Intercomparison, Fusion and Application of FY-3E/WindRAD and HY-2B/SCA Ocean Surface Wind Products for Tropical Cyclone Monitoring

Highlights

What are the main findings?

• This study provides the first comprehensive, full-year intercomparison of Ku-band ocean surface wind vector (OWV) products from FY-3E/WindRAD and HY-2B/SCA, showing high consistency in wind speed and direction.
• A fusion experiment combining FY-3E/WindRAD’s higher resolution with HY-2B/SCA’s wider swath improves the spatial completeness of tropical cyclone (TC) wind field structure and enables instantaneous estimates of the 34 kt wind radius (R34).

What is the implication of the main finding?

• The results indicate that Chinese scatterometer missions can reliably contribute to global OWV monitoring.
• Dual-satellite integration enhances TC monitoring and supports operational marine weather services.

Abstract

Ocean surface wind vector (OWV) is a key variable for ocean remote sensing and tropical cyclone (TC) monitoring. This study presents the first comprehensive intercomparison of Ku-band OWV products from FY-3E/WindRAD and HY-2B/SCA scatterometers using full-year data from 2022 (583,805 spatiotemporal collocations), with both sensors sampling the morning–evening local-time sector in sun-synchronous orbits. Results indicate strong agreement in wind speed (R = 0.95; mean bias −0.47 m/s; RMSE 1.30 m/s) and wind direction (mean bias 0.22°; std 28.13°) for wind speeds ≥ 3.4 m/s (Beaufort scale B3 and above), with the highest consistency across Beaufort scale 3–8 (B3–B8); however, at wind speeds greater than 20.8 m/s (B9) the bias increases. A fusion leveraging FY-3E’s fine resolution and HY-2B’s wide coverage is implemented and applied to Super Typhoon Hinnamnor (2022), enhancing the spatial coverage and structural detail of TC winds. Quadrant 34 kt wind radii (R34) are estimated from the fused wind fields and evaluated against the best-track data from the Joint Typhoon Warning Center (JTWC), showing close agreement during compact, symmetric TC stages but larger differences during structural reorganization. Overall, the findings confirm inter-satellite consistency for the two Chinese scatterometers and demonstrate the practical value of a multi-source fusion approach that benefits TC monitoring, wind radii estimation, and marine weather services.

1. Introduction

Ocean surface wind vector (OWV) is an essential geophysical parameter in meteorology, ocean dynamics, and climate studies, and is critical for tropical cyclone monitoring, numerical weather prediction, offshore wind energy assessment, and marine engineering safety [1,2]. Conventional in situ measurements from buoys and ships are constrained by sparse spatial and temporal coverage, making them inadequate for continuous global monitoring [3,4]. By contrast, satellite-based remote sensing, particularly active microwave scatterometry, provides high-resolution, all-weather, and wide-area observations of global ocean surface winds [5,6].

Since the Seasat satellite first carried the Seasat-A Satellite Scatterometer (SASS) in the late 1970s [7,8], the international ocean wind observing system has developed rapidly, with successive generations of scatterometers such as the AMI-SCAT on ERS-1/2 [9], NSCAT on ADEOS-I [10,11], SeaWinds on QuikSCAT [12], ASCAT on MetOp [13], OSCAT on OceanSat, and CFOSCAT on CFOSat. These missions have collectively advanced global wind observations and highlighted the value of scatterometer data for weather analysis, data assimilation, and marine forecasting [14,15,16,17,18,19,20,21,22,23].

In recent years, China has made significant progress in spaceborne wind observations. Fengyun-3E (FY-3E), launched in July 2021, carries the WindRAD, China’s first operational dual-frequency (Ku/C-band), dual-polarization, conically scanning scatterometer. This instrument enhances global OWV monitoring and provides critical data for dawn–dusk orbit meteorology [24,25]. Validation studies have demonstrated good wind speed and direction accuracy for FY-3E/WindRAD [24,26,27,28]. Meanwhile, the Haiyang-2B (HY-2B) satellite, launched in October 2018 for dynamic ocean environment monitoring, is equipped with a microwave scatterometer (SCA), one of China’s first operational Ku-band instruments [29]. Its wind products have shown high reliability and broad applicability through multi-source validation [30,31,32,33], and intercomparison with international scatterometers (e.g., ASCAT, OSCAT) has further confirmed the consistency of its wind observations [31,34,35]. However, no systematic intercomparison between FY-3E/WindRAD and HY-2B/SCA has yet been conducted.

