Matching ASCAT and QuikSCAT Winds

1 2 3 4 5 Matching ASCAT and QuikSCAT Winds 6 7 8 9 10 11 12 13 14 15 16 17 Abderrahim Bentamy 1, Semyon A. Grodsky 2, James A. Carton 2, Denis Croizé-Fillon 1, Bertrand Chapron 1 Submitted to : JGR-Oceans (rev. 07/29/2011) July 29, 2011 18 19 20 1 Institut Francais pour la Recherche et l Exploitation de la Mer, Plouzane, France 2 21 Department of Atmospheric and Oceanic Science, University of Maryland, College 22 Park, MD 20742, USA 23 24 25 26 Corresponding author: senya@atmos.umd.edu

27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 Abstract Surface winds from two scatterometers, the Advanced scatterometer (ASCAT), available since 2007, and QuikSCAT, which was available through November 2009, show persistent differences during their period of overlap. This study examines a set of collocated observations during a 13-month period November 2008 through November 2009, to evaluate the causes of these differences. The difference in the operating frequency of the scatterometers leads to differences that depend on rain rate, wind velocity, and SST. The impact of rainfall on the higher frequency QuikSCAT is up to 1 ms -1 in the tropical convergence zones and along the western boundary currents even after the rain flagging is applied. This difference from ASCAT is reduced by some 30% to 40% when data for which the multidimensional rain probabilities >0.05 is also removed. An additional component of the difference in wind speed seems to be the result of biases in the geophysical transfer functions used in processing the two data sets and is parameterized as a function of ASCAT wind speed and direction relative to the mid beam azimuth. After applying the above two corrections, QuikSCAT wind speed remains systematically lower (by 0.5 ms -1 ) than ASCAT over cold SST<5 o C. This difference appears to be the result of temperature-dependence in the viscous dumping which differentially impacts shorter waves and thus preferentially impacts QuikSCAT. The difference in wind retrievals also increases in the storm track corridors as well as in the coastal regions where the diurnal cycle of breeze projects on the time lag between satellites. 48 2

49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 1. Introduction Many meteorological and oceanic applications require the high spatial and temporal resolution of satellite [e.g. Grima et al, 1999; Grodsky et al, 2001, Blanke et al, 2005; Risien and Chelton, 2008]. Many of these studies also require consistent time series spanning the lifetime of multiple satellite missions, of which there have been seven since the launch of the first European Remote Sensing satellite (ERS-1) in August 1991. Creating consistent time series requires accounting for changes in individual mission biases, most strikingly when successive scatterometers operate in different spectral bands [e.g. Bentamy et al., 2002; Ebuchi et al., 2002]. This paper presents a comparative study of winds derived from the matching of two such scatterometers: the C-band Advanced SCATterometer (ASCAT) on board Metop-A and the higher frequency Ku-band SeaWinds scatterometer onboard QuikSCAT. Scatterometers measure wind velocity indirectly through the impact of wind on the amplitude of capillary or near-capillary surface waves. This wave field is monitored by measuring the strength of Bragg scattering of an incident microwave pulse. If a pulse of wavenumber k impinges on the surface at incidence angle relative to the surface the maximum backscatter will occur for surface waves of wavenumber of k B 2k sin( ) whose amplitude, in turn reflects local wind conditions [e.g. Wright and Keller, 1971). 0 The fraction of transmitted power that returns back to the satellite is thus a function of local wind speed and direction (relative to the antenna azimuth), and. To date, the most successful conversions of scatterometer measurements into near- surface wind rely on empirically derived geophysical model functions (GMFs) augmented by procedures to resolve directional ambiguities resulting from uncertainties 3

72 73 74 75 76 77 78 79 80 in the direction of wave propagation. Careful tuning of the GMFs is currently providing estimates of 10m neutral wind velocity with errors estimated to be around ±1 ms -1 and ±20 o, [e.g. Bentamy et al., 2002; Ebuchi et al., 2002]. However systematic errors are also present. For example, Bentamy et al. [2008] have shown that ASCAT has a systematic underestimation of wind speed that increases with wind speed, reaching 1 ms -1 for winds of 20 ms -1. The GMFs and other parameters must change if the frequency band being used by the scatterometer changes and historically scatterometers have used two different frequency bands. US scatterometers as well as the Indian OCEANSAT2 use frequencies 81 in the 10.95-14.5GHz Ku-band. In contrast the European Remote Sensing satellite 82 83 84 85 86 87 88 89 90 scatterometers have all adopted the 4-8 GHz C-band to allow for reduced sensitivity to rain interference [Sobieski et al, 1999; Weissman et al., 2002]. To avoid spurious trends in a combined multi-scatterometer data set, the bias between different scatterometers needs to be removed. There have been numerous efforts to construct a combined multi-satellite data set of winds [Milliff et al., 1999; Zhang et al, 2006; Bentamy et al, 2007, Atlas et al, 2011]. These efforts generally rely on utilizing a third reference product such as passive microwave winds or reanalysis winds to which the individual scatterometer data sets can be calibrated. In contrast to such previous efforts, this current study focuses on directly 91 matching scatterometer data. We illustrate the processes by matching the current 92 ASCAT and the recent QuikSCAT winds [following Bentamy et al., 2008]. This 93 matching exploits the existence of a 13 month period November 2008 November 2009 4

