Correction of the Effect of Relative Wind Direction on Wind Speed Derived by Advanced Microwave Scanning Radiometer

Transcription

1 Journal of Oceanography, Vol., pp. 39 to, Correction of the Effect of Relative Wind Direction on Wind Speed Derived by Advanced Microwave Scanning Radiometer MASANORI KONDA 1 *, AKIRA SHIBATA, NAOTO EBUCHI 3 and KOHEI ARAI 1 Department of Geophysics, Graduate School of Science, Kyoto University, Kyoto -8, Japan Earth Observation Research and Application Center, Japan Aerospace Exploration Agency, Tokyo -3, Japan 3 Institute of Low Temperature Science, Hokkaido University, Sapporo -819, Japan Department of Information Science, Faculty of Science and Engineering, Saga University, Saga 8-8, Japan (Received August ; in revised form 7 February ; accepted 7 February ) The sea surface wind speed (SSWS) derived by a microwave radiometer can be contaminated by changes of the brightness temperature owing to the angle between the sensor azimuth and the wind direction (Relative Wind Direction effect: RWD effect). We attempt to apply the method proposed by Konda and Shibata () to the SSWS derived by Advanced Microwave Scanning Radiometer (AMSR) on Advanced Earth Observing Satellite II (ADEOS-II), in order to correct for the RWD effect. The improvement of accuracy of the SSWS estimation amounts to roughly % of the error caused by the RWD effect. Comparison with in situ observation at the Tropical Atmosphere Ocean (TAO) array shows that the root mean square error of the corrected SSWS is 1.1 ms 1. It is found that the impact of the RWD effect on the estimation of the latent heat flux can amount to about 3 Wm on average. We applied the method to the SSWS derived by AMSR for Earth Observing System (AMSR-E) and obtained a similar result. Keywords: ADEOS-II, AMSR, AMSR-E, wind speed, relative wind direction. 1. Introduction Microwave radiometers contribute to monitoring the sea surface wind speed (SSWS) and its variability. For example, the SSWS derived by the Special Sensor Microwave Imager (SSM/I) on the Defense Meteorological Satellite Program (DMSP) satellite is used by many climate studies and for building the sea surface turbulent heat flux data set (Kubota et al., ; Chou et al., 3, ). In general, the microwave radiometer has the merit of simultaneously measuring other geophysical parameters such as water vapor content and sea surface temperature (SST) without time lag. The simultaneous observation of these parameters enables us to obtain information about various phenomena, such as the evaporative cooling of the sea surface. However, there are still some problems in algorithms for the estimation of the SSWS as well as the other geo- * Corresponding author. konda@kugi.kyoto-u.ac.jp Copyright The Oceanographic Society of Japan/TERRAPUB/Springer physical parameters. Several studies (Wentz, 199; Yueh et al., 1999; Meissner and Wentz, ) have pointed out that the brightness temperature (BT) can change according to the angle between the sensor azimuth and the insitu wind direction (, hereafter RWD). The change of the BT due to the RWD allows some errors to creep into the SSWS retrieval. The error approximately amounts to 3 ms 1 (Wentz, 1997; Konda and Shibata, ). Wentz (1997) was the first to take this effect (RWD effect) into account in the operational SSM/I wind retrieval algorithm, where an error curve derived from a harmonic analysis was assumed. However, the error of his analysis is not evaluated well, possibly because of the difficulty in obtaining the footprint of the SSM/I collocated with the ocean vector winds (OVW) under various climatic conditions. Successful operations of microwave scatterometers show a marked improvement in the estimation of the OVW (Liu et al., a, b). Ebuchi et al. () and Freilich and Vanhoff (3) have reported high accuracy of the OVW derived by SeaWinds on QuikSCAT. The 39

