Impact of FY-3D MWRI and MWHS-2 Radiance Data Assimilation in WRFDA System on Forecasts of Typhoon Muifa

Abstract

This study investigates the impact of assimilating FY-3D Microwave Radiation Imager (MWRI) radiance data into the Weather Research and Forecasting (WRF) model, utilizing a 3D-Var data assimilation system, on the forecast accuracy of Typhoon Muifa (2022). The research focuses on the selection of data from different channels, land/ocean coverage, and orbits of the MWRI, along with the synergistic assimilation strategy with MWHS-2 data. Ten assimilation experiments were conducted, starting from 0600 UTC on 14 September 2022, covering a 42 h forecast period. The results show that after assimilating the microwave radiometer data, the brightness temperature deviation in the ocean area was significantly reduced compared to the simulation without data assimilation. This led to an improvement in the accuracy of typhoon track and intensity predictions, particularly for predictions beyond 24 h. Furthermore, the assimilation of land data and single-orbit data (particularly from the western orbit) further enhanced forecast accuracy, while the joint assimilation of MWHS-2 and MWRI data yielded additional error reductions. These findings underscore the potential of satellite data assimilation in improving typhoon forecasting and highlight the need for optimal land observation and channel selection techniques.

1. Introduction

With the application of various advanced remote sensing data in data assimilation providing more accurate initial conditions for numerical models, the performance of numerical weather forecasting has been significantly improved [1,2,3,4,5,6]. Satellite observations facilitate comprehensive monitoring of Earth’s atmospheric conditions, significantly enhancing the precision of numerical weather predictions. Data from the European Centre for Medium-Range Weather Forecasts (ECMWF) indicate that satellite-derived information constitutes over 90% of the data integrated into global numerical forecasting assimilation systems [7]. Progress in satellite sensing technologies has markedly improved the resolution of atmospheric observations in both spatial and temporal dimensions. Satellite-mounted sensors demonstrate strong responsiveness to atmospheric temperature and moisture profiles [8]. Consequently, incorporating data from microwave or infrared instruments through assimilation methods substantially enhances the precision of weather predictions. The penetration capability of microwave radiation through clouds, fog, and water vapor ensures reliable data acquisition under various atmospheric conditions [9,10,11,12,13].

The Fengyun (FY) satellite series, a cornerstone of China’s comprehensive meteorological and environmental monitoring efforts, encompasses two generations of geostationary and polar-orbiting satellites. FY-3D, as the fourth experimental satellite in China’s second-generation polar-orbiting series, plays a pivotal role in advancing these applications. The FY-3D satellite is equipped with 10 advanced remote sensing instruments. In addition to the five inherited instruments, microwave temperature meters, microwave humidity meters, microwave imaging instruments, space environment monitoring instrument packages, and global navigation satellite occultation detectors, the infrared high-spectral atmospheric detection instrument, near-infrared high-spectral greenhouse gas monitoring instrument, wide-angle aurora imaging instrument, and ionospheric photometer have been newly developed and are being carried into space for the first time. The core instrument, the high-resolution spectral imaging instrument, has been significantly upgraded and improved in performance.

The evolution of microwave imaging technology began with the Special Sensor Microwave Imager (SSMI), manufactured by Hughes Aircraft Company in California, USA, and launched on the Defense Meteorological Satellite Program (DMSP) F-8 satellite [8]. This was followed by the introduction of the Special Sensor Microwave Imager and Sounder (SSMIS), also manufactured by the Hughes Aircraft Company, and deployed on the DMSP F-16 and F-17 satellites [14,15]. Subsequently, the Tropical Rainfall Measuring Mission (TRMM) Microwave Imager, built by NASA Goddard Space Flight Center in Greenbelt, Maryland, USA [16] and the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E), developed by the Japan Aerospace Exploration Agency (JAXA) in Japan [17] were placed into orbit on the Aqua satellite. The Advanced Microwave Scanning Radiometer 2 (AMSR2), succeeding the AMSR-E and developed by JAXA in Japan, and mounted on the Global Change Observation Mission—Water 1 (GCOM-W1) satellite [18], is extensively utilized in operational forecasting models. The Microwave Radiometric Imager (MWRI), a key payload on the China Meteorological Administration’s (CMA) Fengyun-3 (FY-3) satellite, inherits many capabilities of these predecessors.

Additionally, the assimilation of MWRI data has been demonstrated to improve the performance of numerical weather prediction (NWP) models. Yang et al. were the first to investigate the impact of assimilating MWRI data from the Fengyun-3B satellite into the Weather Research and Forecasting Data Assimilation (WRFDA) System on the analysis and forecasting of Typhoon Son-Tinh in 2012 [8]. Their study demonstrated that MWRI assimilation significantly improved typhoon structure analysis and forecast accuracy by providing a more precise cloud distribution, a higher correlation with satellite observations, a stronger central sea level pressure (CSLP) and a warm core, and a maximum wind speed analysis closer to optimal tracking data, which substantially reduced forecast errors.

Lawrence et al. (2017) conducted a comprehensive evaluation of the MWRI on the FY-3C satellite, identifying several significant limitations [19]. These include an ascending–descending orbital bias of approximately 2 K across all channels, inter-satellite biases of 4–6 K relative to AMSR-2, radio frequency interference (RFI) affecting the 10.65 GHz channels (notably in descending passes over Europe), and a temporal drift in the ascending–descending bias, which increased from 1 K to 2 K between 2014 and 2016. These issues highlight challenges in data calibration and assimilation for numerical weather prediction and reanalysis applications. Due to these deficiencies, the direct assimilation of MWRI data into NWP systems was severely restricted. In response to these prior findings, Carminati et al. (2021) turned their attention to the MWRI instrument onboard the FY-3D satellite [20]. Through the application of enhanced calibration techniques, the ascending–descending bias was successfully reduced to a minimal 0.17 K. Not only were global biases mitigated, but the standard deviation of observation-minus-background (OMB) differences also showed substantial reductions, highlighting the enhanced performance of the updated instrument. Meanwhile, the potential of FY-3D MWRI radiance data for operational forecasting was explored by Xiao et al. (2020) through its direct assimilation into the Global/Regional Assimilation and Prediction System (GRAPES) using a four-dimensional variational (4DVar) approach [21]. In contrast to a control experiment, the assimilation not only yielded more accurate temperature and humidity analyses but also contributed to a better prediction of the subtropical high’s position. Consequently, the overall forecast accuracy for Typhoon Shanshan was significantly improved.

