A comparison of multiplatform wind products in the South China Sea during summer and autumn in 2019*

Yuxin LIU , Mingsen LIN , Xingwei JIANG , Xiujun SUN , Xiangzhou SONG

1 Key Laboratory of Marine Hazards Forecasting, Ministry of Natural Resources, Hohai University, Nanjing 210098, China

2 National Satellite Ocean Application Service, Beijing 100081, China

3 Key Laboratory of Space Ocean Remote Sensing and Application, Ministry of Natural Resources, Beijing 100081, China

4 Physical Oceanography Laboratory, Ocean University of China, Qingdao 266100, China

5 College of Oceanography, Hohai University, Nanjing 210098, China

Abstract Sea surface wind (SSW) observations from a newly developed “Black Pearl” wave glider,the Chinese-French Oceanography Satellite (CFOSAT), the HY-2A microwave scatterometer, and a recently released high-resolution atmospheric reanalysis (ERA5) are evaluated with respect to in-situ buoy observations (115.46°E, 19.85°N) from the South China Sea. Buoy observations from June to November 2019 are used to evaluate the wind estimates from the diff erent platforms. The comparisons show that the HY-2A and CFOSAT scatterometer wind speeds have mean root mean square errors (RMSEs) of approximately 1.6 and 1.6 m/s, respectively, and the corresponding mean wind direction RMSEs are approximately 19°and 17°, which indicates that these satellite retrievals meet the requirements of design engineering missions.The wind speed and wind direction RMSEs of ERA5 are approximately 1.9 m/s and 33°, respectively. The correlation coeffi cients between the HY-2A, CFOSAT, and ERA5 wind speeds and the buoy observations are 0.86, 0.85, and 0.84, respectively, and the corresponding coeffi cients of the wind direction are 0.98, 0.98,and 0.93, respectively, at a 95% conf idence level. However, the wind sensor in the wave glider provides relatively poor-quality observations compared with the buoy measurements and has higher wind speed and wind direction RMSEs of 2.9 m/s and 50.1°, respectively. Taylor diagrams are utilized to illustrate comprehensive wind comparisons between the multiplatform observations and buoy observations. The results help identify the basic biases in SSWs among diff erent products and enhance conf idence in the future use of SSW data for studies of upper ocean dynamics and climate analysis. Suggestions are also off ered to help improve the design of next-generation wave gliders.

Keyword: sea surface wind (SSW); calibration; South China Sea (SCS); wave glider; f ifth European Centre for Medium-Range Weather Forecasts reanalysis (ERA5); HY-2A; Chinese-French Oceanography Satellite (CFOSAT)

1 INTR O DUCTION

Sea surface winds (SSWs) are important for the study of marine dynamic environments. For example,the basin-scale Ekman pumping associated with the wind stress curl (Pedlosky, 1987; Huang, 2010) drives general ocean circulation. Furthermore, the Ekman current, spiral and transport, which are fundamental processes of upper ocean dynamics, correspond to local wind vectors (Ekman, 1905; Price et al., 1986).The ocean mixing generated by surface winds,especially strong tropical cyclones (TCs, Sriver and Huber, 2007), can enhance vertical heat transport and penetration, helping maintain the global energy cycle.In addition, winds also play a central role in determining air-sea turbulent heat f luxes, namely,latent and sensible heat f luxes (Yu and Weller, 2007;Yu, 2019), and help establish a new water vapor/temperature balance at the air-sea boundary layer in terms of wind turbulence, thereby aff ecting air-sea heat exchanges. The increase in wind speed over the past few decades has resulted in global increases in evaporation and latent heat f lux (Yu and Weller,2007). Moreover, the wind anomaly associated with atmospheric oscillations (Cayan, 1992; Thompson and Wallace, 2000a, b; Song and Yu, 2012) modif ies the air-sea thermal contrast in terms of wind-induced ocean and atmospheric processes. Li and Carbone(2012) also pointed out the importance of the SSW generated by mesoscale sea surface temperature(SST) gradients, which may lead to the initiation of local convection in the Tropical Western Pacif ic and is further organized at the Madden-Julian Oscillation(MJO, Madden and Julian, 1971, 1972) scale (Carbone and Li, 2015). Above all, SSW is used for numerical modeling of the ocean and atmosphere for weather and wave forecasts, biophysical interactions, and climate studies (Zhang et al., 2006).

