“'Matching ASCAT and QuikSCAT Winds”
Review #1
This paper is wonderfully written and is a valiant effort to make several great points about the differences between Ku and C band scatterometer calibrations and winds. The paper begins with by examining error characteristics over all wind speeds and assuming that conclusions based on these statistics apply equally well to all wind speeds. The paper goes on to show that these assumptions are bad, but it is not clear that the bad assumptions have no impact on the findings. Reading the paper a second time suggest that these problems might not be large, however, the authors must verify that in a rebuttal. There are also key details missing in the paper (see major issues) that might change the interpretation of the findings. I would like to see the cold water impacts verified for Northern latitudes as well as Southern latitudes, shown as a function of wind speed - otherwise I could argue that the problem is due to sea state assumptions rather than temperature. I am very impressed with the effort, and would like to see the work modified to be more convincing as mentioned above. My impression is that the technical work should be quite easy, and the biggest chore will be rearranging the paper to avoid the consequences (impression or reality?) of the poor assumptions. Because major changes in the text or the finds are likely, I recommend that this paper be accepted with major revisions, however, I anticipate a relatively easy change and look forward to seeing the revised work.
Below are examples of verication/rebuttal of some of our hypotheses.
In Fig. 4b (new) we illustrate that wind speed difference has only weak dependence on the ASCAT incident angle. In Fig.5 (new) we show that QuikSCAT and ASCAT wind directions agree well. Explanations have been added on p. 12, lines 258-260.
The hypothesis on the impact of the sea state on at high southern latitudes is rejected because the area of negative is displaced southward of the ‘roaring forties’ toward higher latitudes where SST is really cold but winds and waves are not that strong as to the north. This is explained on p. 16, lines 332-337.
Cold SSTs<5C are present south of 60S year around in contrast to high northern latitudes where SSTs<5C occur only in winter (see illustration below) when sea ice reduces the number of wind retrievals. This explanation of the south-north asymmetry in is added on p. 15, lines 325-331.
Figure. Climatological SST in boreal (DJF, top) and austral (JJA, bottom) winter.
We have analyzed (not included in the text) possible impact of the sea state by analyzing the QS-AS wind speed difference at the NDBC buoys that provide wave data. We didn’t find statistically significant impact of waves (significant wave height) on . According to the two-scale theory and observations the wind speed retrievals from particular instrument show some weak dependence on SWH (e.g. Ebuchi 2002). But this effect contributes much less to the QS-AS wind speed difference.
Verification/rebuttal of our hypotheses is further explained in reply to specific comments below.
Major issues
1) Specify which QuikSCAT product is used (JPL or RSS) and provide a version number. During the time this paper was in review, new product came out from RSS and JPL, and a new version was announced by KNMI. Ideally the paper would be redone with the new versions, which have moved closer together. If it is not redone, then the new products must be acknowledged.
QuikSCAT data are the 25km Level2b processed with QSCAT-1 GMF and available from JPL/PODAAC (see p.5, lines 98-110). The Product details are available from the JPL online documentation (see footnote #2 on p.5).
New QuikSCAT product based on the Ku-2011 GMF has been announced recently by the Remote Sensing Systems, but new KNMI product has not been released yet. The text has been modified on. p.6 (lines 118-122) to include results of our preliminary analysis indicating improvements in tropical rainfall detection by the new RSS product. However the difference between QuikSCAT and ASCAT winds doesn’t depend much on the choice of product.
2) Lines 125 and 126: ASCAT and QuikSCAT accuracies are statically similar when taken as averaged over the available wind speeds. That is not true when high and low wind speeds are examined. Please clarify.
Discussions of ASCAT and QSCAT to buoy comparisons are eliminated from the revised manuscript. Instead new Figs. 4 and 5 have been included to justify the choice of parameters of the correction function dW.
