JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 91, NO. C2, PAGES 2233-2240, FEBRUARY 15, 1986 Global Analysis of Ocean Surface Wind and Wind Stress Using a General Circulation Model and Seasat Scatterometer Winds EUGENIA KALNAY AND ROBERT ATLAS Laboratory for Atmospheres, NASA/Goddard SpaceFlight Center, Greenbelt, Maryland We present instantaneous and 15-daytime-averaged fields of surface wind, wind stress, curl of the wind stress, and wind divergence. These fields are derived from the Goddard Laboratory for Atmospheres four-dimensional analysis/forecast cycle,for the period September 6-20, 1978, usingconventional data, satellite temperature soundings, cloud-track winds, and subjectively dealiased Seasat scatterometer winds. 1. INTRODUCTION The importance of determiningsurface fluxesof heat, mois- ture, and momentum for the study of the ocean and its inter- action with the atmosphere, especially in the tropics, cannot be overemphasized. Unfortunately, these fluxes are not easily measured, and because of the lack of shipping routesthrough- out vast oceanic regions, even long-term time means in these areascan only be determined by smoothing and extrapolation of values from better observed regions. Surface wind stress is the single most important factor that determines the world ocean circulation. Estimations of wind stress have been made for certain regions by several authors [e.g., Hellerman, 1967, 1968; Bunker, 1976; Hastenrath and Lamb, 1977; Wyrtki and Meyers, 1976; O'Brien and Golden- berg, 1982]. Recently, two papers have been publishedwith global analyses of the wind stressover the oceans for each month of the year based on the same TDF-11 long data set. Hellerman and Rosenstein [1983] used all of the approxi- mately 35 million ship wind observationsavailable between 1870 and 1976 and computedmonthly averaged wind stress in 2ø x 2ø boxes,with a surface drag coefficient Ca adapted from Bunker [1976], which depends both on wind speed and stabili- ty (T•ir- T•ea)- Han and Lee [1983] produced similar fields, deriving them from the monthly averaged surface wind vector, assuming that the drag coefficientdepends only on wind speed.They modeled the nonlinear dependence of the wind stress on wind speedby assuming a Gaussian distribution of wind speeds and directions,with a standard deviation given by the observed (direction-independent) speed standard devi- ation. O'Brien and Goldenberg [1982] reanalyzed the 1961- 1970tropical Pacificdata of Wyrtki and Meyer, and produced a long-term as well as a year by year monthly wind stress atlas. In order to avoid the uncertainty associated with the drag coefficient, they did not includeeither Ca or the surface air density in their definition of stress. The purpose of this paper is to discuss another method to compute gridded fields of surfacewind and wind stress' the use of a four-dimensional analysis/forecastsystem, and scatterometer-derived surface winds. For the experiments, we utilized the Goddard Laboratory for Atmospheres (GLA) analysis/forecast system, which is similar to the systems used in several operationaland research centers to produce atmo- This paperis not subject to U.S. copyright. Published in 1986by the AmericanGeophysical Union. Paper number 5C0438. spheric analyses. Several experiments had previously been conducted with this system to study the impact of objectively dealiased Seasat scatterometer (SASS)wind data on numerical weather prediction [Baker et al., 1984]. In the present study, we make use of the subjectively dea- liasedSASSwind data set,which was prepared by Woiceshyn, Peteherych, Wurtele, and collaboratorsand is described by Peteherych et al. [1984]. We present both instantaneous and time-averagedanalyses from September 6-20, 1978, which is the only period for which subjectively dealiased SASS winds are available. For the 15 days we assimilatedall available conventionaland satellite temperaturesoundingand cloud- track wind data in addition to the subjectively dealiased SASS winds.While this period of time is extremely limited, it should be sufficient to demonstrate the potential for atmospheric models using global scatterometer data to provide gridded fields of surface wind and wind stress over the world oceans. The G LA analysis/forecast system produces other diag- nostic fields, including surface fluxes of heat and moisture. However, in this paper we confine ourselves to the discussion of surfacewind and stress. In section 2 we briefly describethe GLA system. Section3 containsan instantaneous comparison between our analysis and the hand analysis produced by Pete- herych et al. Sections4 and 5 are devoted to the 15-day average fieldsand their comparison with other availablemean fields, and section 6 has a summaryand conclusions. 2. DESCRIPTION OF THE G LA ANALYSIS/FORECAST SYSTEM The GLA analysis/forecast system consists of an objective analysis scheme that uses the continuity provided by a 6-hour model forecast, integratedfrom a previous analysis. This fore- cast serves as the first guess for the next analysisof all avail- able data within a window of + 3 hours performedat 0600, 1200, 1800, and 0000 UT. Both the forecast model and the objective analysis scheme employ a coarse horizontal grid res- olution of 4ø in latitude by 5 ø in longitude,and nine sigma levels in the vertical. However, the model is of fourth-order accuracy, making it comparable in truncation error to a model with higher resolution. In the GLA objective analysis scheme [Baker, 1983], zonal and meridional wind components,geopotential height, and relative humidity are analyzedon a mandatory pressure sur- face. Vertical consistency is maintainedthrough static stability constraints. The analysis at each level is performed with a successive correction method modified to account for differ- 2233