94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 when both missions were collecting data and when ASCAT processing used the current CMOD5N GMF [Verspeek et al., 2010] to derive 10m neutral wind. 2. Data This study relies on data from two scatterometers, the SeaWinds Ku-band (13.4 GHz, 2.2 cm) scatterometer onboard QuikSCAT (referred to as QuikSCAT or QS, Table 1) which was operational June, 1999 to November, 2009 and the C-band ASCAT onboard the European Meteorological Satellite Organization (EUMESAT) MetOp-A 1 (5.225GHz, 5.7 cm) launched October, 2006 (referred to as ASCAT or AS). QuikSCAT uses a rotating antenna with two emitters: the H-pol inner beam with an incidence angle =46.25 and the V-pol outer beam with an incidence angle of =54. Observations from these two emitters have swaths of 1400km and 1800km, respectively which cover around 90% of the global ocean daily. These observations are then binned into Wind Vector Cells of 25 25 km. QuikSCAT winds used in this study are Level 2b data estimated with the empirical QSCAT-1 GMF [Dunbar et al., 2006] and the Maximum Likelihood Estimator which selects the most probable wind direction. To improve wind direction estimates in the middle of the swath an additional Direction Interval Retrieval with Threshold Nudging algorithm is applied. The winds produced by this technique show RMS speed and direction differences from concurrent in-situ buoy and ship data of approximately 1ms -1 and 23, and temporal correlations in excess of 0.92 [e.g. Bentamy et al., 2002; Bourassa et al, 2003; Ebuchi et al, 2002]. The QuikSCAT wind product includes several rain flags determined directly from the scatterometer 1 http://www.eumetsat.int/home/main/publications/technical_and_scientific_documentation/technical_no tes/ http://www.knmi.nl/scatterometer/. 5

115 116 117 118 119 120 121 122 observations and from collocated radiometer rain rate measurements from other satellites. The impact of rain on QuikSCAT data is indicated by two independent rain indices: rain flag and multidimensional rain probability (MRP). ASCAT has an engineering design that is quite different from QuikSCAT. Rather than a rotating antenna it has a three beam antenna looking 45 o (fore-beam), 90 o (midbeam), 135 o (aft-beam) of the satellite track, which together sweep out two 550 km swaths on both sides of the track. The incidence angle varies in the range 34 o -64 o for the outermost beams and 25 o -53 o for the mid-beam, giving Bragg wavelengths of 3.2-5.1cm 123 and 3.6-6.8 cm. Here we use level 2b ASCAT near real time data distributed by 124 EUMETSAT and by KNMI at 25x25 km 2 resolution. Comparisons to independent 125 126 127 128 129 130 131 132 133 134 135 136 mooring and shipboard observations by Bentamy et al. [2008] and Verspeek et al. [2010] show that ASCAT wind speed and direction has accuracies similar to QuikSCAT. To reevaluate wind accuracy we use a variety of moored buoy measurements including wind velocity, SST, air temperature, and significant wave height. These are obtained from Météo-France and U.K. MetOffice (European seas) [Rolland et al, 2002], the NOAA National Data Buoy Center (US coastal zone) [Meindl et al, 1992], and the tropical moorings of the Tropical Atmosphere Ocean Project (Pacific), the Pilot Research Moored Array (Atlantic), and Research Moored Array for African Asian Australian Monsoon Analysis and Prediction project (Indian Ocean) [McPhaden et al., 1998; Bourles et al., 2008; McPhaden et al., 2009]. Hourly averaged buoy measurements of wind velocity, sea surface and air temperatures, and humidity are converted to neutral wind at 10m height using the COARE3.0 algorithm of Fairall et al. [2003]. 6

137 138 139 140 141 142 143 144 Quarter-daily 10m winds, as well as SST and air temperatures, are also obtained from the European Centre for Medium Weather Forecasts (ECMWF) operational analysis. These are routinely provided by the Grid for Ocean Diagnostics, Interactive Visualization and Analysis (GODIVA) project 2 on a regular grid of 0.5 o in longitude and latitude. 3. Collocated observations Both satellites are on quasi sun-synchronous orbits, but the QuikSCAT local equator crossing time at the ascending node (6:30 a.m.) leads the ASCAT local equator 145 crossing time (9:30 a.m.) by 3 hours. This difference implies that data precisely 146 147 148 149 150 151 152 153 154 155 156 157 158 collocated spatially are available only with a few hours time difference at low latitudes (Figure. 1) [Bentamy et al, 2008]. Here we accept data pairs collocated in space and time where QuikSCAT data is collected within 0< <4 hours and 50 km of each valid ASCAT Wind Vector Cell. For the thirteen month period November 2008 through November 2009 this collocation selection procedure produces an average 2800 collocated pairs of data in each 1 o x1 o geographical bin. The data count is somewhat less in the regions of the tropical convergence zones due to the need to remove QuikSCAT rain-flagged data. The data count increases with increasing latitude as a result of the near-polar orbits, but decreases again at very high latitudes due to the presence of ice. At shorter lags of <3 hours collocated data coverage is still global, but the number of match-ups is reduced by a factor of two. At lags of <1 hours collocated data is only available in the extra tropics (Figure. 1). 2 behemoth.nerc-essc.ac.uk/ncwms/godiva2.html 7