Previous studies on assimilating MWRI data from the FY satellite series have predominantly excluded data over land surfaces, resulting in limited research on the impact of assimilating MWRI data over land on typhoon forecasting. Furthermore, there is a lack of testing for channel sensitivity, with most studies adopting a strategy of simultaneously assimilating channels 3, 5, and 7. The influence of various assimilation approaches for multi-orbit MWRI data on forecasting accuracy also remains uncertain. Beyond the singular assimilation of MWRI data, there is also a paucity of studies exploring the joint assimilation of MWRI data with other microwave instrument data from the same satellite. The aim of this research is to assess the impact of assimilating FY-3D MWRI data and various assimilation approaches on the precision of typhoon analysis and forecasting outcomes. Additionally, a preliminary assessment is conducted to examine the impact of jointly assimilating FY-3D MWRI and MWHS-2 data on the accuracy of analysis and forecasting in typhoon prediction.

2. Data and Methods

2.1. FY-3D MWRI Radiance Data

This study employs FY-3D MWRI Level 1 (L1) data retrieved from the website of the National Satellite Meteorological Center of the China Meteorological Administration (https://satellite.nsmc.org.cn/DataPortal/cn/home/index.html), accessed on 10 August 2024. The scanning configuration of the MWRI closely resembles that of operational conical-scanning microwave imaging instruments currently in orbit, such as SSMIS, AMSR2, and GMI. The microwave imager conducts conical scans over a 1400 km wide terrestrial swath at a zenith angle of 53.1°. The number of ground scanning points varies according to the specific scanning cycle mode employed. The microwave imager features three distinct scanning modes with cycle durations of 1.7 s, 1.8 s, and 2.0 s. Under its default operational setting, the scanning cycle lasts 1.8 s, yielding 266 scanning points per scan line. Within a frequency range of 10.65–89 GHz, the microwave imager is equipped with five frequency channels, each incorporating both vertical and horizontal polarization modes. The dual polarization channels at the lowest frequency show the highest sensitivity to surface emissivity, while those at the highest frequency are most effective in detecting areas of convective precipitation. The 19 V/H channels are sensitive to surface winds and oceanic rainfall, the 24 V/H channels are employed to measure atmospheric water vapor, and the 37 V/H channels are utilized to detect liquid water content within clouds [21].

Table 1. FY-3D MWRI channel specifications.

Channel	Central Frequency	Polarizations	Bandwidth (MHz)	NEΔT (K)	IFOV (km2)	Pange (K)
1	10.65	V	180	0.5	51 × 85	40 × 11.2
2	10.65	H	180	0.5	51 × 85	40 × 11.2
3	18.7	V	200	0.5	30 × 50	40 × 11.2
4	18.7	H	200	0.5	30 × 50	40 × 11.2
5	23.8	V	400	0.5	27 × 45	20 × 11.2
6	23.8	H	400	0.5	27 × 45	20 × 11.2
7	36.5	V	400	0.5	18 × 30	20 × 11.2
8	36.5	H	400	0.5	18 × 30	20 × 11.2
9	89	V	3000	0.8	9 × 15	10 × 11.2
10	89	H	3000	0.8	9 × 15	10 × 11.2

presents an overview of the MWRI channel specifications.

2.2. WRFDA System

The experiments in this research utilized the WRFDA System (version 4.3), developed by the National Center for Atmospheric Research (NCAR), integrated with its three-dimensional variational (3D-Var) data assimilation framework [22]. This system encompasses multiple assimilation approaches, including 3D-Var, 4D-Var, and hybrid assimilation techniques. This research employed the well-established 3D-Var assimilation technique. By iteratively optimizing a cost function, the 3D-Var method facilitates precise estimation of atmospheric conditions. This method produces an enhanced analysis of the atmospheric condition by iteratively optimizing a designated cost function, which is expressed as

$$J\left(x\right)={\left(x-x_b\right)}^T{B}^{-1}\left(x-x_b\right)+{\left[y_0-H\left(x\right)\right]}^T{R}^{-1}\left[y_0-H\left(x\right)\right].$$

(1)

In the formula, the analysis vector is denoted by $x$, the background vector by $x_b$, and the observation vector by $y_0$, while $B$ represents the background error covariance matrix, $H$ the observation operator, and $R$ the observation error covariance matrix. Minimizing the cost function facilitates the identification of the ideal state vector $x$, enabling enhanced consistency between model predictions and observational data [23].

The background error covariance (BEC) matrix $B$ weights background and observation information. The BEC is estimated using the National Meteorological Center (NMC) method, approximating background errors from differences between model forecasts of varying lead times (e.g., 24 h and 12 h forecasts for regional domains) [24]. The WRFDA System applies a control variable transform (CVT) to convert model variables into control variables, making $B$ block-diagonal and imposing balance constraints. This study uses the CV5 option for regional BEC statistics, effective for simulating mesoscale phenomena like typhoons [25]. In CV5, control variables include full fields of the stream function ($\mathsf{\psi }$) and relative humidity (rh) and unbalanced components of velocity potential (${\mathsf{\chi }}_{\mathrm{u}}$), temperature (${\mathrm{T}}_u$), and surface pressure ($p{s}_u$), can be expressed by the following balance equations:

$${\mathsf{\chi }}_{\mathrm{u}}\left(i,j,k\right)=\mathsf{\chi }\left(i,j,k\right)-{\mathsf{\alpha }}_{\mathsf{\psi }\mathsf{\chi }}\left(i,j,k\right)\mathsf{\psi }\left(i,j,k\right),$$

(2)

$${\mathrm{T}}_u\left(i,j,k\right)=T\left(i,j,k\right)-{\displaystyle \sum _{l=1}^{N_k}{\mathsf{\alpha }}_{\mathsf{\psi }T}}\left(i,j,k,l\right)\mathsf{\psi }\left(i,j,l\right),$$

(3)

$$p{s}_u\left(i,j\right)=ps\left(i,j\right)-{\displaystyle \sum _{l=1}^{N_k}{\mathsf{\alpha }}_{\mathsf{\psi }ps}}\left(i,j,l\right)\mathsf{\psi }\left(i,j,l\right),$$

(4)

where $i$ and $i$ are horizontal indices, $k$ and $l$ are vertical sigma levels, and $\mathsf{\alpha }$ represents regression coefficients. Relative humidity is analyzed independently [26].The CV5 BEC is generated via the NMC method using forecast differences from a 1-month period during the 2022 typhoon season (24 h and 12 h forecasts) [26].The 3D-Var assimilation is performed once at 0600 UTC on 14 September 2022, with a ±3 h observation window around the analysis time, incorporating available observations (e.g., MWRI data at 0356 and 0538 UTC). A 42 h free forecast is conducted without cycling.

