Can Sea Surface Waves Be Simulated by Numerical Wave Models Using the Fusion Data from Remote-Sensed Winds?

Abstract

The purpose of our work is to investigate the performance of fusion wind from multiple remote-sensed data in forcing numeric wave models, and the experiment is described herein. In this study, 0.125° gridded wind fields at 12 h intervals were fused by using swath products from an advanced scatterometer (ASCAT) (a Haiyang-2B (HY-2B) scatterometer) and a spaceborne polarimetric microwave radiometer (WindSAT) during the period November 2019 to October 2020. The daily average wind speeds were compared with observations from National Data Buoy Center (NDBC) buoys from the National Oceanic and Atmospheric Administration (NOAA), yielding a 1.66 m/s root mean squared error (RMSE) with a 0.81 correlation (COR). This suggests that fusion wind was reliable for our work. The fusion winds were used for hindcasting sea surface waves by using two third-generation numeric wave models, denoted as WAVEWATCH-III (WW3) and Simulation Wave Nearshore (SWAN). The WW3-simulated waves in the North Pacific Ocean and the SWAN-simulated waves in the Gulf of Mexico were validated against the measurements from the NDBC buoys and the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA-5) for the period June−September 2020. The analysis of significant wave heights (SWHs) up to 9 m yielded a < 0.5 m RMSE with a > 0.8 COR for the WW3 and SWAN models. Therefore, it was believed that the accuracy of the simulation using the two numeric models was comparable with that forced by a numeric atmospheric model. An error analysis was systematically conducted by comparing the modeled WW3-simulated SWHs with the monthly average products from the HY-2B and a Jason-3 altimeter over global seas. The seasonal analysis showed that the differences in the SWHs (i.e., altimeter minus the WW3) were within ±1.5 m in March and June; however, the difference was quite significant in December. It was concluded that remote-sensed fusion wind can serve as a driving force for hindcasting waves using numeric wave models.

1. Introduction

In the past several decades, remote sensing techniques operating at microwave frequencies have been well developed and aimed at sea surface dynamic monitoring. Sea surface waves play a critical role in oceanic dynamics for determining the heat and motion exchange at the air−sea layer. Nowadays, remote-sensed products, including sea surface wind and waves, are operationally released with a 1–2 day delay, and they are valuable sources for analyzing sea climates [1]. With more and more earth observation satellites launching, there is particular importance in their marine applications using remote-sensed data, especially for forecasting and hindcasting atmospheric and oceanic dynamics.

Until now, all-weather winds over global seas have been measured by satellites carrying active microwave scatterometers [2], i.e., quick scatterometers (QuikSCATs) [3], advanced scatterometers (ASCATs) onboard the Metop-A/B/C [4], scatterometers onboard the Haiyang-2B/C/D (HY-2B/2C/2D) satellites [5], and the Chinese–French Oceanography SATellite (CFOSAT) [6], as well as microwave radiometers, such as WindSAT [7]. The spatial resolution of standard ocean wind products is 12.5 km, with a swath of approximately 100 km, and thus, there are gaps on the surface of the ocean. The limitation of scatterometer wind is undetectable under extreme meteorological conditions (>25 m/s), such as tropical cyclones (TCs), because the backscattering signals from sea surfaces suffer a saturation problem similar to synthetic aperture radars (SARs) during co-polarization [8,9]. This is due to both scatterometers and SARs utilizing the same geophysical model function (GMF) for wind retrieval [10,11]. Validation against moored buoys has shown that the accuracy of scatterometer-measured wind speed is approximately 1 m/s [12]. In addition, the spaceborne polarimetric microwave radiometer WindSAT, developed by the Naval Research Laboratory (NRL), has the ability to measure 0.25° gridded sea surface wind vectors with a swath coverage of 350 km [13]. The advantage of WindSAT is that the maximum wind speed could reach 40 m/s in a TC [14]. The wave products (i.e., SWHs) are operationally released by altimeters, i.e., the Topography Experiment (TOPEX)/Poseidon [15], the Jason family [16,17], and the Sentinel-3A/3B [18]. Although altimeter-measured waves are reliable for large-scale analyses, the spatial resolution of approximately 10 km following the satellite footprints does not satisfy the requirement for regional seas. An SAR could provide the wave spectrum with a fine spatial resolution (approximately 3 km), with a swath coverage of 100–300 km [19], and the Surface Wave Investigation and Monitoring (SWIM) [20] module onboard the CFOSAT is designed to obtain the wave spectrum with a spatial resolution of 18 km over global seas, and it also operates in TCs [21,22]. The validation of remote-sensed wind and wave products has been systematically studied; however, the applications for using these data are worthy of further study.