2 errors of the wind speed and the wind direction are respectively within 1.1 ms 1 and 3, compared with in situ observation at ocean buoys (Ebuchi et al., ). Recently, Konda and Shibata () attempted to evaluate the change of the BT due to the RWD effect, by using SSM/I-derived SSWS collocated with wind vector cells derived by a microwave radar scatterometer, SeaWinds on QuikSCAT. They proposed an experimental approach to correcting the RWD effect without knowing the information of the wind direction. The approach, using an experimental look-up table, generally succeeded in reducing the error caused by the RWD effect. However, the data used in their study were obtained under limited environmental conditions, because of the orbital difference between SST/I and QuikSCAT. Since the Advanced Earth Observing Satellite II (ADEOS-II) carried the SeaWinds scatterometer as well as the Advanced Microwave Scanning Radiometer (AMSR), many collocations for various climatic zones were derived, which enables us to extract the RWD effect experimentally. The simultaneous observations of AMSR and SeaWinds are helpful in preparing a matchup data set, covering various kinds of climatic zones and geographical locations, in order to create a look-up table. The AMSR for the Earth Observing System (EOS) Aqua Satellite (AMSR-E) is a slightly modified AMSR sensor and the AMSR-E-derived SSWS also suffers from the RWD effect. It is thus expected that the experimental correction established by the ADEOS-II observations can also be applied to the correction of the AMSR-E-derived SSWS. We describe first the data and the method used in this study in Section. The results of the application of the approach proposed by Konda and Shibata () to the AMSR observation are described in Section 3. An example of the application of the AMSR-E wind speed correction will be shown in Section. The impact of the correction of the RWD effect on estimation of the latent heat flux is evaluated in the final part.. Data and Method We retrieved the SSWS from the measured BTs provided by Japan Aerospace Exploration Agency (JAXA) in a form of the AMSR level 1B data set. Owing to the limited lifetime of ADEOS-II, we used the observations from April to September 3, when both AMSR and SeaWinds data were available. Level 1B data consists of the calibrated AMSR BTs, the footprints of which are on the swath along the orbit. The wind speed is retrieved basically according to the AMSR standard algorithm, which is the modified algorithm of Shibata (199) (Shibata,, see products/pdf/alg_des.pdf). As AMSR does not detect the wind direction, we use the level B of the OVW data derived by SeaWinds on ADEOS-II. The SeaWinds OVW data are available online from the Jet Propulsion Laboratory (JPL) Physical Oceanography Distributed Active Archive Center (PO.DAAC). The wind speed values derived by AMSR are compared with the in situ observations from April to September 3 at the Tropical Atmosphere and Ocean project (TAO) array in the tropical Pacific (McPhaden, 199). We present an example of the application of this method improving the accuracy of the AMSR-E-derived wind speed and evaluate the effect of the correction. AMSR-E data used in this study were obtained in April 3. The SSWS and the impact of the RWD effect are evaluated using the AMSR-E level 1B BTs data and level B geophysical parameters. The AMSR-E-derived SSWS values are validated by the level B product of the SeaWinds on QuikSCAT OVW derived from PO.DAAC. First, we retrieve the raw AMSR SSWS without the RWD correction (hereafter RAWS) by removing the module for the correction of the RWD effect from the AMSR standard algorithm. Next, we apply the method of Konda and Shibata () and evaluate how much the correction is done. The method for correcting the RWD effect on the BTs is described in detail in Konda and Shibata (). In this study, according to Konda and Shibata (), we extract the relationship between the BTs with the vertical (BT3 V ) and the horizontal polarization (BT3 H ) at 3. GHz. As the incident angle of AMSR is fixed at, the vertical component has a weak dependence on the surface roughness (Stogryn, 197; Fung and Eom, 198). We define the index representing the partial shift of the BT3 H caused by the SSWS (PS3). PS3 is determined as the deviation from a linear relationship between the BT3 V and BT3 H under the windless condition (SSWS = ) at each realistic SST value (BBT3 SSWS= ) (see figure of Konda and Shibata ()). PS3 is then obtained from the BT and the level B product of the AMSR-derived SST. PS 3 = BT 3 BBT 3 () 1 H SSWS=. SST In preparation for the correction procedure, we experimentally established a look-up table correlating PS3 with SST, integrated water vapor (IWV), cloud liquid water (CLW), SSWS, and RWD. The value of PS3 in an element in the table is obtained by averaging many collocations of these variables. PS3 PS3 SST, IWV, CLW, SSWS, RWD, T = ( ) ( ) where PS3 T indicates the value in the look-up table. We use the AMSR level B geophysical parameter standard products for SST, IWV, and CLW. The index for 39 M. Konda et al.