159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 Next we explore possible impacts of nonzero time lag in the collocated data by examining the temporal lag decorrelation time of buoy winds. In midlatitudes where the characteristic time separation between collocated ASCAT and QuikSCAT data is less than 2 hours we examine wind measurements from NDBC buoys. The original buoy data is sub-sampled based on the timing of the nearest ASCAT (sub-sample 1) and QuikSCAT (sub-sample 2) overpasses. The mean bias between subsamples at these extra tropical NDBC mooring locations is close to zero with an RMS difference of less than 1ms -1, and a temporal correlation >0.95. Negligible time mean bias at the buoys suggests that time mean differences between collocated satellite wind fields (if any) is the result of differences in the satellite wind estimates rather than the time lag between samples. Similar results are found in an examination of data from the tropical moorings assuming < 4 hr with no mean bias, RMS differences of 1.2 m/s, and temporal correlations >0.92. Finally we examine the impact of the collocation space and time lag using the ECMWF surface winds to simulate space and time sampling of ASCAT and QuikSCAT during the period November 20, 2008 through November 19, 2009. This comparison also shows no significant geographical patterns in the sampling bias. Next we evaluate the accuracy of the satellite winds in the tropics by using this data window to find triply-collocated ASCAT, QuikSCAT, and tropical mooring winds (here we assume the mooring winds are unbiased). Both, QuikSCAT and ASCAT wind speeds ( W QS and W AS ) agree well with collocated mooring wind speeds, W buoy, (Figure 2) with time mean differences less than 1 m/s at all locations. ASCAT has a slight -0.35 ms -1 negative bias averaged over the moorings (Figure 3. In contrast, QuikSCAT mean wind 8

181 182 183 184 185 186 187 188 189 190 191 192 193 speed errors have a mixture of negative and positive biases, roughly mirror the spatial distribution of rainfall and giving a mooring-average bias of -0.15 ms -1. We evaluated the agreement of the time-dependent component of wind speed by examining RMS differences and correlations of the collocated satellite and buoy winds. This RMS difference averaged over all tropical buoys is 1.2 m/s and is well above the mean wind speed bias. The correlation of satellite and tropical buoy winds is 0.8 at = 4 hr, but exceeds 0.9 at zero lag (Figure 4). At higher latitudes, where the winds have longer synoptic timescales, the collocated satellite-buoy wind time series have even higher correlations (0.89 at =4 hours, and 0.96 at less than half an hour, not shown). 4. Results 4.1 Global comparisons Time mean difference of collocated QuikSCAT and ASCAT wind speed ( W W QS WAS, Figure 5) is generally less than 1ms -1, but contains some planetary- 194 195 196 197 198 199 200 201 202 203 scale patterns. In general, W >0 at tropical to midlatitudes where it exhibits a pattern of values in the range of 0.40-0.80 ms -1 resembling the pattern of tropical and midlatitude storm track precipitation. This pattern suggests that even after applying the rain flag there is a residual impact of rainfall on QuikSCAT wind retrievals (as previously examined by, e.g. Weissman et al., 2002]. Enhancing rain flagging by removing cases with MRP > 0.05 (5% of QuikSCAT data) decreases the frequency of positive differences in the tropical precipitation zones by some 30% to 40% and the number of large differences W >1 ms -1 is reduced by 38% (Figure 5). However, the RMS variability of W doesn t change with the enhanced rain flagging. Both with and without additional rain flagging RMS( W ) has a maximum of up to 3ms -1 in the zones of the midlatitude storm 9

204 205 206 207 208 tracks of both hemispheres and approaches 1.8 ms -1 in the tropical convergence zones. These maxima in inter-instrument variability reflect a combination of still unaccounted for impacts of rain events, and impacts of strong transient winds in the midlatitude storm tracks, as well as deep convection events in the tropics. In contrast to the tropics or midlatitudes, time mean W is negative at subpolar 209 and polar latitudes. This is particularly noticeable at southern latitudes where the 210 difference may exceed -0.5 ms -1. This striking high latitude inter-scatterometer 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 difference remains even after applying enhanced rain detection to QuikSCAT (Figure 5). The temporal correlation of QuikSCAT and ASCAT winds is high (>0.9) over much of the global ocean (Figure 5, CORR). It decreases to 0.7 (0.5 at some locations) in the tropical convergence zones. This decrease occurs because in the tropics the space scales of transient deep convective systems approache the scale of a single radar footprint while the temporal scales of transients compare with the time lag between the two satellites [Houze and Cheng, 1977]. In contrast, wind variability at mid and high latitudes has long multi-day synoptic timescales and large spatial scales which result in high correlations among collocated measurements. 4.2 Effect of geophysical model function In this section we attempt to parameterize observed W (after applying enhanced rain flagging to the QuikSCAT data) as a function of geophysical and satellite-earth geometrical factors. The ASCAT GMF is parameterized by a truncated Fourier series of wind direction (DIR) relative to antenna azimuth (AZIM) for each of the antenna beams =DIR-AZIM as 0 B0 B1 cos B2 cos 2, (1) 10

