Ocean surface wind products play a critical role in marine environmental forecasting. Traditional methods for wind observation mainly depend on research vessels, buoys, and meteorological stations. However, when it comes to the entire ocean, wind products have some obvious deficiencies, such as lack of temporal and spatial resolution and others. In this context, the ERS-1 satellite was launched in 1991, carrying advanced microwave scatterometers and other sensors. Since then, scatterometer-derived wind products have been widely used in scientifi c research(Bentamy et al., 2008).

The scatterometer does not measure sea surface wind directly, but is designed to determine the normalized radar cross section(sigma-0)of the sea surface. Then, the value of sigma-0 is used to calculate winds at 10-meter height above that surface.

Three series of scatterometers have been launched to provide sea surface wind products over the last several years(Wang et al., 2013). In addition to the HY-2A Scatterometer(HY-2A SCAT)launched in 2011, the Advanced Scatterometer(ASCAT)aboard Meteorological Operational Satellite A(MetOp-A) and MetOp-B was launched in 2006 and 2012 by the European Space Agency(ESA), and the Oceansat-2 Scatterometer(OSCAT)was launched in 2009 by the Indian Space Research Organization(ISRO)(Rani et al., 2014).

HY-2A SCAT is a rotating pencil-beam radar operating at 13.256 GHz(Ku-b and ). It has a one meter dish refl ector antenna with two beams. The outer beam is VV-polarized and covers a swath of 1 750 km. The inner beam is HH-polarized and covers a swath of 1 350 km. The HY-2A SCAT is expected to increase temporal sampling of scatterometer wind measurements(Wang et al., 2013). At present, the wind products of HY-2A SCAT are delivered with spatial resolution 25 km. For more detail, refer to the National Satellite Ocean Application Service(NSOAS)website(http://www.nsoas.gov.cn).

ASCAT is a C-b and scatterometer with three vertically polarized antennas transmitting pulses at 5.255 GHz. Fan-beam antennas are oriented at 45, 90, and 135 degrees with respect to the satellite track. There are two observation swaths, both about 500-km wide. Current ASCAT wind products are provided at two spatial resolutions over the global oceans, 12.5 km and 25 km(Bentamy et al., 2008; Rani et al., 2014). More information on the ASCAT instrument is available from the European Organisation for the Exploitation of Meteorological Satellites(EUMETSAT)website(http://www.eumetsat.int).

OSCAT is an active microwave sensor designed and developed at ISRO. The instrument is a pencilbeam wind scatterometer operating at Ku-b and of 13.515 GHz. The sensor covers a continuous swath of 1 400 km for the inner beam and 1 840 km for the outer beam. The aim is to provide global ocean coverage and wind vector retrieval with a revisit time of two days(Rani et al., 2014). Currently, two types of wind products with different spatial resolutions are produced, 25 km and 50 km. Refer to the ISRO website(http://www.isro.org/)for more detail.

To facilitate comparison of the three instruments, some major parameters of HY-2A SCAT ASCAT and OSCAT are shown in Table 1(Singh et al., 2012; Wang et al., 2013; http://www.knmi.nl/scatterometer/ oscat_50_prod/; http://www.remss.com/missions/ascat; http://www.nsoas.gov.cn).

Tab. 1 Major parameters of HY-2A SCAT, ASCAT, and OSCAT

Table options

Several works have evaluated the quality of scatterometer-derived wind products. In general, validation of these products has used buoy data, research vessels, and other satellite sensors. Ebuchi et al.(2002)validated QuikSCAT wind products, fi nding that they agreed well with buoy data. Bentamy et al.(2008)evaluated ASCAT measurements via buoy and QuikSCAT wind vector observations. Mathew et al.(2012)compared OSCAT against Jason-2 altimeter wind products. Wang et al.(2013)assessed the initial six-month wind products of HY-2A SCAT, using in situ measurements. In addition to validation, some works have compared performance differences of these scatterometers. Bentamy et al.(2000)intercompared ERS-2 and QuikSCAT winds, revealing that QuikSCAT wind vector estimates compared well with ERS-2 wind products. Rani et al.(2014)compared Oceansat-2 and ASCAT winds with buoy observations, showing that ASCAT winds were slightly more accurate than those of OSCAT.

In this paper, validation and intercomparison of HY-2A SCAT, ASCAT and OSCAT wind products are done using buoy measurements. Section 2 describes the data sources and processing methods. Section 3 gives results of the validation and intercomparison, and a discussion based on those results. Finally, conclusions are provided in Section 4.

