Impact of Assimilating FY-4B/GIIRS Radiances on Typhoon “Doksuri” and Typhoon “Gaemi” Forecasts

Abstract

The Geostationary Interferometric Infrared Sounder (GIIRS) on board FengYun-4B (FY-4B), a Chinese second-generation hyperspectral infrared, enables the provision of critical data for forecasting high-impact weather events such as typhoons. To evaluate the reliability of FY-4B/GIIRS data, this study conducted three comparative assimilation trials for both Typhoon Gaemi and Typhoon Doksuri, assimilating observations from the Infrared Atmospheric Sounding Interferometer (IASI), Advanced Microwave Sounding Unit-A (AMSU-A), and FY-4B/GIIRS, respectively. Results demonstrate that the assimilation of GIIRS observations yields more stable forecasts of the wind field at 300 hPa and 500 hPa compared to AMSU-A and IASI, with biases within ±6 m/s relative to NCEP FNL data. However, GIIRS assimilation produces systematic underprediction of vertical velocity, whereas AMSU-A forecasts align more closely with reanalysis. For track forecasts, the GIIRS-assimilated trajectory exhibits closer alignment with observations than AMSU-A and IASI experiments, maintaining biases below 50 km throughout 48 h forecast period of Gaemi. This study provides valuable experience for the application of FY-4B/GIIRS data assimilation.

1. Introduction

Tropical cyclones (TCs), as one of the most extreme maritime weather events, impact numerous monsoon-affected countries globally with their extremely high intensity and vast coverage. China, as the largest coastal nation along the western North Pacific, is among the countries most severely affected by TC worldwide. Therefore, accurately observing and forecasting information such as typhoon track and intensity holds extremely significant importance.

Among meteorological satellite instruments, microwave instruments/sensors, owing to their good penetrating capability through cloud layers, have long served as a primary observational resource for typhoon monitoring and forecasting [1]. Currently, spaceborne microwave sounding primarily relies on passive microwave radiometers and active microwave scatterometers. Microwave radiometers can retrieve vertical atmospheric temperature and humidity profiles, thereby uncovering the internal structure of typhoons [2]. In 2017, Yu et al. selected Typhoon Rammasun (2409) as a case study to investigate the effects of assimilating Advanced Microwave Scanning Radiometer 2 (AMSR2) microwave imaging data on Typhoon track and intensity forecasts [3]. Their results demonstrated improvements in both track and intensity forecasts. In 2018, Li et al. [4] extended the Weather Research and Forecasting—Three-Dimensional Variational Data Assimilation System (WRF-3DVar) to assimilate FY-3C Microwave Humidity and Temperature Sounder (MWHTS) data. They analyzed track and pressure biases of Typhoon Megi before and after its landfall. In 2021, Shu et al. [5] explored the impact of directly assimilating FY-3D Microwave Humidity Sounder-II (MWHS-2) radiance data on the forecast of Typhoon Mitag. They found that assimilating MWHS-2 radiance data improved circulation patterns of the typhoon, thereby enhancing track forecast accuracy. Meanwhile, microwave scatterometers demonstrate a strong capability in capturing sea surface wind speed and direction [6]. Lin et al. [7] conducted an error analysis of sea surface wind field retrievals from multiple global satellite microwave scatterometers. Their study found that the intrinsic errors for the u and v components of these scatterometer-derived wind fields ranged from 0.5 to 1.3 m·s⁻¹. This result provides a crucial reference for setting error covariance matrices in applications utilizing multi-source satellite scatterometer wind data. Spaceborne microwave instruments serve as critical data sources for typhoon monitoring, as well as for thermodynamic and dynamic studies of typhoons [8].