227 228 229 230 231 232 233 234 where the coefficients B depend on wind speed and incident angle,. This suggests that a correction function dw (to be added to ASCAT winds in order to adjust them to QuikSCAT winds) should depend on similar parameters, dw (,, ). Due to the fixed, three beam observation geometry of ASCAT, the azimuthal dependence of dw is reduced to dependence on ASCAT wind direction relative to the ASCAT midbeam azimuth, DIR AS AZIM 1. In initial experiments we find that dw does not depend on incidence angle (not shown). Empirical estimates of this correction function dw are constructed by binning W AS 235 observed W as a function of wind speed and. Here we begin by considering 236 estimates based on data from latitudes equatorward of 55 o where the time mean W 237 values are positive. Figure 6 reveals a clear dependence on. For ASCAT mid-beam 238 looking in the wind vector direction ( =0 o ) W is small. Looking in the upwind 239 240 241 242 243 244 245 246 247 248 direction ( 180 o ) QuikSCAT wind speed is stronger than ASCAT by 0.5 m/s. Looking in the cross-wind direction ( 90 o ) ASCAT wind speed is stronger than QuikSCAT wind speed by approximately 0.2 ms -1. Similar azimuthal dependencies show up when examining the difference between buoy wind speed and ASCAT wind speed (Figure 7) although data scatter is larger due to the smaller number of satellite-buoy collocations (compare right hand axes in Figures 6 and 7). We next examine the dependence of W on both wind speed and azimuth (Figure 8). Independent of wind direction, QuikSCAT wind speed exceeds ASCAT at both low winds ( W AS <3 ms -1 ) and high winds ( W AS >15 ms -1 ), with a larger difference at high winds [see also Bentamy et al., 2008]. Independent of wind speed, QuikSCAT wind is 11

249 250 251 252 253 254 255 256 257 258 259 stronger than ASCAT if the ASCAT mid-beam is oriented in the upwind direction ( 180 o, compare with Figure 6). Due to sampling errors, binned W are noisy at high winds (for which less data is available) and are not symmetrical in azimuth (Figure 8a). To mitigate the impact of sampling errors we represent dw by its mean and first three symmetric harmonics m 3 m dw P5 ( WAS )cos( m ), (2) m 0 where the coefficients P m ( W ) are assumed to be fifth order polynomials of ASCAT 5 AS wind speed. The polynomial coefficients in turn are estimated by least squares minimization (Table 2). To evaluate the usefulness of this approximation we compare its time mean structure to that of the time mean of W in Figure 9. This comparison shows that most 260 of the spatial structure in the time mean of W is retained in the time mean of dw. 261 Applying the correction function (2) to instantaneous ASCAT winds leads to apparent 262 decrease in the residual time mean wind speed difference (compare W W and QS AS 263 W QS W dw panels in Figure 9). The unmodeled time mean difference includes AS 264 265 266 267 268 269 positive values in the regions of the midlatitude storm tracks, which is to be expected because ASCAT underestimates strong winds (>20ms -1 ). Strong wind events are relatively rare, and the correction function dw is uncertain at those conditions (Figure 8a). Some unmodeled time mean dw is also present in coastal areas such as off Peru and Namibia, where we suspect there are sampling problems resulting from under-resolving of the prominent diurnal land-sea breezes. 12

270 After correction for our modeled dw the global histogram of W W QS W AS 271 272 273 274 275 276 277 278 eliminates the slight 0.18 ms -1 positive bias in W (Figure 9). However large negative tails of the distribution remain, reflecting the presence of large negative values of W at high latitude mentioned previously. Here we explore a physically-based hypothesis for this high latitude negative 4.3 Impact of SST W. 0 As described in the Appendix radar backscatter depends on the spectrum of the resonant Bragg waves. These near-capillary waves have amplitudes that depend on a balance of wave growth and viscous dissipation, and the latter is a function of SST 279 because of the impact of SST on viscosity. We present a linearized form of the 280 281 dependence of ΔW on SST through a Taylor series expansion of the growth-dissipation equation around the space-time mean value of SST ( T 0 =19 o C) 282 W 2[ ( T) ( T0 )]( QS AS ) (3) ( / ) C C W(1 ) a w D c 283 284 where notations and variable definitions are the same as in (A3). W is affected by deviation of local SST from T 0 through kinematic viscosity ( T ) ( T0 ), and 285 inversely by increases in wind speed with some modifications due to changes in a. 286 287 288 289 290 291 These effects are relatively small over much of the ocean but increase at high latitudes due to the increase in kinematic viscosity (Figure 10). The SST dependent bias varies inversely with wind speed and thus is further enhanced south of 60 o S due to the combination of cold SSTs and relatively weak winds. To summarize the impact of SST and winds on ΔW we present the QuikSCAT minus ASCAT collocated differences binned as a function of wind speed and SST 13

