FEBRUARY 1986 D.E. HARRISON AND D. S. GUTZLER 285Variability of Monthly-Averaged Surface and 850 mb Winds at Tropical Pacific Islands D. E. HARRISONCenter for Meteorology and Physical Oceanography, Massachusetts Institute of Technology, Cambridge, M,,I 02139 and Pacific Marine Environmental Laboratory/NOAA, Seattle, WA 98105 D. S. GUTZLER*Center for Meteorology and Physical Oceanography, Massachusetts Institute of Technology, Cambridge, MA 02139(Manuscript received 9 April 1985, in final form 8 July 1985) ABSTRACT We examine the variability of monthly mean winds at 850 mb and the surface from five island stations inthe tropical western Pacific Ocean. Climatological winds and (850 mb-surface) wind shear are evaluated andused to construct time series of monthly mean wind and shear anomalies. Wind variance at 850 mb tends tobe substantially greater than at the surface, and large temporal variations in shear are found. Prominent anomaliesare associated with El Nifio-Southern Oscillation periods. Composite El Nifio event anomalies are examined;it is found that the westerly wind anomalies associated with warm central Pacific sea surface temperatures arestronger at 850 mb than at the surface, and that the anomalous (850 mb-surface) shears are as large as thesurface wind anomalies themselves. Several simple techniques are described to investigate the feasibility of estimating surface wind anomaliesfrom 850 mb wind anomalies. Because strong correlations exist between the zonal winds at these levels, zonalestimate errors can be reduced to ~0.5 m s-' if known shear statistics are included in the estimate algorithm.Estimates which extrapolate cloud level wind anomalies to the surface using only climatological shear are shownto produce much greater surface wind errors. If these results are representative and if accurate monthly meanwinds at 850 mb can be obtained from cloud motion vectors, then very useful low-frequency surface wind fieldscan be derived from cloud motion data.1. Introduction The present interest in the variability of tropical atmospheric winds and the coupled ocean-atmospherephenomenon of the E1 Nifio-Southcrn Oscillation(ENSO) has led to increased study of winds at low levelsin the tropics. Data generally available are: low-levelcloud motion vectors (CMVs) which are typically assigned a nominal height of 850 mb in operational processing, and ship reports, island observations and buoywind recorder data from very near the surface. Thedataset for each level has limitations concerning datacoverage in space and time, duration of records, and/or data quality. Over the tropical Pacific, the two sets have oftenoffered quite complementary spatial coverage, and efforts have been made to combine them (Sadler andKilonsky, 1981, hereafter referred to as SK). Over thepast few years, efforts to provide CMV data have beenmade by the European community and the Japaneseand Indian governments. Global coverage is oftenavailable at present, but CMV data are obtained withdifferent techniques and serious intercomparison remains to be done. Harrison (1984) provides a brief * Present affiliation: CIRES, University of Colorado, Boulder, CO80309.c 1985 American Meteorological Societysurvey of many of the issues pertinent to the use of thevarious types of data. A crucial issue concerns how to adjust the cloudmotion data to an appropriate near-surface level. Sadlerand Kilonsky have used a monthly mean shear fieldconstructed as the difference between climatologicalmonthly mean CMV winds based on about seven yearsof data, and climatological monthly mean surfacewinds based on about 25 years of ship reports (Wyrtkiand Meyers, 1975). For a particular monthly meanCMV field, the climatological shear field is subtractedto produce a first-guess surface wind field, which isthen blended subjectively with the available ship, islandand buoy data into a surface wind field. Wylie andHinton (1982) have described efforts to use a differenttype of empirical shear adjustment, which dependsupon observed shear values for winds from differentdirections, evaluated region by region over the IndianOcean during FGGE. Only the SK approach is in usein an ongoing basis. Because the CMV dataset often provides spatialcoverage and data density far superior to that availablefrom the surface dataset, there is much interest in developing the best possible shear adjustment