The comparison was carried out over a long period, from January 2012 to August 2013. Level-2B wind products with spatial resolution 25 km were required. Owing to the absence of OSCAT Level-2B wind products of resolution 25 km, wind products with resolution 50 km were used instead. In addition, only wind products of the ASCAT aboard MetOp-A were used. All products contained the same measurements of wind speed and direction.

Scatterometer data were obtained from different sources. HY-2A SCAT wind products were from NSOAS. The EUMETSAT Ocean and Sea Ice Satellite Application Facility(OSI SAF)provided ASCAT wind data via the EUMETSAT data center. ISRO reprocessed OSCAT Level-2B wind data, which were obtained from the National Remote Sensing Centre(NRSC).

Buoys furnish continuous wind data for generating 10-minute average wind speed and direction. Qualitycontrolled data can be downloaded from the National Data Buoy Center(NDBC). The buoys are all moored. They are deployed in coastal and offshore waters from the western Atlantic to Pacifi c Ocean around Hawaii, and from the Bering Sea to the South Pacifi c. We chose 43 buoys for this study. Figure 1 shows locations of the selected buoys. The height of anemometers on the buoys is 5 or 10 m above sea level, and all buoys are more than 50 km offshore.

Fig. 1 Locations of 43 moored buoys used in the study

Figure options

Because scatterometer-derived wind products are equivalent neutral wind vectors at a height of 10 meters above sea level, some buoy wind data must be converted. The wind profi le power law shows the relationship between wind speeds at one height and those at another. This law is expressed as

Click Browse formula

The statistical parameters used are bias and rootmean- square error(RMSE). These can be calculated by

Click Browse formula

Click Browse formula

For validation, buoy measurements served as the ground truth. We first established a filtering criterion for matching. Scatterometer-derived wind and buoy wind data were collected only if the spatial difference was less than 0.1 degree and temporal difference less than 5 minutes. Then, considering possible noise in the measurement process, we exercised further quality control, i.e., collecting wind speed and direction data only if both were within five times the standard deviation.

Intercomparison was based on the validation results.The discussion below gives a better understanding of the performance of each scatterometer.

Figure 2 shows a comparison of wind products derived from HY-2A SCAT, ASCAT and OSCAT with NDBC buoy wind data. After data processing, the numbers of collocation for HY-2A SCAT, ASCAT and OSCAT were 6 702, 8 378, and 3 809, respectively.

Fig. 2 Comparison of wind speed and direction observed by HY-2A SCAT (a, b, c), ASCAT (d, e, f) and OSCAT (g, h, i), using NDBC buoy wind data Comparison of wind speed (left) and direction (middle) for data in all buoy wind speed ranges, and direction in buoy wind speed range 5 to 15 m/s (right). Solid diagonal line is 1:1 and dashed line represents regression line. Number of collocation, bias, and RMSE are also shown.

Figure options

In general, there was good agreement between scatterometer-derived and buoy wind speeds(left panels of Fig. 2). The wind direction comparison shown by the middle panels of the fi gure is not satisfactory. However, in the range 5–15 m/s, the results improved. For wind speed, Fig. 2a shows that the RMSE of HY-2A SCAT wind products is 1.23 m/s, with a small negative bias of 0.36 m/s. Clearly, HY- 2A SCAT had good performance in wind speed retrieval, because it was within the accuracy specifi ed in the HY-2A SCAT mission goal. ASCAT performed better than HY-2A SCAT, with RMSE for wind speed 0.86 m/s and bias -0.07 m/s(Fig. 2d). OSCAT had the largest RMSE of 1.54 m/s and greatest negative bias of 0.38 m/s. However, it also met the requirements when compared with the mission goal of 2 m/s. However, for wind direction, only the ASCAT result met expectations, with RMSE 17.97 degrees and bias 1.60 degrees(Fig. 2f). HY-2A SCAT performed slightly worse than the mission goal, with RMSE 22.85 degrees. However, the OSCAT results show poor performance in measuring wind direction. Figure 2 h indicates an RMSE as large as 39.86 degrees, much greater than the mission goal of 20 degrees. This seems inconsistent with the conclusions drawn by Rani et al.(2014), in which the reported wind direction was within that goal. The difference may be mainly caused by different collocation criteria. Scatterometer winds with direction difference greater than 90 degrees relative to the model forecast were not considered for the collocation in Rani et al. Such collocation criteria is rarely seen and may be somewhat unreasonable, because it would undoubtedly improve OSCAT performance. In addition, Rani et al. used only three months of data. This might be too little from a statistical viewpoint to reach a reliable conclusion, in comparison to the present work. If only 5–15 m/s buoy wind speed data were used, the results improved. ASCAT also shows high accuracy, with RMSE 11.29 degrees. The RMSE for HY-2A SCAT declined to 20.14 degrees, nearly meeting mission requirements. Although OSCAT still had poor performance, it was much improved when compared using the entire data range, with RMSE decreasing from 39.86 to 29.20 degrees.