292 293 294 295 296 297 298 299 300 301 302 303 (Figure 11). QuikSCAT wind speed is higher than ASCAT at low (<3 ms -1 ) and high (>12 ms -1 ) winds. In contrast, ASCAT wind exceeds QuikSCAT wind over cold SST<10 C under moderate wind 5 ms -1 <W <10 ms -1. The strongest negative wind speed difference between the scatterometers is found at SST<5 C. In contrast to our expectations from (3) the comparison in Figure 11 indicates that the negative difference between ASCAT and QuikSCAT wind speeds rises in magnitude at moderate winds. The explanation for this behavior likely requires an improved parameterization of the windwave growth parameter and a more detailed radar imaging model (see Kudryavtsev et al., 2005 for further details). 5. Summary The end of the QuikSCAT mission in 2009 and the lack of a follow-on Ku band scatterometer has made it a high priority to ensure the compatibility of QuikSCAT and 304 the continuing C band ASCAT mission measurements. Examination of collocated 305 306 307 308 309 310 311 312 313 314 observations during a 13 month period of mission overlap November 2008-November 2009 shows that there are currently systematic differences in the 10m neutral wind estimates from the two missions. This study is an attempt to identify and model these differences in order to produce a consistent scatterometer-based wind record spanning the period 1999 present. The basic data set we use to identify the differences and develop corrections is the set of space-time collocated satellite wind observations from the two missions and triply collocated satellite-satellite-buoy observations during the 13 month period of overlap. The first part of the study explores the acceptable range of spatial and temporal separations between observations for comparison. Satellite and buoy winds show high 14

315 316 317 318 319 320 321 322 temporal correlation ( 0.8) for time lags of up to 4 hours and thus 4 hours was determined to be the maximum acceptable time separation for winds stronger than 3.5 m/s. Requiring shorter time lags severely limits the number of collocated observations at lower latitudes. Similarly, spatial separations of up to 50 km were determined to be acceptable. An examination of the collocated winds shows a high degree of agreement in direction, but reveals systematic differences in speed that depend on rain rate, the strength of the wind, and SST. The impact in wind speed of rainfall on the higher 323 frequency of the two scatterometers, QuikSCAT, is up to 1ms -1 in the tropical 324 325 326 327 328 329 330 331 332 333 334 335 336 337 convergence zones and the high precipitation zones of the midlatitude storm tracks even after rain flagged data is removed. This 1 ms -1 bias is cut by half by removing QuikSCAT wind retrievals with multidimensional rain probabilities >0.05 even though this additional rain flagging reduces data amount by only 5%. For this reason we recommend the additional rain flagging. We next examine the causes of the differences in collocated wind speed remaining after removing QuikSCAT winds with MRP>0.05. Examination of histograms shows that the differences are partly a function of ASCAT wind speed and direction relative to the mid beam azimuth, a dependence strongly suggesting bias in the ASCAT geophysical model function. This dependence is empirically modeled based on the collocation data, and is then removed. The bias model reduces collocated differences by 0.5 ms -1 in the tropics where it has large scale patterns related to variability of wind speed and direction. After applying the above two corrections, a third source of difference is evident in that QuikSCAT wind speed remains systematically lower (by 0.5 ms -1 ) than ASCAT 15

338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 wind speed at high latitudes where SST<5 o C. An examination of the physics of wind ripples implicates the SST-dependence of wave dissipation, an effect which is enhanced for shorter waves. Again, the higher frequency scatterometer, QuikSCAT, will be more sensitive to this effect. Two additional sources of differences in the collocated observations are also identified. Differences in wind retrievals occur in storm track zones due to underestimation of infrequent strong wind events (>20 ms -1 ) by ASCAT. Differences in wind speed estimates are also evident in the Peruvian and Namibian coastal zones where diurnal land-sea breezes are poorly resolved by the time delays between passes of the two scatterometers (up to 4 hours in this study). Here the differences seem to be a result of the limitations of the collocated data rather than indicating error in the scatterometer winds. Acknowledgements. This research was supported by the NASA International Ocean Vector Wind Science Team. We thank J. F. Piollé and IFREMER/CERSAT for data processing support. The authors are grateful to ECMWF, EUMETSAT, CERSAT, GODIVA, JPL, Météo-France, NDBC, O&SI SAF, PMEL, and UK MetOffice for providing buoy, numerical, and satellite data used in this study. 355 16

