• Abdalla, S., , and Hersbach H. , 2007: The technical support for global validation of ERS wind and wave Products at ECMWF. ECMWF Contract Report to the European Space Agency, 59 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/esa/ESA_abdalla_hersbach_18212-2007.pdf].

    • Search Google Scholar
    • Export Citation
  • Abdalla, S., , and Hersbach H. , 2008: The technical support for global validation of ERS wind and wave products at ECMWF. ECMWF Contract Report to the European Space Agency, 40 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/esa/ESA_abdalla_hersbach_20901-2008.pdf].

    • Search Google Scholar
    • Export Citation
  • Agusti-Panareda, A., , and Beljaars A. , 2008: ECMWF’S contribution to AMMA. ECMWF Newsletter, No. 115, ECMWF, Reading, United Kingdom, 19–25.

    • Search Google Scholar
    • Export Citation
  • Attema, E. P. W., 1986: An experimental campaign for the determination of the radar signature of the ocean at C-band. Proc. Third Int. Colloquium on Spectral Signatures of Objects in Remote Sensing, Les Arcs, France, ESA, 791–799.

    • Search Google Scholar
    • Export Citation
  • Brown, A. R., , Beljaars A. C. M. , , and Hersbach H. , 2006: Errors in parameterizations of convective boundary layer turbulent moment mixing. Quart. J. Roy. Meteor. Soc., 132 , 18591876.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Carswell, J. R., , Donnelly W. J. , , McIntosh R. E. , , Donelan M. A. , , and Vandemark D. C. , 1999: Analysis of C and Ku band ocean backscatter measurements under low-wind conditions. J. Geophys. Res., 104 , (C9). 2068720701.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , and Freilich M. H. , 2005: Scatterometer-based assessment of 10-m wind analyses from the operational ECMWF and NCEP numerical weather prediction models. Mon. Wea. Rev., 133 , 409429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , Schlax M. G. , , Freilich M. H. , , Milliff R. F. , , and Vandemark D. C. , 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303 , 978983.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Donnelly, W. J., , Carswell J. R. , , McIntosh R. E. , , Chang P. S. , , Wilkerson J. , , Marks F. , , and Black P. G. , 1999: Revised ocean backscatter models at C and Ku-band under high-wind conditions. J. Geophys. Res., 104 , (C5). 1148511497.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebuchi, N., , Graber H. C. , , and Caruso M. J. , 2002: Evaluation of wind vectors observed by QuikSCAT/SeaWinds using ocean buoy data. J. Atmos. Oceanic Technol., 19 , 20492062.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fernandez, D. E., , Carswell J. R. , , Frasier S. , , Chang P. S. , , Black P. G. , , and Marks F. D. , 2006: Dual-polarized C- and Ku-band ocean backscatter response to hurricane-force winds. J. Geophys. Res., 111 , C08013. doi:10.1029/2005JC003048.

    • Search Google Scholar
    • Export Citation
  • Figa-Saldaña, J., , Wilson J. J. W. , , Attema E. , , Gelsthorpe R. , , Drinkwater M. , , and Stoffelen A. , 2002: The advanced scatterometer (ASCAT) on the meteorological operational (MetOp) platform: A follow on for European wind scatterometers. Can. J. Remote Sens., 28 , 404412.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Francis, R., and Coauthors, 1991: The ERS-1 spacecraft and its payload. ESA Bull., 65 , 2748.

  • Freilich, M. H., , and Dunbar R. S. , 1999: The accuracy of the NSCAT 1 vector winds: Comparisons with National Data Buoy Center buoys. J. Geophys. Res., 104 , (C5). 1123111246.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hersbach, H., , and Janssen P. , 2007a: Improved scatterometer screening. ECMWF Research Department Memo. R60.9/HH/0735, 24 pp.

  • Hersbach, H., , and Janssen P. , 2007b: Preparation for assimilation of surface-wind data from ASCAT at ECMWF. ECMWF Research Department Memo. R60.9/HH/0750, 24 pp.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., , Stoffelen A. , , and de Haan S. , 2007: An improved C-band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112 , C03006. doi:10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., , and Janssen P. A. E. M. , 2004: The benefit of ERS scatterometer winds in ECMWF’s variational assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 17931814.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Janssen, P. A. E. M., 1991: Quasi-linear theory of wind-wave generation applied to wave forecasting. J. Phys. Oceanogr., 21 , 16311642.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelly, A. K., , Dickinson S. , , McPhaden J. , , and Johnson G. C. , 2001: Ocean currents evident in satellite wind data. 28 , 24692472.

  • Long, A. E., 1984: Towards a C-band radar sea echo model for the ERS-1 scatterometer. Proc. Third Int. Colloquium on Spectral Signatures of Objects in Remote Sensing, Les Arcs, France, ESA, 29–34.

    • Search Google Scholar
    • Export Citation
  • Monin, A. S., , and Obukhov A. M. , 1954: Basic regularity in turbulent mixing in the surface layer of the atmosphere. Geofiz. Inst. Akad. Nauk SSSR, 24 , 163187.

    • Search Google Scholar
    • Export Citation
  • Offiler, D., 1994: The calibration of ERS-1 satellite scatterometer winds. J. Atmos. Oceanic Technol., 11 , 10021017.

  • Portabella, M., , and Stoffelen A. , 2009: On scatterometer ocean stress. J. Atmos. Oceanic Technol., 26 , 368382.

  • Press, W. H., , Teukolsky S. A. , , Vetterling W. T. , , and Flannery B. P. , 1997: Numerical Recipes in Fortran 90: The Art of Scientific Computing. 2nd ed. Cambridge University Press, 933 pp.

    • Search Google Scholar
    • Export Citation
  • Quilfen, Y., , Chapron B. , , Collard F. , , and Vandemark D. , 2004: Relationship between ERS scatterometer measurement and integrated wind and wave parameters. J. Atmos. Oceanic Technol., 21 , 368373.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Song, Q., , Chelton D. B. , , Esbensen S. K. , , Thum N. , , and O’Neill L. W. , 2009: Coupling between sea surface temperature and low-level winds in mesoscale numerical models. J. Climate, 22 , 146164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., 1999: A simple method for calibration of a scatterometer over the ocean. J. Atmos. Oceanic Technol., 16 , 275282.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and Anderson D. , 1997: Scatterometer data interpretation: Estimation and validation of the transfer function CMOD4. J. Geophys. Res., 102 , (C3). 57675780.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012.

  • Verspeek, J., , Portabella M. , , Stoffelen A. , , and Verhoef A. , 2008: Calibration and validation of ASCAT winds, version 4.0. EUMETSAT Ocean and Sea Ice SAF Tech. Note SAF/OSI/KNMI/TEC/TN/163, 31 pp. [Available online at http://www.knmi.nl/publications/fulltexts/calibration_and_validation_of_ascat_winds_4.0.pdf].

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., , and Smith D. K. , 1999: A model function for the ocean-normalized radar cross section at 14 GHz derived from NSCAT observations. J. Geophys. Res., 104 , (C5). 1149911514.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, S-P., , Liu W. T. , , Liu Q. , , Nonaka M. , , and Hafner J. , 2001: Long-range effects of the Hawaiian Islands on the Pacific ocean-atmosphere system. Proc. IGARSS Int. Geoscience and Remote Sensing Symp., Sydney, Australia, IEEE, Vol. 3, 1086–1088.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    Scatterometer winds from ERS-2 and ASCAT as assimilated at ECMWF for the 12-h assimilation cycle of 0000 UTC 9 Dec 2007. The two tracks around 40°W and the track parallel to 10°W concern ASCAT data; the three tracks around 45°W, 20°W, and 0° originate from ERS-2. Large dark barbs indicate actively assimilated winds, and small light barbs indicate rejected (mostly thinned) winds. The gray dashed lines represent streamlines for the ECMWF analysis surface wind field.

  • View in gallery

    (top) Average ocean wind speed (gray scales) and ocean vector wind (arrows) at 10-m height and (middle) average difference and (bottom) standard deviation from equivalent neutral wind speed, for the operational ECMWF short-range forecast at ASCAT location during (left) July 2007 and (right) January 2008.

  • View in gallery

    Average relation between ECMWF neutral and nonneutral wind speed (m s−1) at (left) ERS-2 and (right) ASCAT locations, summed over July 2007 and January 2008. Data points represent wind-vector cell-wise (19 for ERS-2 and 42 for ASCAT) averages in speed bins of 2.5 m s−1 for which at least 100 data points were available. Average difference is 0.20 m s−1 for both ERS-2 and ASCAT.

