Impacts of Bogus Vortex Initialization Using Scatterometer-Derived 34 kt Wind Radii and Centers on Tropical Cyclone Forecasts

Highlights

What are the main findings?

• Scatterometer wind-based bogus vortex initialization improves the initial structure, location, and intensity representation of Typhoon Doksuri (2023).
• This scheme enhances intensity forecasts and better reproduces storm features.

What is the implication of the main finding?

• Scatterometer wind observations are highly valuable for improving typhoon initialization in numerical models.
• Incorporating scatterometer data through bogus vortex initialization can enhance the accuracy of intensity predictions in regional NWP models.

Abstract

This study demonstrates the positive impact of scatterometer wind-based bogus vortex initialization on forecasts of Typhoon Doksuri (2023). In this scheme, the NCEP analysis vortex in the initial conditions is replaced with a bogus vortex. A regression model links the scatterometer wind-derived 34 kt wind radius with the radius of maximum sea-level pressure gradient, a required parameter in Fujita’s bogus formula. The cyclonic circulation center identified in the scatterometer wind field is designated as the typhoon center. The resulting bogus vortex provides a more realistic representation of the low-level circulation, center location, and intensity. Numerical experiments with the WRF model, configured with two-way nested domains (9–3 km) and 115 vertical levels below the model top at 1 hPa, show that the scatterometer wind-bogus scheme effectively improves the initial vortex position and minimum sea-level pressure, slightly enhances track forecasts, and substantially improves intensity forecasts, particularly during rapid intensification and weakening stages of Typhoon Doksuri over the western North Pacific. Furthermore, comparisons with Himawari-9 AHI infrared observations indicate that forecasts with bogus vortex initialization better reproduce the eye, eyewall, and spiral rainband structures than forecasts without it. These results underscore the value of scatterometer observations for improving typhoon forecasts.

1. Introduction

High-precision numerical simulations of typhoons rely on accurate initial conditions. However, due to the scarcity of conventional high-resolution observations over remote ocean areas, the initial vortex in global analyses often exhibits significant discrepancies from the real vortex in terms of center position, scale, and intensity [1,2]. Thus, bogus vortex initialization (BVI) has been widely adopted to address this challenge. In recent years, satellite microwave scatterometers have provided multi-temporal, wide-coverage, near-real-time observations of sea surface winds in typhoon regions, offering a unique advantage for capturing the low-level circulation of typhoons [3,4]. This study evaluates the impact of a BVI scheme based on scatterometer wind observations on typhoon forecasts.

The BVI involves two major steps: (1) constructing a two-dimensional (horizontal) bogus vortex of a prescribed variable (i.e., sea-level pressure or surface wind) using an empirical formula; and (2) deriving three-dimensional structures of other model variables using dynamical constraints (e.g., gradient wind balance, hydrostatic balance) to produce a bogus vortex that approximates a real typhoon at the initial time of forecast [1,5,6,7,8,9,10,11,12,13,14]. Early studies demonstrated that inserting a bogus vortex could reproduce the warm-core structure and eyewall features of mature typhoons [5,8]. However, errors in the initial vortex position can cause the storm center to deviate from its true location, and direct insertion may result in misalignment of the low-level cyclonic circulation. Kurihara [1] proposed an advanced BVI procedure. The analysis vortex in the large-scale analysis is removed to obtain a so-called environmental field. A bogus vortex, whose intensity and structure are adjusted based on a few parameters from observations, is positioned at the observed storm center and added to the environmental field. This approach significantly improved the track and intensity forecasts of Northwest Pacific typhoons [6]. Leslie and Holland [15] examined four commonly used empirical pressure–wind relationships and discussed how differences in radial wind, pressure distribution, and vortex radial profiles affected numerical simulations of typhoon tracks under various environmental conditions. Davis and Low-Nam [16] and Reed and Jablonowski [17] demonstrated that the same BVI scheme can produce different cyclone evolutions under varying model resolutions, pointing out that the model resolution must be appropriately matched to the strength and size of a tropical cyclone to ensure proper vortex development during the spin-up process. Kwon and Cheong [18] applied a spherical high-order filter and analytically defined three-dimensional vortices to enhance the dynamical consistency of the BVI. Rappin [19] proposed a highly configurable framework that allows a flexible adjustment of parameters in the bogus vortex to suit different simulation scenarios. The BVI was also implemented in operational typhoon models at NCEP (HWRF and HMON) to create an improved background field [20]. The background field after BVI was used as an initial field in the HMON model without data assimilation, and for data assimilation in the HWRF model. Liu [12] created a global, ERA5-based tropical cyclone wind-field dataset enhanced by an integrated parametric correction method that incorporated the Willoughby adjustment [21], which imposes gradient wind balance based on the Holland model [22].