356 357 358 359 360 361 362 363 364 365 Appendix. Impact of SST on the Bragg scattering The purpose of this Appendix is to evaluate deviation in retrieved scatterometer wind speed due to changes in SST. This evaluation is done for the pure Bragg scattering and using simplified model of the Bragg wave spectrum. Radar backscatter, 0, is mostly affected by energy of the resonant Bragg wave component. According to Donelan and Pierson [1997] for the pure Bragg scattering 0 regime radar backscatter is a positive definite function of the spectrum of the resonant Bragg wave component whose energy depends on the balance of wind wave growth parameter w and viscous dissipation 0 that is a function of SST, F ( ). For simplicity, we next consider waves aligned with wind and use the Plant [1982] w 366 parameterization 2 w a / wc ( u* / c) where a / w is the air to water density ratio, 367 368 C =32 is an empirical constant, u* CDW is the friction velocity in the air, C D is the neutral drag coefficient that is parameterized as in Large and Yeager [2009]. This w 369 works over wide range of u * / c but fails at small values corresponding to threshold 370 371 372 373 374 375 winds [Donelan and Pierson, 1987]. It should be also corrected by a factor ( 1 c ), where c is the wind-wave coupling parameter to account for the splitting of total wind stress in the wave boundary layer into wave-induced and turbulent components [Makin and Kudryavtsev, 1999; Kudryavtsev et al., 1999) 2 / C ( u / c) (1 ), (A1) w a w * c Viscous dissipation depends on the wavenumber, k, and frequency,, of the Bragg 376 component, 2 1 4 k. It also depends on the water temperature, T, through the 17

377 378 379 380 381 kinematic viscosity, (T ), that explains temperature behavior of the wind-wave growth threshold wind speed [Donelan and Pierson, 1987, Donelan and Plant, 2009]. We also assume that scatterometer calibration, 0 ( W ), corresponds to the globally and time mean sea surface temperature T 0 =19 o C. Then an expected error in retrieved wind speed W ~ due to deviation of local temperature from T 0, T T T0, is calculated by differentiating 382 along ( T, W) const : 383 ~ / T W ( k) / W T / a a / W w w (A2) 384 385 386 387 The first term in (A2) accounts for changes in retrieved wind speed due to changes in the air density [Bourassa et al., 2010]. For chosen w this term doesn t depend on k, thus doesn t contribute to the wind difference between Ku- and C-band. Temperature dependent difference between QS and AS winds is explained by the viscous term in (A2) 388 W ~ ~ 2 ( QS AS ) W( kqs ) W( k AS ), (A3) ( / ) C C W(1 ) a w D c 389 where T ) ( T ). ( 0 390 18

391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 References Atlas, R., R. N. Hoffman, J. Ardizzone, S. M. Leidner, J. C. Jusem, D. K. Smith, and D. Gombos (2011), A Cross-calibrated, Multiplatform Ocean Surface Wind Velocity Product for Meteorological and Oceanographic Applications. Bull. Amer. Meteor. Soc., 92, 157 174. doi: 10.1175/2010BAMS2946.1 Bentamy, A., K B. Katsaros, M. Alberto, W. M. Drennan, and E. B. Forde (2002), Daily surface wind fields produced by merged satellite data. American Geophys. Union, Geophysical Monograph Series, 127, 343-349. Bentamy, A., H-L Ayina, P. Queffeulou, and D. Croize-Fillon (2007), Improved Near Real Time Surface Wind Resolution over The Mediterranean Sea. Ocean Sci., 3, 259-271. Bentamy, A., D. Croize-Fillon, and C. Perigaud (2008), Characterization of ASCAT measurements based on buoy and QuikSCAT wind vector observations, Ocean Sci., 4, 265 274. Blanke, B., S. Speich, A. Bentamy, C. Roy, and B. Sow (2005), Modeling the structure and variability of the southern Benguela upwelling using QuikSCAT wind forcing, J. Geophys. Res., 110,C07018, doi:10.1029/2004jc002529. Bourassa, M. A., D. M. Legler, J. J. O Brien, and S. R. Smith (2003), SeaWinds validation with research vessels, J. Geophys Res., 108(C2), 3019, doi:10.1029/2001jc001028. Bourassa, M. A., E. Rodriguez, and R. Gaston (2010), NASA's Ocean Vector Winds Science Team Workshops. Bull. Amer. Meteor. Soc., 91, 925 928. doi: 10.1175/2010BAMS2880.1 19