In addition to microwave data, hyperspectral infrared data from clear-sky channels can also serve as critical information sources for typhoon forecasting. As early as the 1990s, William L. Smith [9] emphasized the enhancement value of geostationary hyperspectral sounders for Numerical Weather Prediction (NWP) systems. In 2002, following the operational deployment of the first hyperspectral infrared instrument, the Atmospheric Infrared Sounder (AIRS) [10], multiple hyperspectral instruments such as Infrared Atmospheric Sounding Interferometer (IASI) and Cross-track Infrared Sounder (CrIS) were gradually integrated into global data assimilation systems. Many studies have shown that data assimilation from hyperspectral infrared sounders has a positive impact on numerical weather forecasts [11,12,13,14]. However, most infrared hyperspectral sounders are currently on polar-orbiting satellites, which cannot provide continuous, long-term observations of typhoons, posing certain difficulties for typhoon forecasting and monitoring. In the early 21st century, NASA developed the GIFTS (Geosynchronous Imaging Fourier Transform Spectrometer) instrument and completed ground-based testing [15,16,17]. Although the mission was ultimately canceled prior to launch, China’s GIIRS (Geosynchronous Interferometric Infrared Sounder) inherited GIFTS technological legacy. With the launch of the FY-4A satellite in 2016, GIIRS became the world’s first hyperspectral infrared sounder operational in geostationary orbit. Extensive research has validated its scientific value [18,19,20], realizing William L. Smith’s pioneering theoretical vision from the 1990s. A study in 2020 [21] employing atmospheric profile retrieval and assimilation methods with FY-4A/GIIRS, IASI, and CrIS observations conducted comparative evaluations of polar-orbiting and geostationary hyperspectral infrared sounders in numerical weather forecasts. The results revealed enhanced forecast accuracy in quantitative precipitation forecasting (QPF) and typhoon track positioning. Furthermore, the assimilation of FY-4A/GIIRS data demonstrated additional improvements in extreme precipitation forecasting performance. In 2021, Wang et al. conducted research on channel selection and cloud detection impacts for the FY-4A/GIIRS mid-wave bands. Using Typhoon Lekima (2019), they applied the Minimum Residual Method to perform cloud detection on GIIRS data [22]. This method not only identifies cloud presence in field-of-view pixels but also derives effective cloud amount and cloud-top pressure information. In the same year, Yin et al. investigated the assimilation of FY-4A/GIIRS data for Typhoon Maria, finding that high-temporal-resolution data assimilation significantly improved Typhoon intensity and track forecast accuracy [23].

The new-generation GIIRS, hosted aboard the FY-4B satellite, was launched on 3 June 2021 and became operational on 1 June 2022. Compared to the FY-4A/GIIRS, the FY-4B/GIIRS demonstrates enhanced spatial, temporal, and spectral resolution, along with expanded spectral coverage. As of now, the FY-4B/GIIRS has been applied in multiple research domains [24,25,26]. In 2024, Gao et al. [27] conducted a comprehensive evaluation of the FY-4B/GIIRS temperature retrieval products and further examined its diagnostic capabilities for precipitation classification. Regarding typhoon forecasting, in 2024, Yin et al. [28] assimilated FY-4B/GIIRS radiance data into the numerical model for two typhoon cases, demonstrating that the GIIRS assimilation improved track forecast accuracy, with the most pronounced enhancements observed beyond the 60 h forecast time. This study represents the first assimilation application research for FY-4B/GIIRS, offering initial insights into assimilation methodologies. However, it lacks detailed comparative analyses with advanced international hyperspectral infrared instruments (e.g., IASI, CrIS) and microwave sounders (e.g., AMSU-A, Advanced Technology Microwave Sounder, ATMS). The potential advantages of FY-4B/GIIRS for typhoon monitoring and early warning require further validation. Currently, assimilation studies utilizing FY-4B/GIIRS remain limited, and the quality control methods and cloud detection algorithms necessitate further research.

The remainder of this paper is organized as follows: Section 2 details the experimental methodology, including the configuration of GIIRS, AMSU-A, and IASI datasets, followed by a synoptic-scale environmental analysis of Typhoons Doksuri (2023) and Gaemi (2024), focusing on their genesis and intensification processes. Section 3 details cloud detection, quality control, and bias correction procedures. Section 4 evaluates the performance of FY-4B/GIIRS through comparative analyses of experiments on Typhoons Doksuri and Gaemi. Section 5 concludes the study.

2. Materials and Methods

2.1. Experimental Design

Numerical experiments were conducted for Typhoons Doksuri (2023) and Gaemi (2024), with 48 h forecasts spanning from 23 July 2023 12:00 UTC to 26 July 2023 12:00 UTC, and 24 July 2024 00:00 UTC to 27 July 2024 00:00 UTC, respectively. We conducted a comparative analysis of the AMSU-A and IASI instruments with the GIIRS. A spatial thinning procedure with 60 km grid spacing was implemented on the observational datasets to mitigate data redundancy in the experimental framework. The numerical simulations utilized Final Operational Global Analysis (FNL) provided by the National Centers for Environmental Prediction (NCEP) with a horizontal grid resolution of 1° × 1° and 6-hourly temporal resolution to generate the initial and boundary conditions [29]. We added conventional observation (con-obs) data to all three groups of experiments to ensure the stability of the experimental results. The experiments details are described in

Table 1. Experimental arrangement.