procedure(e.g., see WCP, 1983). For short time scales (days andless), it is well established that no systematic shear relationship exists between 850 mb and the surface (e.g.,Kanton and Cole, 1980). But at lower frequencies the286 MONTHLY WEATHER REVIEW VOLUME 11420ON20'S 140*E 160*E 180' 160*W FIG. 1. Locations of the five Western Pacific islands considered in this study.situation improves markedly, and SK have produceduseful monthly mean fields. However, the tropical Pacific undergoes dramatic changes during ENSO periods,during which the utility of a climatological shear relationship is not obvious. For instance, should a shearvalue largely based on a prevailing easterly trade windenvironment, be expected to apply during periods oflight easterlies to moderate westerlies? As part of an ongoing study of the vertical and horizontal structure of low-frequency variability of tropicalwinds, we have used multi-year time series of monthlymean winds ,'it 850 mb and at the surface at five islandsin the central and western tropical Pacific (Fig. 1) toexamine some of these issues. The rawinsonde observations at 850 mb will not necessarily be identical toCMV winds for the same time and place; there arevarious reasons why the two should differ in generaland can diffi:r greatly. Any results based on the 850mb winds will probably represent "best case" resultspossible for CMV winds. Unfortunately, CMV windsgenerally are not available around these islands, because the GOES-W satellite which is the basis for theCMV data does not provide coverage this far west inthe Pacific. First we present the monthly mean seasonal cyclesat the surface and at 850 mb for these islands, andsummarize the seasonal variation of vertical shear.Next, the interannual variability is briefly summarizedand the variance at 850 mb is shown to be much greaterthan at the surface. We then examine the least-square optimum bilinearregression between 850 mb and surface winds (andother linear regressions), and show that it is crucial toknow the ratio of the variances at the two levels inorder to obtain a good surface zonal wind estimate.The surface wind anomalies are then estimated fromthe 850 mb winds using four different procedures, including that of SK. The three others presented hereproduce much better estimates. Furthermore, we showthat the regression coefficients for the zonal wind ateach island are quite consistent, allowing us to formulate a simple, general adjustment scheme whichworks nearly as well as the statistically optimal parameterizations.2. Data Table 1 summarizes the time series that will be usedin this work. The 850 mb monthly means are l~akenfrom Monthly Climatic Data for the World, archivedon tape at the National Center for Atmospheric Research. The surface monthly means were constructedfrom tapes of individual observations in the NationalClimatic Data Center TDF-13 file. The surface data,in somewhat different form, have been described byLuther and Harrison (1984). Gaps exist in the data records at Majuro, Canton,and Johnston, as can be seen by comparing the recordlength and period of record in Table 1. These gaps aredue to missing or suspicious data in the upper air records. They occur eariy in the period at Majuro andCanton, in April 1969 at Johnston, and from November 1967 through November 1969 at Majuro. In~ addition, the period of record prior to January 1964 atTruk was neglected because the surface wind recordcontained a suspicious discontinuity at that point(Luther and Harrison, 1984). The shortest time series is that at Canton, for whichapproximately ten years of (noncontinuous) data areused. Each estimate of the monthly climatological surface and 850 mb winds is based on at least nine samples.Elsewhere, the monthly climatological estimates aretypically based on 12-18 samples.3. The climatological cycle of monthly mean wind and shear Figure 2a shows the climatological monthly mean850 mb vector winds at the islands, plotted bimon~thly.The zonal wind is easterly at all locations in everymonth, and is almost always stronger than the meridional wind. The meridional wind is less than 2 m s-~,except early in the year at Pago Pago. The changethrough the year is greatest at Truk, where the raagnitude of the vector wind varies between 1 and 8 ms-~; at the other islands the range is typically 3-5 ms-I. Typical bimonth-to-bimonth vector magnitudechanges