437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 Bourle`s, B., and Coauthors (2008), The Pirata Program: History, accomplishments, and future directions, Bull. Am. Meteorol. Soc., 89, 1111 1125, DOI:10.1175/2008BAMS2462.1. Donelan, M. A., and W. J. Pierson Jr. (1987), Radar scattering and equilibrium ranges in wind-generated waves with application to scatterometry, J. Geophys. Res., 92(C5), 4971 5029, doi:10.1029/jc092ic05p04971. Donelan, M. A., and W. J. Plant (2009), A threshold for wind-wave growth, J. Geophys. Res., 114, C07012, doi:10.1029/2008jc005238. Dunbar, S., and Coauthors (2006), QuikSCAT Science Data Product User's Manual Overview & Geophysical Data Products, Jet Propulsion Laboratory, ftp://podaacftp.jpl.nasa.gov/alldata/quikscat/l2b/docs/qsug_v3.pdf Ebuchi, N., H. C. Graber, and M. J. Caruso (2002), Evaluation of wind vectors observed by QuikSCAT/SeaWinds using ocean buoy data, J. Atmos. Ocean. Technol., 19, 2049-2069. Fairall, C.W., E.F. Bradley, J.E. Hare, A.A. Grachev, and J.B. Edson (2003), Bulk Parameterization of Air Sea Fluxes: Updates and Verification for the COARE Algorithm, J. Clim., 16, 571 591. Grima, N., A. Bentamy, K. Katsaros, and Y. Quilfen (1999), Sensitivity of an oceanic general circulation model forced by satellite wind stress fields, J. Geophys. Res., 104(C4), 7967-7989, doi:10.1029/1999jc900007. Grodsky, S. A., and J. A. Carton (2001), Coupled land/atmosphere interactions in the West African Monsoon, Geophys. Res. Lett., 28(8), 1503 1506, doi:10.1029/2000gl012601. Field Code Ch 20

460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 Houze, R. A., and C.-P. Cheng (1977), Radar Characteristics of Tropical Convection Observed During GATE: Mean Properties and Trends Over the Summer Season. Mon. Wea. Rev., 105, 964 980. doi: 10.1175/1520-0493(1977)105<0964:RCOTCO>2.0.CO;2 Kudryavtsev, V., V. Makin, and B. Chapron (1999), Coupled sea surface atmosphere model 2. Spectrum of short wind waves, J. Geophys. Res., 104(C4), 7625-7639, doi:10.1029/1999jc900005. Kudryavtsev, V., D. Akimov, J. Johannessen, and B. Chapron (2005), On radar imaging of current features: 1. Model and comparison with observations, J. Geophys. Res., 110, C07016, doi:10.1029/2004jc002505. Large, W. G., and S. G. Yeager (2009), The Global Climatology of an Interannually Varying Air-Sea Flux Data Set, Clim. Dyn., 33, 341-364. Makin, V., and V. Kudryavtsev (1999), Coupled sea surface atmosphere model 1. Wind over waves coupling, J. Geophys. Res., 104(C4), 7613-7623, doi:10.1029/1999jc900006 McPhaden, M. J., and Coauthors (1998), The Tropical Ocean-Global Atmosphere observing system: A decade of progress, J. Geophys. Res., 103(C7), 14,169 14,240, doi:10.1029/97jc02906 McPhaden, M. J., and Coauthors (2009), RAMA: The Research Moored Array for African Asian Australian Monsoon Analysis and Prediction. Bull. Amer. Meteor. Soc., 90, 459 480, doi: 10.1175/2008BAMS2608.1 Meindl, E.A., and G.D. Hamilton( 1992), Programs of the National Data Buoy Center, Bull. Amer. Meteor. Soc., 4, 984-993. 21

483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 Milliff, R. F., W. G. Large, J. Morzel, G. Danabasoglu, and T. M. Chin (1999), Ocean general circulation model sensitivity to forcing from scatterometer winds. J. Geophys. Res., 104(C5), 11 337 11 358, doi:10.1029/1998jc900045 Risien, C.M., and D.B. Chelton (2008), A Global Climatology of Surface Wind and Wind Stress Fields from Eight Years of QuikSCAT Scatterometer Data, J. Phys. Oceanogr., 38, 2379-2413. Rolland, J., and P. Blouch (2002), Les bouées météorologiques, Météo France Publication, La météorologie 39, available online at http://documents.irevues.inist.fr/bitstream/handle/2042/36252/meteo_2002_39_83.pd f Sobieski, P. W., C. Craeye, and L. F. Bliven (1999), Scatterometric signatures of multivariate drop impacts on fresh and salt water surfaces, Int. J. Remote Sens., 20, 2149-2166. Verspeek, J.; A. Stoffelen, M, Portabella, H. Bonekamp, C. Anderson, and J.F. Saldana (2010), Validation and Calibration of ASCAT Using CMOD5.n, IEEE Transactions on Geoscience and Remote Sensing, 48, 386-395, doi: 10.1109/TGRS.2009.2027896 Weissman, D. E., M. A. Bourassa, and J. Tongue (2002), Effects of Rain Rate and Wind Magnitude on SeaWinds Scatterometer Wind Speed Errors, J. Atmos. Oceanic Technol., 19, 738 746. doi: 10.1175/1520-0426(2002)019<0738:EORRAW>2.0.CO;2 Wright, J. W., and W. C. Keller (1971), Doppler spectra in microwave scattering from wind waves, Phys. Fluids, 14, 466-474. 22