Oceanic modeling is a useful technique for research on oceanography. Several numeric models have been developed, i.e., a wave model called the WAve Model (WAM) [23], as have hydrodynamic models, including the Finite-Volume Community Ocean Model (FVCOM) [24], the Stony Brook Parallel Ocean Model (sbPOM) [25], and the Hybrid Coordinate Ocean Model (HYCOM) [26]. The sea surface wind field is the main driving force of sea surface dynamics, i.e., ocean waves, circulation, and eddies. Under this circumstance, a well-calibrated wind field is crucial for oceanic modeling. Several types of wind data derived from meteorological numerical models are convenient to use, i.e., the European Centre for Medium-Range Weather Forecasts (ECMWF) Interim (ERA-Interim) data [27], the Cross-Calibrated Multi-Platform (CCMP) from Remote Sensing Systems (RSS) data [28], and the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) data [29]. In the above data, winds are implemented in two wave numerical models, that is, WAVEWATCH-III (WW3) [30], which was developed by the National Centers for Environmental Prediction (NCEP) of the National Oceanic and Atmospheric Administration (NOAA), and Simulation Wave Nearshore (SWAN) [31], which was developed by the Delft University of Technology. It was revealed in [32] that WW3, using 1° gridded CCMP winds that are assimilated by ERA-Interim reanalysis winds and observations, had the best performance in wave simulation for the Western Pacific Ocean. However, two aspects of employing those winds must be noted: (1) the inevitable error in the meteorological numerical model and (2) the under-estimation in the TCs [33]. These problems could be solved by using remote-sensed wind products that are regarded as real observations. Moreover, whether waves are predictable by numeric wave models using fusion wind derived from radiometers and scatterometers and their levels of accuracy are interesting questions to explore. Up to now, a modeling wave is usually forced by the result of a meteorological model, and remote-sensed products have been implemented for sea surface dynamics prediction by using artificial intelligence methods [34,35]. The applicability of fusion wind from radiometers and scatterometers to force numeric wave models has not been reported; therefore, an experiment utilizing multiple satellites is carried out in order to classify this issue.

The remainder of this paper is organized as follows: Section 2 presents the satellite products, i.e., the wind data from ASCAT, HY-2B, and WindSAT; the wave data from Jason-3; and the observation data from the National Data Buoy Center (NDBC) buoys of the NOAA. The methodology for the fusion of the multiple satellite-measured winds during the period November 2019 to October 2020 is presented in Section 3. Additionally, the model settings of the numeric wave models (i.e., WW3 and SWAN) are briefly introduced in Section 3. The validation of the fusion wind and simulated waves against the NDBC buoy data is exhibited in Section 4, and the hindcasting waves are evaluated with the products from the HY-2B and Jason-3 altimeters in this section. Section 5 presents the conclusions and prospects.

2. Datasets

In this section, the available datasets, including winds from scatterometers and microwave radiometers, waves from altimeters, and observations from NDBC buoys, are briefly described.