FY-3E and HY-2B both operate in sun-synchronous orbits, with nearly coincident overpass times (within about 30 min), and both provide Ku-band wind vector products. This enables high-precision spatiotemporal collocation and intercomparison. Their observing capabilities are complementary: FY-3E/WindRAD offers higher spatial resolution, capturing wind gradients in tropical cyclone (TC) cores and high-wind regions, while HY-2B/SCA, with its wider swath, provides broader spatial coverage that captures the overall TC structure [31,36,37]. Intercomparison of these sensors not only allows objective evaluation of FY-3E/WindRAD data quality and stability but also reveals the impacts of different system designs (e.g., dual-/single-frequency, scanning geometry, retrieval algorithms) [38,39] on wind product consistency.

In the context of multi-satellite constellation observations, data fusion has become an effective strategy for filling observation gaps, improving spatial continuity, and enhancing extreme weather monitoring capability [40,41]. While fusion methods have been widely applied to international scatterometers (e.g., ASCAT, SeaWinds) [42,43], the integration of FY-3E/WindRAD and HY-2B/SCA products remains underexplored. To address this gap, this study conducts the first systematic intercomparison of Ku-band wind speed and direction products from FY-3E/WindRAD and HY-2B/SCA using full-year 2022 data. Furthermore, the potential of near-synchronous dual-satellite fusion is assessed using a case study of Super Typhoon Hinnamnor (2022), with emphasis on its utility for enhanced TC monitoring and refined wind radii estimation [44,45,46].

2. Materials and Methods

2.1. Data

The WindRAD onboard FY-3E is China’s first operational dual-frequency (Ku/C-band) conically scanning scatterometer with dual-polarization capability. It has swath widths of 1300 km/1250 km and provides wind products at spatial resolutions of 10 km and 25 km, with an effective wind speed range of 3–40 m/s [47]. In this study, the Ku-band product (WRADKu, central frequency 13.256 GHz) is used, which is stored in HDF5 format and publicly available from the National Satellite Meteorological Center (https://data.nsmc.org.cn, accessed on 22 September 2025). The scatterometer (SCA) onboard the Haiyang-2B (HY-2B) satellite is one of China’s first operational Ku-band fan-beam scatterometers. Its Level-2B wind field product has a spatial resolution of 25 km, also with a central frequency of 13.256 GHz, and is distributed by the National Satellite Ocean Application Service (https://osdds.nsoas.org.cn, accessed on 22 September 2025). The dataset, in HDF5 format, contains wind speed, wind direction, observation time, geolocation, and quality control information [48].

Since HY-2B/SCA only operates on Ku-band, this study aims to evaluate the consistency and error structure of Ku-band OWV products from FY-3E/WindRAD and HY-2B/SCA, in order to avoid potential errors arising from different frequencies. The analysis covers the entire year of 2022 (from 1 January to 31 December), focusing on the Northwest Pacific and adjacent seas of China. Despite differences in polarization, scanning geometry, and retrieval algorithms [18,49,50,51,52,53] (

Table 1. Technical specifications of FY-3E/WindRAD and HY-2B/SCA.

Parameter	FY-3E/WindRAD	HY-2B/SCA
Orbit type	Sun-synchronous orbit	Sun-synchronous orbit
Equator crossing time	05:30 (descending)/17:30 (ascending)	06:00 (descending)/18:00 (ascending)
Orbital period	~99 min per cycle	~100 min per cycle
Global coverage cycle	Twice daily over global oceans	>90% coverage every 1–2 days
Wind measurement mode	Dual-frequency, dual-polarization, conical scanning scatterometer	Single-frequency, dual-polarization, conical scanning scatterometer
Operating frequency	C-band (5.3 GHz) + Ku-band (13.256 GHz)	Ku-band (13.256 GHz)
Polarization	HH, VV	HH, VV
Calibration	External calibration + onboard calibration	Onboard calibration
Spatial resolution	~10/25 km	~25 km
Swath width	~1250–1300 km	1350 km (H-pol), 1700 km (V-pol)
Wind retrieval model	CMOD7 (C-band), NSCAT-6 (Ku-band), CMOD7 + NSCAT-6 (dual-frequency)	NSCAT-4 GMF model
Wind products	WRADC (C-band), WRADKu (Ku-band), WRADX (dual-frequency)	L2B wind speed and direction (HDF5)
Wind definition	10 m stress-equivalent wind	10 m stress-equivalent wind
Wind speed range	3–40 m/s	3–35 m/s
Quality control	Multi-bit QC flags (e.g., Bit13–Bit16)	Multi-bit QC flags (rain, land contamination, anomalies, retrieval failure)
Wind direction processing	MSS + 2DVAR ambiguity removal	MSS + 2DVAR ambiguity removal
Data format	HDF5	HDF5

Note: MSS = Multiple Solution Sampling; 2DVAR = Two-Dimensional Variational method.

