GEOPHYSICALRESEARCH LETTERS, VOL. 18, NO. 12, PAGES 2249-2251, DECEMBER 1991

INFLUENCEOF SPATIALLYVARIABLE INSTRUMENT NETWORKS ON CLIMATIC AVERAGES C. J. Willmort and S. M. Robeson Centerfor Climatic Research, Department ofGeography, University ofDelaware J. J. Feddema DepartmentofGeography, University ofCalifornia, Los Angeles

Abstract.Instrument networks for measuringsurface air temperature(T) andprecipitation (P)have varied consid- erablyover the last century. Inadequate observing-station locationshave produced incomplete, uneven, and biased samplesof the spatialvariability in climateand, in turn, terrestrialand globalscale averages of T and P havebeen biased. New high-resolutionclimatologies [Legates and Willmort,1990a; 1990b] are intensively sampled and inte- gratedto illustratethe effects of thesenontrivial sampling biases.Since station networks may not representspatial climaticvariability adequately, their ability to represent climatethrough time is suspect.

Introduction

Increasingdependence of the worldeconomy on climate- influencedrenewable resources (e.g., crops) has heightened our interestsin climate,climatic 'variability and climatic forecasting.In recentyea•s, general circulation or global climatemodels (GCMs) have simulated dramatic climatic changescenarios, intensifying scientific interest as well as economicand socialconcerns among the public and gov- ernment.Efforts to verifyor refuteGCM prognostications haveprovided new impetusfor climatologiststo exarnine the instrumentalrecord for evidenceof climatic change. Within this paper, we focuson how well the instrumen- tal recordrepresents areally averaged climate and climatic variability. While a number of climatic variablesare mea- suredroutinely, shelter-height air temperatureand precip- itation are examinedsince they havebeen the mostexten- sivelysampled in both spaceand time. How well temper- ature and precipitationstation networks represent spatial variability is of particular concernbecause ill-conditioned networkscan and have injected bias into spatial averages of climate. Fig. 1. Spatialdistributions of air temperaturestations From a climatologicalstandpoint, the instrumentalrec- in the NCAR surfacestation climatologyfor three se- ord is both incompleteand uneven. It is incompletein lectedyears: (a) 1886,(b) 1920,and (c) 1970.Equal area that there are few variables for which there is sufficient Molleweideprojection allows comparison of stationdensi- large-scalespatial and temporalcoverage for us to make ties betweenregions. confidentstatements about climatic variability. It is un- evenin severalrespects. Spatial and temporalvariability, for example,are better resolved for the recentpast and for or preciselythe instrumentalrecord represents climate at certainregions, such as North America,Europe, and In- the measurement(station) locations, the mostcommonly dia(Figure 1). Overviewsof thehistorical (instrumental) availablevariables, and the temporalvariability that can record[Ellsaesser et al., 1986]usually summarize the avail- be detectedin the instrumentalrecord. Documentederrors able instrumental data as well as what cax•be inferred from that affectclimatic data include: changes or discrepancies those data. Such reviews have focusedon how accurately in observingpractices [Mitchell, 1953; Karl eta!., 1986; Schaaland Dale, 1977],movement and siting of stations Copyright1991 by the AmericanGeophysical Union. [Mitchell,1953; Quinlan eta!., 1987],changes in thelocal environment[Mitchell, 1953; Karl andJones, 1989; Kukla Papernumber 91GL02844 et al., 1986],and problems in the waythat time-averages 0094-8534/91/91GL-02844503.00 are constructedat the stationlocations [Mitchell, 1953;