are 1-2 m s-I except sometimes late in the TABLE 1. Data characteristics for the five island wind stations used in this study. Truk Majuro Canton Pago Pago JohnstortLatitude, longitude 7.5-N, 151.9-E 7.1-N, 171.4-E 2.8-S, 171.4-W 14.3-S, 170.6-W 17.0-N, 169.5-WPeriod of record Jan 64-Dec 78 Jan 59-Dec 78 Jan 57-Aug 68 Apr 66-Dec 78 Nov 62-Dec 78Record length (months) 180 210 121 153 193FEnRU^RY 1986 D.E. HARRISON AND D. S. GUTZLER 287 TRUK s MAJURO$ ?5 H JOHNSTON ICANTON(b)U850[ ~ i ~ I MS-I8 6 4 2 0TRUK gI ; ; i ~ MS-I8 6 4 2 0MAJUROPAGO PAGOCANTONPAGO PAGO(c)TRUKMAJUROJOHNSTON 55CANTON PAGO PAGO ~ i , MS-I U850- US 4 2 0 FIG. 2. (a) Climatological mean 850 mb wind vectors for the months of January, March, May, July, September, andNovember (labeled 1, 3, 5, 7, 9, 11, respectively). (b) As in (a) but for surface wind vectors. (c) As in (a) but for (850 mbsurface) shear vectors.288 MONTHLY WEATHER REVIEW VOLUME 114 year, when 'they can be 4-5 m s-~ as at Truk and Canton. Figure 2b presents the surface winds in analogousfashion. The climatological surface winds shown hereagree quite ,closely with the ship data compiled byWyrtki and iMeyers (1975), despite the differences indata source and periods of record. Note that the mean- winds are generally weaker and less zonal at the surfacethan at 850 rob; the typical annual range is also somewhat smaller. Maximum bimonthly changes are similarto those at 850 mb. The clima~tological vector wind shears between 850 mb and the surface are given in Fig. 2c. They are quite different from island to island. At Pago Pago the shear vectors differ' by less than I m s-t from bimonth to bimonth. At Truk, Johnston, and Canton the shear varies by as much as 4 m s-~ throughout the year, and up to 2 m s-l between bimonths. At Truk and Majuro the shear vectors are nearly colinear throughout the year (angular difference is less than 20-); they vary over about 30- at Pago Pago, 45- at Canton and 120- at Johnston. Another aspect of the shear is the angle between the850 mb and surface wind vectors. The climatologicalshear angle entered in Table 2 was determined by averaging the angular differences between the bimonthlysurface and 850 mb wind vectors shown in Fig. 2 foreach island. Clockwise turning from the surface to the850 mb level (i.e. anticyclonic turning in the NorthernHemisphere and cyclonic turning in the SouthernHemisphere) was defined to be positive, and a turningangle was calculated only if both surface and 850 mb- wind speeds were greater than 1.0 m s-~. Although the magnitude of the climatological turn ing angle varies widely from island to island, the signs are consistent: with Ekman dynamics in the boundary layer, except at Canton. Month-to-month variability in the climatological turning angle (not presented) is typically about half the mean value at each island; the sign of the angle does not change at any particular is land throughout the year. Clearly, the space and time variation of the Climatological shear is substantial, and cannot generally bewell approximated by any simple function of space ortime. Such variations are plausible, because these islands exist in quite distinctly different tropical windregimes, and no island is in the same sort of regimethroughout the year.4. Monthly mean wind variability We shall concentrate on the anomalies, defined, relative to the monthly mean climatological cycle. Figures3-5 present vector time series of anomalous monthlymean wind at 850 mb, the surface, and the 850 robsurface shear, respectively. Table 2 provides a statisticalsummary of these records. From Fig. 3 and Table 2 it is clear that the dominantvariability at 850 mb is in the zonal component; theratios of standard deviations (au/a~,) at 850 mb rangefrom 1.6 at Pago Pago to 4.1 at Canton, and the averageratio for all the islands is about 3 (Table 2). At thesurface, the ratios of zonal to meridional standard deviation are smaller, ranging from 1.0 at Johnston to1.5 at Majuro and Canton. The ratio of zonal standarddeviation at the surface to that at 850 mb is quite constant, varying between 0.47 and 0.63. No similar constancy 'is found between meridional standard de. viations. - The clearest signals at 850 mb are seen at Canton,where the monthly mean variability is spectrally "red".Episodes of substantial duration of significant westerlyanomaly are seen in late 1957-58, late 