), unified quality control and pixel selection criteria were applied to ensure scientific rigor and comparability. Figure 1 illustrates typical spatial coverage and wind field structures from both satellites.

The TC information used in this study is obtained from the IBTrACS v4 (International Best Track Archive for Climate Stewardship) best-track dataset, which integrates records from multiple agencies such as JTWC and CMA, providing global coverage and standardized sources (data access: https://www.ncei.noaa.gov/products/international-best-track-archive, accessed on 22 September 2025). From IBTrACS, the JTWC best track is extracted as the reference for evaluating TC center positions and wind radii, including variables such as center location, intensity, maximum sustained wind, minimum sea level pressure, and quadrant wind radii (e.g., R34/R50). Specifically, IBTrACS v4 provides a 3 hourly series interpolated from the 6 hourly source analyses [54]. Positions are interpolated using splines, whereas non-position variables (e.g., winds, pressure, and quadrant radii) are interpolated linearly, which facilitates precise time matching with satellite overpasses and a finer depiction of TC evolution.

2.2. Methods

2.2.1. Data Preprocessing

To ensure the scientific validity and comparability of the intercomparison, strict quality control procedures were applied. Pixels flagged as invalid due to retrieval failure, GMF fitting anomalies, sea ice contamination, land contamination, rainfall effects, or insufficient sigma-0 were removed, while physically reasonable high-wind and low-wind observations were retained. Edge pixels from the first two and last two columns of both satellites were discarded to avoid scan distortions. Wind speed values were converted from raw storage units (e.g., 0.01 m/s) to standard SI units (m/s). All subsequent analyses were based on these quality-controlled datasets.

2.2.2. Spatiotemporal Collocation

Intercomparison between FY-3E/WindRAD and HY-2B/SCA was performed using a nearest-neighbor matching method constrained by temporal and spatial windows, considering orbit proximity, resolution, and operational feasibility.

• Temporal matching: FY-3E and HY-2B are both in sun-synchronous orbits with local equator crossing times of 05:30/17:30 and 06:00/18:00, respectively. Their observation time difference is typically within ±30 min. Accordingly, a temporal window of ±30 min was applied, and only collocated orbits within this interval were retained.
• Spatial matching: Taking FY-3E/WindRAD pixels as the reference, the nearest neighbor in HY-2B/SCA was identified using great-circle distance. Pairs with separation less than 25 km (corresponding to the scatterometer footprint) were retained. The great-circle distance was calculated as follows:

$$d=R·arccos\left(sin\right({\phi }_1)·sin({\phi }_2) + cos({\phi }_1)·cos({\phi }_2)·cos({\lambda }_2 - {\lambda }_2\left)\right),$$

(1)

where d is the great-circle distance (km), R is the Earth’s mean radius (6371 km), and ${\phi }_1$, ${\phi }_2$, ${\lambda }_1$ and ${\lambda }_2$ are latitudes and longitudes (radians) of FY-3E/WindRAD and HY-2B/SCA pixels.

To further avoid degraded quality at swath edges, the outermost two rows and columns of both datasets were removed.

• Wind direction normalization: To resolve 360° periodic ambiguity, wind direction differences were normalized to the [−180°, +180°] interval:

$$\Delta \theta =mod\left({\theta }_{FY} - {\theta }_{HY} + 180, 360\right) - 180,$$

(2)

where ${\theta }_{FY}$ and ${\theta }_{HY}$ are wind directions from FY-3E/WindRAD and HY-2B/SCA.

Wind direction statistics were computed after excluding low-wind cases, for which direction is poorly defined. Because our analysis is organized by Beaufort classes, we adopt a Beaufort-consistent cutoff of B ≥ 3 (≥3.4 m/s) for the main results.

The resulting collocated dataset includes matched wind speeds, wind directions, and geolocation from both satellites, providing high-consistency samples for inter-sensor validation.

2.2.3. Evaluation Metrics

To assess consistency, multiple statistical metrics were applied to matched wind speed and direction samples.

• Wind speed metrics: Mean Bias, Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Pearson correlation coefficient (R).
• Wind direction metrics: Mean Bias, Standard Deviation (Std), Median Bias, and Median Absolute Bias.

Bias reflects systematic deviation, RMSE and MAE measure overall dispersion, and re-valuates linear consistency. Wind direction metrics capture angular difference distributions under circular conditions. These indicators are widely used in wind field validation [55,56].

Furthermore, wind speed differences were stratified by Beaufort scale based on HY-2B/SCA as the reference (

Table 2. Beaufort scale classification for wind speed.