To gain a better underst and ing of the comparisons of different scatterometers, some statistical parameters for low, medium, and high wind speeds were calculated(Table 2). It is clear that ASCAT performed the best, because RMSE and bias were the smallest for all ranges. By contrast, HY-2A SCAT and OSCAT were less accurate, especially the latter. Furthermore, all scatterometers performed better in the middle range, since both RMSE and bias for observed wind speed and direction were nearly minimum compared with other wind speed intervals. However, there was poor performance of wind direction retrieval for low wind speeds(less than 5 m/s).

Tab. 2 Statistics for comparisons of different wind speed ranges

Table options

Figure 3 shows RMSE for wind speed and direction of HY-2A SCAT, ASCAT and OSCAT versus buoy wind speed. Because the number of collocation for high winds was too small, only buoy wind speeds from 1 to 20 m/s were used. For wind speed, there was no clear dependency upon comparison. Figure 3c shows good ASCAT performance in wind speed retrieval, with RMSE well below the reference line over the entire range. Not much difference is seen between HY-2A SCAT and OSCAT for medium winds. However, HY-2A had poorer performance at low speeds, and OSCAT was worse at high speeds. For wind direction, ASCAT again performed the best. As shown in Fig. 3d, except for low winds, RMSEs for both medium and high winds were under the reference line. HY-2A SCAT performed less satisfactorily compared with ASCAT, because the mission goal was only attained in half the wind speed b and s(Fig. 3b). By contrast, OSCAT gave the worst performance of wind direction retrieval. No wind speed b and met the requirement over the entire range(Fig. 3f). The results also show a high-low pattern in wind direction for the three scatterometers. The RMSE for wind direction decreased monotonically at low wind speed. Except for a few abnormal b and s, an overall weak fl uctuation is observed at medium and high winds. Interestingly, there was a common abnormal b and between 14 and 15 m/s. RMSE for this b and peaked compared with nearby b and s. Further investigation was made of the buoys to see if this was caused by a few r and om buoys, but, the results disproved this conjecture. Coincidentally, according to Chen(2004), the transition from swell dominance to sea wind dominance was at 14.2 m/s, which lies within the abnormal b and . However, at present, no clear relationship has been found between this transition and the suddenly peaked b and , and the reason remains unclear.

Fig. 3 RMSE for wind speed (left) and direction (right) of HY-2A SCAT (a and b), ASCAT (c and d) and OSCAT (e and f) versus buoy wind speed Solid curves with squares and triangles represent RMSE results and number of collocation, respectively. Dashed horizontal line is reference line, which represents mission requirement. RMSE was calculated for every 1 m/s buoy wind speed bin from 1 to 20 m/s .

Figure options

Figure 4 shows the dependence of wind speed and direction residuals on buoy wind speed for HY-2A SCAT, ASCAT and OSCAT. Only buoy wind speed data from 1 m/s to 20 m/s were collected. There was no common systematic dependence for either wind speed or direction residuals on buoy wind speed over the entire range. The wind speed residual for HY-2A SCAT is shown in Fig. 4a. It is obvious that HY-2A SCAT overestimated for low winds. For medium and high winds, there was a negative bias with comparison to buoy wind data. This is consistent with the regression line in Fig. 2a and results in Table 2. The wind direction residual is shown in Fig. 4b. As discussed above, HY-2A SCAT has low accuracy with large st and ard deviation for low winds. For buoy wind speeds stronger than 5 m/s, the accuracy of wind direction retrieval improved greatly. ASCAT also overestimated in wind speed retrieval for low winds and underestimated for high winds(Fig. 4c). Nonetheless, ASCAT performed better, with smaller wind speed bias and st and ard deviation in almost all buoy wind speed intervals. Considering the calculated results(Table 2 and Fig. 3c), we conclude that ASCAT measures sea surface wind speed with high accuracy and stability. Furthermore, as shown in Fig. 4d, we found that ASCAT had satisfactory performance in wind direction retrieval, except for low wind speeds(＜5 m/s). This agrees with the results in Table 2 and Fig. 3d. For OSCAT, wind speed was underestimated for most buoy wind speed bins(Fig. 4e). Although the bias was smaller than HY-2A SCAT in most buoy wind speed intervals, there was a relatively high st and ard deviation. OSCAT shows the same characteristic as HY-2A SCAT and ASCAT, i.e., low accuracy for wind direction(Fig. 4f). However, the OSCAT st and ard deviation in each wind speed interval appears the largest of the three sensors.