Conventional observation platforms mainly include ships and buoys. The SSWs obtained from conventional platforms are very limited, and the spatial and temporal distributions of the data are not uniform; hence, it is diffi cult for conventional SSW data to meet the needs of ongoing research. Since the 1970s, the development of satellite remote sensing has provided a more eff ective technique than conventional means for continuous global SSW observations. As an active microwave radar dedicated to monitoring SSWs at the global scale, the spaceborne microwave scatterometer can achieve relatively accurate retrievals of the SSW speed and direction over a swath of more than 1 000 km and a spatial resolution of 10 km. Accordingly, this instrument represents an important satellite sensor for observing SSWs. Reanalysis products, which are advantageously characterized by long time series and high temporal resolutions, are also an important source of SSW data.The products from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the United States National Centers for Environmental Prediction(NCEP) are commonly used reanalysis data in oceanatmosphere studies. In addition, the wave glider is a newly developed device for ocean observations within the past decade. With their remarkable endurance and survivability, wave gliders have become increasingly popular for research, and related products have been widely used in marine science and engineering (Hine et al., 2009; Manley and Willcox,2010; Daniel et al., 2011; Bingham et al., 2012).

However, calibration is essential to the production and application of SSW products. Comparing and analyzing the errors of SSWs through calibration can help us select a more appropriate SSW dataset and enhance our conf idence in its application. A large number of studies have been conducted to evaluate the SSW data quality by comparing these SSW data with in-situ observations. Previous studies on the quality of scatterometer data have considered the Seasat-A Satellite Scatterometer (SASS, Jones et al.,1982), European Remote Sensing Satellite (ERS)Active Microwave Instrument (AMI, Bentamy et al.,1994), Advanced Earth Observing Satellite (ADEOS)NASA Scatterometer (NSCAT, Bourassa et al., 1997;Ebuchi, 1999; Dickinson et al., 2001), Quick Scatterometer (QSCAT) SeaWinds instrument(Ebuchi et al., 2002; Chelton and Freilich, 2005),Meteorological Operational Satellite Programme(MetOp) Advanced Scatterometer (ASCAT, Verspeek et al., 2010) and HY-2A Scatterometer (SCAT, Li and Shen, 2015; Zhao and Zhao, 2019). Similarly, to evaluate the quality of SSW reanalysis data, the NCEP National Center for Atmospheric Research(NCAR, Ladd and Bond, 2002), NCEP Climate Forecast System Reanalysis (CFSR, Peng et al.,2013), Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA2,Kim et al., 2018), ERA40 (Höglund et al., 2009), and ERA-Interim (Song et al., 2015) datasets have been considered benchmarks for climate analysis.

Most of the abovementioned studies evaluated satellite scatterometers launched many years ago and employed the previous generation of reanalysis data.In contrast, the data from newly developed SSWs such as wave glider observations, data from the Chinese-French Oceanography Satellite (CFOSAT), and the f ifth ECMWF reanalysis (ERA5) dataset have not been fully examined. Especially in the South China Sea (SCS), the applicability of these new data is not yet fully understood. Accordingly, the errors of HY-2A, CFOSAT, ERA5 and wave glider SSW data are analyzed and evaluated with respect to the buoy in this paper, and a reasonable reference and basis are provided for the selection of SSW datasets in the SCS.

