urbanclimatology,1969-1973,WMO Tech. Parkville, Victoria, Australia, 1996.
Note 169, WMO 539, World Meteorol. Organ.,Geneva,1974. Trenberth, K. E., Recent observed interdecadal climate
Parker, D. E., The statisticaleffectsof incompletesamplingof changesin the Northern Hemisphere,Bull. Am. Meteorol.
coherent data series,J. Climatol., 4, 445-449, 1984. Soc., 71, 988-993, 1990.
Parker, D. E., Effects of changingexposureof thermometersat Trenberth, K. E., J. R. Christy, and J. W. Hurrell, Monitoring
land stations,Int. J. Climatol., 14, 1-31, 1994. global monthly mean surface temperatures, J. Clim., 5,
Parker, D. E., P. D. Jones, A. Bevan, and C. K. Folland, 1405-1423, 1992.
Interdecadalchangesof surfacetemperaturesincethe nine- van Loon, H., and J. Williams, The connection between trends
teenth century,J. Geophys.Res., 99, 14,373-14,399, 1994. of mean temperature and circulation at the surface, I,
Parker, D. E., C. K. Folland, and M. Jackson, Marine surface Winter, Mort. WeatherRev., 104, 365-380, 1976.
temperature:Observedvariationsand data requirements, Vinnikov, K. Ya., P. Ya. Groisman, and K. M. Lugina, The
Clim. Change,31,559-600, 1995a. empiricaldata on modernglobalclimatechanges(temper-
Parker, D. E., M. Jackson, and E. B. Horton, The GISST2.2 ature and precipitation),J. Clim., 3, 662-677, 1990.
seasurfacetemperatureand sea-iceclimatology,Clim. Res. Walsh, J. E., The elusiveArctic warming,Nature, 361, 300-
Tech. Note 63, Hadley Cent., Meteorol. Off., Bracknell, 301, 1993.
England, 1995b. Wentz, F. J., and M. Schabel, On the discrepancybetween
Parker, D. E., M. Gordon, D. P. N. Cullum, D. M. H. Sexton, observedin situ surfacewarming and the coolingtrend in
C. K. Folland, and N. Rayner, A new globalgriddedradio- MSU tropospheric temper.atures,Nature,394,661-664,1998.
sonde temperature database and recent temperature Wigley, T. M. L., J. K. Angell, and P. D. Jones,Analysisof the
trends,Geophys.Res.Lett., 24, 1499-1502, 1997. temperature record, in Detecting the Climatic Effects of
Peterson, T. C., H. Daan, and P. D. Jones, Initial selection of IncreasingCarbon Dioxide, edited by M. C. MacCracken
a GCOS surface network, Bull. Am. Meteorol. Soc., 78, and F. M. Luther, pp. 55-90, U.S. Dep. of Energy, Wash-
2145-2152, 1997. ington, D.C., 1985.
Peterson, T. C., et al., Homogeneity adjustmentsof in-situ Wigley, T. M. L., P. D. Jones, and P.M. Kelly, Empirical
atmosphericclimate data: A review, Int. J. Climatol., 18, climate studies: Warm world scenarios and the detection of
1495-1517, 1998a. climatic changeinducedby radiatively active gases,in The
Peterson,T. C., T. R. Karl, P. F. Jamason,R. Knight, and D. R. GreenhouseEffect, Climatic Change,and Ecosystems, pp.
Easterling,The first differencemethod:Maximizing station 271-323, SCOPE series,John Wiley, New York, 1986.
densityfor the calculationof long-termglobal temperature Wigley,T. M. L., P. D. Jones,and S.C. B. Raper, The observed
change,J. Geophys.Res., 103, 25,967-25,974, 1998b. global warming record: What does it tell us?, Proc. Natl.
Reynolds,R. W., A real-time global sea surfacetemperature Acad. Sci. U.S.A., 94, 8314-8320, 1997.
analysis,J. Clim., 1, 75-86, 1988.
Rigor, I. G., R. L. Colony, and S. Martin, Statisticsof surface
air temperatureobservations in the Arctic,J. Clim., in press, P. D. JonesandM. New, ClimaticResearch Unit, University
of EastAnglia,NorwichNR4 7TJ,England,United Kingdom.
Santer, B. D., J. J. Hnilo, T. M. L. Wigley, J. S. Boyle, (p.jones@uea.ac.uk)
C. Doutriaux, M. Fiorino, D. E. Parker, and K. E. Taylor,
Uncertaintiesin observationallybasedestimatesof temper-
S. Martin, Schoolof Oceanography,Universityof Wash-
ature changein the free atmosphere,J. Geophys.Res.,104, ington,Seattle,WA 98195-7940.
6305-6333, 1999. D. E. Parker,HadleyCentre,MeteorologicalOffice, Brack-
Schwerdtfeger,W., L. M. de la Canal, and J. Scholter,Meteo- nell, BerkshireRG12 2SY, England.
rologiaDescriptivadel SectorAntarcticoSudamericano, Publ. I. G. Rigor, Applied PhysicsLaboratory,Universityof
7, 425 pp., Inst. Antarct. Argentino, BuenosAires, 1959. Washington,Seattle,WA 98105-6698.