Satellite microwave scatterometers provide surface wind vectors over the ocean. They played multiple roles in both typhoon research and operational forecasting [23], including early detection and identification of tropical cyclogenesis [24], determination of low-level circulation centers [25,26,27], derivation of characteristic parameters of wind field structure and intensity parameters [28,29,30,31,32,33], and data assimilation to improve typhoon forecasts [34,35,36,37]. The ERS scatterometer winds have been assimilated into the European Centre for Medium-Range Weather Forecasts (ECMWF) 3D-Var and 4D-Var systems since 1996. Assimilation of scatterometer winds has been shown to improve the representation of typhoon structures, producing positive (negative) impacts on intensity and track forecasts of 66% (23%) of cases [35,36]. Chen [38] assimilated QuikSCAT winds into the MM5 3D-Var system and reported improved intensity and track forecasts of Hurricane Isidore, with track improvements partly attributed to repositioning of the cyclone center. Yu [37] demonstrated that assimilation of HY-2A scatterometer winds into the WRF 3D-Var system improved the analysis of intensity, warm-core structure, and moisture fields for typhoon Bolaven (2012). Assimilation of multiple scatterometers (e.g., ASCAT-A/B and OSCAT) had further enhanced the forecast skill of typhoons, reducing the forecast errors of minimum central sea-level pressure by approximately 10–15% [34]. Despite these advances, scatterometer winds have limitations. They are sensitive to heavy precipitation and extreme winds, prone to saturation in high-wind regions, and have discontinuous spatial and temporal coverage. There were efforts to mitigate these limitations using geophysical model functions [39] or strong-wind calibration [40,41].

Studies using scatterometer wind measurements for BVI remain relatively sparse. Chavas and Knaff [42] and Avenas [43] developed a latitudinal-dependent semiempirical model for predicting the tropical cyclone radius of maximum wind based on intensity and latitude from the best-track data, Rmax observations from spaceborne synthetic aperture radar (SAR). The wind radii from an intercalibrated dataset of medium-resolution scatterometers and radiometers were used to revise the model coefficients. Pan [4] employed the 10 m sea surface wind vector products from HY-2B/2C/2D HSCAT, MetOp-B/C ASCAT, and FY-3E WindRAD to construct a bogus vortex of sea-level pressure. They used the Fujita formula [7] to specify a radial profile of sea-level pressure. The radius of maximum pressure gradient required in the Fujita profile was estimated from the 34 kt radius of scatterometer. This study is a follow-on work to implement the scatterometer-deduced bogus vortex into BVI and assess its impact on the forecasts of Typhoon Doksuri (2023).

The remainder of this article is organized as follows: Section 2 describes the data and provides a case overview. Section 3 outlines the methodology. Section 4 shows forecast results. Section 5 provides conclusions.

2. Materials and Case Description

2.1. Sea Surface Wind and AHI Data

The 10 m sea surface wind data used in this study are primarily derived from microwave scatterometers: HSCAT onboard HY-2B/C/D launched by China and ASCAT onboard MetOp-B/C launched by Europe. For example, the HY-2C satellite orbits at ~957 km with a 66° inclination. The HSCAT operates at 13.256 GHz (Ku band) and provides an effective swath of 1400–1800 km [44]. In contrast, MetOp-C is in a sun-synchronous orbit at ~827 km with 98.7° inclination. The ASCAT operates at 5.255 GHz (C-band) with two swaths of 550 km width separated by ~700 km [45]. This study employs Level-2B scatterometer wind products at 25 km × 25 km resolutions. Both HSCAT and ASCAT provide reliable wind speed measurements within the range of 4–24 m s⁻¹, with a typical accuracy of 2 m s⁻¹ for speed and 20° for direction. As mentioned before, scatterometer observations are subject to saturation effects under extreme wind conditions, contamination from rain scattering in precipitating regions, and low signal-to-noise ratios under calm wind conditions (<4 m s⁻¹) [46].