  • View in gallery

    (left) B0 for CMOD5 as function of wind speed υ and incidence angle θ (dB) and (right) the wind speed shift b (m s−1) for which B0(υ + b, θ) based on CMOD5.N coefficients equals B0(υ, θ) using original CMOD5 coefficients. Light horizontal lines mark the incidence-angle range for ERS-2, and dark horizontal lines show the corresponding range for ASCAT (thin for mid and thick for fore/aft antenna).

  • View in gallery

    As in Fig. 4, but for (left) B1 in natural unit and (right) the difference between fit (2).

  • View in gallery

    As in Fig. 5, but for B2.

  • View in gallery

    Bias correction (dB) applied to ASCAT backscatter before wind inversion as function of WVC and antenna (fore is dashed, mid is solid, and aft is dotted–dashed) as used at ECMWF for indicated period.

  • View in gallery

    Additional correction in wind speed (m s−1) after inversion as function of WVC and wind speed based on a comparison study with ECMWF short-range forecast nonneutral winds for the period 1–20 Feb 2007. Minimum and maximum values in speed correction are −0.81 and 1.20 m s−1, respectively.

  • View in gallery

    Difference between CMOD5.N and CMOD5 wind speeds after inversion of (left) ERS-2 backscatter triplets and (right) ASCAT triplets, respectively, accumulated over July 2007 and January 2008. Dots represent averages for data of specific WVC and speed domain (in bins of 2.5 m s−1) for which at least 100 data points were available, and they are displayed as functions of (top) wind speed and (bottom) WVC, respectively.

  • View in gallery

    Histogram between scatterometer wind based on CMOD5.N, for (left) ERS-2 and (right) ASCAT, and collocated ECMWF short-range forecast equivalent neutral wind speed at 10 m. Contours are in steps of 5 dB. Circles denote averages for bins in the x direction, and squares denote averages for bins in the y direction.

  • View in gallery

    As in Fig. 3 and Fig. 9 (top), but for CMOD5.N wind speed vs collocated ECMWF neutral wind.

  • View in gallery

    Anomaly from the global mean of ASCAT departures from ECMWF (top) nonneutral and (bottom) neutral wind accumulated over (left) July 2007 and (right) January 2008 and gridded into boxes on an N80 reduced Gaussian grid (average box size is 1.125° by 1.125°). Winds are based on CMOD5.N. Note a difference in gray scaling with respect to Fig. 2.

  • View in gallery

    The local correlation of stability corrections with ASCAT wind speed departures for (top) nonneutral and (bottom) neutral departures, for subsets collected per N80 reduced Gaussian grid box and accumulated over (left) July 2007 and (right) January 2008.

  • View in gallery

    (left) Local value of the ECMWF short-forecast Charnock parameter collocated at ASCAT location and (right) the local effect on the determination of ASCAT wind speed averaged over January 2008 onto an N80 reduced Gaussian grid.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 151 151 14
PDF Downloads 141 141 14

Comparison of C-Band Scatterometer CMOD5.N Equivalent Neutral Winds with ECMWF

View More View Less
  • 1 European Centre for Medium-Range Weather Forecasts, Reading, United Kingdom
© Get Permissions
Full access

Abstract

This article describes the evaluation of a C-band geophysical model function called C-band model 5.N (CMOD5.N). It is used to provide an empirical relation between backscatter as sensed by the spaceborne European Remote Sensing Satellite-2 (ERS-2) and Advanced Scatterometer (ASCAT) scatterometers and equivalent neutral ocean vector wind at 10-m height (neutral surface wind) as function of scatterometer incidence angle.

CMOD5.N embodies a refit of CMOD5, a C-band model function, which was previously derived to obtain nonneutral surface wind, in such a way that its 28 tunable coefficients lead, for a given backscatter observation, to an enhancement of 0.7 m s−1 in wind speed. The value of 0.7 m s−1 is chosen to be independent of wind speed and incidence angle, and it incorporates the average difference between neutral and nonneutral wind (∼0.2 m s−1) and for a known bias of CMOD5 (∼0.5 m s−1) when compared to buoy wind data. The quality of the CMOD5.N fit is successfully tested for the Active Microwave Instrument (AMI) scatterometer on ERS-2 and ASCAT instrument on Meteorological Operational-A (MetOp-A) for July 2007 and January 2008.

ASCAT and ERS-2 wind speed obtained from CMOD5.N compares well on average with operational neutral wind from the European Centre for Medium-Range Weather Forecasts (ECMWF). In comparison with nonneutral wind, the local, seasonally dependent biases between scatterometer and ECMWF model are reduced. Besides effects introduced by sea state, orography, and ocean currents, a residual stability-dependent bias between scatterometer and neutral wind remains, which is likely connected to a nonoptimality in the ECMWF boundary layer formalism that is reported in the literature.

Corresponding author address: Hans Hersbach, European Centre for Medium-Range Weather Forecasts, Shinfield Park, RG2 9AX Reading, United Kingdom. Email: hans.hersbach@ecmwf.int

Abstract

This article describes the evaluation of a C-band geophysical model function called C-band model 5.N (CMOD5.N). It is used to provide an empirical relation between backscatter as sensed by the spaceborne European Remote Sensing Satellite-2 (ERS-2) and Advanced Scatterometer (ASCAT) scatterometers and equivalent neutral ocean vector wind at 10-m height (neutral surface wind) as function of scatterometer incidence angle.

CMOD5.N embodies a refit of CMOD5, a C-band model function, which was previously derived to obtain nonneutral surface wind, in such a way that its 28 tunable coefficients lead, for a given backscatter observation, to an enhancement of 0.7 m s−1 in wind speed. The value of 0.7 m s−1 is chosen to be independent of wind speed and incidence angle, and it incorporates the average difference between neutral and nonneutral wind (∼0.2 m s−1) and for a known bias of CMOD5 (∼0.5 m s−1) when compared to buoy wind data. The quality of the CMOD5.N fit is successfully tested for the Active Microwave Instrument (AMI) scatterometer on ERS-2 and ASCAT instrument on Meteorological Operational-A (MetOp-A) for July 2007 and January 2008.

ASCAT and ERS-2 wind speed obtained from CMOD5.N compares well on average with operational neutral wind from the European Centre for Medium-Range Weather Forecasts (ECMWF). In comparison with nonneutral wind, the local, seasonally dependent biases between scatterometer and ECMWF model are reduced. Besides effects introduced by sea state, orography, and ocean currents, a residual stability-dependent bias between scatterometer and neutral wind remains, which is likely connected to a nonoptimality in the ECMWF boundary layer formalism that is reported in the literature.

Corresponding author address: Hans Hersbach, European Centre for Medium-Range Weather Forecasts, Shinfield Park, RG2 9AX Reading, United Kingdom. Email: hans.hersbach@ecmwf.int

1. Introduction

Spaceborne scatterometer data provide accurate information on speed and direction of surface wind over the global oceans. Since the launch of the European Remote Sensing Satellite-1 (ERS-1) in July 1991, global coverage of scatterometer data has been available without interruption. Applications vary from near-real-time assimilation into numerical weather prediction (NWP) models and the forcing of ocean models to climate studies accessing the now 18-yr data record.

At the European Centre for Medium-Range Weather Forecasts (ECMWF) scatterometer winds have been assimilated in the operational integrated forecast and assimilation system (IFS) from 30 January 1996 onward. The four-dimensional variational assimilation system at ECMWF allows for a dynamically consistent use of observations. In this way, information of scatterometer surface winds is propagated to the entire troposphere (Isaksen and Janssen 2004). Currently (December 2009), data are used from the Active Microwave Instrument (AMI) scatterometer on board ERS-2 (from June 1996 onward) and from the Advanced Scatterometer (ASCAT) instrument on the Meteorological Operational-A (MetOp-A) platform (from June 2007 onward), whereas the SeaWinds instrument on board Quick Scatterometer (QuikSCAT) has been used from January 2002 until the end of November 2009. Because of the different timing of ascending nodes of these three satellites, until recently, most areas on the globe were covered within a 6-h period. Morning and afternoon were captured by QuikSCAT, whereas noon and midnight are served by ASCAT and ERS-2. An example is given in Fig. 1, which shows the observation of an intense low near the British Isles by ASCAT and ERS-2 around 0000 UTC 9 December 2007.