Experiment Name	Obs	Data Assimilation Method	Typhoon
GIIRS	GIIRS + con-obs	Refinement quality control and 3D cloud method	Gaemi and Doksuri
AMSU-A	AMSU-A + con-obs	WRFDA default method	Gaemi and Doksuri
IASI	IASI + con-obs	WRFDA default method	Gaemi and Doksuri

. In the GIIRS assimilation experiments, we used conventional 45 min resolution observations for Typhoon Doksuri and 15 min resolution observations for Typhoon Gaemi, respectively. (Operational since July 2024, the FY-4B/GIIRS delivers regionally densified observation data targeting high-impact weather processes.) The experimental domain employs a two-way nested grid configuration, with an outer grid resolution of 27 km and an inner grid resolution of 9 km, spanning 34 vertical layers. The inner grid data are updated via interpolation from the outer grid. The experimental domain is illustrated in Figure 1. The experiments were conducted using the WRF-3DVAR system. For parameterization schemes, the following configurations were adopted:

• Longwave radiation: Rapid Radiative Transfer Model for GCMs (RRTMG) scheme [30];
• Shortwave radiation: Dudhia scheme [31];
• Boundary layer: Yonsei University (YSU) scheme [32];
• Surface layer: Monin–Obukhov scheme [33];
• Land surface: Unified Noah Land Surface Model (LSM) [34];
• Cumulus convection: KainFritsch (KF) scheme [35];
• Microphysics: WRF Single-Moment 6-Class Microphysical (WSM6) scheme [36].

2.2. Data Description

During the assimilation process, FY-4B/GIIRS assimilated 41 spectral channels, employing a 3D cloud detection method and a combined quality control approach integrating pixel- and channel-level criteria. IASI and AMSU-A assimilated 166 and 13 channels, respectively, both utilizing the default WRFDA quality control procedures.

2.2.1. FY-4B/GIIRS

The Geostationary Interferometric Infrared Sounder (GIIRS) is the first hyperspectral infrared sounder deployed in geostationary orbit [37] and has high temporal and spectral resolution around the world. The FY-4B GIIRS was developed by the China Aerospace Science and Technology Corporation (CASC) and is operated by the National Satellite Meteorological Center (NSMC). The inaugural GIIRS instrument was successfully launched aboard the FY-4A satellite on 11 December 2016, and officially entered operational service in September 2017. However, due to contamination in 2019, the data quality of GIIRS atmospheric humidity detection significantly degraded [27]. The FY-4B satellite was launched in 2021 and officially replaced FY-4A in operational service in March 2024. Compared to its predecessor, the FY-4B/GIIRS instrument features precision upgrades, including improvements in spectral resolution, spatial resolution and temporal resolution (

Table 2. GIIRS parameters [38].

GIIRS	FY-4B	FY-4A
Spectral Range	LWIR: 680~1130 cm−1 MWIR: 1650–2250 cm−1	LWIR: 700~1130 cm−1 MWIR: 1650–2250 cm−1
Spectral Resolution	<0.625 cm−1	0.625 cm−1
Spatial Resolution	12 km	16 km
Temporal Resolution	45 min	1 h
Radiometric Calibration Accuracy	0.7 K	1.5 K
Spectral Calibration Accuracy	<10 ppm	10 ppm

). Additionally, FY-4B/GIIRS will activate the intensified observation mode for disastrous weather processes (including typhoons), delivering higher-resolution observational data. Compared to the original data, the intensified observation strategy triples the temporal resolution, allowing more data to be assimilated within the same temporal window. Figure 2a,b illustrate the pixel distributions of standard and intensified GIIRS observation data, respectively.

2.2.2. IASI

The Infrared Atmospheric Sounding Interferometer (IASI), on board the European Metop satellites, is a next-generation ultra-hyperspectral infrared atmospheric sounder employing interferometric spectroscopy. The Centre National d’Etudes Spatiales (CNES) is leading the IASI development in association with EUMETSAT. ALCATEL is the instrument Prime Contractor. To mitigate the impact of inter-channel correlations on data assimilation while maintaining assimilation quality, we employed an information entropy-based stepwise iterative method [39] to screen IASI channels, ultimately selecting 167 channels. The channel information is shown in Appendix A 

Table A1. IASI assimilated channels.