3 (a) (b) AMSR U (ms --1 ) SeaWinds SSW - AMSR U (ms -1 ) - - SeaWinds SSWS (ms -1 ) - - Relative wind direction Fig. 1. (a) Density distribution of the relationship between the SSWS derived by SeaWinds and U. The density is expressed by contours, the values of which are superimposed on the individual lines. (b) Mean difference between the SSWS derived by SeaWinds and U is plotted against the RWD. The vertical line indicates the standard deviation of the respective mean values. the SWSS in the look-up table should be independent of the SeaWinds wind speed, considering the application to the AMSR-E-derived SSWS. Therefore, we should not refer to the SeaWinds wind speed in the process of correcting the RWD effect. The RAWS should not be used as the index, either. If we categorized them by using the RAWS, all of the PS3 values collected in each group in the look-up table would have almost the same value. As a result, we would not be able to extract the dependence of PS3 T on the RWD from the look-up table by using the RAWS for the index of the SSWS. We should thus use another index representing the SSWS for the look-up table. According to Hollinger (1971), the wind speed dependence of the horizontal component of the sea surface emissivity decreases with decreasing frequency. In contrast, the vertical component of the emissivity is almost insensitive to the surface roughness at an incident angle of (Stogryn, 197; Fung and Eom, 198). These tendencies can change the relationship between the horizontal and vertical components at different frequencies. In general, it is expected that the BTs are less sensitive to and correlated with the SSWS at lower frequencies than those at 3 GHz. However, the dependence on the RWD should be weaker due to the weaker sensitivity to the surface roughness at the lower frequencies. It is expected that the SSWS values derived from the BTs at the lower frequency are lower in accuracy but higher in RWD dependence than that at 3 GHz. Therefore, we define another index for the SSWS by applying an algorithm the same as that used to derive RAWS to the BTs at the lowest frequency ( GHz). Figure 1(a) shows a comparison of this index (hereafter U) with the collocated SSWS derived by SeaWinds. It indicates that the U is well correlated with the wind speed but has a small bias and a relatively large standard deviation of as much as ± ms 1. These inaccuracies might be attributed to the low spatial resolution at this frequency, in addition to the low sensitivity to the sea surface roughness. Panel (b) in Fig. 1 displays the difference between the U and the SeaWinds-derived wind speed and its dependence on the RWD. It suggests that the average dependence of the U on the RWD is not strong. The approach using the BTs at GHz shows a similar result (not shown). For the purpose of obtaining an alternative index for the wind speed that has a different RWD dependence from RAWS derived from the BTs at 3 GHz, we use the lowest frequency ( GHz) for the index of the SSWS to create the look-up table. The RWD is computed using the ADEOS-II level B swath data for SeaWinds OVW. There is a small time lag between observations of a footprint by AMSR and SeaWinds, owing to the difference in their scanning systems. The lag is usually within 18 seconds. A spatial difference of up to 1. km is allowed as the collocation of the observation cells. The look-up table obtained by the near-simultaneous observation of AMSR and SeaWinds on ADEOS-II is used to correct the BT3 H. The change of the PS3 in the ranges of geophysical parameters of SST, IWV, CLW and U should express the RWD effect appearing in the BT3 H. PS3 PS3 RWD. 3 = ( ) ( ) TR T SST, IWV, CLW, U Correction of the Relative Wind Direction Effect on the AMSR Wind 397

4 Figure shows examples of the look-up table in ranges of the geophysical parameters, which are typical of the tropical, mid-latitude, and cold climatic zones. The value of the PS3 clearly depends on the RWD in each range of geophysical parameters. The most appropriate value of PS3 TR in the table is determined by comparing the PS3 TR obtained from the look-up table with the PS3 directly obtained by Eq. (1). The difference between the selected PS3 TR and that for the crosswind measurement is considered as the RWD effect on the BT3 H. We reduce the value from the raw BT3 H and compute the wind speed again, using the corrected BT3 H. We used version 1 of the AMSR data in this study. There might be a slight difference between the BTs of version 1 and that of the later version. In the same way, the version of the standard algorithm is also corrected now. It is possible that the algorithm used in this study is slightly modified. However, using the earlier version should adequately present an alternative way to correct the RWD effect, as the basic design of the algorithm is not changed in the later version. (a-1) (a-) (a-3) PS3 (K) - - (b-1) (b-) (b-3) - PS3 (K) (c-1) (c-) (c-3) PS3 (K) (d-1) (d-) (d-3) PS3 (K) Fig.. Values of the PS3 at some typical ranges of (top) cold, (second row) mid-latitude, (third row) tropical climate without cloudiness, and (bottom) tropical climate with cloudiness. The ranges of the environmental parameters in the respective regimes (SST, IWV, CLW) are respectively, ( 7 C, mm,. Kgm ), (17 19 C, 1 mm,. Kgm ), ( 7 C, 1 mm,. Kgm ) and ( 7 C, 3 mm,..8 Kgm ). The index of the SSWS (U) increases from (left) 3 ms 1, (center) 7 ms 1, to (right) 9 ms 1. The values are plotted against the RWD. The vertical line in the figure indicates the standard deviation of each value of the PS M. Konda et al.