The brightness temperature of Channel 13 (atmospheric window, central wavelength 10.45 µm, bandwidth 0.3 µm, spatial resolution 2 km) from the Advanced Himawari Imager (AHI) onboard Himawari-9 [47] is utilized to characterize the horizontal distribution of typhoon cloud–rainband structures and to validate the all-sky simulations.

2.2. Best Track and Large-Scale Analysis Data

The International Best Track Archive for Climate Stewardship (IBTrACS, v04r01) provides 6 hourly typhoon parameters, including storm center position (latitude, longitude), minimum sea-level pressure (MSLP, $P_c$), the radius of the outermost closed isobar (ROCI, $R_{out}$), the pressure of the outermost closed isobar (POCI, $P_{out}$), and the 34 kt wind radius in four quadrants ($R_{34kt}$) [48]. These parameters are essential for constructing radial sea-level pressure profiles of bogus vortices. The far-field pressure ($P_{\infty }$), required in Fujita’s formula, is obtained from the International Comprehensive Ocean–Atmosphere Dataset (ICOADS, v3.0.2) [49].

The U.S. National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) analysis is used as the background fields for starting model forecasts. The GFS global analyses are available four times daily, with a horizontal resolution of 0.25° × 0.25° and 57 vertical pressure levels below the model top at 1 hPa [50].

2.3. Typhoon Doksuri (2023)

Typhoon Doksuri, the fifth named tropical cyclone of the 2023 western North Pacific season, was the most intense landfalling typhoon of the year. Figure 1 shows the best track, minimum central sea-level pressure, and maximum sustained wind speed (1 min average). The storm formed east of the Philippines, then moved northwestward and strengthened to a tropical storm (TS) at 0000 UTC 22 July. Doksuri underwent a rapid intensification (RI) [51], reaching peak intensity at 1800 UTC 24 July with maximum sustained winds of 128 kt and MSLP of 926 hPa. It then experienced rapid weakening (RW) [52] before making landfall in Jinjiang, Fujian Province, on 28 July. At landfall, maximum sustained wind had decreased to 93 kt, and MSLP had risen to 945 hPa. The inland storm moved northwestward.

The track and intensity evolution of Doksuri are typical of western North Pacific tropical cyclones, making it a representative case for evaluating the impact of scatterometer-based BVI on numerical forecasts.

3. Methodology of BVI

3.1. Bogus Vortex Construction

The objective three-dimensional axisymmetric bogus vortex initialization scheme based on satellite scatterometer observations consists of four major steps:

• Construct the radial profile of sea-level pressure (SLP) for Typhoon Doksuri using scatterometer-derived 34 kt wind radius together with other parameters from the best track;
• Derive the three-dimensional (3D) axisymmetric bogus vortex from the two-dimensional (2D) axisymmetric SLP using gradient wind balance and an empirical vertical distribution of wind;
• Decompose the GFS analysis into the sum of the analysis vortex and the environmental background field;
• Remove the analysis vortex and merge the 3D axisymmetric bogus vortex with the environmental background field.

• Steps 2–4 follow the method of Kurihara [1].

The radial profile of sea-level pressure is defined:

$$\begin{array}{l}{P}_{bogus}(r)=P_{\infty }-{\displaystyle \frac{(P_{\infty }-P_c)}{{(1+r^2/2R_0^2)}^{\frac{1}{2}}}}, (r\le R_{out})V_g(r)\\ R_0=a{R}_{34kt}^{Scat}-b.\end{array}$$

(1)

where r is the radial distance from the typhoon center, $P_c$ is the central SLP, $P_{\infty }$ is the far-field environmental pressure, $R_{out}$ is the radius of the outermost closed isobar, and $R_0$ is the radius of maximum radial gradient. The value of $R_0$ is estimated from the scatterometer wind-derived radius of 34 kt wind ($R_{34kt}^{Scat}$). Parameters a and b are empirical coefficients that depend on storm intensity [4]. The statistical relationship between $R_0$ and $R_{34kt}^{Scat}$ is derived through an indirect two-step procedure. A baseline relationship is first established by regressing $R_0$ against the best track $R_{34kt}$ for different intensity categories. This baseline is then applied to real-time observations by replacing $R_{34kt}$ with the observed $R_{34kt}^{Scat}$, yielding a more objective and observation-based estimate of $R_0$ for constructing the bogus vortex.

