Spaceborne scatterometers have been the mainsource of sea-surface wind vectors(SSWVs)withhigh spatial resolution since the introduction of theActive Microwave Instrument(AMI)onboard theEuropean Remote Sensing Satellite(ERS-1)in 1991.The successful mission of QuikSCAT has proved thatscatterometer wind data would improve the weatherforecasting, marine weather prediction, marinewarning and analysis applications(Atlas et al., 2001;Chelton et al., 2006; Von Ahn et al., 2006).Scatterometer wind data are assimilated in global orregional Numerical Weather Prediction(NWP)suchas that of the European Centre for Medium-RangeWeather Forecasts(ECMWF), Global and RegionalAssimilation and Prediction System(GRAPES), and Australian Community Climate and Earth-SystemSimulator(ACCESS)(Singh et al., 2012; Vogelzang and Stoffelen, 2012; Shrestha et al., 2013). Casestudies have demonstrated that scatterometer windsare useful in tropical cyclone(TC)analysis and forecasting(Snoeij et al., 2005; Brennan et al., 2009).Improvements of these applications increasinglydem and higher spatiotemporal resolution for windvectors. However, an individual sensor has limitedspatial coverage and revisit time. SSWVs have beenoperationally observed from multiple satellites(Table 1). These observations have initiated the“golden age” since the launch of MetOp-B inSeptember 2012. With the development ofmeteorology and oceanography, SSWs obtained froma single instrument cannot meet the requirements ofthese fi elds. It is urgently needed to combine multipleinstruments to achieve fi ne temporal resolution, widespatial coverage, and high credibility of mergedproducts. Comprehensive use of wind informationfrom four scatterometers, including ASCAT-A/B, OSCAT, and HY2-SCAT, will shorten observationintervals and fi ll data gaps(Zhang et al., 2006a).

Tab. 1 Past and present scatterometer missions

Table Options

Validation studies of scatterometer winds using insitu measurements or NWP model outputs have beenconducted, such as a comparison of Oceansat-2scatterometer wind products with buoy observations(Sudha et al., 2013) and their validation with windsfrom ECMWF, using triple collocation(Chakraborty et al., 2013). Advanced Scatterometer(ASCAT) and SeaWinds winds were assessed by similar methods(Vogelzang et al., 2012), and the fi rst 6 months ofHY2-SCAT winds were compared with in situ data(Wang et al., 2013). However, there have been fewstudies comparing operationally released windproducts of HY2-SCAT, ASCAT, and OSCAT.

Studies of composite analysis of availableinstruments in terms of sampling frequency and spatiotemporal data coverage for specifi c temporalresolutions represent the foundation for comprehensiveapplication(Zhang et al., 2006a). One such applicationis the blended sea winds products based on blending ofobservations from multiple satellites(Zhang et al., 2006b). Nevertheless, given a lack of scatterometerwind information, wind directions of blended windsrely on NWP models such as from the National Centersfor Environmental Prediction(NCEP)or ECMWF.

Around the China Seas(0°–40°N, 105°–135°E), many typhoons and other weather disasters greatlyimpact human activities, especially in coastal areas.Therefore, it is important to improve the capabilitiesof monitoring and forecasting in this region, but NWPrequires wind vector data of high quality. In thispaper, we assess operational wind products from thefour scatterometers and from various wind retrievalalgorithms. The paper is organized as follows. Thewind products with brief introduction to the windvector retrieval algorithms are described in Section 2.Section 3 examines the composite sampling capabilityin terms of revisit time and spatial coverage for fi xedtemporal resolution over the China Seas. Section 4includes calculations of the statistical distributions ofwind speed and directions, and Section 5 showscalculations of the wind component spectra to assessnoise and relative amounts of small-scale informationin wind products from ASCAT, OSCAT, HY2-SCAT and ERA-Interim. Conclusions are presented inSection 6.

Four wind products in orbit format from ASCAT-A, ASCAT-B, OSCAT and HY2-SCAT, plus ERAInterimforecast data with gridded zonal and meridional wind components, were used to assesswind data spatial coverage, revisit time, and erroranalysis. Because the scatterometer may providedifferent wind products due to various retrievalalgorithms or spatial resolution, wind vector datafrom scatterometers with 25-km spatial resolutionwere used. We now proceed to give a brief descriptionof these wind products.