Channel_ID	Frequency	Channel_ID	Frequency	Channel_ID	Frequency	Channel_ID	Frequency
1	650	70	667.25	130	682.25	192	697.75
5	651	71	667.5	131	682.5	193	698
9	652	72	667.75	132	682.75	196	698.75
11	652.5	73	668	134	683.25	198	699.25
13	653	75	668.5	135	683.5	199	699.5
15	653.5	76	668.75	136	683.75	200	699.75
16	653.75	77	669	137	684	201	700
17	654	79	669.5	138	684.25	202	700.25
19	654.5	81	670	139	684.5	205	701
20	654.75	85	671	140	684.75	206	701.25
22	655.25	86	671.25	142	685.25	208	701.75
23	655.5	87	671.5	143	685.5	209	702
24	655.75	88	671.75	145	686	211	702.5
27	656.5	90	672.25	146	686.25	230	707.25
28	656.75	91	672.5	147	686.5	237	709
30	657.25	93	673	148	686.75	239	709.5
32	657.75	94	673.25	149	687	241	710
34	658.25	95	673.5	150	687.25	245	711
43	660.5	97	674	151	687.5	254	713.25
44	660.75	99	674.5	152	687.75	256	713.75
45	661	101	675	153	688	257	714
47	661.5	102	675.25	154	688.25	259	714.5
48	661.75	104	675.75	155	688.5	260	714.75
49	662	105	676	157	689	261	715
50	662.25	106	676.25	159	689.5	262	715.25
51	662.5	107	676.5	161	690	263	715.5
52	662.75	108	676.75	162	690.25	265	716
53	663	109	677	165	691	266	716.25
54	663.25	111	677.5	166	691.25	272	717.75
55	663.5	112	677.75	167	691.5	274	718.25
56	663.75	113	678	169	692	276	718.75
57	664	114	678.25	170	692.25	280	719.75
58	664.25	115	678.5	172	692.75	281	720
59	664.5	116	678.75	174	693.25	282	720.25
60	664.75	118	679.25	180	694.75	283	720.5
61	665	120	679.75	181	695	284	720.75
63	665.5	121	680	182	695.25	285	721
64	665.75	122	680.25	183	695.5	286	721.25
65	666	123	680.5	184	695.75	288	721.75
66	666.25	124	680.75	185	696	296	723.75
68	666.75	125	681	186	696.25
69	667	129	682	191	697.5

. IASI operates across a spectral range of 3.62–15.4 μm, encompassing 8461 channels, each with an equal spectral resolution of 0.25 cm⁻¹ and a radiometric resolution of 0.1–0.5 K. With a swath width of approximately 2000 km, IASI provides global coverage every 12 h, delivering two observation cycles daily [40]. Currently, IASI data are widely integrated into diverse numerical models, offering high-reliability data support for weather forecasting systems across multiple nations and regions [41,42,43].

2.2.3. AMSU-A

The Advanced Microwave Sounding Unit-A (AMSU-A) is a 15-channel microwave sounder primarily designed to retrieve atmospheric temperature profiles in the upper atmosphere [44]. The AMSU-A instrument was jointly developed by multiple companies and is operated by the National Oceanic and Atmospheric Administration (NOAA) of the United States. The channel information of AMSU-A is shown in

Table 3. The channel specifications for AMSU-A. The column *f* denotes the central frequency (GHz) of Channel 9. In the ch_use column, a value of 1 indicates active utilization of the channel in retrieval algorithms. A value of −1 flags channel exclusion due to cloud contamination or quality control failures.

Channel_id	Frequency (GHz)	Ch_Use (−1/1)
1	23.8	−1
2	31.4	−1
3	50.3	−1
4	52.8	−1
5	53.596 ± 0.115	1
6	54.4	1
7	54.94	1
8	55.5	1
9	57.29 = f	1
10	f ± 0.217	−1
11	f ± 0.3222 ± 0.048	−1
12	f ± 0.322 ± 0.022	−1
13	f ± 0.3222 ± 0.010	−1
14	f ± 0.3222 ± 0.0045	−1
15	89.0	−1

[45]. It features a nadir-point horizontal resolution of approximately 40 km and an instantaneous scanning angle of 3.3°. The first AMSUA was launched aboard NOAA-15, a satellite operated by the U.S. National Oceanic and Atmospheric Administration (NOAA), in May 1998. As of 2021, seven such instruments on board the NOAA series and METOP series have been successfully deployed. The AMSU-A Microwave radiation detection instrument is not sensitive to cloud signals, and can penetrate the clouds and obtain the atmospheric state information under the clouds. Microwave sounders provide critical data sources for global numerical weather prediction (NWP) systems.