2.3. Quality Control

This experiment implemented several quality control methods for the MWRI radiance data: (1) All pixels over land were excluded [8]. (2) Due to solar glare interfering with microwave radiation measurements in low-frequency channels, observational data from the 10.6 GHz channel were discarded when the glare angle was less than 25 degrees [8]. (3) In cases where the CRTM cloud model was not utilized, the cloud liquid water path (CLWP) for each channel was examined, and CLWP values exceeding a predefined threshold were rejected [8]. (4) Observational residual check: Pixels with bias-corrected observed brightness temperature deviations exceeding 3 K were eliminated [27]. (5) Observational error check: Pixels with bias-corrected observed brightness temperature deviations exceeding three times the standard deviation of the observational error were removed [27]. (6) Outlier detection: Pixels with observed brightness temperatures below 70 K or above 320 K were excluded [27].

3. Experimental Design

3.1. Summary of Typhoon Event

Typhoon Muifa emerged in the Northwest Pacific at 1400 UTC on 6 September 2022, with an initial central pressure of 1002 hPa. Throughout its development, “Muifa” exerted significant impacts on mainland China, Taiwan, Japan, and other regions (Figure 1). After forming, Typhoon Muifa strengthened rapidly and experienced several intensity variations along its track. Remarkably, Muifa struck China’s coastline four times, a rare event since 1949 and the first documented typhoon to reach Northeast China. Typhoon Muifa had a substantial impact on the Yangtze River Delta and Shandong Province, leading to emergency actions such as railway closures and disruptions to passenger train operations.

On 14 September 2022, at 1230 UTC, Typhoon Muifa first made landfall in Zhoushan, Zhejiang, at the intensity of a severe typhoon. It was the most intense typhoon to strike China in 2022, with a central pressure of 965 hPa and sustained winds reaching 40 m/s. Later that day, at 1630 UTC, it made its second landfall. On 15 September 2022, “Muifa” struck Fengxian, Shanghai, at typhoon strength, setting a record as the most intense typhoon to hit Shanghai in nearly 70 years. Later, on 16 September 2022, at 1600 UTC, it made its third landfall in Laoshan, Qingdao, and the coastal areas of Shandong Province, subsequently weakening into a tropical storm. At 0440 UTC on 16 September 2022, Typhoon Muifa completed its final landfall in Dalian, Liaoning Province, having transformed into an extratropical cyclone. The China Meteorological Administration discontinued its typhoon designation at 1200 UTC on the same day.

3.2. Experiment Configuration

This study employs the Advanced Research WRF (ARW) version 4.3. Figure 2 illustrates the computational domain, and

Table 2. Model configuration and domain settings.

Component	Details
Model Version	Advanced Research WRF (ARW) 4.3
Computational Domain	(31°N, 123°E)
Grid Points	559 × 469
Horizontal Resolution	9 km
Vertical Levels	61 levels, extending to 10 hPa
Initial/Boundary Data	ERA-5 dataset (0.25° × 0.25° resolution)

provides details on the model configuration and domain settings. The CV5 method is utilized to calculate the background error covariance matrix (B). The model incorporates various physical parameterization schemes, including the Kain–Fritsch cumulus convection scheme [28], the Noah Land Surface Model, the Rapid Radiative Transfer Model (RRTM) for longwave radiation [29], the Dudhia shortwave radiation scheme [30], the Yonsei University (YSU) planetary boundary layer scheme [31], and the WRF Single-Moment 6-Class Microphysics (WSM6) scheme [32]. The microwave imager is equipped with Earth observation functions at five frequency bands, namely, 10.65 GHz, 18.7 GHz, 23.8 GHz, 36.5 GHz, and 89 GHz, each incorporating dual polarization modes (vertical and horizontal), yielding a total of ten channels. The spatial resolution of the microwave imager’s channels for Earth observation varies across frequencies, ranging between 10 km and 70 km. Channels 9 and 10 (89.0 GHz) exhibit heightened sensitivity to convective precipitation, whereas channels 1 and 2 (10 GHz) are more significantly influenced by surface emissivity [21].

This study developed 11 parallel experimental setups, as detailed in

Table 3. The comparison of each experiment.

Number	Experiment Name	Data Used	Channel	Land Data	MWRI Orbit
1	CTRL
2	MWRI_3_Ocean	GTS and MWRI	3	Not In Use	Both
3	MWRI_5_Ocean	GTS and MWRI	5	Not In Use	Both
4	MWRI_7_Ocean	GTS and MWRI	7	Not In Use	Both
5	MWRI_357_Ocean	GTS and MWRI	357	Not In Use	Both
6	MWRI_3	GTS and MWRI	3	Use	Both
7	MWRI_5	GTS and MWRI	5	Use	Both
8	MWRI_3_East	GTS and MWRI	3	Use	East
9	MWRI_3_West	GTS and MWRI	3	Use	West
10	MWHS_MWRI_2STEP	GTS, MWRI, and MWHS2	3/11-15	Use	West
11	MWHS	GTS and MWHS2	3/11-15	Use

, to assess the effects of assimilating multiple datasets, including conventional data from the Global Telecommunications System (GTS), FY-3D MWRI, and FY-3D MWHS-2 observations. In this study, the assimilated GTS data includes conventional observations such as radiosonde data (wind, temperature, relative humidity, and pressure), surface observations (wind, temperature, relative humidity, and pressure), ship reports (sea surface temperature and pressure), and aircraft reports (wind and temperature). Furthermore, this research investigated the impacts of sequential joint assimilation of MWRI and MWHS-2 observations. The modeling timeframe extended from 0000 UTC, 14 September 2022, to 0000 UTC, 15 September 2022. The control experiment (CTRL) did not assimilate any data and conducted a 48 h simulation of the typhoon. The background for all assimilation experiments was obtained from a six-hour spin-up forecast initiated at 0000 UTC on 14 September, with assimilation conducted at 0600 UTC on the same day.

Experiments 2–5 were channel sensitivity tests aimed at selecting an optimal MWRI channel configuration. The FY-3D MWRI consists of a total of 10 channels. Channels 9 and 10 (89.0 GHz) exhibit higher sensitivity to convective precipitation, while channels 1 and 2 (10 GHz) are more significantly affected by surface emissivity. This study conducted assimilation experiments using the vertically polarized channels at 18.7 GHz, 23.8 GHz, and 36.5 GHz, designing MWRI_3_Ocean, MWRI_5_Ocean, MWRI_7_Ocean, and a commonly used MWRI channel scheme, MWRI_357_Ocean. Since most of the MWRI data in this study originated from land, two experiments, MWRI_3 and MWRI_5, were designed. The objective was to integrate quality-controlled land data by eliminating data screening steps. This approach was compared with the results from the MWRI_3_Ocean and MWRI_5_Ocean experiments, which demonstrated strong performance in channel sensitivity tests, to assess the impact of land data on assimilation.