5 PS3 (K) PS3 (K) (a) - (c) - PS3 (K) PS3 (K) (b) - (d) - Fig. 3. PS3 TR against RWD in the selected ranges. The ranges of SST, IWV and CLW are (a) 1 3 C, 1 mm,. Kgm, (b) 1 3 C, 1 mm,. Kgm, (c) 1 3 C, 1 mm,.1.1 Kgm, and (d) 17 C, 1 mm,. Kgm. 3. Correction of the RWD Effect on the SSWS Derived by AMSR As indicated in Fig., the value of the PS3 depends on the RWD. The dependence is stronger under high wind speeds than under low ones. Under high wind speeds, the difference of PS3 TR between the upwind and the downwind condition becomes up to K, as shown in panels (a- 3, b-3, c-3 and d-3) in Fig.. This is caused by the inhomogeneous distribution of the slope of ocean facets (Cox and Munk, 19a, b; Ebuchi and Kizu, ). The dependence of the PS3 changes in proportion to the climate condition. For example, the RWD dependence of the PS3 is generally stronger in cold conditions (a-3) than in the tropical ones (c-3 and d-3). As individual geophysical parameters change correlatively in proportion to the climate condition, the low temperature and the low humidity should together contribute to the value of the PS3 TR. The dependence of the PS3 TR on the individual parameters is discussed by comparing the PS3 TR in the selected ranges, which are in the same regime except for only one of the geophysical parameters. Figure 3 shows the RWD dependences of the PS3 TR in ranges where the U values are from 9 to ms 1, and the range of only one of the other geophysical parameters is changed. Al- - (a-1) (a-) (a-3) SeaWinds wind average.1 ms SeaWinds wind average.9 ms -1 SeaWinds wind average 13.3 ms (b-1) (b-) (b-3) SeaWinds wind average 3. ms SeaWinds wind average 8.1 ms SeaWinds wind average 9.8 ms Fig.. Difference between the AMSR-derived SSWS and the collocated SeaWinds SSWS under (a) cold and (b) tropical climatic conditions. The thick line and diamonds indicate the SSWS, which undergoes the correction of the RWD effect. The broken line indicates the RAWS without the correction. The ranges of the physical parameters in panels (a) and (b) are respectively the same as the cold and tropical climatic conditions in Fig., respectively. The U increases from the left panel to the right, i.e. (1) 3 ms 1, () 7 ms 1 and (3) 9 ms 1. The mean SeaWinds SSWS in the individual range is also shown in each panel. The vertical line indicates the standard deviation of the difference between AMSR and SeaWinds SSWS. Correction of the Relative Wind Direction Effect on the AMSR Wind 399

6 Table 1. RWD dependence of the estimation error of the SSWS in four typical climate conditions. The results tabulated here are those where the wind speed index U is from 9. to. ms 1. The dependence is evaluated by the difference between the maximum and the minimum value of the estimation errors in a range, where the parameters (SST, CLW, IWV, U) other than the RWD are fixed. The results with and without the correction of the RWD effect are indicated in cases using AMSR and AMSR-E. The SeaWinds wind speed is referred to as the truth data. Cold condition (a) Mid-latitude condition (b) Tropical condition (c) Cold-cloud condition (d) (maximum minimum)/standard deviation 9. < U <. AMSR Raw wind.9/1.7.87/1.79./1. 3.3/1.31 With correction./. 1.9/.3 1./.7./.7 AMSR-E Raw wind.8/1.3 / /.33/1. With correction 1.3/.3 / / 3.1/1. (a) SST: 7 C, IWV: mm, CLW:. Kgm 1. (b) SST: C, IWV: 1 mm, CLW:. Kgm 1. (c) SST: 7 C, IWV: 1 mm, CLW:. Kgm 1. (d) SST: 3 C, IWV: 13 1 mm, CLW:.1.1 Kgm 1. though these parameters do not always influence the PS3 in a linear manner, Fig. 3 suggests that the low SST and high IWV conditions tend to strengthen the dependence on the RWD, whereas the high CLW conditions hardly change it. The maximum value of the PS3 TR in panel (d) is higher than in the panel (a) by about 1 K. Now that we have obtained the look-up table, we can correct the AMSR-derived SSWS. Note that we do not use any information on the wind direction for the correction. We apply the RWD correction scheme to the AMSR BTs. Figure shows the difference between the AMSRderived SSWS with the RWD correction and the SeaWinds-derived SSWS under cloud-free conditions in cold and tropical climatic zones. Mean differences in the respective ranges of the wind speed are plotted against the RWD. Mean wind speed derived by SeaWinds in each range is shown in individual panels. The result is evaluated by the simultaneously collocated SeaWinds observations. The error of the RAWS without the RWD correction is superimposed by the broken line. The contrast of these values seen in Fig. clearly indicates the significant reduction of the systematic error caused by the RWD effect. As one can easily see from