1963, and late1965-66, which are E1 Nifio periods. No periods ofsustained westerly anomaly exist except during El Nifioevents. At Truk and Majuro the.situation is less clear:sustained westerly anomalies are observed in 1963,-1972, 1976, and 1977 at Majuro, and in 1965, 1967,1968, 1972, 1974, 1976, and 1977-78 at Truk. In addition, periods of sustained easterly anomalies are justas prevalent, but the magnitude of the anomalous easterlies is substantially less than the westerly anomaliesassociated with ENSO. No straightforward relationshipis found at Pago Pago or Johnston, where the timeseries contmn less persistence ~n general.TABLE 2. Turning angle and wind statistics. Truk Majuro Canton Pago Pago JohnstonTurning angleclimatological (850-surface) 22- 13- 16- 224- 5-anomalous (850-surface) 9- 5- 2- -2- 11 ostandard deviation 29- 28- 60- 29- 29-Standard deviation of [850 2.4 1.9 2.5 1.7 2.~ anomalous zonal wind Jsurface 1.1 0.9 1.4 1.1 1.3. 850 0.8 0.7 0.6 1.1 0.8 anomalous meridional wind surface 0.7 0.6 0.9 ' 0.8 1.3Correlation r~(ff~, u'~) 0.90 0.86 0.77 0.70 0.87Correlation r~(~, t/s) 0.53 0.58 -0.01 0.68 0.61FEBRUARY 1986 D.E. HARRISON AND D. S. GUTZLER 289a59606162 8 50 WIND ANOMALY VECTORSMA,U~ CANTON I:~gOPaGOaDHNS1~ ~.~ ~ 66 I 67 ?o ~ - .-= 70 74 .,...."V.." f 74 78 d - 78 /_. . 79 79b EL NINO COMPOSITES ~G. 3. (a) Time series of monthly mean anomalous 850 mb windvectors (monthly mean 850 mb wind minus monthly climatology)for the five islands. Note the scale for u and v components in thecolumn for Pago Pago: each tick mark represents I m s-~. Verticalbars to let~ and right of the time series delineate El Nifio events used Luther and Harrison (1984) have discussed thechanges in these surface wind records associated withE1 Nifio events. A brief comparison of plotted surfaceanomalies (Fig. 4) with the 850 mb anomalies in Fig.3 emphasizes the implications of the statistics displayedin Table 2: at the surface, the variability is generallyless than at 850 mb and is more evenly distributedbetween the zonal and meridional components. Episodes of persistent zonal anomalies at the surface matchup well with zonal anomalies at 850 mb. Persistentmeridional anomalies are evident in the surface recordas well--for example, a southerly anomaly at PagoPago appears in late 1975 and persists through 1976,despite changes in sign of the zonal anomaly. Shear anomalies (Fig. 5) are generally of the samemagnitude as the surface anomalies. Hence, neglect ofanomalies in the 850 mb-surface shear when estimating surface winds from CMV data will inevitably produce estimate errors comparable to the surface anomalies themselves. An average turning angle for wind anomaly vectorswas calculated, considering only months where bothsurface and 850 mb anomalous speeds exceeded 1.0m's-~ (one-third to one-half of all months). As shownin Table 2, this averge turning angle had the same signas in climatology for each island, but individual valuesshowed wide variability despite the rather stringentminimum wind speed criterion. These results indicatethat there is little utility to the notion of a mean turningangle for the extrapolation of CMV data to the surface. As an example of a well-known signal in the surfacewind field, we examined the wind anomalies associatedwith the life cycle of the composite El Nifio event asdefined by Rasmusson and Carpenter (1982). Composite three-month average anomalies of 850 mb wind,surface wind, and (850-surface) shear for the five canonical event phases defined by Rasmusson and Carpenter are shown at the bottom of Figs. 3, 4 and 5,respectively. The composites are centered respectivelyon the September preceding each El Nifio (the "Antecedent'' phase), followed by the subsequent December ("Onset" phase), April ("Peak" phase), September("Transition" phase), and January ("Mature" phase).The five El Nifio events (1957, 1965, 1969, 1972, 1976)for which we have data are delineated by vertical barsin Figs. 3a, 4a and 5a. Qualitatively, the composite surface anomalies depicted here are similar to the Rasmusson and Carpentercomposites at all islands except Pago Pago, where thecomposite anomalies shown here have much smallermagnitudes than the strong southerly anomalies shownby Rasmusson and Carpenter. No consistent, quantitative relationship is evident between the surface andfor composite anomalies shown in (b). (b) Composite El Niho anomalies, representing three-month averages of Antecedent, Onset, Peak,Transition, and Mature phases, labeled A, O, P, T, and M, respectively(see text for details).290 MONTHLY WEATHER REVIEW VOLUME: 114a TRUK 58b A 0 P T Id!,~RI:~CE WIND ANOMALY VECTORS ~a~eO ~, .