As shown in Figure 2, within 3 h of 0600 UTC on 14 September, there were two MWRI scanning orbits available near the center of the typhoon: one at 0356 UTC was located to the east of the typhoon center, and the other at 0538 UTC was located to the west of the typhoon center. The MWRI_3 experiment simultaneously assimilated the MWRI data from both orbits. To select the best assimilation scheme, two sets of experiments, MWRI_3_East and MWRI_3_West, were designed to test the effects when only assimilating the MWRI data from the eastward orbit and when only assimilating the MWRI data from the westward orbit.

The MWHS_MWRI_2STEP experiment involved the joint assimilation of FY-3D MWHS-2 and MWRI data in two steps. First, data from MWHS-2 channels 11–15 were assimilated at 0600 UTC on September 14 [23], providing the background for the subsequent step, which involved assimilating MWRI data at the same time point. The MWHS experiment, assimilating only FY-3D MWHS-2 data, served as a control for MWHS_MWRI_2STEP to investigate the impact of jointly assimilating MWRI data on the MWHS-2 assimilation results. All assimilation experiments performed a 42 h deterministic forecast from 0600 UTC on September 14 to 0000 UTC on September 16 following data assimilation.

4. Results

4.1. Radiance Simulations and Bias Correction

Figure 3 presents the outcomes of four experimental setups, depicting the brightness temperature differences between observations and background (OMB) and between observations and analysis (OMA) after bias correction in the assimilation procedure. Prior to assimilating MWRI radiance data, the absolute values of OMB for most scan points are relatively large, indicating significant discrepancies between the background and observations. After the assimilation of MWRI radiance data, the absolute values of OMA for most scan points are diminished to near 0 K, signifying that the analysis is more consistent with observations than the background, thereby validating the efficacy of the assimilation process. Furthermore, a comparison between Figure 3e,f and Figure 3g,h reveals that the assimilation of MWRI radiance data over land results in mean OMA values for both channels approaching 0 K more closely, indicating the effectiveness of assimilating land-based data. However, the standard deviation increases, which can be attributed to the substantial differences between the brightness temperature values over land and those over the ocean.

Figure 4 shows the scatter of simulated brightness temperatures from four experiments, namely, MWRI_3_Ocean, MWRI_3, MWRI_3_East, and MWRI_3_West, compared with the observed brightness temperatures at the assimilation time. Taking the MWRI_3_Ocean experiment as an example, most of the points in the scatter plot before the bias correction (Figure 4a) are above the diagonal line, indicating that the background brightness temperatures are mostly larger than the observed ones. After bias correction (Figure 4b), the points are more evenly distributed on both sides of the diagonal line, and the absolute value of the mean decreases from 0.742 to 0.458, getting closer to 0. After assimilation, the brightness temperatures of the analysis (Figure 4c) are more concentrated around the diagonal line compared to the background, and the mean, standard deviation, and root mean square error further decrease and approach 0. This indicates that assimilating MWRI radiance data is effective, and the analysis obtained has smaller errors and is significantly better than the background.

After incorporating the land data, a large number of scattered points ranging from 260 K to 300 K appeared in the scatter plot before bias correction (Figure 4d), indicating that the brightness temperature of the land data was significantly higher than that of the ocean data. Compared with the experimental MWRI_3_Ocean, the average value of OMB increased from −0.742 to 0.182, suggesting that the observations from the land data were greater than the background at this time. After bias correction (Figure 4e), the average value of OMB decreased from 0.182 to −0.318, indicating that the error of the land data observation being greater than the background has been corrected. After assimilation processing, the brightness temperature obtained from the analysis (Figure 4f) was more concentrated near the diagonal line compared to the background, and the average value was closer to 0 compared to the background. Moreover, compared with only assimilating ocean data (Figure 4c), the average value of OMA after adding land data was significantly lower, which indicates that the assimilation of MWRI land data is effective.

Figure 4g–i and Figure 4j–l present the results of assimilating data from the two orbits separately. The MWRI_3_East experiment incorporated a larger volume of assimilated data and, following the bias correction and assimilation, exhibited superior performance in terms of the mean, standard deviation, and root mean square error compared to the MWRI_3_West experiment.

Figure 5 presents the outcomes of all radiation data assimilation experiments conducted in this research, encompassing the count of assimilated satellite observations, their mean values, and standard deviations. These experiments were designed to assess the effectiveness of MWRI radiation data within the assimilation framework and the influence of various data selection strategies on the outcomes. When only ocean data are assimilated, the amount of data assimilated in channel 5 is the highest. Before the bias correction, the mean absolute value of OMB in channel 7 is significantly larger, but after the bias correction, the means of both OMB and OMA are the best, close to 0. In terms of the standard deviation, the standard deviations of the three channels significantly decreased after the bias correction, with channel 3 having the lowest standard deviation, indicating that the difference between the simulated values and the average values in the MWRI_3_Ocean experiment is the smallest. After adding the land data, the amount of data assimilated in the MWRI_3 and MWRI_5 experiments increased significantly, and the mean of OMA is closer to 0. However, the standard deviation of OMA also increased, which is caused by the large difference in the brightness temperature between the land and ocean. When only MWRI data from one orbit are assimilated, the amount of data assimilated in the MWRI_3_East experiment is significantly higher than that in the MWRI_3_West experiment. This is because the proportion of land area in the assimilation region on the west orbit is much larger than that on the east orbit, and the data quality on land is poor, with many data being excluded by quality control. Since the data on the east orbit are mainly ocean data, the mean of OMA in the MWRI_3_East experiment is closer to 0, and the standard deviation value is also significantly lower than that in the MWRI_3_West experiment.

4.2. Impact of FY-3D MWRI Data Assimilation

This section systematically evaluates the impact of FY-3D MWRI radiance data assimilation on the track, intensity, and precipitation forecasts of Typhoon Muifa (2022). Through the design of multiple comparative experiments (including different channel selections, land and ocean data assimilation strategies, and comparisons between single-orbit and multi-orbit data), this study aims to reveal the optimization potential and limitations of MWRI data assimilation. Specifically, it focuses on analyzing the regulatory role of ocean surface parameter adjustments during assimilation on the dynamic and thermodynamic structure of the typhoon while also exploring the mechanisms by which land data assimilation improves forecasts after typhoon landfall. The findings provide critical insights for the refined design of satellite data assimilation strategies.