Beaufort Scale	Wind Speed Range (m/s)
0	<0.2
1	0.3–1.5
2	1.6–3.3
3	3.4–5.4
4	5.5–7.9
5	8.0–10.7
6	10.8–13.8
7	13.9–17.1
8	17.2–20.7
9	20.8–24.4
10	24.5–28.4
11	28.5–32.6
12	≥32.7

). This enabled evaluation of FY-3E/WindRAD performance across weak, moderate, and high wind conditions.

2.2.4. Satellite Data Fusion

To enhance spatial coverage and structure reconstruction of wind fields under extreme weather, a fusion experiment was conducted using Typhoon Hinnamnor (No. 2211) as a case study. The procedure included:

• Quality control and preprocessing: Both datasets underwent unified quality control to remove retrieval failures, precipitation, and land contamination.
• Sample selection: Only pixels with wind speed ≥ 10.8 m/s (Beaufort scale 6 or higher) were retained to focus on TC core structures and reduce low-wind noise.
• Resolution harmonization: HY-2B/SCA data were resampled to 10 km resolution, consistent with FY-3E/WRADKu. The resampling was performed using linear interpolation as the primary method. In cases where linear interpolation failed due to insufficient neighboring data points, nearest-neighbor interpolation was used as a fallback. This two-step interpolation approach ensures the method’s resilience to data gaps and maintains the reliability and stability of the data fusion process.
• Data fusion logic: FY-3E was given priority, meaning that if FY-3E wind data was available at a given grid point, no HY-2B data was used at that point. HY-2B data were used to fill uncovered areas only.
• Spatial tolerance: A spatial tolerance of 0.06° (~6 km) was applied, approximately half of the native FY-3E/WindRADKu footprint. This value balances collation density and representativeness errors, and is consistent with commonly adopted spatial collocation windows in previous scatterometer validation and intercomparison studies [56,57].

2.2.5. Estimation of Wind Radius (R34) and Comparison with JTWC

This study focuses on estimating the 34 kt wind radius (R34). In operations, R50/R64 are often inferred from statistical/parameterized relations with R34 and intensity, whereas R34 can be directly and robustly extracted from the fused wind field. TC centers are taken from IBTrACS v4, prioritizing JTWC fixes (CMA as fallback). JTWC best-track analyses are issued at 6 h synoptic times (00/06/12/18 UTC). For time matching, the 3 hourly IBTrACS v4 series is used; off-synoptic timestamps (e.g., 09:00 or 21:00 UTC) therefore correspond to IBTrACS interpolations rather than original JTWC analyses. Fused snapshots are paired with best tracks within ±45 min to cover the ~30 min orbit offset and edge/interpolation effects; the implied translation (~27–54 km) is small relative to the collocation radius (~25 km), grid spacing (10 km), and the R34 scale (~<500 km).

Using the matched center as origin, grid points are partitioned into NE/SE/SW/NW quadrants. For points exceeding 34 kt (17.5 m/s), great-circle distances to the center are computed per Equation (1). To ensure accurate and reliable estimates of the 34 kt wind radius (R34), sensitivity tests were first conducted on key parameters. These parameters include:

• Maximum radius (rmax), ranging from 300 km to 600 km.
• Minimum number of samples per quadrant (minN), ranging from 5 to 40.
• Cumulative distribution function (CDF) percentile, ranging from 80% to 95%.

These tests aimed to identify the optimal combination of parameters that minimizes errors and ensures robustness in the R34 estimation. Based on the sensitivity analysis, the final parameter set selected was CDF at the 90th percentile, rmax at 500 km, and minN at 5, which consistently minimized errors and provided stable and accurate R34 estimates.

A wind speed bias correction was applied for wind speeds ≥ 17.5 m/s (34 kt). For these samples, a ±ΔV/2 correction was applied, where half of the observed wind speed bias was added to the FY-3E data and subtracted from the HY-2B data. A 2 m/s correction, derived from the wind speed bias observed between the two satellites for wind speeds above 17.5 m/s, was applied to samples near this threshold.

Validation time-matches quadrantwise R34 to JTWC radii (units harmonized via 1 nm = 1.852 km) and reports Bias and RMSE, with Bias (km) defined as

$$∆R_{34,q}={\widehat{R}}_{34,q}-R_{34,q}^{JTWC} ,$$

(3)

and the relative bias defined as

$${\mathsf{\delta }}_{34,\mathrm{q}}={\displaystyle \frac{∆R_{34,q}}{R_{34,q}^{JTWC}}}· 100\% ,$$

(4)

Here, ${\widehat{R}}_{34,q}$ is the estimate and $R_{34,q}^{JTWC}$ is the reference.