The remainder of this paper is organized as follows.In Section 2, the datasets, namely, the wave glider observations, HY-2A measurements, CFOSAT measurements, and ERA5 reanalysis, and the data processing method are brief ly introduced, and buoy observations are used to calibrate the SSW estimates from diff erent platforms. Then, errors in themultiplatform observations and a detailed analysis and discussion based on the applicability of SSW data are presented in Sections 3 & 4. Finally, the conclusions and suggestions for future research are given in Section 5.

Table 1 Detailed information of the wave glider, HY-2A, CFOSAT, and ERA5 datasets

2 DATA AND METHOD

2.1 Data

Five types of SSW data are used in this article:buoy observations, “Black Pearl” medium-size wave glider observations, HY-2A measurements, CFOSAT measurements, and ERA5 reanalysis data. In general,the in-situ data provided by the World Meteorological Organization Information System (WMO WIS) are assimilated to ERA5 to improve the predictive ability in numerical weather predictions. Satellite scatterometer data are also assimilated into ERA5,including AMI, ASCAT, Oceansat-2 scatterometer(OSCAT), and SeaWinds. However, the near-coast buoy observations and satellite achievement used in this study have not been assimilated to ERA5. Thus,HY-2A, CFOSAT, buoy, and ERA5 can be regarded as independent resources for comparison. It is expected that newly achieved data products can be assimilated into reanalysis systems to improve the ability to predict ocean and climate states in the future. The ERA5 reanalysis has assimilated more proven wind products such as ASCAT. In this study,the wind in ERA5 will be compared with buoy observations, instead of the ASCAT wind. The detailed information of the four SSW datasets to be evaluated is shown in Table 1.

2.1.1 Wave glider

The “Black Pearl” medium-size wave glider used in this study was an ocean mobile observation platform with autonomous navigation capability, and it was developed by Ocean University of China and Tianjin Polytechnic University, China. It has many advantages. Its kinetic energy was generated from waves. The energy used for measurement, control,navigation, and communication systems was generated in solar technology. The observations could be conducted in a long period and long distance. It was very tolerant to extreme sea conditions.Observation jobs could be completed by various sensing equipment, such as sonar, weather stations,underwater cameras, conductivity, temperature, and depth (CTD) sensors, hydrophones, and biological or chemical detection instruments. Measurements of airsea interface environment could be performed on large spatial scales via an iridium communication system, a global positioning system, and an autonomous navigation control system (Sun et al.,2019). In addition, a micro weather station for the“Black Pearl” called AirMar 200WX-IPX7 was used,and the data were transmitted by the iridium communication system. The wind speed accuracy of the sensor was 0.5 m/s±10% (0-5 m/s) or 1 m/s ±5%(5-40 m/s) in resolution of 0.1 m/s, and the wind direction accuracy was 5° (2-5 m/s) or 2° (5-40 m/s)in resolution of 0.1° (Li et al., 2017). The observations from the wave glider temporally ranged from June 15 to August 30, 2019, and the trajectory of the wave glider is shown in Fig.1 as the solid black line.

2.1.2 HY-2A

Fig.1 The trajectory (black line) of the wave glider from June 15 to August 30, 2019, and the location of buoy SF304 (purple square) affi liated with the Ministry of Natural Resources (MNR) in the SCS

The HY-2A satellite was successfully launched in China on August 16, 2011, and is equipped with a radar altimeter, a microwave scatterometer, a scanning microwave radiometer, and a calibrated microwave radiometer. In particular, the microwave scatterometer has a swath width exceeding 1 700 km, a spatial resolution of 25 km×25 km, a wind speed measurement range of 2-24 m/s, a wind speed measurement accuracy of 2 m/s, and a wind direction measurement accuracy of 20° (Jiang et al., 2012). Since the experimental period of the wave glider spanned from June 15 to August 30, 2019, the HY-2A data within the period from June 15 to November 31, 2019, were used to match the observation period and guarantee a suffi cient number of matching samples. HY-2A data were obtained from National Satellite Ocean Application Service (NSOAS, https://osdds.nsoas.org.cn/#/?lang=en).