4.2.1. Impacts on Analyses

Variations in geopotential height can help correct the position of a typhoon. Figure 6 illustrates the variations in 500 hPa geopotential height between the analysis and background across the four experimental groups. In all experiments, the typhoon center lies within a region of positive increments, with values ranging from 0 to +20 gpm, slightly weakening the low-pressure intensity. Meanwhile, surrounding areas exhibit negative increments ranging from −5 to −20 gpm. As shown in Figure 6a, the increment at the typhoon center is approximately +10 to +15 gpm, with a small area of stronger positive increment (+15 to +20 gpm) to the west, and a large area of negative increment to the northeast (red rectangular area). These increments are concentrated along the typhoon track, indicating that the assimilated data effectively improved the 500 hPa height structure of the typhoon. This enhancement strengthens the pressure gradient southwest and northeast of the typhoon center, potentially influencing the simulated typhoon track and causing an eastward shift. This interpretation is supported by the track forecast results presented in the Channel Selection Section (Figure 7).

After incorporating land surface data, as shown in Figure 6b, the area of negative increments in the northeast expands further, and the region of positive increments to the northwest (blue rectangular area) slightly expands as well. This suggests that the inclusion of MWRI land data further improves the 500 hPa geopotential height based on the MWRI_3_Ocean experiment. Figure 6c,d show the differences between the two MWRI orbit datasets. Data from the eastern orbit primarily impact the negative increment area northeast of the typhoon, whereas data from the western orbit affect the positive increment zone across Guangdong and Fujian Provinces.

To delve deeper into the impact of data assimilation on the typhoon’s three-dimensional structure, this study first conducted a preliminary analysis of the model’s background field. As shown in Figure 7c, the sea level pressure distribution of the model’s background field reveals a relatively loose typhoon structure. The low-level wind circulation, although formed, is weak and asymmetric, with the wind speed on the eastern side significantly higher than that on the western side, which may be a result of the prominent influence of the land. Building upon this, we performed a detailed comparison of the vertical wind field structure in the core region of the typhoon before and after assimilation, plotting vertical cross-sections of wind speed for the CTRL and MWRI_3_Ocean experiments.

In the CTRL experiment, the wind structure in the typhoon core exhibits a clear vertical discontinuity. Specifically, a narrow, low-wind-speed area with a core of nearly 0 m/s exists in the low-level layer from 850 hPa to 950 hPa. As altitude increases, this low-wind-speed area shifts westward and becomes disconnected from the upper-level low-wind-speed area between 300 hPa and 500 hPa, failing to form a vertically continuous eye structure. Furthermore, the upper-level wind field does not show good symmetry around the typhoon center. The highest wind speed regions, with speeds of approximately 40 m/s, appear at 800–900 hPa near 28.52°N, 121.75°E and at 800–850 hPa near 29.02°N, 123.25°E.

In stark contrast, the MWRI_3_Ocean assimilation experiment significantly improves the typhoon’s vertical wind structure. The central low-wind-speed area extends upwards, from approximately 800 hPa to 400 hPa, forming a vertically continuous and symmetric typhoon eye structure with the minimum value located at about 750 hPa. This indicates that the assimilation of MWRI radiance data effectively bridged the gap between the low-level and high-level wind in the CTRL experiment, reconstructing a more realistic typhoon eye. In terms of the wind speed distribution, the low-level wind speed at the typhoon center in the MWRI_3_Ocean experiment is slightly enhanced, and the symmetry of the wind is markedly improved. It is worth noting that while the high-wind-speed region at 28.52°N, 121.75°E shifts upwards to near 750–850 hPa, its outer layer contracts and weakens slightly. Meanwhile, the high-wind-speed region at 29.02°N, 123.25°E shifts to a lower layer, with the highest wind speed at about 850–950 hPa, and its outer wind intensity also slightly decreases. These adjustments suggest that the assimilation leads to a more concentrated energy distribution and a more compact core wind field structure.

In summary, the assimilation of MWRI radiance data effectively enhances vertical coupling, improving the vertical consistency and symmetry of the typhoon’s three-dimensional structure, particularly by reconstructing the eyewall. This provides a more accurate and physically consistent initial condition for subsequent forecasts.

4.2.2. Impacts on Forecast

Channel Selection

Figure 8 displays the observed track of Typhoon “Muifa” from 0600 UTC on 14 September 2022 to 0000 UTC on 16 September 2022, spanning 42 h, alongside the predicted tracks from each experimental group and the distance errors between the predicted and observed tracks (the data is from CMA). Specific track error values for each time step are provided in

Table 4. The track error (km) for five experiments in the channel selection.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	6.27	45.41	31.77	101.15	119.57	172.10	257.03	312.41	130.71
MWRI_3_ocean	12.96	56.30	105.90	112.33	153.20	167.85	201.67	218.61	128.60
MWRI_5_ocean	6.27	46.98	94.15	112.33	134.67	162.39	193.76	279.53	128.76
MWRI_7_ocean	6.27	56.30	94.15	112.33	143.51	171.01	205.47	242.46	128.94
MWRI_357_ocean	6.27	56.30	94.15	112.33	134.67	171.01	193.76	286.25	131.84

. The average track error for the CTRL experiment is 130.71 km. After assimilating MWRI radiance data, the track errors for all four experimental groups are significantly larger than those of the control experiment within the first 24 h of the forecast. This may be explained by the fact that when simulations begin from the initial time of typhoon landfall, demonstrating the effectiveness of data assimilation becomes more challenging, as the CTRL experiment already generates a well-simulated typhoon system. However, as shown in Figure 8a, after 12 h of forecasting, the CTRL experiment exhibits a noticeable track deviation, with the trajectory veering westward and penetrating deeper inland, likely due to the typhoon’s landfall. In the assimilation experiments, the track deviations observed in the CTRL experiment are corrected, with the tracks of all four assimilation experiments aligning more closely with the actual track. This is likely due to the optimized geopotential height after the assimilation of MWRI radiance data, which leads to a more accurate steering flow for the typhoon’s movement. This improvement in forecast accuracy can also be physically linked to the superior initial conditions produced by the assimilation process. During the last 18 h of the prediction period, the track errors in the assimilation experiments are consistently lower than those in the CTRL experiment. Over the 42 h period, the average track error for the MWRI_3_ocean experiment is 128.60 km, while that for MWRI_5_ocean, MWRI_7_ocean, and MWRI_357_ocean is 128.76 km, 128.94 km, and 131.84 km, respectively.