2.3. Two Super Typhoons

Typhoon Doksuri (International Designation: 2305), the fifth typhoon of 2023, formed over the western North Pacific Ocean on 21 July 2023, and intensified into a super typhoon around 23 July [46]. On 26 July, Doksuri made landfall in the Philippines and continued its northward trajectory. It weakened to a severe typhoon after traversing the Bashi Channel, but re-intensified into a super typhoon upon entering the South China Sea on 27 July. On 28 July, Doksuri made landfall in Jinjiang Fujian Province, China, at super typhoon intensity.

Typhoon Gaemi (International Designation: 2403) [47] formed over the ocean east of the Philippines at 14:00 UTC on 20 July 2024. It intensified into a super typhoon by 08:00 UTC on 24 July and maintained this intensity while executing a looping motion over the water northeast of Taiwan Island. Around 00:00 UTC on 25th July, it made landfall in Nan’ao Township, Yilan County, Taiwan Province, as a strong typhoon. After landfall, it crossed the northern part of Taiwan Island and entered the Taiwan Strait, where its intensity gradually weakened. Around 19:50 UTC on 25th July, it made landfall again along the coast of Xiuyu District, Putian City, Fujian Province, at the typhoon level.

Typhoons Doksuri and Gaemi caused severe disasters to the countries along their paths. The residual atmospheric disturbances they generated triggered extreme heavy rainfall, causing serious damage to human lives and property. Figure 3a,b show the tracks and intensity information of the two typhoons, respectively.

3. Pre-Experiment Processing

3.1. Channel Selection

In the GIIRS experiment, we only assimilated 41 selected long-wave infrared channels, according to the experience from other NWP centers; for example, initial assimilation at ECMWF only used channels in the longwave CO₂ band, as channels in other bands are noisier than the longwave CO₂ band [48]. Analysis of their weighting function peaks reveals that the assimilated channel signals originate from distinct atmospheric layers, exhibiting lower inter-channel correlations. This aligns with the fundamental assumption in assimilation methodology requiring uncorrelated observations (Figure 4a). These channels are mainly temperature-sounding channels. The peaks of the temperature-sounding channels are located near 500 hPa and 800 hPa (Figure 4b), providing a strong capability for detecting temperatures in the middle and lower atmosphere. In the experiment, only the clear-sky radiance data from these channels are assimilated.

3.2. Quality Control and Bias Correction

Quality Control (QC) of observational data constitutes a paramount step in the data assimilation workflow, with its efficacy directly determining the accuracy of the resultant analysis field [49]. In the GIIRS assimilation experiment, a hybrid quality control method was applied. We screened the GIIRS data through two approaches: pixel detection and channel detection. For details, see

Table 4. Quality control.

Ordinal	Detection Method	Quality Control Options	Quality Control Criteria	Mark
1	Pixel Detection	Underlying Surface Detection	surf_type = 1	1
3		Scanning Angle Detection	satzen > 70	1
4		Scan Array Limb Effect Detection	iscanpos = 1–16, 30–35, 62–67, 94–99, 113–128	1
5	Channel Detection	Outlier Channel Rejection	channel = 33–34, 65, 70, 72, 77, 79, 89	1
6		Channel Rejection	channel = 1–35, 37, 39–40, 42, 44, 45, 49–53, 55, 58, 60, 62–101, 103, 106, 108, 113–115, 118, 123, 125–129, 131, 133–140, 132, 143, 146–153, 158, 159, 161–167, 169–256, 258–276, 278–721	1
7		Extreme Value Detection	rad < 150 or rad > 350	1
8		Bi-weight Robust Quality Control	Z > 3.0	1
9		(OMB) Threshold Control	|drad| > 1.5 or |drad + bias| > 2.0	1

.

Given the significant susceptibility of the GIIRS to cloud interference [50], this study adopts the three-dimensional (3D) cloud detection method proposed by Yan et al. [51] in 2023 for FY-4A GIIRS. This method calculates the impact of clouds on channel radiance brightness temperatures using the infrared radiative transfer equation. By integrating the cloud mask (CLM) and cloud top pressure (CTP) products from the co-mounted Advanced Geostationary Radiation Imager (AGRI) on the same platform, it transforms two-dimensional (2D) cloud distributions into 3D spatial representations of cloud radiative effects at varying altitudes on specific channel radiance transmission. Figure 5a displays the distribution of GIIRS observations after 60 km thinning. Figure 5b illustrates the data retained after 3D cloud detection, quality control, and 60 km thinning. Figure 5c shows the volume of assimilated data following 2D cloud detection, quality control, and thinning. Compared to traditional 2D cloud detection methods, 3D cloud detection retains partially cloudy pixels minimally affected by cloud contamination, thereby improving data utilization.