3. Results

Based on the aforementioned quality control and spatiotemporal collocation methods, this study systematically evaluates the consistency of Ku-band wind field products from FY-3E/WindRAD and HY-2B/Scatterometer (SCA). The results section focuses on statistical analyses of wind speed and wind direction, exploring key factors that influence the agreement between the two satellite products. In addition, a representative case of data fusion is presented to illustrate the complementary strengths of the two sensors during a typhoon event.

3.1. Wind Speed Comparison

Using pixel-level collocated data from the entire year of 2022, a total of 583,805 valid pairs of wind speed observations were obtained. The results (Figure 2) show a high level of agreement between the two sensors, with a Pearson correlation coefficient (R) of 0.95, a mean bias of −0.47 m/s, and a root mean square error (RMSE) of 1.30 m/s. These values indicate that the wind speed retrieved by FY-3E is slightly lower than that from HY-2B. The slope and intercept of the least squares regression line are 0.89 and 0.46 m/s, respectively, suggesting a systematic underestimation by FY-3E in the moderate to high wind speed range. Most of the sample points are densely distributed around the 1:1 reference line, with only a small number of outliers. Overall, the two wind speed products exhibit good consistency and high operational applicability across the primary wind speed range.

To further characterize the statistical distribution of differences between FY-3E/WRADKu and HY-2B/SCA wind speeds, a probability density function (PDF) analysis was conducted (Figure 3). Consistent with the results above, FY-3E/WRADKu tends to underestimate wind speeds relative to HY-2B/SCA. The median bias is −0.30 m/s, and the standard deviation is 1.21 m/s. The skewness and kurtosis are −1.09 and 8.80, respectively, indicating a sharp-peaked, heavy-tailed distribution with a slight leftward skew. Most of the differences lie within ±3 m/s, with a minimum of −19.90 m/s and a maximum of 10.70 m/s. These few extreme outliers have limited impact on the overall distribution. In summary, the FY-3E/WRADKu and HY-2B/SCA products demonstrate good consistency within the main operational range, although slight negative biases and a small number of large negative deviations are present in the FY-3E/WRADKu product.

3.1.1. Bias Under Different Wind Speeds

To assess wind speed differences under various wind conditions, the collocated samples from 2022 were categorized according to the Beaufort wind scale (B0–B12). Figure 4 presents the mean bias and standard deviation (±1σ) of wind speed differences between FY-3E/WRADKu and HY-2B/SCA across all Beaufort levels, along with the corresponding number of samples per category (right y-axis).

The majority of samples are concentrated in the moderate wind force range (B3–B6, 3.4–13.8 m/s), accounting for over 48% of the total. Specifically, the B4 and B5 categories contain 147,074 and 141,174 matched samples, respectively, with mean biases within ±0.25 m/s and standard deviations ranging from 0.74 to 0.82 m/s. These results indicate strong agreement between the two sensors under moderate wind conditions.

In the low wind range (B0–B1, 0–1.5 m/s), FY-3E/WRADKu tends to report higher wind speeds than HY-2B/SCA, with biases of +1.52 m/s for B0 and +0.68 m/s for B1. This discrepancy is likely due to differences in backscatter calibration between the two instruments, especially at low wind speeds.

In the high wind range (B9–B12, ≥20.8 m/s), the number of matched samples is extremely limited, accounting for less than 0.5% of the total. For instance, only four pairs fall into the B12 category throughout the year, and the observed bias (−13.5 m/s) lacks statistical robustness due to insufficient sample size. Although B9 and B10 categories have slightly more samples (>1000), they also exhibit increased negative bias (−1.84 m/s for B9 and −2.54 m/s for B10). This is likely due to calibration differences in the backscatter response of the two instruments at higher wind speeds, with FY-3E/WindRAD using the NSCAT-6 model for Ku-band retrieval, while HY-2B uses the NSCAT-4 GMF model. Therefore, the ±2 m/s wind speed bias correction, applied in the R34 estimation process for wind speeds ≥ 17.5 m/s, was considered to address this bias.

3.1.2. Monthly Bias Variation

Figure 5 further illustrates the monthly statistics of wind speed differences between FY-3E/WRADKu and HY-2B/SCA throughout 2022, including the mean bias and standard deviation (Std). The bias is defined as the wind speed of FY-3E/WRADKu minus that of HY-2B/SCA.