The minimum sea level pressure (MSLP) and maximum surface wind speed (MSWS) directly reflect the intensity and development trends of a typhoon. The MSLP error is defined as the simulated value minus the observed value. A positive MSLP error indicates that the model’s simulated typhoon central pressure is higher than the actual observed pressure, suggesting that the typhoon intensity is underestimated. Conversely, a negative MSLP error indicates that the simulated pressure is lower, suggesting that the typhoon intensity is overestimated. The MSWS error is also defined as the simulated value minus the observed value. A positive MSWS error indicates that the model’s simulated maximum surface wind speed is higher than the actual observed speed, suggesting that the typhoon intensity is overestimated. Conversely, a negative MSWS error indicates that the simulated speed is lower, suggesting that the typhoon intensity is underestimated. As shown in

Table 5. MSLP error (hPa) for five experiments in the channel selection.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	38.08	27.72	15.98	7.46	2.41	4.19	4.99	6.91	13.47
MWRI_3_ocean	39.86	28.83	16.89	8.24	2.06	4.09	4.44	5.03	13.68
MWRI_5_ocean	39.84	28.83	16.82	8.25	1.8	3.78	4.33	4.91	13.57
MWRI_7_ocean	39.84	28.81	16.9	8.34	1.85	3.87	4.16	4.77	13.57
MWRI_357_ocean	39.87	28.86	16.94	8.39	1.83	3.95	4.37	5.04	13.66

and Figure 9, within the first 18 h of the forecast, the MSLP error of the CTRL experiment is lower than that of the five assimilation experiments. However, after 18 h, the positive impact of data assimilation becomes evident, with all assimilation experiments exhibiting lower MSLP errors than the CTRL experiment. This later improvement may be attributed to the significant optimization of the typhoon’s core vertical structure after assimilating MWRI radiance data, which provides a superior initial field and enables the model to more accurately capture the typhoon’s evolution in later forecasts. Nevertheless, over the entire 42 h forecast period, the CTRL experiment has the lowest average MSLP error at 13.47 hPa, while the errors of the assimilation experiments exceed that of the CTRL experiment. A potential reason for this is that while the MWRI data assimilation optimizes the typhoon’s vertical structure, it might slightly enhance the typhoon’s central pressure, leading to a slight deviation in the initial intensity forecast and thus increasing the average error over the entire forecast period. Regarding MSWS errors, as presented in

Table 6. MSWS error (m/s) for five experiments in the channel selection.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	−15.83	−18.5	−12.16	−5.54	−4.94	−5.98	−6.27	−7.64	−9.61
MWRI_3_ocean	−15.83	−17.4	−13.16	−5.38	−5.39	−5.26	−5.62	−4.15	−9.02
MWRI_5_ocean	−15.83	−17.31	−13.05	−5.08	−5.15	−4.27	−5.55	−3.66	−8.74
MWRI_7_ocean	−15.83	−17.36	−13.1	−4.92	−5.37	−4.51	−5.72	−5.89	−9.09
MWRI_357_ocean	−15.83	−17.39	−13.24	−4.9	−5.24	−4.72	−5.01	−6.21	−9.07

and Figure 9, the CTRL experiment has the largest average error at −9.61 m/s, whereas the MWRI_5_ocean experiment has the smallest average error at −8.74 m/s. Overall, according to a comprehensive analysis of errors in the track, MSLP, and MSWS, the MWRI_3_ocean and MWRI_5_ocean experiments perform relatively well, though their average errors are not significantly different from those of the CTRL experiment, indicating a need for further improvements.

Land Surface and Ocean Data

The experiments discussed in this section aim to investigate the impact of incorporating MWRI land data into the assimilation process during the forecast of Typhoon “Muifa”. Figure 10a depicts the comparison of track prediction outcomes for two experimental groups: one incorporating land data from MWRI channels 3 and 5 and the other using solely MWRI ocean data (MWRI_3_ocean and MWRI_5_ocean). Figure 10b presents the discrepancies between the predicted and observed tracks for each group.

Table 7. The track error (km) for five experiments in the land surface and ocean data testing trial.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	6.27	45.41	31.77	101.15	119.57	172.10	257.03	312.41	130.71
MWRI_3_ocean	12.96	56.30	105.90	112.33	153.20	167.85	201.67	218.61	128.60
MWRI_3	12.96	56.30	94.15	103.58	126.93	151.89	186.93	230.79	120.44
MWRI_5_ocean	6.27	46.98	94.15	112.33	134.67	162.39	193.76	279.53	128.76
MWRI_5	6.27	56.30	94.15	102.45	126.93	162.39	186.93	275.48	126.36

lists the precise track errors for each experiment. According to Table 7, incorporating land data into MWRI channels 3 and 5 decreases the mean track forecasting error. For channel 3, the average track error decreases from 128.6 km to 120.44 km, and for channel 5, it decreases from 128.76 km to 126.36 km. These results indicate that assimilating land data has a positive impact on track forecasting.

As shown in Figure 11 and

Table 8. MSLP error (hPa) for five experiments in the land surface and ocean data testing trial.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	38.08	27.72	15.98	7.46	2.41	4.19	4.99	6.91	13.47
MWRI_3_ocean	39.86	28.83	16.89	8.24	2.06	4.09	4.44	5.03	13.68
MWRI_3	39.85	28.75	16.86	8.22	1.72	3.65	3.92	4.34	13.41
MWRI_5_ocean	39.84	28.83	16.82	8.25	1.8	3.78	4.33	4.91	13.57
MWRI_5	39.87	28.86	16.87	8.25	1.74	3.75	4.21	4.97	13.56

, during the 30 h period from 1406 UTC to 1518 UTC, the MSLP forecast error for MWRI_3 was consistently lower than that for MWRI_3_Ocean, with the average forecast error decreasing from 13.68 hPa to 13.41 hPa. The forecast accuracy surpassed that of the CTRL experiment, indicating that the assimilation of land data from channel 3 had a relatively significant positive impact on MSLP forecasts. However, no notable difference was observed in MSLP forecasts between the two groups of experiments for channel 5. Regarding MSWS forecast errors, as shown in

Table 9. MSWS error (m/s) for five experiments in the land surface and ocean data testing trial.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	−15.83	−18.5	−12.16	−5.54	−4.94	−5.98	−6.27	−7.64	−9.61
MWRI_3_ocean	−15.83	−17.4	−13.16	−5.38	−5.39	−5.26	−5.62	−4.15	−9.02
MWRI_3	−15.83	−17.34	−13.23	−5.34	−4.98	−4.16	−5.84	−4.95	−8.96
MWRI_5_ocean	−15.83	−17.36	−13.1	−4.92	−5.37	−4.51	−5.72	−5.89	−9.09
MWRI_5	−15.83	−17.42	−13.04	−5.05	−5.17	−4.20	−5.57	−4.67	−8.87

, after assimilating land data, the error for channel 3 decreased from −9.02 m/s to −8.96 m/s, and for channel 5, it decreased from −9.09 m/s to −8.87 m/s. The accuracy of MSWS predictions for both channels increased, indicating that incorporating land data positively influenced MSWS forecasting outcomes.

