The impact of an air-sea roughness parameterization over the ocean that more closely matches recent observations of air-sea exchange is examined in the NASA Goddard Earth Observing System, version 5 (GEOS-5) atmospheric general circulation model. Surface wind biases in the GEOS-5 AGCM are decreased by up to 1.2m/s. The new parameterization also has implications aloft as improvements extend into the stratosphere. Many other GCMs (both for operational weather forecasting and climate) use a similar class of parameterization for their air-sea roughness scheme. We therefore expect that results from GEOS-5 are relevant to other models as well.
 The interaction between the ocean surface and the lowest levels of the atmosphere is a crucial component of any atmospheric GCM. The exchange of momentum, moisture, and sensible heat between the ocean and atmosphere occurs on spatial and temporal scales far finer than any GCM can directly simulate. Many models therefore rely on Monin-Obhukov Similarity Theory (MOST) to specify air-sea exchange as a function of bulk winds, temperature, and humidity. Early attempts at quantifying the exchange coefficients underlying MOST were conducted under conditions far removed from that actually experienced in the ocean [e.g., Charnock, 1955; Large and Pond, 1981]. Nevertheless, generations of atmospheric models have relied on these earlier measurements for tuning their air-sea roughness scheme. For example, GEOS-5 currently implements Large and Pond  for moderate and strong winds and Kondo  for weak winds [Helfand and Schubert, 1995]. See Table 1 for a description of the schemes in a range of models.
Table 1. Momentum Exchange Over Oceans in Different GCMsa
 More recent in-situ observations have improved our understanding of air-sea exchange over deep ocean waters, especially over high wind regions like the Southern Ocean. In particular, recent field campaigns have measured turbulent exchange over the Southern Ocean, over the Gulf Stream, and over the North Atlantic in high wind speeds [e.g., Edson, 2008, also manuscript in preparation, 2011; Yelland et al., 1998; Edson et al., 2007; Banner et al., 1999]. These field campaigns have found that the Charnok parameter appears to increase with wind speed beyond 10 m/s, so that a parameterization based on Charnock  or Large and Pond  underestimates the drag on surface winds [Fairall et al., 2003, section 3c]. Recent observations of air-sea exchange imply that the current air-sea roughness scheme in GEOS-5 produces too little drag on surface winds in the range of wind speeds common in the Southern Ocean.
 Accurate climatologies of surface winds over ocean regions were not available when the current Large and Pond -based parameterization in the GEOS-5 model was created, but satellite-based climatologies of surface winds are now available [e.g., Chou et al., 2003]. These satellite based climatologies suggest that surface winds in the GEOS-5 model are too strong over the Southern Ocean and off the coast of Asia in the North Pacific (Figures 1a–1c and Figure S1 in the auxiliary material). As surface winds over the Southern Ocean drive present and future oceanic uptake of CO2 [Downes et al., 2011; Matebr and Hirst, 1999], it is important to accurately simulate surface climate in this region. The GEOS-5 model is not alone in its poor representation of Southern Ocean surface wind; Barnes and Hartmann  find that the latitude of the Southern Hemisphere jet maxima varies by over 5° in the Coupled Model Intercomparison Project (CMIP3) ensemble, and that such a bias has implications for the response of a GCM to doubled CO2. Regional models also have difficulty capturing mesoscale turbulent surface fluxes [Renfrew et al., 2009].
 This paper discusses efforts to reduce this bias in GEOS-5 by updating the air-sea roughness parameterization from Helfand and Schubert . Section 2 describes changes made to the model and Section 3 presents results. As other atmospheric GCMs base their air-sea roughness parameterization for momentum exchange on similarly old data, we expect that the reduction in model bias shown here might be common to other GCMs as well.
2. Change to Scheme
 We first describe the air-sea roughness scheme in GEOS-5 before discussing the changes made to increase the surface friction. GEOS-5 contains 72 vertical levels, with approximately 8 in the boundary layer [Rienecker et al., 2008]. Like many atmospheric GCMs, GEOS-5 uses MOST to describe momentum, heat, and moisture flux coefficients in terms of bulk quantities (e.g., zonal wind, specific humidity, and temperature) in the model. The wind stress vector at the surface can be expressed as
where ρa is the air density, CD the transfer coefficient for momentum, vs the surface wind speed, Δ[u, v] is the difference between the ocean and atmosphere surface wind vector, and u* the friction velocity. MOST computes u* (and CD) as a function of bulk parameters via the following equations:
where z0 is the roughness length, κ is the Von-Karman constant, A1 through A5 are tunable parameters used to match the air-sea roughness scheme to observations, and ΨMO(ζ z0) is controlled by stability of the air column above. After an initial guess is made at CD (in practice CD assuming neutral stability), equation (2) is solved iteratively until a new value for CD has been reached consistent with the actual stability.
 Previously, the A1 through A5 coefficients were chosen to interpolate between the reciprocal relation of Kondo  for weak winds and the piecewise linear relation of Large and Pond  for moderate to large winds. The key change described by this paper is that the values for the A1 through A5 coefficients are changed so that the roughness length is increased for a given friction velocity. Neither the formulation for ΨMO(ζ z0) nor the coefficients at low wind speeds is changed. For very strong winds (e.g., hurricanes), roughness length no longer increases with wind speed (A. Molod and G. Partyka, The Impact on GEOS-5 Hurricane Forecasts of Limiting Ocean Roughness, submitted to Journal of Climate, 2011). See Table 2 for the coefficients used. Runs with the old polynomial for z0 are referred to as CONTROL, and runs with the new polynomial for z0 are referred to as NEW.