2.1.3 CFOSAT

CFOSAT was jointly developed by China and France and was successfully launched on October 29, 2018. The satellite is equipped with two payloads:the Surface Waves Investigation and Monitoring(SWIM) instrument and SCAT. SCAT is the f irst rotating fan-beam scatterometer for the acquisition of global ocean vector wind observations, and SWIM is the f irst space radar dedicated to measuring the directional spectrum of ocean waves through a method involving real-aperture radar with six incidents. The combination of the SCAT sensor with rotating fan beams and the SWIM instrument with multi-incident beams is of great value for investigating the processes of marine dynamic environments and surface scattering characteristics. SCAT has a swath width exceeding 1 000 km, and it takes approximately 3 d to cover the globe (Earth Observation Portal,2020, https://directory.eoportal.org). There are two CFOSAT SSW products with spatial resolutions of 25 km×25 km and 12.5 km×12.5 km, and the latter high-resolution dataset is selected in this paper. In addition, the wind speed measurement range is 4-24 m/s, the wind speed measurement accuracy is 2 m/s, and the wind direction measurement accuracy is 20° (National Satellite Ocean Application Service,2020a). The CFOSAT data employed herein range from June 15 to November 31, 2019. CFOSAT data were obtained from NSOAS (https://osdds.nsoas.org.cn/#/?lang=en).

2.1.4 ERA5

ERA5 was released in 2016 and is the f ifthgeneration reanalysis of global climate and atmospheric data provided by the ECMWF. Compared with the previous versions of the reanalysis product,ERA5 provides a higher spatial resolution, hourly estimates of atmospheric variables, and a consistent representation of uncertainty. Moreover, ERA5 uses more satellite observations compared to earlier products. Transitioning from ERA-Interim to ERA5 represents an improvement in the overall quality and level of detail (Hersbach et al., 2019). The 10-m SSW of ERA5 is selected for this study. The spatial resolution is 0.25°×0.25°, and the temporal resolution is 1 h. Consistent with the observation periods of the wave glider and scatterometer data, the ERA5 data range from June 15 to November 31, 2019.

Table 2 The excluded numbers of observations

2.1.5 Buoy

Because HY-2A, CFOSAT, and ERA5 all provide global wind f ield data, a buoy that can match the trajectory of the wave glider must be selected. Hence,a 6-m medium-sized ocean data buoy, buoy SF304 that was deployed by the Ministry of Natural Resources in the SCS, was selected for this research. Buoy SF304 mainly carried out observations of the wind speed,wind direction, air temperature, air pressure, humidity,precipitation, wave height, wave period, wave direction, sea surface temperature, and ocean current,among other variables. The location of the buoy was 115.46°E/19.85°N as shown as a purple square in Fig.1. The buoy’s anemometer was an R. M. Young wind speed and wind direction sensor. The observed wind speed was in hours, and the measurement range was 0-100 m/s. The wind speed accuracy was±0.3 m/s, and the wind direction accuracy was ±3°.The observational height of the anemometer was 7.4 m.

2.2 Processing method

2.2.1 Data processing

First, the quality of the data needs to be controlled.According to the wind speed observation range of each SSW product, records outside the observation range and the wind vector cell (WVC) marked by rainfall in the scatterometer observations were excluded. The sigma values used in each WVC for the retrieval were ensured to be suffi ciently high quality.Furthermore, the wind speed and wind direction observations from the buoy and wave glider were checked manually to ensure the credibility and rationality of the data. In addition, when the diff erence of wind directions between the scatterometer and buoy was more than 45°, the scatterometer estimate was rejected as having poor WVC ambiguity selection(Plagge et al., 2012). Due to the poor representativeness of low wind speeds and the quality control process of scatterometer, a slightly higher proportion of samples were not used for calibration. The excluded numbers of observations are shown in Table 2.