A scatterometer is a microwave radar that emits pulses at a well-defined frequency and polarization to the earth’s surface. A backscatter is recorded, from which information on the sensed surface can be obtained. In polar regions, characteristics and age of sea ice can be determined. Over land, information on soil moisture can be extracted, while over open water, an estimate of the surface vector wind can be obtained. It is this latter application that has been known longest. The main physical process is based on Bragg scattering where backscatter is related to the intensity of surface water waves with wavelengths that are comparable to that of the emitted pulse. By choice of the scatterometer wavelength in the centimeter range, the strength of gravity–capillary surface waves is sensed. In turn, these are determined by the local surface stress or, in effect, the local surface wind condition. Because backscatter response also depends on the relative angle between the incident pulse and capillary wave direction, information on wind direction can be extracted as well.

Although the physical mechanism is known in principle, for practical applications the relation between wind and backscatter is provided by an empirical geophysical model function (GMF). Such a model function is the result of a large collocation study between observed (aircraft and/or spaceborne) scatterometer backscatter with in situ buoy data and/or NWP model data. Backscatter should correlate closely with surface stress. Because this quantity cannot be obtained accurately from in situ observations, in practice readily available winds converted to a standard height of 10 m are used instead.

In the usual assumption of a constant (turbulent) stress layer near the surface, which is subject to a form of Monin–Obukhov stability theory (Monin and Obukhov 1954), the relation between stress τ = ρau*u* (where ρa is the air density, u* is the friction velocity, and u* is its magnitude) and wind u(z) at height z is given by
i1520-0426-27-4-721-e1
where κ = 0.4 is the von Kármán constant and z0 is the roughness length. In principle, this relation depends on variations in atmospheric stability (expressed by the stability function ΨM and Obukhov length L), air density, and ocean current uoc. In the ECMWF formulation, z0 depends on the kinematic viscosity ν = 1.5 × 10−5 m2 s−1 for light wind and on a Charnock relation for strong wind as
i1520-0426-27-4-721-e2
Here, αM = 0.11, g = 9.81 m s−2 is the gravitational acceleration, and αch depends on the sea state (Janssen 1991). This introduces a sea-state dependency on the relation between stress and surface wind as well.

For the C-band (5.3 GHz) AMI scatterometers on board ERS-1 and ERS-2 and ASCAT on board MetOp-A, a branch of model functions with common name CMOD (C-band model) was developed in the past. These are all based on collocation studies with observed and/or modeled wind at 10-m height.

For the Ku-band frequency (13.4 GHz) at which the scatterometer on board QuikSCAT operates—as well as the SeaWinds and National Aeronautics and Space Administration (NASA) Scatterometer (NSCAT) instruments on board the past Advanced Earth Observing Satellite-II (ADEOS-II) and ADEOS-I platforms, respectively—GMFs have been traditionally trained on neutral equivalent winds (Wentz and Smith 1999; Freilich and Dunbar 1999; Ebuchi et al. 2002; for an overview, see Chelton and Freilich 2005). Such winds (denoted hereafter as neutral) represent the wind at 10-m height for given surface stress in case the marine boundary layer was neutrally stratified. They can be estimated from real winds (denoted hereafter as nonneutral) at 10-m or buoy observation height by transformation of such winds to surface stress by taking account of additional information on atmospheric stability and subsequently transforming back to 10-m wind by the neglect of these stability effects. As a result, the relation between neutral wind and stress (or u*) is given by (1) with ΨM = 0. Although the intermediate values for surface stress may depend sensitively on the details of the surface layer model used for the transformation, the final differences between neutral and nonneutral wind appear more robust. An example is discussed in detail by Portabella and Stoffelen (2009). The rationale is that the model dependencies in the transformation from (nonneutral) wind to stress are largely undone when transforming back from stress to (neutral) wind. Therefore, by using neutral wind, the case-dependent effect of stability can be incorporated without having to know actual values for stress.

On average, the marine boundary layer is weakly unstable, and the global average neutral wind appears ∼0.2 m s−1 stronger than the nonneutral wind (see, e.g., Brown et al. 2006). Stability is to a large extent determined by the difference between surface air temperature and sea surface temperature (SST). Locally, average neutral effects correlate with the typical weather regime in relation to (the slower) SST component, whereas fluctuations correlate more with variations in the (faster) meteorological component. An example of this is presented in Fig. 2, where statistics of stability effects, as calculated in the IFS system, which uses relations (1) and (2) (see IFS documentation, available online at http://www.ecmwf.int/research/ifsdocs), are summarized for ECMWF short-range (3–15 h) forecast model wind for July 2007 and January 2008.1 In Northern Hemispheric summer (middle-left panel) a stable boundary layer exists in the Arctic region, the Hudson Bay, and the area east of Newfoundland. It is induced by warm off-land and or southerly maritime wind over cooler ocean water. In these areas, neutral winds can be lower than nonneutral winds by 0.5 m s−1. The middle-right panel shows a similar, though less intense, effect off Patagonia over the summer Southern Hemisphere. Northern Hemispheric winter clearly shows the signature of an enhanced SST in the Gulf Stream and the Kuroshio Extension. In both hemispheres, winds are more unstable in winter and less unstable during summer, although the difference is largest in the Northern Hemisphere. The bottom panels of Fig. 2 show locally that fluctuations around the mean can be larger than the average difference. For example, this is the case in the storm tracks, where the movement of weather systems alternates the presence of stable and unstable conditions. On the other hand, fluctuations are small in the regions of steady trade winds.

Various studies have confirmed the response of scatterometer data to surface stress (see, e.g., Chelton et al. 2004 for a detailed description for QuikSCAT data). Hence, to address the effect of stability in (1), it is desirable to derive a C-band GMF for neutral wind rather than nonneutral. This will be the objective in this study. This model function, which is called CMOD5.N, is based on the CMOD5 model function by a refit of its 28 tunable coefficients.

In section 2, the most popular CMOD models are briefly reviewed, and the requirement for CMOD5.N will be formulated. The refit of the 28 coefficients is described in section 3. The quality of the fit for scatterometer data from ERS-2 and ASCAT will be the subject of section 4. In section 5, a detailed comparison is made between CMOD5.N scatterometer winds and both neutral and nonneutral ECMWF winds. Although a positive correlation is found between stability corrections and the departure of scatterometer wind from model wind, large local and seasonally dependent biases remain. This article ends with a discussion on possible causes for such biases (section 6).

2. The C-band CMOD family

The development of the CMOD family was initialized by the requirement of a GMF for the C-band AMI scatterometer on board ERS-1, which was launched by the European Space Agency (ESA). The AMI (Attema 1986; Francis et al. 1991) obtains backscatter measurements from three antennas (fore, mid, and aft), illuminating a swath of 500 km, in which 19 nodes, or wind-vector cells (WVCs), define a 25-km product. From these backscatter triplets, two similarly likely, nearly antiparallel wind solutions can be retrieved.

All CMOD models provide an empirical functional relation in which the dependency of normalized backscatter σ0 on wind speed υ, wind direction χ, and incidence angle θ is described as
i1520-0426-27-4-721-e3
Here, ϕ = χα is the angle between wind direction and scatterometer azimuth look angle (both measured from the north), coefficients ci (which are subsets from a larger set c) shape the terms Bi, and p is a parameter. The dependency on wind direction is described by only two harmonics. The dominant term B0 sets the speed scale for a given measurement. The upwind–crosswind asymmetry B2 allows for the determination of wind direction, whereas B1 attributes to resolve a remaining 180° ambiguity in wind direction.

The first model, CMOD2 (Long 1984), was a prelaunch model function that was based on airborne data. After the launch of ERS-1, it soon appeared inadequate, and a replacement, called CMOD4, was developed by Stoffelen and Anderson (1997) using actual data from ERS-1. Besides a reformulation, the innovative part was the introduction of p = 1.6, rather than unity. It effectively avoids the need in (3) for higher harmonics B3, B4, etc. After a selection exercise between various model function candidates (Offiler 1994), ESA decided to base the operational ERS User Fast-delivery Wind product (UWI) on this model function, and it still is for ERS-2 today.