~ . 4 - ~. - ..- .1~ ~, ~.? L%- I~ El. NINO COMPOSITESCOMI~ITESPAGOP~ JOHNSTON 5859 596O 6061 - 61 '62 6265 63 - 67 69~ ?1 ?t ?$ 74 ?~ ?? ?~ ?~AopTM- A- 0. p T M 1~o. 4. (a, b)/is in Fig. 3a, b but for monthly mean anomaloussurface wind vectors. (c) Composite El Nifio surface wind veCtorstaken from Rasmusson and Carpenter (1982) at gridpoints adjacentto the five island.,; studied here.(SSO-SURFACE) SHEAR ANOMALY VECTORS TRUK MAJUR; CANTON PAGOPAGO JOHNSTON -4ACOMPOSITES 'NFIG. 5. As in Fig. 3 but for monthly mean anomalous (850 mb-surface) shear vectors.FEBRUARY I986 D.E. HARRISON AND D. S. GUTZLER 291850 mb vector anomalies. However, it is clear that the850 mb anomalies are typically larger, particularly sofor the Antecedent phase and for the largest anomalies.In other words, anomalous surface and 850 mb windsare associated with anomalous shear. A direct extrapolation of 850 mb wind anomalies to the surface usingthe climatological shear would systematically overestimate the true surface anomalies. Since the equatorialocean-atmosphere system is quite sensitive to anomalous surface wind convergence, an "observed" surfacewind field prepared with 850 mb level anomaliesbrought directly to the surface (as in SK) could lead tovery misleading interpretations.5. Estimating surface wind anomalies from 850 mb winds In this section we consider how best to use information from 850 mb, just above the planetary boundary layer, to estimate the variation of surface windanomalies. We use an elementary statistical approachand find substantial improvement over the methodof SK. First, consider a least-square bilinear regression ff~e = auU'8 + buv~ (la) V[e= avu'8 + &fig. (lb)Here ~e and Vie are our estimates of surface zonal andmeridional anomalous wind, ~ and v~ respectively, interms of the known anomalies d8 and v~ at 850 mb.The constants are determined empirically and are presented in Table 3. It is found that au varies between0.42 and 0.55 and is typically much larger than bu.Because u'8 typically dominates v~ it seems that ff~e isbest determined primarily by ds. Useful estimation ofv requires knowledge of both ds and v~, at least forCanton and Truk. The bilinear regressions provide a very good estimateof ~, as shown in Table 4. The standard deviations ofthe time series of estimates are generally slightly lessthan the observations (cf. Table 2), but greater thanTABLE 3. Empirical regression coefficients for (1) and (2) as follows.(1) bilinear regression: u',, = auu'8 + &,v~; ff~e = a,,vl8 + bvv'8(2) linear regression: ~, = affs; v[~ = bv'a(3) simplified parameterization: use = - u's ;(4) SK parameterization: use = u~; v[~ = Truk Majuro Canton Pago Pago Johnstona,, 0.42 0.42 0.42 0.43 0.55b~ 0.10 0.21 -0.10 -0.15 0.00a~ O. 13 0.06 -0.0 ! -0.03 0.06b~ 0.50 0.50 -0.02 0.52 0.92a 0.42 0.42 0.42 0.44 0.55b 0.49 0.51 -0.02 0.53 0.98the root-mean-square errors. For example, at Majurothe time series of observed surface zonal wind has astandard deviation of 0.9 m s-~. The bflinear estimate(la) yields a time series with a standard deviation of0.8 m s-~, and rms errors of 0.4 m s-I. Correlationcoefficients between the estimated and observed zonalanomalies (not shown) range from 0.71 at Pago Pagoto 0.91 at Truk. As suggested .above, neglecting v~ inestimating ff~ has very little effect; estimate and errorstandard deviations for the linear parameterizationstypically only vary by 0.1 m s- ~ from the bilinear values. To express this more precisely, if the meridionalcomponent is neglected, it is easily shown that the optimum (least-square) linear prediction of ~ in termsof ds has the form F ,~u(S) ] _ /~se: r.L~Z~SJ~,(2)where r~ is the correlation coefficient between ~ andu'8. The quantity on the right-hand side of (2) definesthe regression coefficient a in Table 3; the corresponding quantity for meridional winds is the coefficient b.From Table 3, it can be seen that ru is generally near1 (i.e., ~ and d8 are highly correlated), and [a~(s)/a~(8)]~ 0.5, so that a varies only between 0.42 and 0.55from island to island. Hence, a very good estimate of~ can be obtained from the simple formula ~se ~ 0.5d8. (3)Notice that the SK adjustment is equivalent toU';e = u'8 (4a)lfse = O~ (4b)so that we expect surface zonal wind anomaly estimatesmade this way to be too large by roughly a factor oftwo, at least for the area being considered here. Comparisons of time series of the surface zonal windestimate error, using (la), (2), (3), and (4a) in Table 4clearly illustrate these points. Figures 6 and 7 showtime series of u[~ using (la), (3), and (4a) for Majuroand Truk, respectively. The traces based on (1 a) and(3) are quite similar, and the errors are typically 0.5 ms-~ or less. Errors tend to persist for the largest timesduring ENSO periods. The SK estimate (4a) producesmuch !arger errors than either (la) or (3). Errors associated with the surface meridional component estimates are relatively larger. The standard deviations of the estimates are typically no larger thanthe error standard deviations, even for the bilinearregressions. At Canton, the surface meridional windcomponent is completely uncorrelated with the 850mb wind (Table 2), so the estimates have no skill whatsoever. In general, the surface-850 mb coupling ismuch weaker for the meridional component. Coincidentally, the meridional analogue to (3)292 MONTHLY WEATHER REVIEW VOLUME 114 T^BL- 4. Standard deviations of the time series of estimated surface wind anomalies [(u'~)m and (ff~)u2] and error standard deviations , [(ff~e - ff~):]m and [(ff~e - v~)2]m for the parameterizations defined in Table 3.Truk Majuro Canton Pago Pago JohnstonZonal windMeridional windZonal windMeridional wind Estimates1 1.0 0.8 1.0 0.8 1.2 1.0 0.8 1.0 0.7 1.3 1.2 1.0 1.2 0.8 1 .(4 2.4 1.9 2.5 1.7 2.(I 0.5 0.4 0.0 0.6 0.82 0.4 0.4 0.0 0.6 0.83 0.4 0.3 0.3 0.5 0.e~4 0.8 0.7 0.6 1.1 ErrorsI 0.5 0.4 '0.9 0.7 0.6- 2 0.5 0.5 0.9 0.8 0.63 0.5 0.5 0.9 0.8 0.64 1.4 1.2 1.7 1.2 1.11 0.5 0.5 0.9 0.6 1.02 0.6 0.5 0.9 0.6 1.03 0.6 0.5 0.9 0.6 1.14 0.7 0.6 1.1 0.8 1.0 ~e = 0.sv~ (5)works nearly as well as the bilinear fit (lb), but in thiscase the correlations rv are smaller than ru while thestandard deviation ratios are more nearly 1, so whenmultiplied together the factor of 0.5 is preserved. Thatis, (5) can be used to extrapolate 850 mb wind anomalies to the surface with nearly as much skill as (lb),but the fraction of variance explained by either (1 b) or(5) is small. The significance of these results was tested by recalculating the regression coefficients separately for thefirst and second halves of each island's time series. Values of au and a for each half record at each island werewithin 10% o.f the values listed in Table 3 with theexception of Pago Pago, where a~ was calculated to be0.38 and 0.52 for the first and second halves, respectively. Values of b~ were consistently small except forthe second half record at Pago Pago, for which it was-0.29. The meridional coefficients av, b~, and b weremuch less reproducible; in some cases even the signsof these coefficients changed from one-half of the recordto the other.6. Concluding remarks We have shown that ~very strong correlations existbetween monthly mean zonal wind anomalies at 850mb and those at the surface in the central and westerntropical Pacific. A bilinear recession between surfaceand 850 mb winds leads to correlations between zonalsurface wind anomaly estimates and true zonal surfacewind anomalies of about 0.8 and error standard deviations of 0.5 to 0.9 m s-~. For surface zonal windanomaly estimates it is satisfactory to neglect the meridional wind anomaly entirely and to take ~ = 0.su'~.The SK procedure appears to produce zonal wind errors roughly twice as large as can be obtained usingthis expression. The bilinear estimate for vie is somewhat superiortoany simple relationship like (5), but no estimate examined here does as well for v[e as is possible for ff~e.Still, an accurate estimate of only the zonal wind isvaluable, because much of the variance of surface windsis contained in the zonal component. The lack of skillin obtaining reliable meridional surface wind estimatesimplies that surface convergence fields may be suspect,however. The apparent universality of the factor of 0.5 in (3)is a surprising outcome of this study. Does it also holdfor other locations over the tropical ocean? No simplephysical justification for this value or for its generalityis evident. However, considering that these five islandsreceive widely