The wind directions determined by HY-2A and CFOSAT are def ined as 0° due north and increasing clockwise, whereas those observed by the buoy, wave glider, and ERA5 are expressed according to the meteorological convention, that is, def ined as 0° due south and increasing clockwise. Therefore, there is a 180° diff erence between the wind directions of HY-2A and CFOSAT and those of the other platforms. To unify the wind directions among the various datasets,the def inition of the wind direction according to the meteorological convention was used in this study.

The buoy and wave glider measured the actual wind speed at heights of 7.4 m and 1.5 m above the sea surface, respectively. For HY-2A, CFOSAT, and ERA5, the reference height of the wind vector was 10 m above the sea surface. Thus, the SSW data were converted to 10-m neutral winds using the Liu-Katsaros-Businger (LKB) model (Liu et al., 1979;Fairall et al., 2003).

The temporal resolution of the buoy was 1 h, and the spatial resolution of the CFOSAT data was 12.5 km; thus, the spatial and temporal windows for matching these data were selected as 12.5 km and 30 min, and the same windows were used for the HY-2A and ERA5 data. In contrast, the temporal resolution of the wave glider was 10 min. Therefore, a time window of 5 min was employed to match the wave glider data with the buoy data. Moreover, since there was only one buoy site (SF304) along the trajectory of the wave glider, the spatial window was increased to 100 km to prevent an excessively short observation duration of the collected samples.

2.2.2 Statistical metrics

To quantitatively evaluate the accuracies of the diff erent multiplatform products, several statistical metrics were applied to the collected samples. The Bias (Eq.1) and root mean square error (RMSE, Eq.2)were used to evaluate the uncertainty in the wind products, whereas the Pearson correlation coeffi cient(R, Eq.3) was used to analyze the correlations between the wave glider, HY-2A, CFOSAT, and ERA5 data and the buoy values:

Fig.2 Scatterplots of the wind speeds in the HY-2A, CFOSAT, wave glider, and ERA5 datasets and the corresponding observations from buoy SF304

3 RESULT AND DISCUSSION

3.1 Wind speed/direction error

Fig.3 Scatterplots of the HY-2A, CFOSAT, wave glider, and ERA5 wind directions and the corresponding observations from buoy SF304

Figure 2 shows scatterplots of the HY-2A,CFOSAT, wave glider, and ERA5 wind speeds with the corresponding buoy observations. The solid line at slope of 1 in each plot indicates that the wind speeds of the two platforms are equal to each other. Evidently,the HY-2A, CFOSAT, wave glider, and ERA5 wind speeds are all below 20 m/s. According to the wind speed error statistics, the wind speed biases of the HY-2A, CFOSAT, wave glider, and ERA5 datasets are -0.4, 0.5, -1.3, and -0.4 m/s, respectively, while the corresponding RMSEs are 1.6, 1.6, 2.9, and 1.9 m/s, and the correlation coeffi cients with the buoy observations are 0.86, 0.85, 0.63, and 0.84,respectively. In general, the RMSEs of the wind speed observed by HY-2A and CFOSAT are less than 2 m/s,and their observation accuracies reach the satellite design indexes. Moreover, the quality of the wind speed data from ERA5 is equivalent to the data quality from the two satellites, and the wind speed evaluation results of all three are similar. In contrast, the wind speed error of the wave glider is large, and the wind speed is lower than that of the buoy.

Similarly, Fig.3 depicts scatterplots of the HY-2A,CFOSAT, wave glider, and ERA5 wind directions with the corresponding buoy observations. Clearly,the wind directions observed by HY-2A, CFOSAT,and ERA5 are distributed mainly in the directions of 45° (southwesterly) and 225° (northeasterly).According to the statistical results of the wind direction errors, the HY-2A, CFOSAT, wave glider,and ERA5 wind direction biases were -1.4°, -1.9°,15.5°, and -0.16°, respectively, while the corresponding RMSEs were 18.9°, 17.3°, 50.1°, and 33.4°, and the correlation coeffi cients with the buoy observations were 0.98, 0.98, 0.76, and 0.93, respectively. In general, the RMSEs of the HY-2A and CFOSAT wind directions were both less than 20° and their observationaccuracies satisf ied the satellite design requirements.However, since ERA5 samples with a large direction deviation relative to the buoy observations were not f ilter out, the RMSE of the ERA5 wind direction was relatively large. There was also a signif icant deviation in the wind direction observed by the wave glider.