2.1. Remoted-Sensed Winds

The task of a scatterometer is to measure global wind fields. The first spaceborne scatterometer mission was SeaWinds on the QuikSCAT satellite, which operated during 2002−2012. Since 2007, as the successors of QuikSCAT, the ASCATs onboard the Metop-A/B/C satellites have been releasing official products. Typically, ASCAT winds have a 12.5 km spatial resolution and two 550 km wide swath coverages in ascending/descending directions. The HY-2B scatterometer has also been releasing wind products along the same track since 2018. As concluded in [5], the HY-2B geophysical data records (GDRs) after systematic corrections using the MOE determination method have high accuracy. Scatterometer-based winds are valuable sources for the development and validation of SAR wind retrieval algorithms [11,36,37] due to their high levels of quality and fine spatial resolutions. The main limitation of ASCATs and the HY-2B is that they are unable to detect strong winds (>25 m/s). The spaceborne polarimetric microwave radiometer (i.e., WindSAT, which operated from January 2003 to November 2020) had the ability to monitor winds of up to 40 m/s [38]. Although the spatial resolution of WindSAT was 25 km, its swath coverage reached 350 km, allowing it to obtain wind fields in TCs, which benefited the research on TC dynamics [39,40]. Figure 1 depicts the wind maps from the ASCATs onboard the Metop-A and HY-2B scatterometers and the maps from WindSAT for 7 September 2020 between (10°N, 100°E) and (60°N, 180°E). In this case, it can clearly be observed that there are gaps between the along-track swaths. Thus, it was necessary to reconstruct the wind field by fusing the three types of remote-sensed data.

2.2. Waves from Altimeters and ECMWF

The SEASAT launched in 1978 was the first satellite to carry an altimeter [41], and it demonstrated the feasibility of sea surface height measurements. Since the 1980s, spaceborne radar altimeters have been rapidly developed worldwide, e.g., the Jason-2/3 belonging to the Ocean Surface Topography Mission (OSTM) and the HY-2 satellite constellation [42]. Altimeter-measured sea surface heights have spatial coverages of 10 km and follow the satellites’ footprints. The SWHs are estimated by the sea surface heights and the large-scale current speeds are derived from the sea surface height anomalies. Therefore, altimeters create highly detailed measurements for sea surfaces, which allow researchers to gain insights into ocean circulation and climate change. The cooperative satellite Jason-3 was launched in 2016 by international agencies (NOAA, CNES, and EUMETSAT), and the well-calibrated measurements from Jason-3 are popular and are used for the cross-validation of multiple altimeters [5,43]. The monthly average SWHs from the HY-2B and Jason-3 products during the period November 2019 to October 2020 were used for conducting the error analysis of the hindcasting waves using two numerical models. As an example, the along-track SWH map from Jason-3 on 15 May 2020 between (10°N, 100°E) and (60°N, 180°E) is shown in Figure 2. The ECMWF has continuously provided atmospheric and oceanic data for scientific research since 1979. Typically, two types of waves are available in the ECMWF datasets, i.e., ERA-Interim and ECWWF reanalysis (ERA-5). Although the spatial resolution of ERA-5 is a 0.25° grid, which is relatively coarser than the 0.125° grid of ERA-Interim, the accuracy of ERA-5 is greater, and it was combined with the observations of the operational system. Therefore, the waves from ERA-5 were also used for the analysis of the modeling results. Figure 2 shows the along-track SWH map from the HY-2B and Jason-3 altimeters, which was taken on 15 May 2020 between (10°N, 100°E) and (60°N, 180°E).

2.3. NDBC Buoys

The NDBC moored buoy networks in coastal waters provide real-time measurements of meteorological and oceanographic processes occurring directly on the sea surface [44], i.e., wind and waves. These recorded data are openly accessed for scientific research worldwide. Furthermore, its routine, continuous, and on-scene observations are critical for the error correction of weather and ocean forecasting, as well as for the validation of scatterometer, altimeter, SAR [45], and CFOSAT SWIM [46] data. In this work, the NDBC buoys in the East Pacific Ocean and the Gulf of Mexico were collected for the year 2020, and they were used to confirm the applicability of the fusion wind and hindcasting waves from the WW3 and SWAN models. We note that the wind data from the ASCATs and WindSAT were measured at a 10 m height above the sea surface, and thus, the following equation was employed to convert the sea surface wind speeds from the buoys into values at 10 m heights by using the logarithmically variable wind profiles [47]:

$$U(z_{10}) = U(z)\ln (\frac{z_{10}-z_0}{z-z_0}),$$

(1)

where U(z) is the buoy-measured wind speed at height z, U(z₁₀) is the wind speed at a 10 m height, and z₀ is the roughness length (assumed to be constant of 1.52 × 10⁻⁴).