ASCAT, aboard the MetOp satellite series, isdesigned to measure SSW speed and direction. Windproducts from both MetOp-A and MetOp-B havebeen produced operationally by the EuropeanOrganisation for the Exploitation of MeteorologicalSatellites(EUMETSAT)since February 2007 and June 2013, respectively. The EUMETSAT Ocean and Sea Ice Satellite Application Facility(OSI SAF)produces ASCAT 25- and 12.5-km wind products and a 12.5-km coastal wind product from both ASCAT-A and ASCAT-B(OSI SAF, 2013). OSI SAF ASCAT-A and ASCAT-B 25-km wind products are obtainedusing the EUMETSAT order download client. ASCATalways have three backscatter observations whichdiffer by 45 degrees in azimuth for each WVC; this isquite optimal for wind inversion(Stoffelen and Portabella, 2006). The ASCAT wind products use theMaximum Likelihood Estimation(MLE)algorithmfor wind inversion, and the cone distance minima inthe wind direction domain are quite narrow and welldefi ned(Stoffelen, 1998; Marcos, 2002). The twodimensionalvariational ambiguity removal(2DVAR)technique is performed to the selection of the mostprobable surface wind vector among the ambiguoussolutions(Vogelzang et al., 2009). The performanceof 2DVAR with meteorological balance constraintswas tested and optimized for ERS data and QuikSCATdata. It was found to be superior to other schemes(Stoffelen et al., 2010; Marcos, 2002).

CMOD5.n is used as the Geophysical ModelFunction(GMF)for ASCAT(Hersbach, 2010). Moredetailed information is found in the KNMI scatterometerpages(http://www.knmi.nl/scatterometer/).

The OSCAT instrument was aboard the Oceansat-2polar satellite, launched by the Indian Space ResearchOrganization(ISRO)on 23 September 2009. Serviceswere discontinued following an irrecoverableinstrument failure on 20 February 2014. However, wind data of 50-km resolution are available from 1January 2010 to 20 February 2014, and 25-km productsfrom 1 July 2013. We collected 25-km wind products(version 1.4, effective 28 May 2013)from the NationalRemote Sensing Centre(www.nrsc.gov.in). Theretrieval algorithm is based on a criterion of minimumnormalized st and ard deviation(NSD)of derived windspeed using the OSCAT-specifi c GMF(Gohil et al., 2008, 2013). Directional ambiguities are removedusing the Directional Stability and Conservation ofScattering(DiSCS)algorithm(Gohil et al., 2010).

The HY2-SCAT instrument is aboard the HY-2Apolar satellite, and its products have been operationallyreleased by the National Satellite Ocean ApplicationService(NSOAS)since October 2011. The windproducts are retrieved using MLE algorithm withNSCAT-2 GMF. An ambiguity removal algorithmbased on a circular median fi lter is used to to removewind direction ambiguities. NCEP winds are used asa background fi eld during wind direction ambiguityremoval(AR)(Yang et al., 2014).

The ERA-Interim reanalysis data set is the latestglobal atmospheric reanalysis produced by theEuropean Centre for Medium-Range WeatherForecasts(ECMWF)(Dee et al., 2011). The ECMWFERA-Interim daily atmospheric datasets include 6-hanalysis fi elds available at 00:00, 06:00, 12:00 and 18:00 UTC, and forecast fi elds that are produced fromanalysis fi elds beginning at 00:00 and 12:00 UTCwith 3-, 6-, 9- and 12-h steps. A recent study comparedsea surface winds from the ERA-Interim with thatcollected from eight buoys deployed in the Yellow and East China Seas, and results showed that theERA-Interim wind data agree well with the buoy data(Song et al., 2014).

Moreover, the ERA-Interim data are producedseparately with its own set of data assimilationobservations which is different from the observationsused in the operational model of ECMWF. The nearreal-time ASCAT- or OSCAT- derived winds areassimilated in the ECMWF operational model, butneither is used in ERA-Interim(Persson and Grazzini, 2007 ; Dee et al., 2011).

For the comparison studies in Section 4, only ERAInterimforecast data(wind components at 10-m zonal(U) and meridional(V))with resolution of 0.25°×0.25°along latitude and longitude were collected. Besides, scatterometer-wind data of 25-km resolution wereused to avoid sampling errors caused by variablespatial resolutions. To make the winds comparableamong all fi ve data sources, we collected data duringthe same three months(July, August and Septemberin 2013)over the China Seas. Global data were usedin the wind spectra analysis, owing to data limitationsfor the China Seas.