The results show that FY-3E/WRADKu consistently exhibits a slightly lower wind speed compared to HY-2B/SCA across all months, with monthly mean bias ranging from −0.53 m/s (December) to −0.38 m/s (January). The standard deviation remains relatively stable, fluctuating between 1.14 and 1.27 m/s. The smallest biases were observed in January (−0.38 m/s, Std = 1.18 m/s) and May (−0.39 m/s, Std = 1.15 m/s), while the largest negative bias occurred in December (−0.53 m/s). Each month contained more than 94,000 valid collocated samples, with January having the highest number (261,965 pairs), ensuring the statistical robustness of the analysis.

Overall, no significant systematic monthly deviations were observed in the intercomparison, indicating that the FY-3E/WRADKu wind speed product maintains good seasonal stability and consistency. This suggests that it holds potential for applications in multi-source wind field fusion and long-term climate-scale studies.

3.2. Wind Direction Comparison

3.2.1. Annual Bias Distribution

Differences in wind direction between FY-3E/WRADKu and HY-2B/SCA were evaluated by constructing a linear histogram of wind direction bias for the entire year of 2022, as shown in Figure 6. The x-axis represents the wind direction bias, and the y-axis shows the probability density (or frequency) of matched samples within each bias interval. Wind direction statistics were computed after excluding low-wind cases, where wind direction is poorly defined due to the calibration issues in backscatter response. For reliable results, a Beaufort-consistent cutoff of B ≥ 3 (≥3.4 m/s) was applied for the main analysis, excluding wind speeds below this threshold.

The circular mean bias of the collocated samples was 0.22°, with a circular standard deviation of 28.13°, indicating no significant systematic deviation in wind direction between the two sensors. The relatively large dispersion suggests substantial variability among individual samples, which may be attributed to factors such as differences in retrieval algorithms and instantaneous wind variability. The overall distribution is approximately symmetrical, showing no clear directional skew. Additionally, results for wind speeds ≥ 4 m/s and for the full sample set were also evaluated and included in the Supplementary Materials (Figures S1 and S2), confirming that the conclusions remain consistent.

3.2.2. Wind Direction Bias Under Different Wind Speeds

Figure 7 presents the circular mean and standard deviation (±1σ) of wind direction bias between FY-3E/WRADKu and HY-2B/SCA under different Beaufort wind force levels (B0–B12), along with the number of matched samples for each category.

In terms of circular means, wind direction bias across all wind speed levels is close to zero, ranging from −8.06° to +3.19°, indicating no significant systematic deviation. Regarding standard deviation, the lower wind speed categories (B0–B2, corresponding to 0–3.3 m/s) exhibit large dispersion, with standard deviations of 98.86°, 97.97°, and 74.90°, respectively. This reflects substantial uncertainty in wind direction differences under weak wind conditions, primarily due to the ambiguous definition of wind direction at low speeds and increased sensitivity of the retrieval algorithms.

As wind speed increases to moderate and strong levels (B4–B7, corresponding to 5.5–17.1 m/s), the standard deviation significantly decreases to the range of 17.49° to 28.44°, suggesting a marked improvement in observational consistency. This indicates that wind direction retrievals from the two satellites are more stable under typical marine wind conditions. Category B6 (10.8–13.8 m/s) exhibits the smallest wind direction bias of the year (Bias = +1.08°, Std = 17.49°).

In higher wind speed categories (B8 and above), the number of matched samples decreases rapidly (e.g., only 4 pairs in B12). While some categories show a slight increase in standard deviation, the limited sample sizes reduce the statistical representativeness of these results. The overall distribution characteristics are well represented in Figure 7.

By combining the results from both wind speed and wind direction comparisons, the FY-3E/WRADKu and HY-2B/SCA wind field products exhibit a high degree of consistency within the commonly observed wind speed range (B3–B8). Although certain deviations exist in low and high wind speed conditions, their statistical characteristics and underlying physical mechanisms are consistent with known limitations of Ku-band scatterometers. These findings provide a solid foundation for future efforts in dual-satellite wind field data fusion, particularly in applications related to extreme weather events such as tropical cyclones.

3.3. Application of Wind Field Fusion in TC Monitoring

To evaluate the practical capability of the dual-satellite fused wind field in typhoon monitoring, this study selects Super Typhoon Hinnamnor (2022), characterized by its complex life cycle and intensity variability, as a representative case. This typhoon underwent a series of canonical processes, including rapid intensification, an eyewall replacement cycle (ERC), merger with a tropical depression, a V-shaped track reversal, and re-intensification over the East China Sea, ultimately evolving into a large-scale circulation system (Figure 8a) [58,59]. Figure 8 overlays best-track data from both JTWC and CMA to provide a consistent cross-agency background and center position reference; all subsequent comparative evaluations of wind radii in this study are conducted within the JTWC framework, with TC center positions prioritized from JTWC and supplemented by CMA data only when missing [60].