3. Methodology

In this section, the fusion data method for assessing global winds using scatterometers and WindSAT is introduced. Then, the model settings of the WW3 and SWAN are exhibited, i.e., the forcing fields and selections of the parametrizations.

3.1. Method for Data Fusion and Evaluation Metrics

Because the along-track ASCAT, HY-2B, and WindSAT products are available in ascending and descending directions, the spatial resolution of the fusion wind fields over global seas is 12 h. In the literature, the spatial interpolation approach has been applied for multiple remote-sensed product fusions. In this study, the following interpolation algorithm was employed, denoted as a Cressman interpolation:

$$\overrightarrow{U}(\mathrm{lat},\mathrm{lon}) = {\overrightarrow{ U}}_0+\frac{1}{n}{\sum }_{k=1}^n\frac{D_k^2-R^2}{D_k^2+R^2}\mathsf{\Delta }{\overrightarrow{U}}_{10},$$

(2)

The spatial resolution of fusion wind was assumed to be a 0.125° grid. The advantage of this method is that the interpolation result can be adjusted by distance under the condition of ensuring accuracy, indicating that the spatial resolution of fusion wind can be artificially selected. In order to ensure that there were available data on all grids, the radius, R, was taken as 50 km, and there were n-element-referred wind vectors (${\overrightarrow{U}}_k$) from the scatterometers (ASCATs, HY-2B, and WindSAT) in both latitude and longitude directions. ${\overrightarrow{U}}_0$ is the initial ASCAT or HY-2B scatterometer wind vector nearest the grids, D_k is the spatial distance, measured in km, between the individual, referred and initial winds, $\mathsf{\Delta }{\overrightarrow{U}}_{10}$(=${\overrightarrow{U}}_0-{\overrightarrow{U}}_k$) is the difference between the individual referred and initial wind vectors, and $\overrightarrow{U}(\mathrm{lat},\mathrm{lon})$ represents the fusion wind vector at a gird of the latitude and longitude. We note that the WindSAT winds were only used at wind speeds smaller than 25 m/s, as applied above the interpolation, due to the winds from scatterometers suffering the saturation problem under greater wind-speed conditions. At wind speeds greater than 25 m/s, the measurements from WindSAT were directly used.

Three metrics, i.e., the bias, the root mean square error (RMSE), and the correlation coefficient (COR), were chosen to investigate the performances of the fusion data. The calculation formulations used were as follows:

$$\mathrm{Bias} = \frac{1}{n}{\sum }_{k=1}^n{X}_k-Y_k ,$$

(3)

$$\mathrm{RMSE} = \sqrt{\frac{1}{n}{\sum }_{k=1}^n{\left(X_k-Y_k\right)}^2 },$$

(4)

$$\mathrm{COR} = \frac{{\sum }_{k=1}^n\left(X_k-{\stackrel{\mathrm{-}}{X}}_k\right)\left(Y_k-{\stackrel{\mathrm{-}}{Y}}_k\right) }{\sqrt{{\sum }_{k=1}^n{\left(X_k-{\stackrel{\mathrm{-}}{X}}_k\right)}^2{\sum }_{k=1}^n{\left(Y_k-{\stackrel{\mathrm{-}}{Y}}_k\right)}^2 }},$$

(5)

where the matrices X and Y represent the evaluated and referred data, respectively; n is the sum of the samples; and ${\stackrel{\mathrm{-}}{X}}_k$ and ${\stackrel{\mathrm{-}}{Y}}_k$ represent the averages of the evaluated and referred data, respectively.