Although the paper does not focus on the scatterometer accuracy, we have checked it (not shown in the paper revision) by comparing satellite retrievals with in-situ buoy data matchups. This analysis confirms that both, QS and AS have similar accuracy of wind velocity retrievals. Statistical parameters characterizing these comparisons are shown in below. Here Rms is the root mean square deviation, r and r² are scalar and vector correlation coefficients. Remarkably similar ASCAT and QuikSCAT error statistics is found at buoy wind speed 5m/s <W< 10m/s. At low winds (< 5m/s) both, ASCAT and QuikSCAT tend to overestimate wind speed. Such behavior may be related to low wind speed distribution. Both scatterometers underestimate wind speed at W >10m/s, but the underestimation is more evident for ASCAT.
3) Line 136: what assumptions where made regarding currents, and what wave (roughness length) parameterization was chosen from the COARE options? Why where these choices made?
Both wave and surface current options are turned off in the COARE3.0 algorithm because only a few buoys used in this study provide such measurements. This is explained on p.7, lines 145-149.
4) Lines 159 to 174 and 181 to 189: This analysis misses the point of cal/val. The biases must be examined as a function of wind speed, otherwise spurious interpretations are likely.
We agree and have investigated bias as a function of various parameters including wind conditions and buoy locations. For instance, the following Figure shows the time mean difference between NDBC buoy wind speed occurring within 2 hours (sub-sample 1 and 2 described on p. 8) as a function of buoy wind speed. The biases are close to zero. About 92% of biases are within –0.1 m/s and 0.1 m/s. No significant trend as a function of wind speed range and/or buoy number (location) is found. This result is briefly explained on p. 9 (lines 181-183). We also have examined the geographical distribution of sampling bias using ECMWF winds sampled at QS and AS overpass times. These results are explained on p. 9 (lines 188-193), see also footnote #5.
5) Lines 175 to 182: Are biases in the individual comparisons to buoy removed prior to estimating random error characteristics? There are considerable biases in buoy observations (see NDBC documentation on acceptable biases). Are these biases responsible for the results? Are the large tropical currents responsible for the results?
Reply:
Direct comparisons of satellite and buoy wind speed have been eliminated. Tropical currents impact the apparent wind seen by a buoy while a scatterometer measures wind velocity relative to the surface current. The tropical currents are of different directions. The NEC and SEC are westward while the NECC is eastward. This should mitigate the impact of currents when averaged over all tropical buoys. Impact of currents on individual buoy wind could be up to 0.5m/s based on the tropical currents magnitude (Kelly et al., JAOT 2005).
In the present study, both ASCAT and QuikSCAT tend to slightly underestimate wind speeds with respect to the tropical buoy data. ASCAT exhibits stronger bias than QuikSCAT. However, recent study (see presentation of A. Plagge at the last IOVWST meeting) surprisingly indicates that impact of currents on ASCAT winds is weaker than on QuikSCAT. Therefore, the ASCAT wind speed bias (that is stronger than QS wind speed bias) cannot be related to the surface current only.
Errors in the historical NDBC buoy wind records are related to changes in wind sensors and measurement height ( http://www.ndbc.noaa.gov/improvements.shtml ). Since the mid-1980's (that includes the QS and AS era) R.M. Young Model 05103 has been the standard wind sensor. All buoy wind sensors are calibrated before deployments to ensure that the reported speed lies within 1 m/s of the wind tunnel standard. Changes in the measurement height are accounted for by computing the 10m neutral wind using the COARE3.0.
Errors in instantaneous NDBC wind data are produced by various wave-related impacts such as the anemometer "pumping", etc. Although these errors are quite hard to quantify, their impact on the 13-month time mean is assumed to be small. Accuracy of off-line NDBC buoy data (used in this study) is about 0.50m/s/10deg in wind speed/direction, respectively. Accuracy of off-line data is higher than that of NRT data.
http://www.ndbc.noaa.gov/rsa.shtml
http://www.ndbc.noaa.gov/NDBCHandbookofAutomatedDataQualityControl2009.pdf
Errors in instantaneous buoy data are mitigated by using of hourly averaged post-processed (off-line) buoy data as explained p. 7 (lines 145-148).