Scatterometers have different swath widths and hence require different periods to observe the entireearth. However, daily coverage of each satellite forthe China Seas is nearly constant. To determinepassage times of every scatterometer over the ChinaSeas, we analyzed every orbital track in July 2013.Only passes with at least 3 000 valid Wind VectorCells(WVCs)in the area of interest were considered.Figure 1 shows the time(UTC)distribution of eachinstrument every day. We see that ASCAT-A and ASCAT-B observe the earth with similar orbits and very short time difference, Two periods each day, from 5:30 to 8:30 and 17:00 to 21:00, have noobservations over the China Seas. Relatively suffi cientobservations were made near 0:00 and 12:00 and there was no simultaneous cross observation betweenASCAT, OSCAT and HY2-SCAT available.

Fig. 1 Passage time (UTC) of spaceborne scatterometers over China Seas during July 2013

Figure Options

Since the discontinuation of OSCAT data inFebruary 2014, winds from HY2-SCAT as acomplement to ASCAT winds became more importantto continuous observation of SSWVs, and two periods(3–9 and 15–21)will have no data available beginningthat month.

To explore the advantages of combined observationswith multiple satellites, it is necessary to assessmerged temporal resolution and spatial coverage.

Average data on the 0.25°grid were examined atvarious temporal resolutions(3-, 6- and 12-hourly)inJuly 2013. In the calculation, if the spatial resolutionwas 0.25° and temporal resolution 3(6/12)h, datawith 1.5-h(3/6 h)separation were selected. Datacoverage for various time intervals is listed in Table 2.This shows that spatial coverage of combined windsobserved by the four scatterometers over the ChinaSeas varied greatly. The two lowest percentages for3-h intervals are 11.0% at 6:00 and 7.9% at 18:00.This is consistent with the two blank periods shown in Fig. 1. Average data coverage was about 92.8% for12-h intervals at 12:00 and 90.7% at 24:00.

Tab. 2 Percentage of available data coverage for 0.25° oceanic boxes on 2 July 2013

Table Options

To study the contribution of combined observationsat the sampling grids for a specifi c time interval, theaverage data availability within one month(July2013)at 12:00 were analyzed. The result is shown in Fig. 2 for time intervals 3 h(a), 6 h(b), and 12 h(c).The maximum value in the dataset is 2.72 for theaverage per sampling grid in one month. Withincreasing time interval, more observations frommultiple sensors are included, and this is evident in Fig. 2, from left to right. There were fewer observationsin the western South China Sea and more in the Bohai, Yellow, and East China Seas. We conclude that thereis still a large data gap for 3-h intervals, and for 6hours except for the region around the Yellow and East China Seas. There are suffi cient observations for12-h intervals, with more than a single passage overmost of the China Seas. The slanting tracks shown in Fig. 2 indicate the paths of the polar-orbiting satellites.

Fig. 2 Average data availability in 1 month (July 2013) at 12:00 for time intervals 3 h (a), 6 h (b), and 12 h (c) in July 2013

Figure Options

Here we evaluate the quality of the fourscatterometer wind products in detail. Allscatterometer wind products retrieved from ASCAT-A, ASCAT-B, OSCAT and HY2-SCAT are composed ofraw swath data of WVCs in the China Seas for fourmonths(July, August, September and October in2013). Since most of OSCAT-25km products inOctober are not available from ISRO, only threemonthobservations were used. The ERA-Interimwinds were used as reference to study errorcharacteristics of wind fi elds. Although ASCAT windproducts contain forecast winds at each WVC butfrom ECMWF operational model which differentfrom ERA-Interim. Then the ERA-Interim forecastwinds data were interpolated according to longitude, latitude and time of the spaceborne observations, using linear 3D-interpolation. Mean absolutedifferences of wind direction with respect to ERAInterimforecast winds were calculated as a functionof ERA-Interim forecast wind speed in 1 m/s bins, asshown in Fig. 3. Corresponding biases and st and arddeviations of wind speed are shown in Fig. 4. Both in Fig. 3 and 4, only bins with at least 100 valid matchedpoints were considered. Besides, the typhoon Soulikwas present in our study area from July 10 to July 13, then more abnormal observations may occur on thismonth.