2249 2250 Willmort et al.: SpatiallyVariable Networks and Climatic

Edwards,1987]. Whether the availablestation networks trial meanas that obtainedfrom the entirehigh-resolution canfaithfully represent areas (particularly large areas) has climatology;however, marked discrepancies occur. Early beenincompletely examined. Tacit neglectof spatialrep- in the record(1880s to 1920s),when less than 500 NCAR resentativenessproblems by climatologistsis illustratedby stationsare available,estimates of terrestriallyaveraged the paucityof maps showingpertinent station networks, temperatureare highly variableand differ from the 'true' stationdensities, or spatialbiases [Ellsaesser et al., 1986]. value(the horizontaldashed line in Figure2) by as much as0.6 øC.Later in the record,the estimatesappear to ap- TerrestrialAir Temperature proachan equilibriumvalue that is about0.2 øC higher than the 'true' average. Of all climatic elements,mean air temperatureis the Moststudies of globaltemperature change do not explic- most commonlyused to characterizethe overall climate. itly addressthe problemsof unevenspatial sampling. How- Althoughmuch attention has been paid to errorsat in- ever,in a studyNorthern Hemisphere temperature change, dividualtemperature stations, perhaps even more uncer- usingan augmentedNCAR data set, an attempt to esti- tainty lies in the uneven distribution of stations and how mate the effectsof incompletespatial coverage was made this distributionhas changedthrough time. Data con- [Joneset al., 1986]. Insteadof varyingthe stationloca- tainedin the NCAR surfacestation climatology [Spangler tions and keepingthe underlyingfield fixed (as we have andJenne, 1988] commonly form the coreof globaltemper- done),an examinationof the differencebetween a fixed aturechange research [Jones et al., 1986;Hansen and Lebe- grid and a time-variablegrid was made, while the station deft, 1987]. While the NCAR archivewas createdso that locationsand temperature data alsochanged through time. spatial coveragewas as uniform as possible,dearly both Temperaturedeviations from stationaverages, rather than the numberand distributionof stationschange through- temperatureitself, also were used. Although thesedif- out the periodof record(Figures 1 and 2). ferencesmake direct comparison with our resultsdifficult, To examine the effects of uneven station distributions, largevariability (> 0.5 øC) occursearly in the time series the yearly NCAR. temperature station networksare eval- of Joneset al. [1986]as well. uatedusing a high-resolution(17,986 terrestrial stations) clima]tology[Legates and Willmort, 1990a]. The ability TerrestrialPrecipitation of each yearly N CAR temperature network to represent terrestrial-scalevariability is evaluatedby samplingfrom the high-resolutionfield at the NCAR stationlocations, in- Precipitationis probablythe mostextensively sampled terpolatingto a regular grid, and numericallyintegrating climatevariable. The numberof raingagesworldwide, to obtaina terrestrialaverage (Figure 2). The interpola- for instance,likely exceeds 120,000 [Mintz, 1981]while tionsare made using a spherically-based,weighted-average monthlyaverages are availablefor well over 24,000ter- algorithm[Willmort et al., 1985]that is reliable[Bussi&es restrialstations [Legates and Willmort, 1990b]. In spiteof and Hogg,1989]. Ideally, temperatures sampled at NCAR numerousobservations, station location bias still produces station locationsshould produce nearly the same terres- markedover- or underestimates ofregional (and ultimately global)averages of .As with air temperature stations,precipitation gages tend to be locatednear human 15.1 , -2500 settlements,with high concentrations of gages in Europe, NorthAmerica, India, and Japan. Few gages are located o / 2000 in the desertsor mountainregions. • 14.914.7 o i co Re-samplingagain is usedto illustratethe effectsof un- (13 o •_ o o . • evenstation networks on large-scale spattally averaged pre- :2 o • v \ l 1500 --• 14.5 ' cipitation.A high-resolutionannual mean precipitation field[Legates and Willmort, 1990b] is uaed to representthe • 1•.3 • 1000 'true'spatial variability in precipitation.Numerical (spa- tial)integration ofthe annual average field can be thought • 14.1 ...... b%• ...... of asthe 'real'globally averaged rate of precipitation.As • ....• 500 • 13.9 o __•...... before,the ability of eachyearly NCAR raingage network ..•-.-'- to representterrestrial average precipitation can be exam- 13.7 , ,,, , , , , .,...... , inedby sampling from the •true' field at thegage locations, interpolatingto a regulargrid [Willmort et al•, 1985]and Ye• thennumerically integrating to obtainestimates of spa- Fig. 2. Sevenpoint Gaussianfilter (solidline) applied tiallyaveraged precipitation. Yearly NCAR networks again to annualaverage terrestrial (excluding Antarctica and are of interestbecause they haveformed the basesof sew Green]and)air temperatureestimates (circles) computed eralstudies of temporalvariability in .spa•tiallyaveraged kom of precipitationstatistics [Bradley et al., 1987;Diaz et al., high-resolutionclimatology [Legates and Willmort, 1990a]. 1989]. Samplepoints were the yearly stationlocations in th• Raingagenetworks within the NCAR archive are highly NCARsurface station climatology. Total number of NCAR variablefrom year to year(Figure 3). Sparsestation net- [emperatures•a•ions also is given (dotted line). A bestes- works,particularly in the late 1800'sto the !ate 1920's, timateof the%rue' value of •errestrial average air temper- producesizable overestimates ofterrestrially averaged pre- amre(using allof the stations inthe climatology) isshown cipitation (•sa,). Losses of importantstations in Oceania as •he horizontal dashed line. duringWorld War II leadto underestimationof Pr in the Willmortet al.' SpaticilyVariable Networks and Climatic 2251

lOlO 3000 ß S. L. Grotch,Global climatictrends revealed by the 99O recordeddata, Reviews of Geophysics,2•, 745-792, 1986. 970 2500 c Hansen,J. andS. Lebedeff,Global trends of measuredsur- o 950 face air temperature,Journal of GeophysicalResearch, o 930 - 92, 345-413, 1987. . 2000 03 910 Jones,P. D., S. (7. B. Raper, R. S. Bradley, H. F. Diaz, •9o; P.M. Kelly,and T. M. L. Wigley,Northern hemisphere