3.2. Model Settings of the Numerical Wave Models

Traditionally, wave research mainly uses on-scene observations, theoretical solutions, and statistical analyses. However, these methods are difficult to apply to wave forecasting and hindcasting predictions. Numerical simulation provides strong technical support for solving this problem. Since the 1960s, wave numerical models have gradually been developed, and their difficulties have been overcome by using the third-generation WAM model, i.e., nonlinear wave energy transmission and fragmentation, which was released in the 1980s. In recent years, based on the theoretical achievements in ocean waves, the third-generation wave models represented by WW3 and SWAN have also been continuously developed, and they follow the basic principles of the WAM. The scheme of the computational grid is different for the WW3 and SWAN models, i.e., a rectangle grid is employed in the WW3 and an unstructured grid is employed in the SWAN. Therefore, WW3 is suitable for wave simulation in large-scale oceans and has high computational efficiency, while SWAN has a good performance in regional seas due to its ability to meet indented coastlines.

The key of WAM Is Its solution to the wave propagation balance equation, which illustrates changes in wave energy in space and time, as follows:

$$\frac{D}{Dt}\left[N\left(k,\mathsf{\theta };x,t\right)\right]=\frac{E_{\mathrm{in}}+E_{\mathrm{bot}}+E_{\mathrm{db}}+E_{\mathrm{nl}}}{\sigma },$$

(6)

where N represents the wave action density spectrum in terms of the wave number, k; the polar direction, θ; the space, x; and the time, t; $\sigma$ is the intrinsic frequency; and the matrix, E, represents the change in the energy intensity, i.e., the wind input, E_in; the bottom fraction, E_bot; the wave decay term associated with wave-breaking (E_db); and the non-linear dissipation caused by wave−wave interactions (E_nl). The details of these sources are delicately described in the technical manuals for WW3 [48] and SWAN [49] users. According to [30], it has been found that sea surface currents and sea levels have an influence on wave simulations. Here, 0.125° gridded fusion wind at intervals of 12 h and daily average Copernicus Marine Environment Monitoring Service (CMEMS) sea currents and sea levels, with a spatial resolution of a 0.08° grid, were treated as the forcing fields. The water depths were derived from the General Bathymetry Chart of the Oceans (GEBCO). Figure 3 illustrates the water depths of the United States coastal waters overlaid by the geographic locations and identification numbers of the buoys used in this study. As a case study, Figure 4a,b depict the daily average CMEMS sea currents and sea levels map for 7 September 2020.

The parameterization scheme for the WW3 model [30,50] included the switch ST6 package for the input/dissipation source terms, the switch TR1 package, the switch GMD2 package for the wave−wave interactions, and the switch FLD2 and BT1 packages for the wave breaking and bottom friction, respectively. The parameterization scheme for the SWAN model [32] included WESTHuysen, that is, the propagation scheme that combines the wind and non-linear dissipation, and the switches QUADrupl, TRIad, and BREakinge for the wave−wave interactions and the wave bottom friction. In addition, other settings were default-employed, i.e., a directional resolution of 24 regular azimuthal directions with a 15° step and a computational resolution of 300 s time-steps in radial and latitudinal directions. The output of WW3 was an SWH with a 0.125° grid spatial resolution and 6 h intervals, and for the SWAN simulation, the SWH was an unstructured grid of approximately 1 km, with 6 h intervals.

The flowchart of the research framework is illustrated in Figure 5.

4. Results

The validation of the fusion wind against the NDBC buoys is first presented. Then, the hindcasting waves by the two numeric models are compared with the measurements from the NDBC buoys in the North Pacific Ocean and the Gulf of Mexico. Lastly, the error analysis is presented, where we evaluated the hindcasting waves with the monthly average products from the altimeters, including those of the HY-2B and Jason-3.