Fig. 3 Mean wind direction difference of MetOp-A (blue), MetOp-B (magenta), HY-2A (red), Oceansat-2 (black) with respect to ERA-Interim forecast winds, calculated in 1 m/s bins of ERA-Interim forecast wind speed

Figure Options

Biases of wind direction for all scatterometer windproducts in Fig. 3 decrease with increasing wind speedthrough 10 m/s and fall to less than 20° with windspeed greater than 5 m/s. The biases of ASCAT and OSCAT are less than 20° with slight variation(lessthan 5°)with wind speed greater than 6 m/s. However, biases of the HY2-SCAT show strong variation withwind speed in excess of 13 m/s, and greater than 20°with wind speed 14 m/s. This abnormality of HY2-SCAT persisted even if much more wind data wereused. This is attributed to a retrieval problem thatremains to be resolved.

The curves in Fig. 3 terminate at different windspeeds, which implies differences of wind speedProbability Distribution Function(PDF)among thefour scatterometer datasets, which can also be seen atthe lower parts of Fig. 4a–d. There are more data athigh wind speeds in ASCAT-derived winds than inthose of OSCAT and HY2-SCAT.

Fig. 4 Wind speed bias and standard deviation referenced to the average of scatterometer-derived and ERA-Interim forecast winds as a function of the average wind speed, in 1 m/s bins of wind speed Histograms of scatterometer wind speeds are shown in lower portions of each panel.

Figure Options

Figure 4 shows mean differences of wind speedbetween scatterometer measurements and the averagewind speed of scatterometer and ERA-Interimforecast winds as a function of the average windspeed. The bias of wind speed for all scatterometerwind products decreases as wind speed increases to7 m/s. Biases of ASCAT-A and ASCAT-B are verysimilar, with correlation coeffi cient 0.97(Fig. 4a–b), which are consistent with those of Verspeek et al.(2013). The biases of OSCAT winds(Fig. 4c)arepositive and slightly fl uctuate around 1 m/s at windspeed from 5 m/s to 20 m/s. Both wind speeds derivedfrom OSCAT and HY-2 show large st and ard deviationsover the full wind speed range(Fig. 4c–d), and thest and ard deviations are typically around 2 m/s and increasing substantially at high wind speeds. Furtherstudies should be done to improve retrieval algorithmsof these two sensors, especially for high wind speeds.

Figure 4a and b show that wind speeds of ASCATproducts are also overestimated at wind speed below15 m/s. But the biases are obviously decreasing tonegative with the increasing wind speed above 15 m/s.While some other studies of comparison show that theASCAT winds are good agreements with buoy data(OSI SAF, 2013). Figures 3 and 4d reveal that windspeeds and directions of HY2-SCAT products arenoisy at low(＜5 m/s) and high(>15 m/s)wind speeds.Compared with the other three products ofscatterometer wind data, there are especially largedifferences at high wind speeds, indicating a need forimprovement in the HY2-SCAT wind retrievalalgorithm. Biases of OSCAT wind direction are lessthan 20° for speeds above 5 m/s. Wind speeds ofOSCAT and HY2-SCAT products are overestimated(positive biases). To some extent, the higher bias(shown in Fig. 4d)at high wind speeds from HY2-SCAT is due to the NSCAT-2 GMF. According thestudies of Verhoef and Stoffelen(2012), the windspeeds are overestimated above 15 m/s in NSCAT-2 and a linear downscaling of such wind speeds wasapplied to refine it.

The differences between the four scatterometerswind products(especially in wind direction domain)are relative to their antenna geometry and windretrieval scheme. ASCAT-A and ASCAT-B containvery few ambiguity removal errors, because the MLEminima in ASCAT are very narrow. In the majority ofcases, inversion of the ASCAT data leads to twosolutions 180 degrees apart. The other two windproducts are obtained from rotating pencil beamscatterometers and are prone to ambiguity removalerrors in the nadir part of the swath. HY-2A uses avery simple ambiguity removal scheme, which willfrequently fail. Therefore the performance of thisproduct is worst. OSCAT has a slightly moresophisticated ambiguity removal method, and therefore its performance is in-between(Vogelzang et al., 2009; Gohil et al., 2010; Stoffelen et al., 2010).