In the original version of the paper we evaluate the scatterometer accuracy by comparing with off-line wind data from TAO, PIRATA, and RAMA buoys. Freitag et al (2001; see NOAA Technical Memorandum OAR PMEL-119) indicate that for wind speed ranging between 4m/s and 9m/s, buoy sensors tend to slightly overestimate winds. However such overestimation does not exceed 0.15m/s. They also indicate that buoy accuracy, defined as rms difference, is about 0.30m/s. Obviously, results shown in the old Figures 2 and 3 cannot be related to buoy wind speed biases only.
Text modification: Direct comparisons of satellite-buoy wind speeds have been eliminated along with old Figs. 2 and 3.
Text has been also modified on p. 7 (lines 145-148) to explain that only hourly averaged post-processed (off-line) buoy measurements are used as an input to the COARE3.0. We feel that detailed description of buoy data errors is beyond the scope of this paper.
6) Fig. 5: What quality control was used on the ASCAT data?
Reply: ASCAT quality is based on the quality flag for each WVC (field 89). All quality flag bits, except #11 (wind speed is greater than 30m/s) and #12 (wind speed is lower than 3m/s), are required to be 0.
Text modification: Explanations have been added on p. 7 (lines 133-137).
7) Lines 232 & 233: A very recent paper by Ebuchi (2011, IGARS) shows that the angular dependence of ASCAT errors is a function of wind speed. That suggests a similar dependence should be in the difference between ASCAT and QuikSCAT winds.
Reply: We model the difference between QuikSCAT and ASCAT as a function of both wind direction (relative to ASCAT beam azimuth) and wind speed (as shown in Figure 8). The problem with ASCAT wind direction retrievals at low winds (reported by Ebuchi, 2011) is also evident in the new Fig. 5a as an increase in the standard deviation of ASCAT and QSCAT wind directions (natural variability of weak winds contributes as well). But the time mean wind directions from the two instruments agree well (Fig. 5a).
Text modification: Fig. 5 has been added to illustrate QS and AS wind direction. Relevant discussions have been added on pp. 12-13 (lines 261-268).
8) Figures 6 and 7: In Figure 6 the wind directions appear to be multiplied by 107, and in Figure 7 they are in thousands of degrees! Something is seriously wrong with either the values or the caption.
Reply: The right y-axis corresponds to wind direction histogram (as noted in Fig. 6 legend) rather than wind speed difference.
Text modification: Explanation is added to the Figure 6 caption.
9) How many of the findings mentioned in 290 to 300 come from interpretational issues mentioned above? A more detailed study is needed to state these (very interesting) findings with such confidence.
Reply: Some information is provided in the new Figs. 3 and 4. It is shown that wind speed bias doesn’t depend on the ASCAT incident angle (to within 0.2 m/s), and wind directions from the two instruments agree well. Exploring of alternative explanations and hypotheses have been conducted using various collocated data from scatterometers (ASCAT and QuikSCAT), ECMWF analyses, buoys, and altimeters (Jason-1/2 and Envisat). Using sub-sampled buoy and ECMWF data we analyzed and rejected the hypothesis that time mean difference between QS and AS winds is explained by systematic time lag between the two satellites. Using altimeter and NDBC buoy data we analyzed a link between and SWH. This latter hypothesis has been also ruled out. The present paper focuses only on the wind speed difference as a function of wind speed, relative wind direction, and SST that contribute significantly as we think these are the largest effects. We leave other issues such as wave state dependence to future studies.
Minor issues
10) Lines 214 to 217: Do the low wind speeds and substantial noise play a role in the reduced correlations? Basic statistics suggest that correlations should also drop because of the low signal to noise ratio.
Explained on p.11 (lines 227-236).
11) Line 251: 'data' is plural: change is to are.
Done.
(It seems like ‘data’ is either singular or plural in English. It is only plural in Latin.)
12) Appendix: (1) define 'c'. (2) clarify the logic behind equation A2.
‘c’ is the wave phase speed (explanation is added on p.20, line 427).
(A2) is based on the Bragg approximation, , where S is the Bragg wave spectrum that depends on ) (p.20, line 425). If is the only parameter depends on, then an error in wind retrieval () can be evaluated by differentiating along =const, i.e. SST-induced changes in radar backscatter, , are interpreted as changes in wind speed