varying amounts of precipitation, andthat the rainfall rate at each island has dramatic interannual and seasonal variability (particularly associatedwith ENSO), it seems unlikely that local convectiveactivity is responsible. Should (3) prove to be a generalresult, it will pose an interesting result to be explainedby tropical boundary layer theory. However, simple schemes such as (3) only workFEBRUARY 1986 D.E. HARRISON AND D. S. GUTZLER 293MAJURO Us (Est.) ERRORS5960616263646~6E70717274 -I O +1 -I O ~,1 -I O *1 (o) (bl (-) FIG. 6. Time series of zonal wind estimate errors (u,e - us) atMajuro for (a) Sadler and Kilonsky parameterization 4 in Table 3,(b) simplified parameterization 3 in Table 3 and (c) bilinear regressionI in Table 3.when applied to wind anomalies and not to the windfield itself. It is readily seen from Fig. 2 that the climatological shear varies from month to month andTRUK Us (Est.) ERRORS64656667686970717273?47576 77 78 79-I 0 ,H -I O ~.1 -I 0 + I (a) (b) (-)FIG. 7. As in Fig. 6 but for Truk.from island to island, so that the climatological analogue of (3) u-~ = 0.5~does not hold generally with any useful degree of accuracy. We have seen that strongly anomalous periods tendto have strongly anomalous shear between 850 mb andthe surface, as might have been anticipated. The SKadjustment scheme for cloud motion vectors will tendto produce the largest surface wind errors during thesedynamically important periods. A very pleasing findingof this work is that surface errors based on the regressions presented are not significantly greater than normal during El Nifio events.294 MONTHLY WEATHER REVIEW VOLUME 114 To the extent that cloud motion vector anomaliesreproduce 85,0 mb wind anomalies and also that (2) or(3) are applicable and the needed coefficients known,CMV anomaly data appear able to usefully augmentother surface: zonal wind anomaly data. But the SKprocedure appears likely to lead to surface wind errorsmuch larger 'than can be obtained using (2) or (3). It is important to keep in perspective the limitedspatial coverage of our data when attempting to generalize these results. The relevance of (3) to the remainder of the tropical Pacific is not known. However,it has been shown here to be useful for a wide rangeof meteorological environments and its apparent universality deserves to be examined further with otherdata. This work does not begin to address other issuespertinent to a determination of how best to use CMVdata to document surface wind variations. Obviouslimitations include the use of monthly mean data, thelack of data between the surface and 850 rob, and theabsence of contemporaneous surface data. There is aclear need for detailed comparison of individual andtime-averaged CMV data and rawinsonde data. Wehope that the international TOGA program may provide the impetus for further data collection and studyon the many issues of relevance. Acknowledgments. This work was supported by NSFGrant OCE83-01787 to M.I.T. and by TOGA andEPOCS Grants to P.M.E.L.REFERENCESHarrison, D. E., 1984: Ocean surface wind stress. Large-Scale Oceanographic Experiments and Satellites, C. Gautier artd M. Fieux, Eds., Reidel 99-115.Kantor, A. J., and A. E. Cole, 1980: Wind distributions and interlevel correlations, surface to 60 km. Air Force Geophysics Laboratory, Environ. Res. Paper No. 713, 115 pp. [AFGL-TR-80-0242].Luther, D. S., and D. E. Harrison, 1984: Observing long-period fluc tuations of surface winds in the tropical Pacific: Initial results from island data. Mon. Wea. Rev., 112, 285-302.Rasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical sea surface temperature and surface wind fields associated with the Southern Oscillation/El Nifio. Mon. Wea. Rev., 110, 354 384.Sadler, J. C., and B. J..Kilonsky, 1981: Tradewind monitoring using satellite observations. Rep. UHMET 81-01, Department c f Me teorology, University of Hawaii, 23 pp.World Climate Program, 1983: Interim ocean surface wind da~a sets, No, 68. ICSU/WMO International TOGA Office, WMO Sec retariat, Geneva, Switzerland.Wylie, D. P., and B. B. Hinton, 1982: The wind stress patterns over the Indian Ocean during the summer monsoon of 1979. J. Phys. O~ean., 12, 186-199.Wyrtki, K., and G. Meyers, 1975: The Wade wind field over the Pacific Ocean, Part 1: The mean field and the mean annual variation. Hawaii Institute of Geophysics, Rep. HIG-75-1, 25 pp.