Figs. and 3, Fig. shows that the RWD effect is stronger in high wind conditions than in low ones. The correction works significantly in high wind conditions (panels (a-), (a-3), (b-) and (b- 3)). In contrast, the correction hardly works in the low wind conditions, as the RWD effect itself is not significant, compared with the standard deviation (panels (a-1) and (b-1)). These features are common both in the cold and tropical conditions. In contrast, as was suggested by Fig. 3, the difference between the broken line in panel (a- 3) and that in (b-3) shows that the RWD effect becomes stronger in cold conditions. The RWD effect on the wind speed is evaluated by the difference between the maximum and the minimum error. Table 1 shows the result of application of the method according to Konda and Shibata () together with RAWS in the range of U from 9 to ms 1. The strong RWD dependence of the RAWS, which is more than. ms 1, shows that the RWD effect on the estimation is serious in this range of U. Note that the realistic wind speed tends to be larger than the index U (Fig. 1). The improvement indicated in Table 1 shows that as much as % of the RWD effect is reduced by the correction used in this study. The climatic condition does not seem to affect the result, although the error reduction is slightly larger in tropical conditions. Table 1 also shows an example of the improvement in cloudy conditions; the error is reduced although the ratio of improvement is smaller than in the cloud-free condition. The improvement of the accuracy due to the RWD correction is also found in a comparison with in situ measurements of the SSWS. The corrected AMSR-derived SSWS is compared with the in situ observation at TAO arrays from April to September, 3. As observation at the TAO array is recorded every minutes, the comparison is made using the closest record in the TAO array. A spatial difference of km between the buoy location and the footprint of AMSR is permitted. The value of the root mean square (RMS) of the errors is 1.1 ms 1 from 9 collocations. This value is smaller than that of RAWS (1.3 ms 1 ). M. Konda et al.

7 AMSR-E - QuikSCAT - (a) (b) (c) AMSR-E - QuikSCAT - AMSR-E - QuikSCAT - - QuikSCAT wind average.9 (ms -1 ) QuikSCAT wind average 8.7 (ms -1 ) QuikSCAT wind average 8.3 (ms -1 ) - - Fig.. Difference of the AMSR-E-derived SSWS from the collocated SSWS derived by SeaWinds on QuikSCAT. The ranges of the physical parameters of the SST, IWV and CLW are 7 C, 3 mm,. Kgm. The range of U increases from (a) ms 1, (b) 7 ms 1 and (c) 9 ms 1. The other conventions in the figure are the same as in Fig... Application to the Wind Speed Derived by AMSR-E We try to apply the method presented in the previous section to the estimation of the SSWS by AMSR-E. AMSR-E is on EOS-Aqua launched in May. As the radiometer frequencies of AMSR-E are the same as AMSR except for.3 and.8 GHz, almost the same method can be used. The dependence and variability are almost the same as those in the PS3 of AMSR. As EOS-Aqua does not carry a scatterometer, the evaluation of the correction is done by the match-up data set of AMSR-E and SeaWinds on QuikSCAT. Owing to the orbital difference between these satellites, the geographical location of the collocated data is limited only in latitudes higher than in both hemispheres (not shown). Figure shows an example of the difference between the SSWS of AMSR-E and SeaWinds on QuikSCAT in April 3. Panels in Fig. show the results of the correction of the RWD effect on the AMSR- E SSWS at three levels of the wind speed index of U. The evaluation is made only in cold conditions because the match-up data exist only in high latitudes. As the number of the match-up data is small, the range of the IWV displayed in Fig. is expanded to 3 mm. The RWD dependence of the RAWS is quite large, as was seen in the case of AMSR. The large standard deviation is noticeable. The SSWS with the correction shows a smaller dependence, which indicates that the correction works significantly. The difference between the maximum and the minimum error in the range of U from 9 to ms 1 is also shown in Table 1 as an indicator of the RWD effect, in the same manner as for the AMSR-derived SSWS. The values in Table 1 show that the influence of the RWD effect is noticeable in the RAWS and that this correction can improve the accuracy to the almost same level as the AMSR-derived SSWS. However, the correction under cloudy conditions is worse than that of the AMSR. The decrease of the difference between the maximum and the minimum error (from.3 to 3. ms 1 ) AMSR-E - estimation of the wind direction QuikSCAT Fig.. Example of the density distribution