The fused dataset effectively integrates the high-resolution advantage of FY-3E/WRADKu with the wide-swath coverage of HY-2B/SCA, simultaneously enhancing spatial coverage and improving the characterization of the typhoon’s wind field structure (Figure 9c,f,i,l). A comparison between the estimated wind radii (${\widehat{R}}_{34,q}$) based on the 34-knot (17.5 m/s) stress-equivalent wind threshold and the JTWC best-track data ($R_{34,q}^{JTWC}$) indicates good overall agreement, albeit with quadrant- and time-dependent discrepancies (Figure 10). The all-quadrant mean bias ($\overline{∆R_{34,q}}$) for the four representative synoptic times (t1: 30 August 21:00, t2: 31 August 21:00, t3: 3 September 09:00, t4: 4 September 21:00) was −34 km, +36 km, +35 km, and −52 km, respectively, reflecting the response of the fused snapshots to different structural phases. The time-aggregated quadrant mean biases ($∆{\overline{R}}_{34,NE}$, $∆{\overline{R}}_{34,SE}$, $∆{\overline{R}}_{34,SW}$, ${∆\overline{R}}_{34,NW}$) were −30 km, −18 km, +13 km, and +21 km, respectively (Figure 10e).

The spatiotemporal distribution of these biases is closely linked to the typhoon’s structural and intensity evolution, primarily manifested in the following four stages:

(1)

Relatively Symmetric Structure: t1 (near 30 August 21:00). During this period, the vortex was compact and relatively symmetric. The closest agreement between ${\widehat{R}}_{34,q}$ and $R_{34,q}^{JTWC}$($\overline{∆R_{34,q}}$ = −24 km; Figure 10a) demonstrates the high reliability of the fused product under quasi-steady conditions.

(2)

Rapid reorganization and outer-wind expansion: t2 (near 31 August 21:00). Following the ERC, the vortex underwent axisymmetrization with a more compact inner core, reduced azimuthal asymmetry, and a steepened radial wind-speed gradient. Under these quasi-steady, more regular conditions, the fused product and $R_{34,q}^{JTWC}$ estimates show near-neutral agreement ($\overline{∆R_{34,q}}$ ≈ +1 km; Figure 10b). This is consistent with the robustness of our point-set fusion with de-duplication and CDF-percentile detection; when large-scale azimuthal asymmetry (dominant k = 1–2) weakens, the influence of isolated strong-wind samples and sampling randomness is minimized, keeping quadrant biases within tens of kilometers and indicating convergence between the algorithmic estimate and best-track values.

(3)

Track reversal and asymmetric reorganization: t3 (near 3 September 09:00). During the V-shaped track reversal within weak steering and a saddle-shaped pressure field, the fused wind map reveals a pronounced south-west (SW)-side enlargement, yielding the largest quadrant-scale positive deviation among the four cases ($∆R_{34,SW}$ ≈ + 140 km; Figure 10c). Other quadrants are much smaller in magnitude (near-neutral to weakly positive/negative), indicating an asymmetric reorganization primarily along the SW–NE axis, but dominated by SW-flank expansion at this time. Notably, the $R_{34,q}^{JTWC}$ used here corresponds to the nearest IBTrACS v4 entry around 09:00 UTC (FY-3E overpass at 08:48 UTC); since IBTrACS provides non-positional variables via 3 h linear interpolation, the $R_{34,q}^{JTWC}$ at 09:00 UTC likely underrepresents the instantaneous outer-wind expansion. In contrast, our point-set fusion with de-duplication (FY-3E priority with HY-2B gap-filling) and CDF-percentile detection resolves the instantaneous SW-flank coverage more completely. Therefore, the SW discrepancy mainly reflects interpolation smoothing and sampling completeness, rather than algorithmic overestimation.

(4)

Re-intensification and quadrant redistribution: t4 (near 4 September 21:00). After re-intensification over the East China Sea, the overall mean bias turned negative $\overline{∆R_{34,q}}$ = −68 km; azimuthal asymmetry was evident, with a large negative deviation in the SE quadrant ($∆R_{34,SE}$ ≈ −109 km, ${\mathsf{\delta }}_{34,\mathrm{S}\mathrm{E}}$ ≈ −24.7%), while other quadrants showed smaller departures ($∆R_{34,NE}$ ≈ −71 km, ${\mathsf{\delta }}_{34,\mathrm{N}\mathrm{E}}$ ≈ −16.1%; $∆R_{34,SW}$ ≈ −19 km, ${\mathsf{\delta }}_{34,\mathrm{S}\mathrm{W}}$ ≈ −6.3%; $∆R_{34,NW}$ ≈ −73 km ${\mathsf{\delta }}_{34,\mathrm{N}\mathrm{W}}$ ≈ −26.4%; Figure 10e). From t1 to t4, $R_{34,q}^{JTWC}$ increased by about 369%, 449%, 285%, and 231% in the NE, SE, SW, and NW quadrants, respectively, suggesting quadrant redistribution of the 34 kt wind radii rather than a uniform expansion. The same interpolation caveat applies at 21:00 UTC, so departures reflect both smoothing and rapid evolution.