Bias correction is a critical stage in satellite data assimilation. This experiment employs the variational bias correction (Var-BC) method. Originally proposed by Dee [52] in 2005, Var-BC has been implemented in numerous global operational assimilation systems over the past two decades. Figure 6 shows the probability density function (PDF) distributions of observation minus background (OMB) differences for channels 38, 84, and 112 before and after bias correction. The weighting function peaks for these channels are located at 207 hPa, 490 hPa, and 825 hPa, respectively. Following bias correction, systematic errors for channels were effectively mitigated, with the OMB statistics demonstrating a closer alignment with the unbiased Gaussian distribution assumption inherent to variational assimilation frameworks.

4. Results

We compared typhoon wind field forecasts from the three experiments against NCEP FNL data, and evaluated their 48 h track forecasts against observed best tracks. The NCEP FNL dataset provides 1° × 1° gridded atmospheric parameters with 6-hourly temporal resolution, including geopotential height, wind fields, and vertical temperature profiles essential for TC verification [52]. Reanalysis data integrate diverse observational sources and model-based physical processes, demonstrating a high degree of physical consistency across both temporal and spatial dimensions. Their use as a reference dataset holds significant value for scientific research [23,27,53,54,55,56,57]. The assimilation times for typhoons Doksuri and Gaemi were 12:00 UTC on 23 July 2023 and 00:00 UTC on 24 July 2024, respectively. The forecast analysis time was set to T + 0 immediately following data assimilation.

4.1. Wind Field Analysis

We evaluated the forecast performance of AMSU-A, GIIRS and IASI experiments with respect to wind fields. Figure 7 and Figure 8 show the wind fields and wind field biases at 300 hPa and 500 hPa for Typhoon Doksuri, respectively. Figure 9 and Figure 10 show the wind fields and wind field biases at 300 hPa and 500 hPa for Typhoon Gaemi, respectively.

At 300 hPa, the wind forecast fields of the three experiments for Typhoon Doksuri are relatively similar but deviate from the FNL value. The main bias is to the eastern side of the typhoon’s center, where the forecasted wind speed is lower than the FNL values, with a significant difference. The maximum absolute difference is close to 10 m/s. The forecasts from AMSU-A and FY-4B/GIIRS have smaller differences from the FNL values (Figure 7e,f), while the IASI forecast has a larger error (Figure 7g). Additionally, in the northern part of the typhoon area, the IASI forecast shows another significant low-value bias area in the wind field.

At 500 hPa, AMSU-A, GIIRS, and IASI experiments all provide better forecasts for the high-value area in the eastern part of the typhoon, but fail to capture the maximum wind speed in the western part. The forecasted wind field has a significant difference from the true field in the southeast side of the typhoon area, with bias greater than 10 m/s. This may be attributed to the location of this region at the junction of island chains and the ocean, where the complex structure of the atmospheric underlying surface affected the simulation performance of the boundary layer parameterization scheme, consequently compromising the model’s forecast accuracy [58,59,60]. The three experimental forecasts for Typhoon Doksuri accurately display the main location of the typhoon and the wind field information around the eyewall. While all three instruments effectively captured central position and eyewall wind field characteristics of the typhoon, their accuracy diminished in forecasting peripheral wind speeds.

Figure 9 shows the FNL experiment wind field, AMSU-A experiment wind field, GIIRS experiment wind field, IASI experiment wind field, and the differences between the FNL and the wind fields from the three experimental forecasts at 300 hPa for Typhoon Gaemi at 00:00 UTC on 24 July 2024. The forecasted wind fields from the three experiments exhibited significant discrepancies at 300 hPa. From Figure 9e, the AMSU-A experiment displayed a pronounced high-wind zone in the northwest of the typhoon center but underestimated wind speeds on the eastern side. The GIIRS experiment showed a similar spatial distribution of high winds to AMSU-A but with magnitudes closer to observations. As shown in Figure 9g, in the IASI experiment, IASI demonstrated improved performance, accurately capturing the orientation of high-wind zones while maintaining wind speed errors within ±4 m/s.

Figure 10 display the wind speed forecast fields of Typhoon Gaemi at 500 hPa from the three experiments, as well as the difference fields between these forecasts and the FNL data. All experiments show the wind fields closely aligned with FNL data. AMSU-A forecasts slightly overestimated wind speeds in the core of Typhoon Gaemi, whereas IASI and GIIRS performed better, with the bias from −2 to 4.