Fig. 4 Dependence of wind speed (left) and direction (right) residuals (SCAT minus buoy) on buoy wind speed for HY-2A SCAT (a and b), ASCAT (c and d) and OSCAT (e and f) wind products. Number of collocation, bias, and standard deviation were calculated for every 1 m/s buoy wind speed bin from 1 to 20 m/s.

Figure options

Figure 5 shows the dependence of wind speed and direction residuals on cross-track location of wind vector cells for HY-2A SCAT, ASCAT and OSCAT wind products. The number of collocation for each wind vector cell less than 10 was discarded. Figure 5a and b shows no systematic dependency on cross-track location for ASCAT. The same held for HY-2A SCAT and OSCAT. However, this result does not confi rm the conclusion reached in other research that the new beam geometry with two conically rotating pencil beams results in less accuracy of QuikSCAT in measuring winds at near nadir and the outer swath(Stiles et al., 2002; Bourassa et al., 2003), because HY-2A SCAT and OSCAT are also rotating pencil beam radars. Although DIRTH techniques have been used with OSCAT to reduce relatively large uncertainty at near nadir and the outer swath, the result should still be less accurate at the outer swath. In addition, our results correspond to those from validation of HY-2A SCAT(Wang et al., 2013) and OSCAT(Singh et al., 2012; Sudha and Prasada Rao, 2013). Based on this, we doubt that various types of noise appear during the measurement and data preprocessing processes, and the algorithms for wind retrieval should be improved.

图 5 中文标题 Fig. 5 Dependence of wind speed (left) and direction (right) residuals (SCAT minus buoy) on location of wind vector cells for HY-2A SCAT (a and b), ASCAT (c and d) and OSCAT (e and f) wind products Number of collocation, bias, and standard deviation were calculated for cross-track position of each cell.

Figure options

Figure 6 shows seasonal dependence of wind speed and direction residuals for HY-2A SCAT, ASCAT and OSCAT wind products. Season refers here to the Northern Hemisphere. As shown in the Fig. 6, no clear seasonal variation for either wind speed or direction of any scatterometer. There was underestimation of wind speed, except for overestimation in winter by ASCAT. HY-2A SCAT and ASCAT had a relatively large negative bias in summer. However, there was a weak fl uctuation for OSCAT(Fig. 6e). Regarding wind direction, all three scatterometers underestimated in all seasons. OSCAT had the smallest bias in autumn relative to the other seasons(Fig. 6f). There was no similar characteristic for either HY-2A SCAT or ASCAT; their performance remained relatively steady over the four seasons.

Fig. 6 Seasonal dependence of wind speed (left) and direction (right) residuals (SCAT minus buoy) for HY-2A SCAT (a, b), ASCAT (c, d) and OSCAT (e, f) wind products Number of collocation, bias, and standard deviation were calculated for each season (season referring to Northern Hemisphere).

Figure options

The validation results show that the HY-2A SCAT, ASCAT and OSCAT all performed well in wind speed retrieval, with RMSEs all less than 2 m/s. However, for wind direction, although ASCAT met the requirement, it had poor performance in wind direction retrieval, especially for low winds. Analysis of the dependence of RMSE on buoy wind speed showed a high-low pattern of wind direction. A common abnormal b and was found between 14 and 15 m/s, at which the RMSE of wind direction suddenly peaked. Further research into this fi nding should be undertaken, since the reason remains unclear. There was no systematic dependence of either wind speed or direction residuals on buoy wind speed and location of wind vector cells over the entire range. We did not fi nd the relatively large uncertainty at near nadir and the outer swath for HY-2A SCAT and OSCAT that has been reported in other studies. Also, no clear seasonal variation was observed for any scatterometer.

and stability in both wind speed and direction retrieval, and its wind product can be effectively applied in scientifi c research. Compared with ASCAT, wind products of the recently launched HY-2A SCAT are less accurate. Nevertheless, its measurements nearly met the technical target, and it has potential for wide use in the future. OSCAT wind products are somewhat less satisfactory, especially in wind direction retrieval. Substantial improvements are required if the quality of wind direction measurement is to be enhanced.

The authors thank the NDBC, NSOAS, EUMETSAT and NRSC for providing buoy data and HY-2A SCAT, ASCAT and OSCAT wind products. In addition, the authors thank ZHANG Haifeng, JIANG Haoyu, WANG Zhixiong, and ZHAO Yong for their help in downloading data and their advice on improving the data processing algorithm.