4.1. Validation of Fusion Wind

The ASCATs onboard the Metop-A/B/C satellites, the HY-2B scatterometer, and the WindSAT winds during the period November 2019 to October 2020 were fused by using the above interpolation method. Figure 6 shows the fusion wind map for 6 September 2020 at 12:00 UTC, where there were maximum wind speeds of up to 45 m/s, and some gaps and swath patterns were observed. This problem could be solved using products from more scatterometers. The 12 h average wind speeds from the fusion data for June−September 2020 were collected and compared with the measurements from the NDBC buoys that were marked in Figure 3. We noted that the distance between the grids and the available NDBC buoys was less than 5 km. The statistical analysis showed a 1.66 m/s RMSE, with a 0.81 COR for the wind speed, as depicted in Figure 7. Although the temporal resolution of the fusion wind was coarser than that of the atmospheric numerical simulation (i.e., it was 1 h for the ERA-5), the accuracy of the fusion wind was very satisfactory. Moreover, the swath patterns and the spatial resolutions of the fusion data could be further improved by using the multiple satellites in orbit, i.e., the ASCATs, HY-2C/2D, and Sentinel-3 satellites, such that the wind structure could be observed in detail. Thus, it was believed that the fusion wind field was reliable for this study.

4.2. Validation of Hindcasting Wave

As mentioned above, the WW3 performs well for large-scale oceans (i.e., the Western Pacific Ocean [51] and the Arctic Ocean [52,53]), and the SWAN is typically employed for wave simulations in coastal waters [54]. Therefore, the waves were simulated by using the WW3 model for the North Pacific Ocean in order to confirm the applicability of the fusion wind. Simultaneously, the waves in the Gulf of Mexico were simulated using the SWAN model. As examples, Figure 8a,b depict the corresponding SWH maps, i.e., the simulations of the WW3 model for the East Pacific Ocean for 07:00 UTC on 7 September 2020 and those of the SWAN model for the Gulf of Mexico for 06:00 UTC on 3 September 2020. Similarly, the ERA-5 wave maps are shown in Figure 9. Judging from a qualitative perspective, the patterns of the modeled waves were consistent with those of the ERA-5 waves, and in particular, the hindcasting waves had more details.

Figure 10a,b provide comparisons of the WW3-simulated SWHs with the observations of the NDBC buoys and the ERA-5 waves for June–September 2020, respectively, and the comparisons of the SWAN-simulated SWHs are exhibited in Figure 11. The statistical analysis yielded a < 0.5 m RMSE with a > 0.8 COR for SWHs up to 9 m. The accuracy levels of the simulations by using the WW3 and SWAN models were comparable with those forced by the numeric atmospheric models [32], i.e., the ECMWF and CCMP models. As reported in [32], the RMSE of SWH between the SWAN-simulated and the Jason-2 altimeter is about 0.7 m, which is worse than that in this study. However, the RMSE of SWH between the WW3-simulated and the Jason-2 altimeter is about 0.2 m. It is supposed that this is due to the lack of matchups at extreme sea states (SWHs > 6 m) [32]. Therefore, we think the fusion wind speed from radiometers and scatterometers is suitable to predict the waves from numeric wave models in near real time, which does not rely on the results from meteorological models.

4.3. Evaluation with the Altimeter Products

As mentioned in the Introduction, WW3 has good efficiency for wave simulations over global seas, and the influence of wave hindcasting by the WW3 model on sea surface temperatures (SSTs) was assessed in our recent study [30]. Although there were some gaps in the fusion wind field data, the simulations by WW3 for the gaps were filled using linear interpolation. Figure 12 shows the monthly average SWH maps for July 2020 that were derived from two sources over global seas, i.e., the WW3 simulations and the average values from the HY-2B and Jason-3 products. It can be observed that the wave patterns from the WW3 simulation are consistent with those from the altimeters. Again, this indicates that the fusion wind derived from the remote-sensed products was suitable for simulating waves using the WW3 model.