Although the CMOD4-based wind product meets ESA’s original instrument requirements, it could be improved in several ways. Besides a negative speed bias that largely depends on incidence angle and wind speed (induced by a nonoptimal description for B0), there was independent evidence from field experiments (Carswell et al. 1999; Donnelly et al. 1999; Fernandez et al. 2006) for inadequacies in the formulation of B1 and B2 as well. Most of these issues were resolved by the development of CMOD5 (Hersbach et al. 2007). CMOD5 gives rise to a more uniform performance across the AMI swath. Improvements are especially obtained for extreme cases, and CMOD5 extends the dynamical range for C-band scatterometer data from 24 to 35 m s−1.

The CMOD5 model function had been derived on the basis of a collocation study between ERS-2 AMI triplets and ECMWF short-range forecast winds. Here, it was assumed that these ECMWF winds represent an unbiased reference for surface winds over the global oceans. However, being a function of time (i.e., depending on the ECMWF model version), biases in these fields are known to exist (for a comparison with QuikSCAT winds, which have been tuned to neutral buoy winds; see, e.g., Chelton and Freilich 2005). A triple collocation with buoys (which are generally believed to provide unbiased estimates of the ground truth) in the North Atlantic and North Pacific for a 1-yr period (August 1998–July 1999) showed that both CMOD5 and ECMWF short-range forecast winds were biased low by about 0.35 m s−1. This bias was found to be uniform: that is, independent on wind speed itself (see Fig. 13 of Hersbach et al. 2007). A similar study by Abdalla and Hersbach (2007), in which the collocation period was extended to two 3-yr periods, one period for ERS-1 and one period for ERS-2, displayed a reasonably flat bias of −0.55 m s−1 for ERS-2. For ERS-1, a wind speed–dependent bias of on average −0.24 m s−1 emerged. The difference of 0.31 m s−1 indicates a difference in calibration between ERS-1 and ERS-2. An analysis performed by Portabella and Stoffelen (2009) for ERS-2 data, which included the Tropical Atmosphere Ocean (TAO) and Prediction and Research Moored Array in the Tropical Atlantic (PIRATA) buoy network in the tropics, indicated a negative bias for CMOD5 of around 0.50 m s−1.

Although all analyses presented above confirm that CMOD5 is biased low, they disagree in the exact amount. For ERS-2 estimates vary between −0.35 and −0.55 m s−1. A value of −0.45 ± 0.10 m s−1 seems reasonable. It appeared that such a difference can be incorporated in the original CMOD5 formulation by a refit of its 28 coefficients. As a result, winds inverted by this retuned model function, called CMOD5.4, appear 0.48 m s−1 stronger than winds based on CMOD5. Details may be found in Abdalla and Hersbach (2007). CMOD5.4 is currently used in the operational assimilation system at ECMWF since June 2007 for both ERS-2 and ASCAT (e.g., for the case displayed in Fig. 1).

For the near-real-time level-2 ASCAT product as disseminated by the data provider [the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT)], initially a slightly different approach in wind inversion was followed. The ASCAT instrument (Figa-Saldaña et al. 2002) is of similar design to AMI, the main differences being a different range of incidence angles and two sets of antennas, providing two swathes. In each swath, 21 WVCs form a 25-km product (i.e., 42 in total). The transformation of backscatter to wind is delivered by the Royal Netherlands Meteorological Office (KNMI) within the EUMETSAT Ocean and Sea Ice Satellite Application Facility (SAF) framework. Up to 20 November 2008, winds were based on CMOD5. After inversion, 0.5 m s−1 was added to wind speed. That model function was called CMOD5.5. From 20 November 2008 onward, the CMOD5.N model function as described in this document is used.

To summarize, there is a prelaunch model function CMOD2; an operational ERS-2 model function CMOD4; and an improved model function CMOD5, which has two very similar derivatives CMOD5.4 and CMOD5.5 that both correct for a bias of ∼−0.5 m s−1 in CMOD5. All are tuned for nonneutral wind.

As discussed in the introduction, the global average neutral wind is around 0.2 m s−1 stronger than the global average nonneutral wind. To investigate this difference when limited to cases covered by ERS-2 and ASCAT and its dependency on the value of wind speed as well, ECMWF winds are collocated with ERS-2 and ASCAT data for the periods of July 2007 and January 2008 (i.e., as in Fig. 2). ECMWF winds are bilinearly interpolated to the scatterometer observation location; in time, the closest field (resolution of 3 h) is used. Data are only regarded where ECMWF fields for sea ice fraction and SST have local values below 10% and above 273.15 K, respectively. Statistics between ECMWF neutral and nonneutral winds are determined for different speed bins (2.5 m s−1 each) and for each scatterometer wind-vector cell. Average differences are presented in Fig. 3. From this, it is seen that the dependency on wind speed is low for the locations sensed by ERS-2, whereas for ASCAT the difference between neutral and nonneutral wind speed is lower for winds above 10 m s−1. The variation at strong wind speed, where statistics is lower, depends on the details of the sampling of the underlying conditions. Besides, the average dependency of neutral effects on wind speed is probably within the uncertainty of similar dependencies of the CMOD5 bias. Therefore, in agreement with Portabella and Stoffelen (2009), it is proposed that an unbiased CMOD5.N model function for neutral winds should provide winds that are 0.7 m s−1 stronger than obtained by CMOD5, independent on wind speed (and of course incidence angle). This amount is the sum of the bias in CMOD5 (0.5 m s−1) and stability correction (0.2 m s−1). This will be achieved by a suitable refit of the original 28 CMOD5 coefficients, rather than to add 0.7 m s−1 after wind inversion.

3. Determination of the coefficients for CMOD5.N

The objective is to retune the 28 CMOD5 coefficients such that an inversion scheme will deliver winds that are 0.7 m s−1 stronger. This can be achieved by demanding that, for given wind speed, CMOD5 should give exactly the same backscatter as CMOD5.N for a wind that is 0.7 m s−1 stronger. Given formulation (3), this leads to the following set of conditions:
i1520-0426-27-4-721-e4
Here, c and c′ are the original and retuned set of coefficients, respectively, and the indices show the range of coefficients that define each term. Values of B0, B1, and B2 as a function of wind speed and incidence angle are displayed in the left panels of Figs. 4 –6. Details on the CMOD5 formulation may be found in Hersbach et al. (2007). Of course, relation (4) cannot be satisfied for all combinations within the in practice encountered range of wind speed υ and incidence angle θ. For each term, therefore, a cost function Ji is to be defined, and a fit involving the relevant part of c′ has to be performed:
i1520-0426-27-4-721-e5
Here, δi expresses a departure from the ideal state, and the angle brackets denote some weighted average over a certain domain in wind speed and incidence angle.
It would be tempting to define δi just as the amount at which (4) is violated:
i1520-0426-27-4-721-e6
For B0, though, one should realize that its dynamical range is over a few orders of magnitude. In decibel units (dB ≡ 10 logσ0), for example, B0 ranges from −30 dB for light wind at θ = 55° to 0 dB for hurricane wind speed at θ = 20°. Typical observation error in backscatter is a few percent, or around 0.2 dB. Therefore, a certain misfit in σ0, compared to observation error, might be insignificant for one case but huge for another case. An alternative would be to define departures δi in decibel units. Although now a more balanced weight is given in terms of observation error, one should realize how the wind inversion scheme would project a normalized deviation from (4) back to a shift in wind speed. As can be seen from the left panel of Fig. 4, at low wind speed, a small shift in wind speed corresponds to a larger shift in decibels than at strong wind speed. Therefore, given a similar misfit in decibels, the effect on wind speed will be small for light winds and large for strong winds. Because it is the homogeneity in the inverted wind product that ultimately matters, it makes more sense to choose δi equal to the shift in wind speed b that is required to mend relation (4):
i1520-0426-27-4-721-e7
In this way, deviations from (4) are automatically transformed appropriately to wind speed space. For B1 and B2, the exact choice for departure δi is less critical, because wind speed dependency is mainly determined by B0. For convenience, though, (7) can be used for these terms as well.

Cost function (5) is minimized over a domain in wind speed from 1 to 50 m s−1 (in 1 m s−1 steps) and from 17° to 66° (1° steps) for incidence angle. This covers both the range of ERS-2 and ASCAT (horizontal lines in Figs. 4 –6). For winds of 25 m s−1 and stronger, the weight is reduced by a factor of 4. In this way, the behavior at extreme wind acts as a constraint without dominating the total cost. At local extrema of Bi (i.e., where ∂Bi/∂υ = 0), a solution for b is less sharply defined or may not even exist. For this reason, points that are within 1 m s−1 of an extremum are excluded. For B0, this only occurs for extreme wind speed at low incidence angle; however, for B1 and B2, extrema are located at more moderate wind speed.