Table 2. Coefficients for MOST Scheme Equation Relating u* to z0a
z0 = + A2 + A3u* + A4u*2 + A5u*3.
u* < 0.0632456
0.2030325E – 5
0.0632456 < u* < 0.381844
0.381844 < u*
u* < 0.0632456
0.2030325E – 5
0.0632456 < u*
Figure 2a shows CDN10m as a function of 10m wind speed for the old and new coefficients and in observations. The drag coefficient has been increased beyond the average suggested by the most recent observations but within the uncertainty. We chose the highest drag coefficient justified by the observations to achieve the maximum impact on the GEOS-5 wind bias. Any further increase would distance us from the range of observational uncertainty. Note that the drag coefficient in Community Atmosphere Model (CAM/CCSM) (dashed red line) appears to be too small. CAM has not upgraded its scheme since version 2.0 (Kiehl et al.  versus Neale et al. [2010, section 4.11.2]). Figure 2b compares modeled output roughness length and friction velocity for the NEW and CONTROL runs. As expected, surface roughness dramatically increases with the new coefficients. Figure 2b also includes curves of z0 = αCharnoku*2/g [Charnock, 1955] but with different values of the Charnok parameter αCharnok. Older measurements suggest values of αCharnok∼0.011 to αCharnok∼0.018 [see Fairall et al., 2003, section 3c]. Newer observations [Edson, 2008, also manuscript in preparation, 2011] would imply a higher αCharnok.
 This change has been implemented in GEOS-5. Several GEOS-5 atmosphere-only simulations with the old and new coefficients were performed to examine the impact of the increased drag:
 1. 2 × 2.5 degree 30 year runs with interactive stratospheric chemistry,
 2. 2 × 2.5 degree 12 year run without interactive stratospheric chemistry,
 3. 1 × 1.25 degree 25 year run without interactive stratospheric chemistry,
 4. a series of 1/4 degree 5-day forecasts.
 All simulations showed similar impact of the new roughness parameterization, and we will focus here on results from the 30 year run with stratospheric chemistry. CONTROL and NEW differ only in the air-sea roughness scheme; all other models settings are fixed. A Student-T two-tailed test is used to assess statistical significance. Each year is taken as one degree of freedom. Surface winds and surface stress from Version 2 of the Goddard Satellite-Based Surface Turbulent Fluxes (GSSTF) Data [Chou et al., 2003] are used to validate the model. We now address the impact of this change in the air-sea roughness parameterization on bulk quantities in the model.
 We now discuss how the change in friction influences the momentum budget in the model. The exchange coefficient for momentum increases over most oceanic regions, with the strongest increase over the Southern Ocean (Figure 3b). Biases in surface wind are reduced across the ocean regions in response to the altered surface roughness coefficients (Figures 1c and 1d). Winds over the Southern Ocean decrease by over 1m/s, but winds are reduced over most ocean covered regions. Near surface geopotential height anomaly biases are also reduced (not shown), consistent with the wind speed improvement.
Figures 3c–3f show zonal surface stress on the ocean. Changes in surface stress are smaller than changes in either CD or wind speed, as might be expected from equation (1). Namely, the decrease in wind speed and increase in CD largely balance each other, so that their product is nearly constant. Nevertheless, the changes are significant in the Southern Hemisphere, whereby surface stress on the Southern Ocean is increased while surface stress further equatorward is decreased. The change is particularly strong in the Indian Ocean/Australia region. Biases in the control run are partially ameliorated. Runs in which the atmosphere is coupled to a full ocean are planned in order to understand the potential impact on the ocean circulation.
 These changes in surface stress imply anomalous eddy momentum flux convergence aloft, as vertically averaged must balance surface friction for a steady state surface jet [Held, 1975; Vallis, 2006, section 12.1]. Figure 4 shows that poleward momentum flux is increased throughout the upper troposphere, as implied by the dipole of surface stress. Eddies are fluxing more momentum poleward in order to counteract the weakening of the surface jet. E. A. Barnes and C. I. Garfinkel (Barotropic eddy-jet coexistence and the response to surface friction, submitted to Journal of Atmospheric Sciences, 2011) are investigating this change in momentum flux in much more detail. Associated with this change in momentum flux are statistically significant improvements in extratropical forecasting skill (not shown).
 The old air-sea roughness scheme in GEOS-5 is based on 30-year old observational data, but newer data suggests seas are rougher. Associated with the old parameterization are overly strong surface winds. By incorporating more recent observations of air-sea exchange into the model's air-sea roughness scheme, we have improved the surface climate in the GEOS-5 AGCM. Preliminary results indicate that the improvement is present at resolutions up to 1/4 degree.
 Modifying the air-sea roughness parameterization leads to statistically significant changes in cloud distribution, heat flux, stratospheric ozone, and planetary wave driving of the stratosphere. Presentation of these changes, a discussion of the surface moisture and sensible heat budgets, will be reported in detail in a future paper. The microphysics scheme in all runs considered does not include interactive aerosols; preliminary results indicate that including interactive aerosols along with this change in surface roughness leads to large changes in sea salt aerosol concentration and subsequent cloud formation.
 Other atmospheric GCMs appear to use a similar scheme to parametrize the exchange of momentum, heat, and moisture with the ocean. We expect that biases in these other models might be reduced if these models were retuned to more closely match available observations.
 This work was supported by the NASA grant NNX06AE70G. The authors thank J. Edson for making available data from Edson  and his manuscript in preparation, Larry Takacs for performing the model simulation in forecast mode, Andrew Eichmann for performing the 1 degree model simulations, and our anonymous reviewers for their comments.
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.