Table 3 Comparison results of the wind speed in summer and autumn

Table 4 Comparison results of the wind direction in summer and autumn

The wave glider observations were conducted mainly during the boreal summer, while the other observations lasted from June to November, covering the summer and autumn. Thus, the results were compared for summer and autumn (Tables 3 & 4)between which the air-sea boundary conditions can be diff erent. Based on this clarif ication, it was found that the HY-2A, CFOSAT, and ERA5 show poorer quality in the summer and better quality in the autumn.According to the above analysis of diff erent seasons,the observation time of the wave glider is in the summer, which may be one of the reasons for the poor quality.

The SCS lies in the tropical and subtropical regions off the southern part of the Asian continent. Compared with other sea areas, it is characterized by a signif icant tropical marine climate with short spring and autumn seasons, long summers, and prevailing southwestnortheast winds. The SCS is typically aff ected by the Asian southwest monsoon from June to August (with the autumn transition period in September) and the northeast monsoon period from October to March.According to the wind direction rose charts depicted in Fig.4, except for the wave glider, the wind direction distributions of HY-2A, CFOSAT, and ERA5 are similar to that of the buoy observations.

The reason for this discrepancy between the wave glider and buoy observations is that buoy SF304 was the only buoy deployed along the wave glider trajectory, and only 14 d (July 22 to August 4) of wave glider observations within 100 km of the buoy can be matched with the buoy observations. Therefore, the wave glider wind directions cannot represent the actual wind direction distribution during this period.Figure 4 shows that the wind speeds during the southwest monsoon are relatively weak (generally lower than 10 m/s), while those during the northeast monsoon have a higher intensity, and the wind direction is relatively stable. The abovementioned results may be caused by cold winter waves in the northern part of the SCS accompanied by high wind speeds. Moreover, the wind direction in summer is not as stable as that in winter due to the frequent occurrence of TCs.

3.2 Generalized analysis

To fully ref lect the evaluation results of the multiplatform SSWs and clarify the results, three statistical metrics, namely, the standard deviation(SD), RMSE, andRwere selected and plotted on standardized Taylor diagrams (Figs.5 & 6) for comparison with the buoy observations. The radial distance from the dot (buoy) represents the ratio of the SD of the SSW data to that of the buoy observations,where the SD of the data relative to the SD of the buoy observations represents the ability to simulate the center amplitude of the SSW data. The closer the SD ratio is to 1, the better is the simulation ability.The RMSE is the diff erence in magnitude between the buoy observation and the SSW data and is represented in the diagrams as a dotted green semiarc with the buoy point at the center of the circle. The closer the SSW data point is to the buoy, the closer are the values.Rwas determined by the azimuth of the SSW data. When the SSW data are consistent with the buoy observations, the closer is the SSW point to the buoy on the x-axis and the higher is the correlation between the SSW data and the buoy observation at that time,bearing the problem of the degrees of freedom.

Fig.4 Wind roses of the following pairs of sensors: collected buoy SF304 and HY-2A, collected buoy SF304 and CFOSAT,collected buoy SF304 and wave glider, and collected buoy SF304 and ERA5

Fig.5 Taylor diagram of the wind speed based on the HY-2A, CFOSAT, wave glider, ERA5, and buoy SF304 observations

Fig.6 Taylor diagram of the wind direction based on the HY-2A, CFOSAT, wave glider, ERA5, and buoy SF304 observations

Fig.7 Daily mean SSW based on the HY-2A, CFOSAT, wave glider and ERA5 data from July to August 2019