The wind forecast biases across the three experiments varied with altitude and typhoon location. Larger biases were concentrated in high-wind zones near the typhoon center, while peripheral wind fields and trailing regions were better forecasted. GIIRS demonstrated relatively stable performance in terms of typhoon horizontal wind speed forecasting, likely due to differences in satellite orbits, scanning patterns, and assimilation channels.

Figure 11 depicts the vertical wind fields of Typhoon Doksuri forecasted by the three experiments and the simulated wind fields from the FNL data at 1200 UTC on 23 July 2023. The four figures on the left show the mean tangential wind of the typhoon, while the four figures on the right show the mean radial wind of the typhoon. In Figure 11a,b, for Typhoon Doksuri, AMSU-A demonstrates the closest alignment with the FNL wind field. In Figure 11c,d GIIRS simulations reveal a weakened circulation core alongside overestimated outflow and underestimated inflow compared to FNL. As shown in Figure 11e,f, IASI exhibits substantial biases with an anomalously intensified wind field structure, including lower-tropospheric tangential winds exceeding 40 m/s that erroneously suggest Category 2 typhoon intensity, inconsistent with the NCEP FNL data (a similar result for Typhoon Gaemi is omitted).

In addition to the study of the wind field, we also evaluated the performance of the three experiments in forecasting the typhoon’s relative humidity field and temperature field, with detailed experimental results provided in the Supplementary Materials. Experimental findings revealed that the AMSU-A experiment demonstrated favorable forecasting results for both typhoon elements. The GIIRS experiment achieved a forecasting performance largely comparable to the AMSU-A experiment, showing results closely aligned with the reanalysis data simulations. In contrast, the IASI experiment performed less effectively than both the GIIRS and AMSU-A experiments across the two sets of results, exhibiting more pronounced discrepancies from the reanalysis data, particularly in the relative humidity forecasts.

4.2. Track Error Analysis

We conducted 48 h typhoon track forecasts for the three experiments and the control experiment (CTL) with corresponding trajectory error analysis. In the control experiment, only conventional observational (Con_OB) data were assimilated, while all other parameter configurations remained identical to those used in the instrument assimilation experiments. Figure 12 depicts a comparative analysis of forecast tracks versus observed best tracks for both typhoons across the three experiments, while Figure 13 illustrates forecast track bias of the three experiments and CTL for the two typhoons. Results demonstrate that all three experiments and Con_OB captured the fundamental movement trends of the typhoon, though with observable variations. Compared to the simulation results of the three instruments, the CTL showed significant differences in its performance for Typhoon Doksuri and Typhoon Gaemi. For Typhoon Doksuri, the track forecast bias in the control experiment increased over time, whereas for Typhoon Gaemi, the track bias remained relatively stable, with performance close to that of the GIIRS assimilation. This discrepancy may be attributed to the fact that Typhoon Doksuri remained over open ocean during the forecast period, where observational data in the typhoon region were relatively sparse. Specifically, the GIIRS experiment initialized the typhoon position closer to observations (tcdata.typhoon.org.cn) [61,62,63] at the forecast outset. During the 12–24 h forecast lead time, discernible biases emerged in its forecasted trajectory, but beyond 24 h, the GIIRS forecast progressively converged toward the actual path. The GIIRS experiment exhibited smaller overall track errors, maintaining bias predominantly below 50 km throughout the typhoon forecast period for Gaemi. Furthermore, it achieved lower RMSE values compared to other instruments as shown in

Table 5. The RMSE of track errors for the two typhoons, forecasted by different instruments. The bold type in the table indicates the smallest RMSE of track forecast for each of the two typhoons.

Experiment	Doksuri	Gaemi
AMSU-A	133.23 km	133.95 km
GIIRS	115.34 km	62.13 km
IASI	135.84 km	90.22 km
CTL	172.96 km	63.91 km

, collectively highlighting the significant potential of FY-4B/GIIRS in advancing typhoon track forecast capabilities.