Spectral analysis of spatial structures in thescatterometer products is done as an advancedanalysis tool to determine product characteristics.Wind component spectra may be used to detect noise and assess the relative amount of small-scaleinformation(Vogelzang et al., 2012). Using windspectra, we can also evaluate the spatial representation(smoothness) and noise level in a wind product. Windcomponent samples are selected along the satellitetrack, and several fi xed indexes in the cross-trackdirection are included to reduce spectrum noise. Thespectrum is calculated for every sample, and allspectra are averaged as a function of spatial frequency.

Spectra are usually plotted on a log-log scale inwhich the logarithm of spectral density in m ³ /s ² isgiven as a function of wave number(Stoffelen et al., 2010). Figure 5 shows spectra of the zonal(u ; Fig. 5a) and meridional(v ; Fig. 5b)wind components for theaforementioned 25-km wind products. Figure 5 alsoshows that the ERA-Interim winds spectrum declinedmuch faster than any other scatterometer spectra. Thisis because the ERA-Interim winds suppresses smallscales, which may not be benefi cial for prediction atmedium time ranges(Vogelzang et al., 2012).

Fig. 5 Spectra of ASCAT-KNMI (red), OSCAT-ISRO (blue), HY-2 NSOAS (magenta), and ERA-Interim (black) wind products with 25-km spatial resolution, for zonal (left) and meridional (right) wind components Plots cover the period of July 2013.

Figure Options

HY2-SCAT product spectra have greater density and tend to become horizontal as spatial frequencyincreases. This means that this data product containsmore small-scale variance than other products and itsspectral variance is not linear, which does not satisfythe wind turbulence law of k ^{- 5/3} . According to Nastrom and Gage(1985) and the turbulence theory ofKolmogorov, the wind spectra should follow this lawfor scales smaller than about 500 km. Conclusively, the spectral analysis confi rms the reasoning in theprevious point: noise at large wavenumbers is causedby ambiguity removal errors. Again, ASCAT performsbest and HY-2A performs worst, with OSCAT inbetween.

In this work, we assessed wind products obtainedfrom four microwave scatterometers for the ChinaSeas, including ASCAT-A/ASCAT-B, OSCAT and HY2-SCAT. We determined composite samplingcapabilities, statistical biases of wind speed and direction, and wind component spectra.The analysis of revisit time by combinedobservations shows that two daily periods, 5:30–8:30UTC and 17:00–21:00 UTC, had no observations ofthe China Seas. The data gap exp and ed to 3:00–9:00UTC and 15:00–21:00 UTC because of thediscontinuation of OSCAT data in February 2014. Weconclude that there remain large data gaps for 3-hintervals, and for 6-h intervals except in the regionnear the Yellow and East China Seas. There aresuffi cient observations for 12-h intervals, with morethan a single passage over most of the China Seas.Winds from HY2-SCAT as complements to ASCATwinds are important for continuous observation ofSSWVs. The assessment of composite spatialsampling revealed that there are fewer observations inthe western South China Sea, and more in the Bohai, Yellow, and East China Seas.

Comparison between scatterometer wind and ERA-Interim forecast data revealed that, wind speeds and directions of HY2-SCAT products were noisy atweak(＜5 m/s) and strong(＞14 m/s)wind speeds.Wind direction biases of OSCAT were less than 20° atwind speeds in excess of 5 m/s. Speeds of OSCAT and HY2-SCAT products were overestimated(positive biases). For ASCAT wind products, the winddirection show lower bias and variation; but the biasesof wind speed are positive and tend to negative atwind speed above 15 m/s. Wind component spectrawere analyzed to detect noise and assess the relativeamount of small-scale information. HY2-SCAT winddata showed some signifi cant noise at high frequencycompared to other three products of scatterometerwind, and it is necessary to study the performance ofthe HY-2A wind product when a more sophisticatedwind retrieval algorithm like MSS in combinationwith 2DVAR would be used. Future topics of researchwill be validation using in-situ data, comparisons ofpassive(radiometers such as WindSat) and active(scatterometer)winds, and blended wind vectorproducts using multiple-scatterometer data.

We greatly appreciate EUMETSAT, ISRO, and NSOAS for their free datasets. We thank thescatterometer team of the Royal Netherl and sMeteorological Institute(KNMI)for their suggestions and discussions about detailed questions, especiallyfor Anton Verhoef’s knowledge about the dataset wehave used. We really appreciate three anonymousreviewers for their helpful comments and valuableinterpretations.