The radial profile of tangential wind can be derived from the radial profile of sea-level pressure [53]:

$$V_g(r)={\left({\displaystyle \frac{r}{\rho }}{\displaystyle \frac{\partial P_{bogus}(r)}{r}}+{\displaystyle \frac{{f_c}^2{r}^2}{4}}\right)}^{1/2}-{\displaystyle \frac{r\left|f_c\right|}{2}}$$

(2)

where $V_g$ is tangential wind speed, $f_c$ is the Coriolis parameter at the latitude of the typhoon center ($f_c=2\omega \sin {\phi }_c$), and $\rho$ is constant air density ($\rho$ = 1.15 kg m⁻³).

The 3D wind field is obtained by applying an empirical vertical weighting function $F(P)$ to:

$$V(r,P)=F(P)V_g(r)$$

(3)

The 3D geopotential field $\Phi$ is obtained from $V(r,P)$ by solving the gradient wind balance equation at each of the pressure levels:

$${\displaystyle \frac{V^2(r,P)}{r}}+f_{c}V(r,P)-{\displaystyle \frac{\partial \Phi (r,P)}{\partial r}}=0$$

(4)

The temperature $T$ is derived from $\Phi$ based on the hydrostatic equation:

$$T=R{\displaystyle \frac{\partial \Phi (r,P)}{\partial \ln \sigma }}$$

(5)

where $R$ is the gas constant, $\sigma$ is a dimensionless defined as $P/P_s$, and $P_s$ represents the surface pressure. Relative humidity ($RH$) is assumed to be near saturation at the vortex center and reduced to a lower value at a specified radius $R_{out}$; $RH$ values at intermediate grid points (r within 0–$R_{out}$) are obtained by linear interpolation [8].

3.2. Decomposing GFS Analysis

Following the Geophysical Fluid Dynamics Laboratory (GFDL) scheme [1,54], the GFS analysis of variable h is decomposed into

$$h=h_B+h_D$$

(6)

where h_B is the background field and h_D is the disturbance field. They are derived from h:

$$h_B(i,j)={\overline{h}}_B(i,j)+K_m\left({\overline{h}}_B(i,j-q_n)+{\overline{h}}_B(i,j+q_n)-2{\overline{h}}_B(i,j)\right)$$

(7)

$${\overline{h}}_B(i,j)=h(i,j)+K_m\left(h(i-q_n,j)+h(i+q_n,j)-2h(i,j)\right)$$

(8)

where $K_m$ denotes a filter operator that removes all waves shorter than 900 km, q_n represents distances away from the i and j grid varying with n.

The outer boundary of the analysis vortex $r_f$ is defined as $r_f=1.25r_o$, where $r_0$ is determined from the azimuthal mean tangential wind ${\overline{V}}_D(r)$ in the 850-hPa disturbance field, subject to the conditions ${\overline{V}}_D(r_0)\le 6$ m s⁻¹, $-\partial {\overline{V}}_D(r_0)/\partial r <4\times {10}^{-6}$ s⁻¹, or ${\overline{V}}_D(r_0)\le 3$ m s⁻¹. Cylindrical filtering is applied to obtain the analysis vortex from the disturbance field:

$$h_{av}(r,\theta )=h_D(r,\theta )-\left(h_D(r_f,\theta )E(r)+{\overline{h}}_D(r_f)\left(1-E(r)\right)\right), 0\le r\le r_f$$

(9)

where ${\overline{h}}_D(r_0)={\displaystyle \frac{1}{2\pi }}{\displaystyle \oint h_D(r_0,\theta )d\theta }$ and

$$E(r)={\displaystyle \frac{\exp \left(-{(r_f-r)}^2/l^2\right)-\exp \left(-r_f^2/l^2\right)}{1-\exp \left(-r_f^2/l^2\right)}},0\le r\le r_f$$

(10)

where $l$ is set to $r_0$/5.