A minimization package is used that is based on the Powell conjugate method as provided by Press et al. (1997). It does not require derivative information. Line searches needed within this method, as well as the line search required for the determination of b in (7), are based on a Brent method. Some coefficients are either fixed or recalculated by hand in case this makes sense from the start. In total, 11 coefficients are fitted for B0, 4 are fitted for B1, and 9 are fitted for B2.

The resulting set c′ is displayed in Table 1. The normalized cost function (5) before (i.e., using c′ = c) and after minimization is displayed in the second and third columns of Table 2. Cost (5), which is now evaluated for choice (6), is displayed in the fourth and fifth columns. Here, (6) is in decibel units for B0, and situations close to extrema are now included in the average (5). Note that it only serves a diagnostic purpose: that is, this cost was not used in the minimization. The sixth and seventh columns of Table 2 show the maximum possible effect on backscatter (explored over all wind speeds, wind directions, and incidence angles) that could be induced by the misfit in B0, B1, and B2.

For B0, the result of the minimization is satisfactory. The weighted average deviation is 0.046 m s−1. A detailed view of b (see right panel of Fig. 4) shows that, for a large part in the (υ, θ) plane, CMOD5.N is expected to give winds that are within 0.02 m s−1 from 0.7 m s−1. Only, for light winds at a lower incidence angle, deviations are larger, and it is this region that contributes most heavily to the cost. For ASCAT and ERS-2, wind speed is mainly determined by the fore and aft beams. Because incidence angles for these beams start at 37° and 25°, respectively, deviations in wind retrieval are expected to be within a few percent of 1 m s−1 (especially for ASCAT). Deviation in backscatter (fifth column in Table 2) is on average 0.01 dB. This is on the discretization level of σ0 reported by the data provider (i.e., 0.01 dB); therefore, it is very small. For all winds above 15 m s−1, the deviation is below 0.01 dB. The largest deviation (0.85 dB) occurs for low wind speed at low incidence angle. Because here sensitivity of backscatter on wind speed is maximal, its effect on inverted wind speed will be limited. In fact, this was the reason for the choice of (7) rather than (6).

For B1 and B2, the reduction in cost is less than that for B0. One reason for this is that these terms are relatively slowly varying functions of wind speed. A small error in σ0 is transformed back to a large shift in wind speed. As mentioned above, though, these terms do not set the wind speed scale. The (diagnostic) reduction in cost when calculated in terms of deviations (6) is larger, especially for B2 (fifth versus fourth columns of Table 2). The right panels of Figs. 5 and 6 show that the differences from relation (5) are small and that the maximum effect on backscatter is very small (column 7 of Table 2). Note that, for B1, usage of CMOD5 coefficients would only give a maximum error of 0.04 dB to start with.

4. Validation of the CMOD5.N fit

The optimization of the CMOD5.N coefficients as performed in the previous section does not involve any wind inversion. The rationale is that, as long as the cost function (5) is used in combination with the choice of departures (7), any inversion scheme should retrieve winds that are on average 0.7 m s−1 stronger compared to CMOD5. In this section, this will be verified for the wind products from ERS-2 and ASCAT as inverted and assimilated at ECMWF. Here, such inversions are based on the PRESCAT algorithm (Stoffelen and Anderson 1997). It uses a precalculated lookup table of CMOD backscatter as function of wind speed (1–60 m s−1 in bins of 0.5 m s−1), relative wind direction (bins of 5°), and incidence angle (15°–69° in 1° bins). As a result, the inverted wind product is discretized into steps of 0.5 m s−1 in speed and 5° in direction. Winds below 1 m s−1 are rejected. Two almost equally likely wind solutions are obtained, which are usually nearly antiparallel. The analysis given below will only involve the solution that best matches the ECMWF wind field. Prior to wind inversion, there is the possibility to correct backscatter; after inversion, wind speed can be corrected as well. These can compensate for inconsistencies in the calibration of backscatter by the data provider and for residual errors in the applied CMOD function. For ERS-2, such corrections are not used; for ASCAT, they are.

The full calibration of the ASCAT backscatter product as provided by EUMETSAT was finalized on 2 December 2008. Comparison of measured backscatter with backscatter obtained from collocated ECMWF wind subjected to CMOD5.4 (known as ocean calibration; see Stoffelen 1999) reveals differences up to −1 dB that mainly appear as a function of incidence angle. This deviation reflects a difference in calibration between ERS-2 and ASCAT. To use CMOD5.4 for ASCAT, a correction in backscatter space is applied prior to wind inversion. At ECMWF, correction tables were changed each time EUMETSAT had updated calibration. The top and middle panels of Fig. 7 display tables used at ECMWF during June 2007 and January 2008. The bottom panel shows the correction used at ECMWF since the finalized calibration. Note that the applied corrections are similar to those used at KNMI for the production of the level-2 ASCAT product (Verspeek et al. 2008). After wind inversion, comparison with ECMWF wind still shows some residual speed biases. They mainly occur at high wind speed for some wind-vector cells. At the moment, it is not clear whether these are connected to residual effects from incomplete calibration or whether these indicate wind speed–dependent inaccuracies in CMOD5(.4). The correction in wind speed as used at ECMWF since the operational implementation of ASCAT on 12 June 2007 is shown in Fig. 8. Details may be found in Hersbach and Janssen (2007b). A 1-yr collocation study (July 2007–June 2008) between ASCAT and ERS shows that both wind products as described above are very well intercalibrated (section 4.5 of Abdalla and Hersbach 2008).

ERS-2 and ASCAT winds based on CMOD5 and CMOD5.N are each inverted for July 2007 and January 2008. From this, it appears that, for both ERS-2 and ASCAT, CMOD5.N winds are on average 0.69 m s−1 stronger than CMOD5. Scatter is low (0.32 m s−1 for both instruments) and is mainly induced by the discretization into 0.5 m s−1 bins for each product. When stratified with respect to wind speed and scatterometer WVC, results are consistent. This is summarized in Fig. 9, where each point represents the mean a of the following conditional averages:
i1520-0426-27-4-721-eq1
where
i1520-0426-27-4-721-e8
Here, x and y denote CMOD5 and CMOD5.N wind speeds, respectively, and Si are predefined speed bins of 2.5 m s−1 and given WVC and the angle brackets denote set-wise averages. Only subsets with 100 winds or more are plotted. Averaging (8) is necessary to disentangle (pseudo) biases introduced by the discretization of wind speed bins into 0.5 m s−1, which is not small compared to the objective speed difference of 0.7 m s−1 and affects results at low and high wind speed. A discussion on the usage of weighted bin averages may be found in section 3.1 of Hersbach et al. (2007).

For ASCAT, the results displayed in Fig. 9 are based on the inclusion of correction in backscatter, but excluding the postinversion correction in wind speed. These latter corrections are determined per wind speed bin. Therefore, when CMOD5.N and CMOD5 winds fall in different bins, different (noninterpolated) corrections are applied, which would clutter the real difference between the two model functions.

As a function of wind speed (top panels of Fig. 9), results are similar for ERS-2 and ASCAT. For light winds, wind correction is a few percent of 1 m s−1 less, especially for ERS-2. Correction is maximal for winds around 5 m s−1; then decreases slightly; and, for winds above 15 m s−1, increases again. These variations, which are all within a few percent of 1 m s−1, are in agreement with what was anticipated during the CMOD5.N fit and are reflected in the right panel of Fig. 4. As a function of wind-vector cell (lower panels of Fig. 9), variation is small for ASCAT; for ERS-2, winds are slightly more enhanced at low WVC.

Differences in wind direction between CMOD5.N and CMOD5 appear small (not shown). Relative standard deviations of 4.2 for ERS-2 and 3.3° for ASCAT are mainly the result of discretization.