ß 1500 L. surfaceair temperaturevariations: 1851-1984, Journal 870 ¸ ß _D of Climateand AppliedMeteorology, 25, 161-179, 1986. 850 ß Karl, T. R., C. N. Williams, Jr., P. J. Young,and W. M. 850 1 ooo D ß Wendland,A modelto estimatetime of observationbias 810 associatedwith monthlymean maximum, minimum, and 790 50O meantemperatures for the United States,Journal of Cli- 1890 19'10" ! 9•50 19'50 19'70 19 Year mate and AppliedMeteorology, 25, 145-160, 1986. Karl, T. R. and P. D. Jones,Urban bias in area-averaged Fig. 3. As in Figure2, exceptthe variableof interestis an- surfaceair temperaturetrends, Bulletin of the American nualaverage gage-corrected precipitation (ram/year) from MeteorologicalSociety, 70, 265-270, 1989. Legatesand Willmort [1990b]. Kukla, G., J. Gavin, and T. R. Karl, Urban warming, Jour- nal of Climate and AppliedMeteorology, 25, 1265-1270, 1986. 1940's. From 1950 to the present,the NCAR networks Legates,D. R. and C. J. Willmort, Mean seasonaland spa- adequatelyresolve Pt, although(in part)this is dueto tial variabilityin globalsurface air temperature,Theo- offsettingregional over- and underestimates. Since these retical and Applied Climatology,,il, 11-21, 1990a. network-inducederrors are nontrivial, they may obscure Legates,D. R. and C. J. Willmott, Mean seasonaland the identificationof real climatic changewithin the rain- spatial variability in gauge-corrected,global precipita- gagerecord. tion, International Journal of Climatology,10, 111-127, Conclusions !990b. Mintz, Y., A brief review of the presentstatus of global Evaluationof climate and climatic changefrom the in- precipitationestimates, in Atlas, D. and O. W. Theile strumental record is subject to severalbiases. Problems (eds.)Report of the Workshopon PrecipitationMeasure- associatedwith instrumentalbias (measurementerror, in- mentsfrom Space,NASA Goddard SpaceFlight Center, strumentexposure changes, temporal samplingbias, ur- Greenbelt,MD, p. 1,1981. banheating effects) have received much attention in recent Mitchell, J. M., Jr., On the causesof instrumentallyob- years.Our analyses,in addition,show that the unevenspa- servedsecular temperature trends, Journal of Meteorol- tial coverageof temperatureand precipitation networks has ogy, I0, 244-261, 1953. introducedmarked variability and bias into the large-scale Quinlan,F. T., T. R. Karl, and C. N. Williams, Jr., United climatic record. States historical climatologynetwork serial temperature and precipitationdata, NDP-019. Carbon Dioxide In- Acknowledgements.Fruitful discussionsand collabora- formationAnalysis Center, Oak Ridge National Labora- tionwith DavidLegates ()regard- tory, 1987. ingthe influenceof stationnetworks on spaticilyaveraged Schaal, L. A. and R. F. Dale, Time of observation bias climaticvariables are most appreciated.Financial support and ','Journal of AppliedMeteorology, for this researchwas made availableby the National Aero- I6, 215-222, 1977. nauticsand SpaceAdministration through grants NAGS- Spangler,W. M. L. and R. L. Jenne, World Monthly 853 and NAGW-1884. SurfaceStation Climatology,National Center for Atmo- sphericResearch, Scientific Computing Division, 14pp, References 1988. Willmort, C. J., C. M. Rowe, and W. D. Philpot, Small- Bradley,R. S., H. F. Diaz,J. K. Eisheid,P. D. Jones,P. scaleclimate maps' A sensitivity analysisof somecom- M. Kelly,and C. M. Goodess,Precipitation fluctuations mon assumptionsassociated with grid-point interpolL- overNorthern Hemisphere land areas since the mid-19th tion and contouring,American Cartographer,1œ, 5-16, century,"Science, 2oø7, 171-175, 1987. 1985. Bussi•.res,N. and W. Hogg,The objective analysis of daily rainfallby distanceweighting schemes on a mesoscale C. J. Wilknott and S. M. Robeson, Center for Cli- grid,Atmosphere Ocean, 27, 521-541,1989. matic Research,Department of Geography,University of Diaz,H., R. S. Bradley,and J. K. Eischeid,Precipitation ,Newark DE 19716. fluctuationsover global land areas since the late 1800's, J. J. Feddema,Department of Geography,University of Journalof GeophysicalResearch, 9•, 1195-1210,1989. California,Los AngelesCA 90024. Edwards,H. B., Samplingtheory applied to measurement andanalysis oftemperature forclimate studies, Journal (ReceivedJuly 18, 1991; ofClimate and Applied Meteorology, 26,731-736, 1987. RevisedSeptember 18, 1991; Ellsaesser,H. W., M. C. MacCracken,J. J. Walton,and AcceptedOctober 17, 1991)