According to the above evaluation principles, the quality evaluations of the multiplatform SSW data are consistent with the results of the scatterplot analyses.Specif ically, in terms of the wind speed, HY-2A is slightly better than ERA5, ERA5 is slightly better than CFOSAT. Although there is little diff erence among the three, wave glider’s results are worse than those of other platforms. It should be pointed out that after 6 years of extended service, HY-2A has a larger proportion of excluded data in the quality control stage (Table 2), and the 12.5-km product of CFOSAT has a lower quality compared to the 25-km product(Lin et al., 2019); thus, the evaluation results may f luctuate within a relatively short period. In terms of the wind direction, CFOSAT is slightly better than HY-2A, HY-2A is slightly better than ERA5, and both are better than wave glider. The causes of the wave glider’s relatively poor performance will be examined in the next section.

3.3 Analysis of the wave glider results

As mentioned above, the quality assessment of the wave glider observations was unsatisfactory. The reasons for this outcome were analyzed. According to the observation of the wave glider, the virtual anchorage phase of the wave glider lasted from June 15 to July 17, 2019, and the navigation observation phase of the wave glider spanned from July 17 to August 30, 2019. Due to poor continuity of observations from the wave glider between June 15 and June 30, 2019, the period from July 1 to August 30, 2019, was selected for analysis, and the trajectory of the wave glider during this interval is shown in Fig.7. In Fig.7, the arrows represent the daily wind velocity (red arrows represent the wave glider, green arrows represent HY-2A, cyan arrows represent CFOSAT, and dark blue arrows represent ERA5).During this period, the starting point is located in the dense trajectory area (approximately 117°E). After sailing west to 115.5°E, the ship sailed east and turned northwest at approximately 119°E. It is obvious from Fig.7 that the quality of wave glider observations is poor and the wind speed is lower than those of the observations from other platforms during this period,which supports the above evaluation results.Furthermore, after heading east to 117.5°E, wind direction of the wave glider was relatively consistent with those determined by the other three platforms.However, due to the unavailability of suitable buoys and other in-situ observations for calibration, these glider-determined wind directions cannot be ref lected in the results.

4 DISCUSSION

The above experiments demonstrate that there are few diff erences in the accuracy evaluations among HY-2A, CFOSAT, and ERA5 in the SCS. A more appropriate SSW dataset should be selected according to the requirements, such as the temporal resolution and spatial resolution. Among the above datasets, the time interval of the ERA5 data is 1 h, which is suitable for research and applications requiring a high temporal resolution. The entire ERA5 dataset from 1950 to the present is expected to be available for use in 2020.The low antenna rotation speed on CFOSAT helps to ensure high redundancy for the back-scattering coeffi cient in azimuth data, which can further improve the spatial resolution of the measured unit to 12.5 km by employing the resolution enhancement technique(Wang et al., 2019); therefore, CFOSAT is more suitable for research and applications that demand relatively small-scale information. Additionally,compared with HY-2A, CFOSAT is more suitable for near-shore SSW measurements. However, since buoy SF304 is far from the coast, the advantage of CFOSAT for near-shore measurements was not ref lected in the results. Of course, the combination of observations from CFOSAT, HY-2A/B, ASCAT-A/B/C, and other scatterometers would greatly improve the observation frequency and help meet the demands for a better temporal resolution. At this stage, the quality of wave glider data is not stable enough because the comparison is constructed between a moving platform and a f ixed buoy location. However, the platform of wave glider has numerous advantages, including excellent maneuverability, a long operating duration,and a controllable f light path. In addition, wave gliders can track and observe phenomena under extreme sea conditions such as TCs.