5. The effect of stability on scatterometer departures from ECMWF wind

For the two periods July 2007 and January 2008, ERS-2 and ASCAT winds based on CMOD5.N are collocated with ECMWF neutral winds, using the setup as described in the last part of section 2. Short-range forecast winds are used rather than analysis winds to minimize the effect of correlation with the operationally assimilated ERS-2 and ASCAT product. Summed over both periods, the entire globe and all WVCs, the comparison is good, as can be seen in Fig. 10. For ERS-2, the overall bias is small (−0.03 m s−1). Although ASCAT winds are somewhat lower (bias of −0.11 m s−1), one should realize this may be due to the difference in coverage between ERS-2 (nonglobal) and ASCAT (global). As a function of WVC, bias varies between −0.26 and 0.10 m s−1 for ERS-2 and between −0.19 and −0.04 m s−1 for ASCAT. A more detailed view is presented in Fig. 11. Each point represents the average as defined in Eq. (8) for a set (of at least 100 data points) restricted to a specific WVC and certain speed bin (width of 2.5 m s−1). From this, it is seen that variations as functions of wind speed and WVC are mild. Only for extreme winds, ERS-2 wind speed is somewhat lower; for ASCAT, there are internode differences. The latter may be tempered by an appropriate update of the wind speed correction table, as displayed in Fig. 8. However, statistics for extreme winds are low.

The relative standard deviation between CMOD5.N and ECMWF neutral winds (1.30 m s−1 for ERS-2 and 1.27 m s−1 for ASCAT) appears slightly lower than the standard deviation with ECMWF nonneutral winds (1.35 m s−1 for ERS-2 and 1.30 m s−1 for ASCAT). This indicates that, indeed, some of the scatter between scatterometer and ECMWF wind can be explained by the effect of stability. To investigate this further, maps of average departures for ASCAT data are presented in Fig. 12 and are to be compared to the middle panels of Fig. 2. Similar results are obtained for ERS-2 in areas of data coverage (not shown). To undo the effect of the difference between neutral and nonneutral wind of 0.20 m s−1, in these maps, global average values have been subtracted. It is seen that, at various locations, relative bias levels have improved. For example, this is the case for the stable areas in summer in the Hudson Bay, east of Newfoundland, and large areas along the summer Antarctic Circumpolar Current (ACC). Connected to unstable stratification in winter, relative biases in the Kuroshio Extension and around the Gulf Stream have slightly diminished. Although departures have been reduced in various areas, stability-dependent biases do remain.

In addition to a reduction of local average biases, the local random error between scatterometer and model wind appears smallest for neutral winds. This can be seen as the result of a positive local correlation between scatterometer departures and stability corrections on shorter (synoptic) time scales. Let dn and d denote the departure of CMOD5.N wind from ECMWF neutral and nonneutral wind speeds respectively, and η denote the correction in stability:
i1520-0426-27-4-721-e9
The local correlation C between departures and stability corrections can be expressed as
i1520-0426-27-4-721-e10
i1520-0426-27-4-721-e11
where σx2 = 〈(x − 〈x〉)2〉, C(x, y) = (〈xy〉 − 〈x〉〈y〉)/(σxσy) are the usual definitions and the angle brackets denote location-wise averages. Validity of relation (10) can be easily verified by the substitution dn = dη in at the rhs and similarly d = dn + η in σd2 for (11). The fact that, for most areas, σdn appears slightly lower than σd means that C(d, η) must be positive. Note that correlation and standard deviation are insensitive to integral biases, such as the average value of ∼0.2 m s−1 for η. Maps for location-wise correlations are displayed in Fig. 13. Indeed, the correlation between nonneutral departure and stability correction is positive almost everywhere, with values that can exceed 0.6 locally. Largest correlation is typically found in areas where the fluctuations in stability are largest (bottom panels of Fig. 2). The correlation for neutral departures is much lower. For July 2007, the global average no longer shows a correlation. Locally, some positive and negative correlations remain. For January 2008, there is still a residual global positive correlation (0.10). Nonneutral correlation (of 0.19) for this month is also larger than for July 2007 (0.11).

6. Discussion

The initial experience with the CMOD5.N geophysical model function is promising. In addition to providing a model function that relates to neutral wind rather than nonneutral wind, CMOD5.N resolves a reported bias in the literature of 0.5 m s−1 for CMOD5 when compared to buoy data.

For ERS-2 and ASCAT, winds inverted on the basis of CMOD5.N compare well on average with ECMWF short-range neutral forecast wind. As expected from the belief that scatterometer data are sensitive to surface stress, the agreement with neutral wind appears better than with nonneutral model wind. Locally, the average seasonally dependent departures are tempered in several areas around the globe, and the relative local standard deviation is slightly reduced. The mostly positive local correlation between scatterometer departures from model wind and corrections for atmospheric stability has reduced.

Although stability corrections usually work in the proper direction, a residual stability-dependent difference between scatterometer and neutral model wind remains. These could be due to stability-dependent errors in the ECMWF boundary layer formulation. For example, as pointed out by Brown et al. (2006) and Song et al. (2009), in the ECMWF model formulation (as well as in the Met Office model), momentum mixing seems too strong in the stable boundary layer and too weak under unstable conditions. In Brown et al. (2006), a triple collocation study between ECMWF, QuikSCAT, and winds from a number of buoys platforms from the U.S. National Data Buoy Center (NDBC) indicated that the relative bias between scatterometer and neutral buoy wind did not show a clear dependence on stability. As a consequence, neutral ECMWF winds under stable conditions are thought to be too strong, under unstable conditions too weak, and this shows up as a residual bias when compared to neutral scatterometer winds.

At first sight, the finding of the present analysis may oppose a triple collocation study performed by Portabella and Stoffelen (2009) between ERS-2, buoy data, and 40-yr ECMWF Re-Analysis (ERA-40; Uppala et al. 2005) wind. That study concludes that scatterometer winds are as close to real (nonneutral) winds as to neutral winds. Note that, at most buoy locations used in that study, stability effects are found to be small in the present work as well. As observed by Kelly et al. (2001) and acknowledged by Portabella and Stoffelen (2009), the effect of ocean current in (1) is significant at many tropical buoy locations. This may clutter the effect of stability on scatterometer departures. No buoy data were available over the ACC, where large stability effects are found in summer. Only in the area east of Newfoundland, a number of buoy data were available, which may have been too little to stand out. The ERA-40 winds for 1999–2000 as used in Portabella and Stoffelen (2009) are of considerably lower quality than the operational ECMWF winds for December 2007 and July 2008, as used in the current study, which may have masked the effect of stability as well.

The difference between neutral and nonneutral wind only explains part of the average bias between scatterometer and ECMWF model winds. As is seen from Fig. 12, there are several areas in which the local biases between scatterometer and model wind do not seem to be connected with stability. Examples are orographic effects (Xie et al. 2001; Chelton et al. 2004) and above-mentioned current effects (Kelly et al. 2001). In the equatorial Pacific, for instance, Fig. 12 displays a dipole of positive and negative bias, which may be connected to the presence of the South Equatorial Current and the Equatorial Counter Current.

In the Gulf of Guinea, Fig. 12 displays large departures as well. Here, ECMWF model winds consistently seem to underestimate wind speed by 1–2 m s−1, where wind speed is generally low to start with (∼4 m s−1). Long-term monitoring of scatterometer data at ECMWF shows that these biases exist at least since the availability of ERS-1 data in 1992. The alleged underestimation of model wind speed is confirmed by wind data from the Soul PIRATA buoy at (0°0 N, 0°0 E) (J. Bidlot 2008, personal communication). A more detailed view on this bias shows that the model winds lack convergence in the intertropical convergence zone (ITCZ) in this region. Between May and August, the negative bias usually appears milder, for the simple reason that the ITCZ is then located more over West Africa. During that period, a shift of the ITCZ in the ECMWF model appears responsible for incorrect precipitation forecasts in the region studied by the African Monsoon Multidisciplinary Analyses (AMMA) project (see, e.g., Agusti-Panareda and Beljaars 2008).

Scatterometer winds seem consistently stronger in the vicinity of land and sea ice. One explanation is a lack in the land–sea gradient in ECMWF wind speed resulting from the horizontal resolution (25 km). Near the poles, contamination of scatterometer data by ice would lead to erroneously high winds. However, the screening used at ECMWF, at which the data used in Fig. 12 are also subjected, is rather conservative (Hersbach and Janssen 2007a) and should take care of most of ice problems.