The initial condition for the forecast experiment with BVI (Exp_BVI) is defined:

$$h_{Exp\_BVI}^{IC}=w{h}_{bogus}+(1-w)\left(h_E+(h_D-h_{av})\right), w=\left\{\begin{array}{ll}\cos ({\displaystyle \frac{\pi }{2}}{\displaystyle \frac{r}{R}}),& r<r_f\\ 0,& r\ge r_f\end{array}\right.$$

(11)

where w(r) is an empirical weighting function suggested by Mathur [8] to ensure continuity at the vortex boundary $r_f$ when the bogus vortex is merged with the background field within a 40 km buffer zone ($r_f\pm 20$ km). The technical procedure of the tropical cyclone (TC) bogus vortex initialization (BVI) scheme adopted is described as follows (Figure 2).

3.3. Steering Flow

The large-scale environmental steering flow is the primary driver of typhoon motion [55]. It is defined as the mean horizontal wind vector within a radius of 500 km from the typhoon center, calculated after the removal of the analysis vortex, at 300, 400, 500, 600, 700, and 850 hPa [56]:

$$\begin{array}{l}{u}_{steering flow}={\displaystyle \frac{75u_{300}+100u_{400}+150u_{500}+175u_{600}+175u_{700}+150u_{850}}{825}}\\ v_{steering flow}={\displaystyle \frac{75v_{300}+100v_{400}+150v_{500}+175v_{600}+175v_{700}+150v_{850}}{825}}\end{array}$$

(12)

where $u_P$ and $v_P$ (P = 300, 400, 500, 600, 700, 850) are the zonal and meridional wind components of the background field at the pressure level P, respectively.

3.4. WRF Model Configuration

The Weather Research and Forecasting (WRF, v4.1.2) model is employed for typhoon simulations [57]. A two-way nested configuration is adopted, with the outer domain (D1) covering the track of Doksuri at 9 km resolution (750 × 580 grid points) and the inner domain (D2) at 3 km resolution (481 × 481 grid points). The model uses 115 vertical levels with the model top set at 1 Pa, allowing detailed representation of tropospheric structure and key physical processes. The model initial conditions are derived from the NCEP Global Forecast System (GFS) analysis at a horizontal resolution of 0.25°, and the lateral boundary conditions are updated every six hours using GFS forecasts.

Physical parameterizations include the Morrison two-moment microphysics [58], Kain–Fritsch cumulus (applied only in D1) [59], Dudhia shortwave radiation [60], RRTM longwave radiation [61], Noah land surface model [62], YSU planetary boundary layer scheme [63], and the Monin–Obukhov surface layer scheme [64].

We conducted two control simulations without the BVI scheme using the WRF model driven by NCEP GFS data (CTRL), and two BVI simulations with the scatterometer-based BVI incorporated into the NCEP GFS data employed in CTRL (Exp_BVI). They consist of two pairs of CTRL and Exp_BVI. The first pair was initialized at 1800 UTC 23 July, and the second pair was initialized at 0000 UTC 24 July. All four simulations made 72 h forecasts.

3.5. All-Sky Simulation

The Radiative Transfer for TOVS (RTTOV) fast radiative transfer model is widely used for simulating satellite brightness temperatures. In this study, RTTOV v13.0 [65] is used to simulate Himawari-9 AHI channel-13 brightness temperatures, with WRF forecast fields as input.

The input variables include 3D fields of pressure, temperature, and water vapor mixing ratio, cloud fraction, cloud liquid water content, and cloud ice content, as well as 2D fields of surface pressure, 2 m temperature, 2 m humidity, 10 m winds, surface temperature, and surface type from WRF output.

The input parameters of the satellite include satellite zenith angle, satellite azimuth angle, solar zenith angle, and solar azimuth angle.

3.6. Scatterometer-Based TC Center Positioning Method

The scatterometer-based center-positioning algorithm was proposed by [26] and tested by Pan [4]. It consists of the following four steps:

• Identify the maximum wind speed location (C_max).
• Search for candidate centers: within a 3° × 3° domain centered on C_max, extract all local minima of wind speed (C_min).

3.

Construct the direction vectors: for each candidate C_min point, define eight two-component direction vectors at the surrounding grid points $\overrightarrow{q_i}\hspace{0.33em}(i=1,\dots ,8)$. Depending on the observed wind direction at each point—falling within one of the intervals 0° ± 22.5°, 45° ± 22.5°, 90° ± 22.5°, 135° ± 22.5°, 180° ± 22.5°, 225° ± 22.5°, 270° ± 22.5° or 315° ± 22.5°—assign the corresponding vectors $\overrightarrow{q_i}$ as (1, 1), (1, 0), (1, −1), (0, −1), (−1, −1), (−1, 0), (−1, 1), and (0, 1), respectively.