5 CONCLUSION

SSW data from HY-2A, CFOSAT, ERA5 and a wave glider were evaluated and analyzed in reference to a buoy in the SCS during summer and autumn in 2019. The main conclusions are as follows. After quality control, the wind speed RMSEs of the HY-2A and CFOSAT scatterometer data are less than 2 m/s,and the wind direction RMSE is less than 20°, which meet the requirements of satellite missions. According to a wind speed evaluation, HY-2A is slightly better than ERA5 and CFOSAT, and CFOSAT shows diff erent qualities for unknown reasons, which will be further diagnosed. According to a wind direction evaluation, CFOSAT is slightly better than HY-2A,and HY-2A is better than ERA5. HY-2A, CFOSAT,and ERA5 all have similar performances in the SSW data evaluation, and the wind speeds and directions have similar statistical results, which are consistent with the seasonal characteristics of the SCS. Buoys are generally regarded as one of the most reliable observation platforms, although they are limited in the number and distribution of stations. ERA5 provides hourly SSW data over long time series,which can meet the increasing demands of ongoing climate research. However, it lacks the ability to analyze small-scale phenomena. Scatterometers have a higher spatial resolution and have advantages in large-scale ocean phenomena observations, but because the diurnal variation information cannot be captured eff ectively by a single satellite, network observations with multiple scatterometers will compensate for the inadequate temporal resolution,thereby enabling more practical needs to be satisf ied.Wave gliders can track ocean phenomena at long temporal and spatial scales; however, as a new type of observation device, the high-frequency SSW observations of wave gliders are less accurate compared to the buoy measurements. Gliders need to be used with strict quality control at present, but these devices will no doubt have great application prospects in the future.

In summary, calibration is an important and key step in promoting the application of marine data. This study can be regarded as the f irst step in performing wind data comparisons in China’s marginal seas.Through the validation in this paper, the spatial and temporal distribution and accuracy of the SSW characteristics and the generality of the SSW observations from a multiplatform in the SCS were evaluated within a limited time of about 6 months.However, in future studies, long-term comparisons of SSW and calibration of the satellite data in this region will be further conducted with the accumulation of satellite achievement and in-situ observations. Due to the limited availability and eff ectiveness of in-situ observations, such as those provided by buoys, the evaluation results are inevitably aff ected by representative errors in the data. There are inevitable systematic observational uncertainties because of the sensor accuracies on buoys; therefore, the quality of the buoy data has a great inf luence on the validation results. In addition, the diffi culty of matching the wave glider observations with the buoy observations indicates that more attention should be paid to the quality control and manual verif ication processes using the limited collected data. At a scale of 10 km,serious errors may occur in the reanalysis data. In this case, the buoy and satellite observation data can be used as a reference for the validation, as a supplement to the reanalysis data, and as a method for correcting errors in the reanalysis data.

Chinese scientists have made great eff orts towards wave glider advancements so that this technology can be used in marginal seas (e.g., SCS, Sun et al., 2019).In this paper, the comparisons demonstrate that the observed SSW quality based on the wave glider information is suffi cient relative to the buoy observations, although the observation positions change over time. This study found that wave gliderbased observations require additional improvements.In the future, wave gliders will be designed with more accurate sensors that can coordinate with the attitudes of the moving trajectory. An improved accuracy of air-sea variables can be obtained with more accurate sensors and equipment that present reliable stability and long survivability. For a better understanding of the observational uncertainties of wave gliders, a comprehensive comparison project that includes a combination of the satellite retrievals, buoy observations, and atmospheric and oceanic reanalysis data will be promoted. In the next step, the air-sea variables in cooperative multiple wave gliders will be compared with those based on the Bailong buoy(Freitag et al., 2016), which is already planned and under construction. It is expected that the wave glider air-sea observations will provide a stable and comprehensive data source for marine studies and for forecasting and prediction.

6 DATA AVAILABILITY STATEMENT

The datasets generated and/or analyzed during the current study are available from the corresponding author on reasonable request.

7 ACKNOWLEDGMENT

The authors would like to thank National Satellite Ocean Application Service (NSOAS) for HY-2A and CFOSAT scatterometer data, the National Marine Environmental Forecasting Center (NMEFC) for buoy data, and the European Centre for Medium-Range Weather Forecasts (ECMWF) for ERA5 reanalysis data.