Furthermore, an error assessment was systematically conducted by comparing the WW3-modeled waves with monthly average products from the altimeters. A seasonal analysis of the monthly average error between the WW3 model and the altimeters is shown in Figure 13 for March 2020, June 2020, September 2020, and December 2019. We note that the gaps for the Southern Ocean were related to a lack of data. It was found that the differences in the SWHs (the altimeter results minus the WW3 results) were within ± 1.5 m for March and June. The differences for September became large for the coastal waters. We determined that this was likely caused by the extreme waves induced by the strong winds of the cyclonic season. The differences for December were significant—up to ± 3 m—especially for latitude > 40°N and latitude < 40°S. In the winter, the coverage of the sea ice gradually grew in the Arctic Ocean, whereas it was reduced in the Antarctic Ocean. Under this circumstance, the reliability of remote-sensed products deviated in the winter, resulting in a large error in the simulation. This is because the measurements from the Jason-3 altimeter are only used here, leading to apparent gaps in Figure 13d, which can be improved by abundant data in the future.

5. Conclusions

At present, sea surface dynamics, i.e., wind and waves, are operationally monitored using remote-sensed techniques, i.e., scatterometers, polarimetric microwave radiometers, and SAR and CFOSAT. The accuracy of these products was systematically investigated. Oceanic modeling primarily relies on atmospheric numerical simulations, and thus, the inherent error for the winds is inevitably included in the modeling techniques, e.g., the under-estimation by ECMWF for TCs [55]. With the abundant remote-sensed data, an interesting question is whether remote-sensed products in near real time could be applied to forecasting and hindcasting, as these do not rely on atmospheric numerical simulations. There is little research on fusion wind from remote-sensed products used in forcing numeric wave models, i.e., WW3 and SWAN. The main purpose of our work was to classify the above issue through conducting an experiment in 2019−2020.

In our work, the wind products from the ASCATs onboard the Metop-A/B/C satellites, the HY-2B scatterometer, and the WindSAT spaceborne polarimetric microwave radiometer for the period November 2019 to October 2020 were fused by using the classic interpolation method, denoted as Cressman interpolation. The validation of the fusion wind speed against the on-scene observations from the NDBC buoys showed a 1.66 m/s RMSE with a 0.81 COR; therefore, the fusion wind derived from the remote-sensed products was reliable for this study. However, gaps and swath patterns for the Southern Hemisphere were still observed in the fusion data, which could be further improved by using more products from other scatterometers onboard satellites (e.g., the CFOSAT satellite) and Soil Moisture Active Passive (SMAP) radiometers. The 0.125° gridded wind at 12 h intervals, daily CMEMS currents, and sea levels at 0.08° grids were treated as the forced fields in the wave simulations by the WW3 and SWAN models. Comparisons of the SWHs for June−September 2020 between the simulations of the two models and the NDBC observations showed that a < 0.5 m RMSE with a COR of > 0.8 was achieved by the WW3 and SWAN models. The wave products from the HY-2B and Jason-3 altimeters were also collected. The simulations by the WW3 model were compared with the monthly average values from the HY-2B and Jason-3 altimeters over global seas. The seasonal analysis of the monthly average error showed that the differences in the SWHs (the altimeter data minus the WW3 results) were within ± 1.5 m for March and June. However, the differences were quite significant for December, especially for latitude > 40°N and latitude < 40°S. It was concluded that the fusion wind data derived from multiple remote-sensed products were potential sources for hindcasting waves using numeric wave models. It is necessary to figure out whether the distortion caused by environmental factors (i.e., SST anomaly) should be further considered so as to improve the accuracy of fusion wind.

Since 2018, the HY-2 satellite constellation has provided synchronous wind products from scatterometers and microwave radiometers. In the near future, the measurements from the HY-2 satellite constellation will be assimilated with TC wind data provided by numeric models and SAR retrieval results [56,57], and this will consider the asymmetrical and eccentrical patterns of the horizontal winds. Furthermore, the assimilated wind field will be further tested for TC wave simulations by the WW3 and SWAN models.