of the relationship between the RWD derived by the SeaWinds OVW and rough estimation of the RWD obtained through the correction of the RWD effect on the AMSR-E-derived SSWS. An example using the data in April 3 is shown. indicates that only 3 to % of the RWD effect is reduced by the correction. Further analysis will be needed using the data for a longer period and using other in situ observations. As we mentioned in our description of the method, the approach used in this study makes a rough estimation of the RWD through the process of the correction. This secondary product can be an indicator of whether the correction succeeds or not. As this approach cannot determine which side of the orbit the RWD is in, the sign of the RWD obtained does not make sense. Figure shows 3 3 Correction of the Relative Wind Direction Effect on the AMSR Wind 1

8 (a) AMSR latent heat flux (Wm - ) (b) AMSR latent heat flux (Wm - ) With the RWD correction 3 With the RWD correction 3 low wind ( - ms -1 ) 3 Without the RWD correction high wind ( - ms-1) 3 Without the RWD correction Fig. 7. Density distribution of the relationship between the instantaneous latent heat fluxes derived by using the AMSR-derived SSWS with and without the RWD correction. The distributions of the data, whose SSWS with the correction is (a) low ( ms 1 ) and (b) high ( ms 1 ), are indicated by using the one month observation in August 3. a density distribution of the relation between the RWD derived by SeaWinds on QuikSCAT and the RWD obtained by this method. The markedly dense distribution reaches out from the corners towards the center of the figure until the RWD approaches about. Figure suggests that the correction can detect the upwind condition of the sensor as well as the crosswind condition. The sparse distribution over the vertical zone, where the RWD derived by QuikSCAT is near, should indicate that cases of the downwind condition (RWD = ) are few in this data. On the other hand, the sparse distribution near the middle of each axis indicates that the mistaken choice between the upwind and the downwind case is rare.. Discussion and Conclusions We attempted to apply the method proposed by Konda and Shibata () for the correction of the RWD effect on the SSWS retrieved by AMSR. The RWD was computed by the angle between the sensor azimuth angle of AMSR and the collocated OVW observed by SeaWinds on ADEOS-II. We computed the PS3 as the index of the partial change of the BTs with horizontal polarization at 3 GHz. We experimentally made up a table of the PS3 (PS3 T ) in terms of geophysical parameters of SST, IWV, CLW, U, and RWD, in advance of the correction. The SST, IWV, and CLW are the standard products of AMSR. U, which is obtained by the BTs at GHz, is used for the index of the SSWS instead of the AMSR-derived RAWS or the SeaWinds-derived SSWS. The reason for using U is that the index for the SSWS in the look-up table should have a weak dependence on the RWD. As the horizontally polarized BTs generally show a marked decrease in ocean surface roughness with decreasing frequency, the BTs at the lowest AMSR frequency should have a weaker sensitivity to the RWD than BT3 H. From the point of view that the index for the SSWS in the look-up table should have a RWD dependence that differs from PS3 as much as possible, we use the lowest frequency. As AMSR and AMSR-E both have another low-frequency channel at GHz, the index derived by BTs at GHz (U) might be useful. The RWD dependence of U is slightly worse than that of U (not shown), whereas the spatial resolution of BT at GHz is better than that at GHz. Further analysis will be needed to determine which index is more appropriate. This is to be addressed in our next study. In the correction process, we can determine the PS3 from the BT3 H and the BT3 V. This value is independent of those used for making up the look-up table. On the other hand, values of the PS3 T are obtained from the look-up table in the range of the SST, IWV, CLW and U at individual footprints (PS3 TR ). The PS3 TR is then given as the function of the RWD. A PS3 TR nearest to the PS3 obtained from the BTs is chosen as the most appropriate value, according to Konda and Shibata (). The difference between the PS3 TR in the cross-wind condition and the selected PS3 TR is reduced from the raw BT3 H in order to obtain the corrected BT3 H. The corrected SWSS is obtained by computing with use of the corrected BT3 H. We validated how much the RWD effect is corrected in the selected ranges of the much-up data set with the SeaWinds OVW. As much as % of the inaccuracy caused by the RWD effect is reduced both in the SSWSs M. Konda et al.