In summary, the simultaneous assimilation of MWRI ocean and land radiance data resulted in higher forecast accuracy for the track, MSLP, and MSWS compared to assimilating ocean data alone. It seems that the MWRI_3 experiment performed the best. The potential reason for this advantage might lie in the relatively lower emission rate of the third channel (18.7 GHz). This property enhances its sensitivity to surface features like ocean winds and rain while reducing atmospheric water vapor interference, allowing for better penetration in the cloudy environment of a typhoon [8,33].

Diverse Orbital Data

This section examines the impact of assimilating MWRI data from two distinct orbits in two separate instances within a three-hour window centered on 0600 UTC, 14 September 2022, on the forecasting accuracy of Typhoon Muifa. In the Land Surface and Ocean Data Section, the scheme used in experiment MWRI_3, which produced the best forecast, is applied to assimilate the MWRI data from both the east and west sides of Typhoon Muifa’s center. Figure 12 presents a comparative analysis of the 42 h forecast results from three assimilation experiments with different single-orbit data schemes. Figure 12a displays the typhoon track diagrams, while Figure 12b shows the corresponding track error curves. The specific errors of the track of each group of experiments are shown in

Table 10. The track error (km) for four experiments in diverse orbital data experiment.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	6.27	45.41	31.77	101.15	119.57	172.10	257.03	312.41	130.71
MWRI_3	12.96	56.30	94.15	103.58	126.93	151.89	186.93	230.79	120.44
MWRI_3_West	12.96	46.98	94.15	103.58	134.67	160.58	174.88	218.81	118.31
MWRI_3_East	6.27	56.30	94.15	112.33	134.67	151.89	182.68	218.81	119.64

. As can be seen from Figure 12a, the forecast tracks of the three MWRI assimilation experiments are very close, and only in the last 12 h of the forecast, the tracks of MWRI_3_West and MWRI_3_East are slightly eastward compared to MWRI_3 and are closer to the observed track. As can be seen from Table 10, the average forecast track errors of the MWRI_3_West and MWRI_3_East experiments are both smaller than those of the MWRI_3 experiment with the assimilation of the MWRI data from both orbits simultaneously. Among them, the MWRI_3_West experiment performs best, with an average track error of 118.31 km, followed by the MWRI_3_East experiment, with an average track error of 119.64 km.

Figure 13 illustrates the prediction errors for the MSLP and MSWS across three MWRI assimilation experiments and the CTRL experiment in this study. Precise error values are detailed in

Table 11. MSLP error (hPa) for four experiments in diverse orbital data experiment.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	38.08	27.72	15.98	7.46	2.41	4.19	4.99	6.91	13.47
MWRI_3	39.85	28.75	16.86	8.22	1.72	3.65	3.92	4.34	13.41
MWRI_3_West	39.86	28.87	16.88	8.27	1.72	3.66	4.02	4.93	13.53
MWRI_3_East	39.83	28.80	16.86	8.19	1.83	3.78	3.99	4.57	13.48

and

Table 12. MSWS error (m/s) for four experiments in diverse orbital data experiment.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	−15.83	−18.5	−12.16	−5.54	−4.94	−5.98	−6.27	−7.64	−9.61
MWRI_3	−15.83	−17.34	−13.23	−5.34	−4.98	−4.16	−5.84	−4.95	−8.96
MWRI_3_West	−15.83	−17.35	−13.12	−5.31	−5.30	−4.45	−5.69	−3.87	−8.87
MWRI_3_East	−15.83	−17.32	−13.25	−5.63	−5.00	−4.48	−5.69	−4.05	−8.91

. As depicted in Figure 13, the MSLP and MSWS forecast outcomes show no notable differences among the three MWRI assimilation experiments. From Table 11 and Table 12, it can be observed that for the average error of the MSLP forecast, experiment MWRI_3 performs the best with an average error of 13.41 hPa, followed by MWRI_3_East with an average error of 13.48 hPa, and MWRI_3_West ranks last with an average error of 13.53 hPa. As for the MSWS forecast, MWRI_3_West has the lowest average error of −8.87 m/s, followed by MWRI_3_East with an average error of −8.91 m/s, and MWRI_3 has the largest average error of −8.96 m/s.

In summary, MWRI_3_West performs best in the track forecast and MSWS forecast, MWRI_3 performs best in the MSLP forecast, and MWRI_3_East ranks second in all three forecast accuracies, showing relatively stable performance.

4.3. Impact of Joint Assimilation of MWHS-2 and MWRI Data on Assimilation

This section focuses on the synergistic assimilation effects of the FY-3D MWHS-2 and the MWRI, investigating the enhanced capabilities of multi-source satellite data fusion for typhoon forecasting. Through a two-step joint assimilation strategy (first assimilating MWHS-2 upper-level humidity data, followed by MWRI surface and lower-level radiance data), the study conducts a comparative analysis of the overall improvements in typhoon track and intensity predictions.

4.3.1. Impacts on Analyses

Figure 14 presents a comparison of the geopotential height increments at 500 hPa between the MWHS experiment (assimilating only MWHS-2 data) and the MWHS_MWRI_2STEP experiment (two-step joint assimilation of MWHS-2 and MWRI data), along with their differences (Figure 14c). In the MWHS experiment (Figure 14a), a positive increment of +10 to +15 gpm is observed near the typhoon center, with a negative increment of −10 to −15 gpm on the northeastern periphery. This indicates that conventional humidity data slightly weakens the low-pressure intensity at the typhoon core while enhancing the peripheral pressure gradient, potentially leading to an eastward track deviation (the MWHS experiment track error is 108.56 km, as shown in Figure 15).