In summary, the discrepancies between the fused product and the JTWC best-track analysis primarily originate from the fundamental differences between instantaneous, detailed observation and temporally smoothed, interpolated analysis. This is particularly evident during rapid evolution phases such as ERC, merger, and abrupt track changes, where the fused wind field demonstrates superior sensitivity to structural changes and a greater potential for resolving asymmetry. It is acknowledged that the fused estimates still face potential limitations, including scatterometer signal saturation in heavy rain, center-fixing uncertainties, and differences in interpolation and sampling methodologies. While these were mitigated through a ±45 min pairing window, robust P90 percentile estimation, and quality control, residual biases may persist during periods of very rapid evolution.

4. Discussion

This study presents the first comprehensive intercomparison of FY-3E/WindRAD and HY-2B/SCA Ku-band ocean surface wind vector (OWV) products using a full year of coincident observations. The two scatterometers show strong consistency: wind speed attains a correlation of 0.95 with a mean bias of −0.47 m/s and an RMSE of 1.30 m/s; wind direction shows no meaningful systematic offset, with an annual circular mean of about 0.22° and a standard deviation of 28.13°. Agreement is strongest for Beaufort scale 3–8. Slight overestimation at low winds, and slight underestimation at high winds, likely due to calibration differences in backscatter response between the two instruments. FY-3E/WindRAD using the NSCAT-6 model for Ku-band retrieval, while HY-2B/SCA uses the NSCAT-4 GMF model. As such, the ±2 m/s wind speed bias correction, applied in the R34 estimation process for wind speeds ≥ 17.5 m/s, was considered to address this bias. Monthly biases remain seasonally stable, indicating robustness for multi-sensor fusion and long-term climate applications.

A key application demonstrated in this study is the use of dual-satellite fusion for typhoon monitoring, as exemplified by Super Typhoon Hinnamnor. By combining FY-3E’s finer resolution with HY-2B’s wider swath, the fused OWV field provides a more complete depiction of both the core and the outer circulation and enables estimation of the 34 kt wind radii (R34). The fused R34 agrees well with JTWC best-track radii in symmetric stages, while clear, structure- and phase-dependent differences appear during rapid reorganization such as eyewall replacement and track reversal. These differences reflect the contrast between instantaneous, observation-based snapshots and temporally smoothed best-track analyses, highlighting the fusion product’s value for capturing fast-evolving asymmetries.

However, this study has some limitations. First, the retrieval challenges in extreme wind conditions, especially for high wind speeds (>25 m/s), are still a major concern. These challenges are partly due to limited sample sizes and the potential limitations of the wind retrieval models under extreme conditions. Additionally, the potential impact of precipitation on Ku-band data retrieval remains an unresolved issue. While this study focuses on a like-for-like Ku-band intercomparison between FY-3E/WRADKu and HY-2B/SCA, future work will address this limitation by integrating dual-frequency measurements (e.g., using both Ku-band and C-band) to mitigate precipitation effects. Another limitation is the lack of ground-based validation, which would enhance the accuracy of the intercomparison. Results from satellite-ground validation will be presented in a separate manuscript. An independent study is being prepared, based on a longer buoy data series, to validate FY-3E/WRADKu and FY-3E/WRADC OWV. Lastly, this study focuses on R34 estimation, but future work will expand the evaluation to R50 and R64 radii and assess the potential forecast impact through data assimilation experiments using the fused OWV fields. It should be noted that the current study is primarily based on data from the Northwest Pacific; future work will extend this approach to other regions for broader applicability.

5. Conclusions

This study demonstrates the strong consistency between FY-3E/WindRAD and HY-2B/SCA Ku-band ocean surface wind vector (OWV) products, with high correlation in wind speed and direction, particularly in moderate wind conditions. The fusion of FY-3E’s higher resolution and HY-2B’s wider swath enhances tropical cyclone monitoring, providing more comprehensive coverage of both the core and outer circulation. The fused 34 kt wind radii (R34) estimates show good agreement with JTWC best-track radii, although differences emerge during rapid reorganization phases, highlighting the advantages of the fusion approach for capturing fast-evolving asymmetries.