In the horizontal wind field analysis, the FY-4B/GIIRS assimilation experiment demonstrated superior consistency with reanalysis data and exhibited enhanced stability compared to AMSU-A and IASI. However, it shows systematic underprediction in vertical velocity forecasts. For 48 h track forecasting, FY-4B/GIIRS delivers exceptional performance with minimized track errors, particularly after assimilating intensified GIIRS observations, where track errors of Typhoon Gaemi became significantly smaller than those in AMSU-A and IASI experiments. This can be attributed to two key factors: Firstly, FY-4B/GIIRS boasts outstanding instrument specifications. Although the FY-4 series satellites are geostationary, they achieve higher spatial resolution. Compared to polar-orbiting satellite instruments, GIIRS provides a larger field of view and higher temporal resolution, enabling more data to be assimilated into forecasting systems. Secondly, through quality control and cloud detection, we retained the high-altitude channels at the typhoon center and the cloud-free peripheral channels of FY-4B/GIIRS, providing high-precision observational data. This approach has enhanced the accuracy of typhoon wind forecasts.

5. Discussion

In operational systems, satellite data are often combined to obtain richer atmospheric profile information. Numerous studies have shown [64,65,66] that assimilating multi-source observational data improves initial field accuracy, thereby enhancing forecast performance. Accurate forecasting of the lower atmosphere is critical for human productivity and safety, yet hyperspectral instruments face inherent limitations in observing this layer. In contrast, microwave sensors penetrate clouds and provide richer atmospheric information near the surface compared to infrared instruments. Here, we conducted a preliminary effort to jointly assimilate AMSU-A and GIIRS observations, with comparative analysis against a control experiment without data assimilation. The assimilation was initialized at 00:00 UTC on 24 July 2024 with a 24 h forecast lead time, corresponding to the imminent landfall phase of the typhoon.

Figure 14 and Figure 15 present comparative analyses of forecast performance for 2 m temperature and 10 m wind speed (respectively) for Typhoon Gaemi. Compared to the control experiment, the combined assimilation of both datasets yields significant improvements. The joint assimilation experiment shows that the temperature field forecast biases to within ±1 °C. Additionally, substantial enhancements were observed in 10 m wind field forecasts, particularly within and around the typhoon regions. The control experiment underestimated wind speeds in both the typhoon core and its surroundings, whereas the assimilated forecasts showed marked improvements, with overall bias constrained to ±2 m/s.

6. Conclusions

This study focuses on the impact of assimilating data from the FY-4B/GIIRS on typhoon forecasting, with comparative analyses against AMSU-A and IASI. Numerical experiments were conducted for two super typhoons, Doksuri and Gaemi. Through comprehensive comparisons of wind fields and typhoon trajectories, the study elucidates the influence of FY-4B/GIIRS assimilation on typhoon forecasting and delivers valuable implementation references for operational applications.

The results demonstrate that numerical forecasts assimilating FY-4B/GIIRS data exhibit advantages in element fields and typhoon trajectory forecast compared to those utilizing AMSU-A and IASI. The GIIRS experiment demonstrates skillful forecasting of the horizontal wind field at 300 hPa and 500 hPa levels, and shows better stability compared with IASI and AMSU-A experiments. The wind speed biases in most areas were within the range of −6 m/s to 6 m/s. However, the GIIRS experiment demonstrates inferior forecast skill in vertical velocity compared to AMSU-A, while its forecasts for tangential and radial wind speeds exhibit discernible discrepancies relative to FNL results. Beyond its performance in wind field forecasting, the GIIRS experiment also demonstrated relatively stable forecasting results for the 500 hPa temperature and 850 hPa relative humidity fields of the typhoons. This indicates GIIRS advantage in capturing atmospheric information within the mid-to-upper levels of typhoons. The GIIRS experiment performed well in 48 h track forecast. Compared to AMSU-A, IASI and CTL experiments, the GIIRS experiment delivered superior performance with smaller track errors and greater statistical reliability in forecasting two typhoons. Additionally, assimilating intensified GIIRS observations demonstrated a pronounced positive impact on typhoon track forecast accuracy. Furthermore, this study conducted extended research on improving typhoon forecasting through the joint assimilation of AMSU-A and FY-4B/GIIRS, while exploring broader application prospects for FY-4B/GIIRS. These efforts have accumulated operational experience in multi-source data assimilation.

While FY-4B/GIIRS demonstrated promising performance in this study, its application efficacy was partially constrained by the intrinsic limitations of hyperspectral infrared instruments in probing the lower-tropospheric atmosphere. Additionally, the default cloud detection and quality control protocols applied to IASI and AMSU-A in the experiments may affect the stability of the two experiments’ outputs. Currently, FY-4B/GIIRS assimilation is limited to clear-sky radiance channels. Future work will prioritize the development of observation error covariance matrices tailored for cloud-affected scenarios, aiming to unlock the full potential of FY-4B/GIIRS under all-sky conditions.