505 506 507 Zhang, H.-M., J. J. Bates, and R. W. Reynolds (2006), Assessment of composite global sampling: Sea surface wind speed, Geophys. Res. Lett., 33, L17714, doi:10.1029/2006gl027086. 508 23

509 Table 1 Orbital parameters Satellite QuikSCAT METOP-A Recurrent period 4 days 29 days Orbital Period 101 minutes 101 minutes Equator crossing Local Sun time at Ascending node 6:30 a.m 9:30 a.m. Altitude above Equator 803 km 837 km 510 511 512 513 Inclination 98.62 o 98.59 o Table 2 Coefficients m a i of 5 m m i 5 ( WAS ) ai ( WAS ) i 0 P in (2) which parameterizes the difference between QuikSCAT and ASCAT wind speeds in terms of ASCAT wind speed ( W AS ) and wind direction relative to ASCAT mid-beam azimuth ( ). Columns represent 514 coefficients a for angular harmonics cos( m ) in (2), m=0,1,2,3. m i 515 1 cos cos 2 cos 3 Factor 516 1 1.6774 0.1427 0.0894 0.2229 x 1 W AS -0.7974 0.2091 0.2806-0.2903 x 1 2 W AS 3 W AS 4 W AS 5 W AS 1.3854-0.7235-0.6268 0.8371 x 0.1-1.1416 0.7916 0.5320-0.9592 x 0.01 0.4476-0.3579-0.1906 0.4681 x 1e-3-0.6307 0.5709 0.2322-0.8181 x 1e-5 24

517 518 519 520 521 522 523 524 Figure 1. Number of collocated QuikSCAT and ASCAT data on a 1 o x1 o grid during November 2008-2009 for time separations less than that shown in the left-upper corner. 25

525 526 527 528 529 Figure 2. Difference between time mean satellite and buoy 10m neutral wind speed (ms -1 ) for QuikSCAT (QS,top) and ASCAT (AS,bottom). 26

530 531 532 533 534 Figure 3.Scatter diagram of satellite and buoy 10m neutral wind speed for ASCAT (AS) and QuikSCAT (QS).Symbols represent time mean collocated winds at buoy locations. 27

535 536 537 538 539 540 Figure 4. Lagged correlation of collocated satellite and buoy winds as a function of time separation between satellite and buoy data. Solid line shows the median value and vertical bars are bounded by 25 and 75 percentiles of correlation for different buoys. 28

541 542 543 544 545 546 547 548 Figure 5. Mean difference between collocated QuikSCAT and ASCAT wind speed ( W ), their standard deviation (STD), and temporal correlation (CORR). (Left) only rain flag is applied to QuikSCAT, (right) rain flag and multidimensional rain probability (MPR<0.05) are both applied. 29

549 550 551 552 553 554 555 556 Figure 6. Difference between collocated QuikSCAT (QS) and ASCAT (AS) wind speed as function of ASCAT wind direction relative to the mid beam azimuth (DIR ASC - AZIM1). Data is averaged for all wind speeds and binned into 10 o in the relative wind direction. Downwind direction corresponds to 0 o. Wind direction histogram is shown in gray against the right axis. 30

557 558 559 560 561 562 Figure 7. The same as in Figure 6, but for buoy minus ASCAT collocated wind speed difference. 31

563 564 565 566 567 568 Figure 8. Collocated QuikSCAT-ASCAT neutral wind speed difference (m/s) binned into 1m/s by 10 o bins in ASCAT wind speed and wind direction relative to the ASCAT mid- beam azimuth DIR AS AZIM 1. (a) Binned data, evaluated with data from 55S -55N to exclude high latitude areas of negative W (see Figure 5). (b) data fit by equation (2). 32

569 570 571 572 573 574 575 Figure 9. (top panel) Wind speed difference ( W W ). QuikSCAT rain flag and MRP<0.05 are applied (the same as in Figure5). (2nd panel) Wind speed difference after applying the correction function dw (, ). (3rd panel) Spatial distribution of time W AS mean dw. (lower panel) Histogram of wind speed difference before (gray bar) and after (empty bar) applying of dw. QS AS 33

576 577 578 579 580 581 582 Figure 10. (top panel) Time mean inverse wind speed from ASCAT, 1 W AS. (2 nd panel) Difference in kinematic viscosity (m 2 s -1 ) of the sea water relative to the viscosity evaluated at global mean SST (19 o C). (third panel) QuikSCAT-ASCAT wind speed difference after applying the correction function dw (the same as in Figure 9). (bottom panel) SST dependent difference between QuikSCAT and ASCAT wind speed W (T) based on (Eq. 3) at =45 o. 34

583 584 585 586 587 588 589 590 Figure 11. Collocated QuikSCAT-ASCAT neutral wind speed difference (ms -1 ) binned 1 o in SST and 1m/s in wind speed from the ECMWF analysis collocated data. There are on average 128,000 collocations in each bin. Only wind speed difference based on more than 2,000 collocations is shown. 35