9 derived by both AMSR and AMSR-E, as shown in Table 1. Results in the different climatic regimes do not change so much. As was pointed out by Konda and Shibata (), the RWD effect is strong in the high wind regime. The correction is successful under cloudy conditions as well as cloud-free ones. The corrected SSWS derived by AMSR is compared with the in situ measurement at TAO arrays from April to September in 3. The RMS of the errors is 1.1 ms 1 from 9 collocations. The uncertainty of the SSWS estimation should have a serious impact on the computation of the latent heat flux, because the role of the SSWS for the latent heat flux shows a marked increase with increasing SSWS (Fairall et al., 199). Therefore, the instantaneous latent heat flux derived by the microwave radiometer suffers from the RWD effect, especially at a large value. The correction of the RWD effect on the SSWS should contribute to improvements in monitoring the global distribution and change of the air-sea thermal exchange. We show the impact of the correction of the RWD effect on the instantaneous estimation of the latent heat flux by contrasting the latent heat fluxes using the SSWS with and without the correction of the RWD effect. As AMSR observes the SST, IWV, and SSWS without a time lag, the estimation of the instantaneous latent heat flux at the footprints of AMSR is possible using a bulk formula, ( ) ( ) Q= Lρ C C q q U, p e s a where L indicates the latent heat of evaporation, ρ, the density of the atmosphere, C p, the isobaric specific heat, C e, the bulk coefficient, q, the specific humidity, and U, wind speed at a height of m. The surface level specific humidity is related to the mixing ratio, which is estimated from the IWV by an experimental equation proposed by Liu and Niiler (198). The TOGA-COARE version.b bulk coefficient for the latent heat (Fairall et al., 3) is computed with some assumptions about the typical oceanic atmosphere. Panels of Fig. 7 shows an example of the density distribution of the relationship between the latent heat flux with and without the correction of the RWD effect during August 3. Results using data from the other months are almost the same. Panels (a) and (b) respectively show the relationships in low and high wind conditions. The difference is evidently enlarged more in the high wind speed condition than in low wind. The contrast between these conditions indicates that the contamination by the RWD effect grows in the high wind condition, and therefore in the large latent heat flux. The root mean square of the impact of the correction for the latent heat flux is determined by the difference from the no correction line. The average values in the low and high wind conditions are 11.1 Wm and 8. Wm, respectively. Looking closely at the high wind speed area in the figure, one sees that the difference of the latent heat flux is not evenly distributed across the diagonal line indicating no correction. This might be attributable to the discrete correction of the SSWS by the look-up table. The results of this study indicate that the method proposed by Konda and Shibata () can be successfully applied to the correction of the systematic error embedded in the SSWS derived by AMSR and AMSR-E. However, some room for improvement remains. The RWD dependence of PS3 is at most K (Fig. ), while that of the BT is as small as 3 K at horizontal polarization. Calibration errors of the BT and their temporal drift can influence the accuracy of the look-up table. Continuous calibration of AMSR-E will be important in maintaining stable measurements. Another major possibility is further improvement of the standard products other than the wind speed, as they are needed for the method in this study for establishing and consulting the look-up table. In particular, the SST and the wind speed are obtained iteratively in the standard AMSR algorithm. The value of PS3 might be affected by the accuracy of the SST. Another major problem in this approach is the relationship between the AMSR PS3 and AMSR-E PS3. In this study, we applied the same approach by adding a small bias (. K) generated probably due to the scaling down of the antenna size. It will be necessary to evaluate the relationship in detail in a future study. Acknowledgements Level 1B brightness temperature and Level geophysical parameters of AMSR and AMSR-E were compiled and provided by Earth Observation Research and application Center, JAXA. SeaWinds on QuikSCAT Level B ocean wind vectors in km swath grid of SeaWinds on ADEOS-II (JPL SeaWinds Project) were supplied by the NASA Jet Propulsion Laboratory, PO.DAAC. We are grateful to Dr. M. J. McPhaden, who is the director of TAO project office, Pacific Marine Environmental Laboratory, NOAA, from which TAO temporal high resolution data were obtained. This study is supported by the Joint Research Program with JAXA Study of the improvement of the ocean surface wind retrieval algorithm by using the comparison of the collocated brightness temperature and sigma- and Study of the relationship between the and brightness temperatures derived by AMSR and AMSR-E. References Chou, S.