After the joint assimilation processing (Figure 14b), the positive increment range at the typhoon center decreased to +5 to +10 gpm, the negative increment range on the northeast side expanded to −15 to −20 gpm, and a new +5 to +10 gpm positive increment appeared on the northwest side. The difference plot (Figure 14c) further shows the emergence of a new positive increment on the northwest side and the deepening of the negative increment on the northeast side, indicating that assimilating the MWRI data further suppressed the eastward deviation trend (the trajectory error of MWHS_MWRI_2STEP dropped to 105.01 km).

4.3.2. Impacts on Forecast

Figure 15 illustrates the impact of joint MWHS-2 and MWRI data assimilation on the track forecast of Typhoon Muifa. As shown in Figure 15a, the forecast track of the MWHS_MWRI_2STEP experiment aligns more closely with the observed track than that of the MWHS experiment (

Table 13. The track error (km) of four experiments.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	6.27	45.41	31.77	101.15	119.57	172.10	257.03	312.41	130.71
MWRI_3_West	12.96	46.98	94.15	103.58	134.67	160.58	174.88	218.81	118.31
MWHS	20.77	62.07	105.89	105.38	139.27	148.70	143.20	143.19	108.56
MWHS_MWRI_2STEP	6.27	53.92	101.70	116.55	129.48	148.70	118.89	164.62	105.01

). The MWHS_MWRI_2STEP experiment achieved an average track error of 105.01 km, which is 3.55 km lower than the MWHS experiment (108.56 km) and outperforms the MWRI_3_West experiment (118.31 km) that assimilated only MWRI data. However, the track error of MWHS_MWRI_2STEP increased to 164.62 km at 1600 UTC compared to the MWHS experiment (143.19 km), as indicated by a westward shift in the forecast position (Figure 15a).

Figure 15. Joint Assimilation Experiment: (a) The track of Typhoon Muifa from 0600 UTC on 14 September 2022 to 0000 UTC on 16 September 2022 and (b) the corresponding track error values, with the x-axis representing dates and hours.

Figure 15. Joint Assimilation Experiment: (a) The track of Typhoon Muifa from 0600 UTC on 14 September 2022 to 0000 UTC on 16 September 2022 and (b) the corresponding track error values, with the x-axis representing dates and hours.

Figure 16 and

Table 14. MSLP error (hPa) for four experiments in joint assimilation experiment.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	38.08	27.72	15.98	7.46	2.41	4.19	4.99	6.91	13.47
MWRI_3_West	39.86	28.87	16.88	8.27	1.72	3.66	4.02	4.93	13.53
MWHS	40.80	28.05	15.63	7.74	1.44	2.75	3.04	4.10	12.94
MWHS_MWRI_2STEP	40.17	29.14	16.9	8.42	1.71	3.29	3.72	2.48	13.23

and

Table 15. MSWS error (m/s) for four experiments in joint assimilation experiment.

Day Hour	1406	1412	1418	1500	1506	1512	1518	1600	Average
CTRL	−15.83	−18.5	−12.16	−5.54	−4.94	−5.98	−6.27	−7.64	−9.61
MWRI_3_West	−15.83	−17.35	−13.12	−5.31	−5.30	−4.45	−5.69	−3.87	−8.87
MWHS	−25.39	−16.17	−12.05	−6.15	−4.24	−4.93	−5.92	−4.81	−9.96
MWHS_MWRI_2STEP	−15.83	−16.58	−13.24	−5.23	−4.00	−4.23	−4.20	−0.10	−7.93

compare the MSLP and MSWS errors across experiments. The MWHS_MWRI_2STEP experiment exhibited an average MSLP error of 13.23 hPa, slightly higher than the MWHS experiment (12.94 hPa) but notably better than the GTS experiment (13.50 hPa). This discrepancy may stem from limited sensitivity of MWRI data to sea surface parameters, constraining corrections to the pressure. In contrast, the MWHS_MWRI_2STEP experiment achieved a significantly lower average MSWS error (−7.93 m/s) compared to the MWHS (−9.96 m/s) experiment. This improvement highlights the role of joint assimilation in refining the vertical distribution of humidity and temperature within the typhoon core (Figure 16), thereby enhancing the simulation of maximum wind speeds. For instance, the MSWS error at 1600 UTC for MWHS_MWRI_2STEP was merely −0.10 m/s (Table 15), demonstrating precise capture of the typhoon’s decay phase.

In summary, the MWHS_MWRI_2STEP experiment outperforms the standalone MWHS and MWRI_3_West experiments across most metrics. Specifically, it reduces the average track error by 3.55 km compared to MWHS and by 13.30 km relative to MWRI_3_West while also decreasing the MSWS error by 2.03 m/s in comparison to MWHS and by 0.94 m/s in comparison to MWRI_3_West. Although its MSLP error is slightly higher than that of MWHS (by 0.29 hPa), it remains competitive and exhibits improvements in the later stages of the forecast. The two-step assimilation approach effectively leverages the complementary strengths of MWHS-2 (upper-level humidity) and MWRI (surface and lower-level data), producing a more balanced and accurate analysis, thereby enhancing forecast skill and highlighting the value of multi-instrument synergy.

5. Conclusions

This study demonstrates that assimilating FY-3D satellite MWRI radiance data, particularly from channels 3 and 5 (18.7 GHz and 23.8 GHz, respectively), improves the accuracy of track and intensity forecasts for Typhoon Muifa, particularly for predictions beyond 24 h. This result is consistent with findings from previous studies, which highlighted the value of MWRI data assimilation in enhancing typhoon track and intensity predictions [8,21].

Building on prior research, we incorporated land observation data for assimilation, optimized orbital assimilation strategies, and implemented rigorous quality control measures (such as bias correction and outlier removal). These improvements further enhanced the effectiveness of MWRI data assimilation in typhoon track and intensity forecasting, with particularly notable performance in forecasts after typhoon landfall. Additionally, the combined assimilation of the MWRI and MWHS-2 integrates complementary information from upper-level humidity and surface observations, demonstrating a synergistic enhancement effect.

Although challenges remain, such as elevated observation-minus-analysis (OMA) standard deviations in land data assimilation and limited MWRI sensitivity to sea surface parameters, these factors may introduce uncertainties and indicate areas requiring further improvement. Nevertheless, the findings highlight the critical importance of strategic channel selection, effective data source integration, and rigorous quality control in improving satellite data assimilation for numerical weather prediction.

Subsequent studies should prioritize improving the sensitivity of the MWRI to surface parameters, refining land data assimilation methods, and incorporating typhoon bogus data assimilation schemes to address potential significant discrepancies between the initial EC data and actual typhoon intensity. This would ensure a more accurate representation of the initial field. We will also explore these techniques across a wider array of typhoon scenarios to ensure reliability and applicability.