Another source for biases could be given by sea-state dependencies (Quilfen et al. 2004). In coastal areas with on average off-land wind conditions, ocean surface waves are younger and steeper, with a higher than average value for Charnock (Janssen 1991) and according to (2) an enhanced surface roughness. For example, the left panel of Fig. 14 shows enhanced average values for ECMWF short-range forecast Charnock (collocated to ASCAT) in the typically short-fetch areas of the North American East Coast, North Sea, and Baltic Sea for January 2008. The empirically determined CMOD5 GMF does not explicitly account for sea-state effects; for this reason, the effective relation between wind and backscatter will be based on a representative value αGMF for Charnock. Because ocean waves at higher wind speed are more likely connected to smaller dimensionless duration-limited growth, they will on average display an enhanced roughness. For that reason, the typical Charnock parameter will be a progressive function of wind speed and so will αGMF. For a given wind speed, there is a large variation in the value of Charnock. An example is given in Fig. 6 of Portabella and Stoffelen (2009), which shows statistics for ECMWF ERA-40 during 1999–2000. The right panel of Fig. 14 shows the potential impact on ASCAT wind speed. Based on a 1-yr collocation between ASCAT data and ECMWF Charnock between May 2007 and April 2008, αGMF(υ) is determined as a function of ASCAT wind speed υ. For the data displayed in Fig. 14, friction velocity u* is estimated from (1) on the basis of CMOD5.N wind speed υ and effective αGMF(υ). The obtained u* is then transformed back to neutral wind using the actually collocated value of model Charnock. Following (1) and as seen in Fig. 14, an enhanced (reduced) Charnock leads to a reduced (enhanced) scatterometer wind speed. Averaged over January 2008, short-fetch areas display a typical reduction in ASCAT wind speed of 0.2–0.3 m s−1, while an enhancement of similar amplitude occurs over the open Atlantic. Although not clearly visible from the bottom-right panel of Fig. 12, a reduction in ASCAT wind speed would temper positive neutral departures close to land in the Gulf of Mexico, near Greenland and Norway, and in the Baltic Sea.

Portabella and Stoffelen (2009) do not find a dependency of average scatterometer departure as a function of the ERA-40 Charnock parameter. However, the conclusion that this would imply that sea-state effects are insignificant for the translation from wind to stress does necessarily hold. For a given Charnock parameter and scatterometer wind speed, a sea-state dependency will result in an expected average departure from (neutral) model wind that is a function of the deviation of α from αGMF(υ). Stratified to a certain value of Charnock, the resulting subset will (because of the large scatter between α and υ) contain a range of wind speeds. Cases of strong wind correspond to a lower than average Charnock value for that wind, whereas lighter winds relate to a higher than average α. Their effect on departures is opposite and will largely counterbalance when averaging over all cases in the stratified subset. As a result, average departures stratified to Charnock are expected to display hardly any relation, regardless of whether sea-state effects are important. This illustrates that the assessment of sea-state effects on scatterometer departures is delicate, which is beyond the scope of this paper.

In conclusion, a number of issues may influence the comparison between scatterometer wind and NWP wind. A variety of model errors may largely contribute. With respect to the interpretation of scatterometer observations, the relation between stress and wind [as, e.g., provided in (1)] may depend, among other effects, on atmospheric stability, sea state, and ocean currents. This paper shows a beneficial effect of stability corrections on scatterometer departures and supports a preference for the interpretation of scatterometer data in terms of neutral wind rather than nonneutral wind. From 20 November 2008 onward, the ASCAT wind product disseminated by EUMETSAT/KNMI is based on CMOD5.N (available online at http://www.knmi.nl/scatterometer).

Acknowledgments

The work presented in this document was funded by ESRIN (Project Ref. 22025/08/I-EC). The author would like to thank Dudley Chelton, Peter Janssen, Ad Stoffelen, Jean Bidlot, Peter Bechtold, Martin Köhler, Anna Agusti-Panareda, Martin Miller, and two reviewers for their useful discussions and/or suggestions.

REFERENCES

  • Abdalla, S., , and Hersbach H. , 2007: The technical support for global validation of ERS wind and wave Products at ECMWF. ECMWF Contract Report to the European Space Agency, 59 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/esa/ESA_abdalla_hersbach_18212-2007.pdf].

    • Search Google Scholar
    • Export Citation
  • Abdalla, S., , and Hersbach H. , 2008: The technical support for global validation of ERS wind and wave products at ECMWF. ECMWF Contract Report to the European Space Agency, 40 pp. [Available online at http://www.ecmwf.int/publications/library/ecpublications/_pdf/esa/ESA_abdalla_hersbach_20901-2008.pdf].

    • Search Google Scholar
    • Export Citation
  • Agusti-Panareda, A., , and Beljaars A. , 2008: ECMWF’S contribution to AMMA. ECMWF Newsletter, No. 115, ECMWF, Reading, United Kingdom, 19–25.

    • Search Google Scholar
    • Export Citation
  • Attema, E. P. W., 1986: An experimental campaign for the determination of the radar signature of the ocean at C-band. Proc. Third Int. Colloquium on Spectral Signatures of Objects in Remote Sensing, Les Arcs, France, ESA, 791–799.

    • Search Google Scholar
    • Export Citation
  • Brown, A. R., , Beljaars A. C. M. , , and Hersbach H. , 2006: Errors in parameterizations of convective boundary layer turbulent moment mixing. Quart. J. Roy. Meteor. Soc., 132 , 18591876.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Carswell, J. R., , Donnelly W. J. , , McIntosh R. E. , , Donelan M. A. , , and Vandemark D. C. , 1999: Analysis of C and Ku band ocean backscatter measurements under low-wind conditions. J. Geophys. Res., 104 , (C9). 2068720701.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , and Freilich M. H. , 2005: Scatterometer-based assessment of 10-m wind analyses from the operational ECMWF and NCEP numerical weather prediction models. Mon. Wea. Rev., 133 , 409429.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Chelton, D. B., , Schlax M. G. , , Freilich M. H. , , Milliff R. F. , , and Vandemark D. C. , 2004: Satellite measurements reveal persistent small-scale features in ocean winds. Science, 303 , 978983.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Donnelly, W. J., , Carswell J. R. , , McIntosh R. E. , , Chang P. S. , , Wilkerson J. , , Marks F. , , and Black P. G. , 1999: Revised ocean backscatter models at C and Ku-band under high-wind conditions. J. Geophys. Res., 104 , (C5). 1148511497.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Ebuchi, N., , Graber H. C. , , and Caruso M. J. , 2002: Evaluation of wind vectors observed by QuikSCAT/SeaWinds using ocean buoy data. J. Atmos. Oceanic Technol., 19 , 20492062.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Fernandez, D. E., , Carswell J. R. , , Frasier S. , , Chang P. S. , , Black P. G. , , and Marks F. D. , 2006: Dual-polarized C- and Ku-band ocean backscatter response to hurricane-force winds. J. Geophys. Res., 111 , C08013. doi:10.1029/2005JC003048.

    • Search Google Scholar
    • Export Citation
  • Figa-Saldaña, J., , Wilson J. J. W. , , Attema E. , , Gelsthorpe R. , , Drinkwater M. , , and Stoffelen A. , 2002: The advanced scatterometer (ASCAT) on the meteorological operational (MetOp) platform: A follow on for European wind scatterometers. Can. J. Remote Sens., 28 , 404412.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Francis, R., and Coauthors, 1991: The ERS-1 spacecraft and its payload. ESA Bull., 65 , 2748.

  • Freilich, M. H., , and Dunbar R. S. , 1999: The accuracy of the NSCAT 1 vector winds: Comparisons with National Data Buoy Center buoys. J. Geophys. Res., 104 , (C5). 1123111246.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Hersbach, H., , and Janssen P. , 2007a: Improved scatterometer screening. ECMWF Research Department Memo. R60.9/HH/0735, 24 pp.

  • Hersbach, H., , and Janssen P. , 2007b: Preparation for assimilation of surface-wind data from ASCAT at ECMWF. ECMWF Research Department Memo. R60.9/HH/0750, 24 pp.

    • Search Google Scholar
    • Export Citation
  • Hersbach, H., , Stoffelen A. , , and de Haan S. , 2007: An improved C-band scatterometer ocean geophysical model function: CMOD5. J. Geophys. Res., 112 , C03006. doi:10.1029/2006JC003743.