-H., E. Nelkin, J. Ardizzone, R. M. Atlas and C.-L. Shie (3): Surface turbulent heat and momentum fluxes over global oceans based on the Goddard Satellite retrievals, version (GSSTF). J. Climate, 1, Chou, S.-H., E. Nelkin, J. Ardizzone and R. M. Atlas (): A comparison of latent heat fluxes over global oceans for four Correction of the Relative Wind Direction Effect on the AMSR Wind 3

10 flux products. J. Climate, 17, Cox, C. and W. Munk (19a): Statistics of the sea surface derived from sun glitter. J. Mar. Res., 13, Cox, C. and W. Munk (19b): Measurements of the roughness of the sea surface from photographs of the sun s glitter. J. Opt. Soc. Am.,, Ebuchi, N. and S. Kizu (): Probability distribution of surface wave slope derived using sun glitter images from Geostationary Meteorological Satellite and surface vector winds from scatterometers. J. Oceanogr., 8, Ebuchi, N., H. C. Graber and M. J. Caruso (): Evaluation of wind vectors observed by QuikSCAT/SeaWinds using ocean buoy data. J. Atmos. Oceanic Technol., 19, 9. Fairall, C. W., E. F. Bradley, D. P. Rogers, J. B. Edson and G. S. Young (199): Bulk parameterizations of air-sea fluxes in TOGA COARE. J. Geophys. Res., 1, Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev and J. B. Dobson (3): Bulk parameterization of air-sea fluxes: Updates and verification for the COARE algorithm. J. Climate, 1, Freilich, M. H. and B. A. Vanhoff (3): The relationship between winds, surface roughness, and radar backscatter at low incidence angles from TRMM Precipitation Radar measurements. J. Atmos. Oceanic Technol.,, 9. Fung, A. K. and H. J. Eom (198): A reevaluation of Stogryn s apparent temperature theory over the sea surface. IEEE Trans. Antennas Propagat., 3, 1 3. Hollinger, J. P. (1971): Passive microwave measurements of sea surface roughness. IEEE Tans. Geosci, Electron., 9, 19. Konda, M. and A. Shibata (): An experimental approach to correct the microwave radiometer wind speed affected by the change of the brightness temperature due to the relative wind direction. Geophys. Res. Lett., 31, L93, doi:.9/gl1987. Kubota, M., N. Iwasaka, A. Kizu, M. Konda and K. Kutsuwada (): Introduction of Japanese ocean flux data sets with use of remote sensing observations (J-OFURO). J. Oceanogr., 8, 13. Liu, W. T. and P. P. Niiler (198): Determination of monthly mean humidity in the atmospheric surface layer over oceans from satellite data. J. Phys. Oceanogr., 1, Liu, W. T., X. Xie, P. S. Polito, S.-P. Xie and H. Hashizume (a): Atmospheric manifestation of tropical instability wave observed by QuikSCAT and Tropical Rain Measuring Mission. Geophys. Res. Lett., 7, 8. Liu, W. T., H. Hu and S. Yueh (b): Interplay between wind and rain observed in Hurricane Floyd. EOS Trans., 81, 3. McPhaden, M. J. (199): The Tropical Atmosphere Ocean (TAO) array is completed. Bull. American Meteorological Society, 7, Meissner, T. and F. Wentz (): An updated analysis of the ocean surface wind direction signal in passive microwave brightness temperatures. IEEE Trans. Geosci. Remote Sens.,, Shibata, A. (199): Determination of water vapor and liquid water content by an iterative method. Meteorol. Atmos. Phy.,, Shibata, A. (): AMSR/AMSR-E sea surface wind speed algorithm. EORC Bulletin/Technical Report Special Issue on AMSR Retrieval Algorithms, 9,. Stogryn, A. (197): The apparent temperature of the sea at microwave frequencies. IEEE Trans. Antennas Propagat.,, Wentz, F. J. (199): Measurement of oceanic wind vector using satellite microwave radiometers. IEEE Trans. Geo. Remote Sens., 3, Wentz, F. J. (1997): A well-calibrated ocean algorithm for SSM/ I. J. Geophys. Res.,, Yueh, S. H., W. J. Wilson, S. J. Dinardo and F. F. Li (1999): Polarimetric microwave brightness signatures of ocean wind directions. IEEE Trans. Geosci. Remote Sens., 37, M. Konda et al.