    • Search Google Scholar
    • Export Citation
  • Isaksen, L., , and Janssen P. A. E. M. , 2004: The benefit of ERS scatterometer winds in ECMWF’s variational assimilation system. Quart. J. Roy. Meteor. Soc., 130 , 17931814.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Janssen, P. A. E. M., 1991: Quasi-linear theory of wind-wave generation applied to wave forecasting. J. Phys. Oceanogr., 21 , 16311642.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Kelly, A. K., , Dickinson S. , , McPhaden J. , , and Johnson G. C. , 2001: Ocean currents evident in satellite wind data. 28 , 24692472.

  • Long, A. E., 1984: Towards a C-band radar sea echo model for the ERS-1 scatterometer. Proc. Third Int. Colloquium on Spectral Signatures of Objects in Remote Sensing, Les Arcs, France, ESA, 29–34.

    • Search Google Scholar
    • Export Citation
  • Monin, A. S., , and Obukhov A. M. , 1954: Basic regularity in turbulent mixing in the surface layer of the atmosphere. Geofiz. Inst. Akad. Nauk SSSR, 24 , 163187.

    • Search Google Scholar
    • Export Citation
  • Offiler, D., 1994: The calibration of ERS-1 satellite scatterometer winds. J. Atmos. Oceanic Technol., 11 , 10021017.

  • Portabella, M., , and Stoffelen A. , 2009: On scatterometer ocean stress. J. Atmos. Oceanic Technol., 26 , 368382.

  • Press, W. H., , Teukolsky S. A. , , Vetterling W. T. , , and Flannery B. P. , 1997: Numerical Recipes in Fortran 90: The Art of Scientific Computing. 2nd ed. Cambridge University Press, 933 pp.

    • Search Google Scholar
    • Export Citation
  • Quilfen, Y., , Chapron B. , , Collard F. , , and Vandemark D. , 2004: Relationship between ERS scatterometer measurement and integrated wind and wave parameters. J. Atmos. Oceanic Technol., 21 , 368373.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Song, Q., , Chelton D. B. , , Esbensen S. K. , , Thum N. , , and O’Neill L. W. , 2009: Coupling between sea surface temperature and low-level winds in mesoscale numerical models. J. Climate, 22 , 146164.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., 1999: A simple method for calibration of a scatterometer over the ocean. J. Atmos. Oceanic Technol., 16 , 275282.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Stoffelen, A., , and Anderson D. , 1997: Scatterometer data interpretation: Estimation and validation of the transfer function CMOD4. J. Geophys. Res., 102 , (C3). 57675780.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131 , 29613012.

  • Verspeek, J., , Portabella M. , , Stoffelen A. , , and Verhoef A. , 2008: Calibration and validation of ASCAT winds, version 4.0. EUMETSAT Ocean and Sea Ice SAF Tech. Note SAF/OSI/KNMI/TEC/TN/163, 31 pp. [Available online at http://www.knmi.nl/publications/fulltexts/calibration_and_validation_of_ascat_winds_4.0.pdf].

    • Search Google Scholar
    • Export Citation
  • Wentz, F. J., , and Smith D. K. , 1999: A model function for the ocean-normalized radar cross section at 14 GHz derived from NSCAT observations. J. Geophys. Res., 104 , (C5). 1149911514.

    • Crossref
    • Search Google Scholar
    • Export Citation
  • Xie, S-P., , Liu W. T. , , Liu Q. , , Nonaka M. , , and Hafner J. , 2001: Long-range effects of the Hawaiian Islands on the Pacific ocean-atmosphere system. Proc. IGARSS Int. Geoscience and Remote Sensing Symp., Sydney, Australia, IEEE, Vol. 3, 1086–1088.

    • Search Google Scholar
    • Export Citation

Fig. 1.
Fig. 1.

Scatterometer winds from ERS-2 and ASCAT as assimilated at ECMWF for the 12-h assimilation cycle of 0000 UTC 9 Dec 2007. The two tracks around 40°W and the track parallel to 10°W concern ASCAT data; the three tracks around 45°W, 20°W, and 0° originate from ERS-2. Large dark barbs indicate actively assimilated winds, and small light barbs indicate rejected (mostly thinned) winds. The gray dashed lines represent streamlines for the ECMWF analysis surface wind field.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 2.
Fig. 2.

(top) Average ocean wind speed (gray scales) and ocean vector wind (arrows) at 10-m height and (middle) average difference and (bottom) standard deviation from equivalent neutral wind speed, for the operational ECMWF short-range forecast at ASCAT location during (left) July 2007 and (right) January 2008.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 3.
Fig. 3.

Average relation between ECMWF neutral and nonneutral wind speed (m s−1) at (left) ERS-2 and (right) ASCAT locations, summed over July 2007 and January 2008. Data points represent wind-vector cell-wise (19 for ERS-2 and 42 for ASCAT) averages in speed bins of 2.5 m s−1 for which at least 100 data points were available. Average difference is 0.20 m s−1 for both ERS-2 and ASCAT.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 4.
Fig. 4.

(left) B0 for CMOD5 as function of wind speed υ and incidence angle θ (dB) and (right) the wind speed shift b (m s−1) for which B0(υ + b, θ) based on CMOD5.N coefficients equals B0(υ, θ) using original CMOD5 coefficients. Light horizontal lines mark the incidence-angle range for ERS-2, and dark horizontal lines show the corresponding range for ASCAT (thin for mid and thick for fore/aft antenna).

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 5.
Fig. 5.

As in Fig. 4, but for (left) B1 in natural unit and (right) the difference between fit (2).

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 6.
Fig. 6.

As in Fig. 5, but for B2.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 7.
Fig. 7.

Bias correction (dB) applied to ASCAT backscatter before wind inversion as function of WVC and antenna (fore is dashed, mid is solid, and aft is dotted–dashed) as used at ECMWF for indicated period.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 8.
Fig. 8.

Additional correction in wind speed (m s−1) after inversion as function of WVC and wind speed based on a comparison study with ECMWF short-range forecast nonneutral winds for the period 1–20 Feb 2007. Minimum and maximum values in speed correction are −0.81 and 1.20 m s−1, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 9.
Fig. 9.

Difference between CMOD5.N and CMOD5 wind speeds after inversion of (left) ERS-2 backscatter triplets and (right) ASCAT triplets, respectively, accumulated over July 2007 and January 2008. Dots represent averages for data of specific WVC and speed domain (in bins of 2.5 m s−1) for which at least 100 data points were available, and they are displayed as functions of (top) wind speed and (bottom) WVC, respectively.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 10.
Fig. 10.

Histogram between scatterometer wind based on CMOD5.N, for (left) ERS-2 and (right) ASCAT, and collocated ECMWF short-range forecast equivalent neutral wind speed at 10 m. Contours are in steps of 5 dB. Circles denote averages for bins in the x direction, and squares denote averages for bins in the y direction.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 11.
Fig. 11.

As in Fig. 3 and Fig. 9 (top), but for CMOD5.N wind speed vs collocated ECMWF neutral wind.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 12.
Fig. 12.

Anomaly from the global mean of ASCAT departures from ECMWF (top) nonneutral and (bottom) neutral wind accumulated over (left) July 2007 and (right) January 2008 and gridded into boxes on an N80 reduced Gaussian grid (average box size is 1.125° by 1.125°). Winds are based on CMOD5.N. Note a difference in gray scaling with respect to Fig. 2.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 13.
Fig. 13.

The local correlation of stability corrections with ASCAT wind speed departures for (top) nonneutral and (bottom) neutral departures, for subsets collected per N80 reduced Gaussian grid box and accumulated over (left) July 2007 and (right) January 2008.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Fig. 14.
Fig. 14.

(left) Local value of the ECMWF short-forecast Charnock parameter collocated at ASCAT location and (right) the local effect on the determination of ASCAT wind speed averaged over January 2008 onto an N80 reduced Gaussian grid.

Citation: Journal of Atmospheric and Oceanic Technology 27, 4; 10.1175/2009JTECHO698.1

Table 1.

CMOD5 and CMOD5.N coefficients.

Table 1.
Table 2.

Square root of cost Ji in wind speed space (m s−1) and backscatter space (dB) averaged over wind speed from 1 to 50 m s−1 (steps of 1 m s−1) and incidence angle from 17° to 60° (steps of 1°), at the start (CMOD5 coefficients) and end (CMOD5.N coefficients) of the minimization.

Table 2.

1

Only ECMWF winds were used that collocate with ASCAT observations and where sea ice fraction was below 10% and SST was above 273.15 K. This restriction is not essential for the discussion.

Save