4.

Determine the cyclone center: take the absolute sum of the eight vectors to obtain $Q=\left|{\displaystyle \sum _{i=1}^8\overrightarrow{q_i}}\right|$ for each candidate. The C_min point with the smallest $Q$ value is determined to be the TC center (C_TC).

Figure 3 provides two schematic illustrations of TC-center positioning based on scatterometer wind data at 0506 UTC 23 July and 0054 UTC 24 July 2023. Figure 3a,b show the horizontal distributions of HY-2C scatterometer wind speed (color shading) and vector (black arrow), along with the scatterometer wind-derived locations of C_max (triangle), C_min (circle), and C_TC (cross) at these two times. The scatterometer data coverage at 0506 UTC 23 July is notably better than that at 0054 UTC 24 July 2023. For convenience, Figure 3c,d provide schematic illustrations of the scatterometer-based TC-center-positioning method described above for these two observation times. Three local minimum wind speed locations (C_min) are identified within the 3° × 3° box centered on C_max. The center positions of Typhoon Doksuri at both times are clearly determined, characterized by a local wind-speed minimum and a coherent cyclonic circulation surrounding them.

4. Results

4.1. Scatterometer Data Availability for Typhoon Doksuri

There were 21 scatterometer swaths that passed Typhoon Doksuri during its lifecycle from 1800 UTC 21 to 0600 UTC 28 July 2023. This count excludes swaths that did not capture Doksuri’s center and full circulation (i.e., those with coverage less than the R_34kt-radius circle). Figure 4 shows the scatterometer-derived center locations (colored symbols) and the best track (black symbols) of Typhoon Doksuri over this period, with TC intensity categories also indicated. It is evident that the scatterometer-derived track follows the best track reasonably well. A slight southward bias of the scatterometer-derived centers is noted prior to 0000 UTC 23 July, when Doksuri was still a tropical storm.

Examples showing scatterometer swaths and wind observations, along with the corresponding NCEP GFS analysis model outputs within and around Doksuri, are shown in Figure 5. Spatial distributions of scatterometer sea surface wind vectors (black arrows) and wind speed (color shading) from HY-2C at 1754 UTC 22 July, MetOp-C at 1222 UTC 23 July, and HY-2B at 2153 UTC 23 July are compared with six-hourly NCEP GFS analysis at 1800 UTC 22 July, 1200 UTC 23 July, and 0000 UTC 24 July 2023, respectively. The MetOp-C swath (Figure 5c) is substantially narrower than the HY-2B/C swaths (Figure 5a,e). Overall, the TC wind structures compare well between the scatterometer and GFS fields, though differences are evident in eye size, eyewall radius, and 34 kt wind radius. The eye and eyewall correspond to the minimum and maximum wind-speed regions, respectively. For example, the eye in the HY-2C scatterometer winds at 1754 UTC 22 July is larger than that in the GFS analysis (Figure 5a,b), whereas the opposite is true for the HY-2B winds at 2153 UTC and the GFS analysis. The eyewall radius in the HY-2B winds at 2153 UTC 23 July is smaller than in the GFS analysis, while the 34 km wind radius from HY-2B at the same time is also smaller than in the GFS analysis (Figure 5e,f).

4.2. Bogus Vortex Initialization Results

Tropical cyclones in large-scale analysis fields are often too weak, too small in size, or misplaced. To forecast Typhoon Doksuri, initialized at 1800 UTC 23 July 2023, we first removed the analysis vortex from the GFS analysis at this time, following the procedure described in Section 3.2. Figure 6 shows the 850-hPa horizontal wind field from the GFS analysis at 1800 UTC 23 July 2023, including its disturbance field, analysis vortex, and environment field. The GFS analysis shows a well-structured vortex, since Typhoon Doksuri had reached a category-2 hurricane intensity (see Figure 1b). In fact, the minimum SLP was 964 hPa in the GFS analysis, which is very close to the best track value of 965 hPa at 1800 UTC 23 July 2023.

The sea-level pressure and 10 m wind fields of the GFS analysis vortex are shown in Figure 7 (left panels) and are compared with those of the bogus initial vortex (right panels in Figure 7). The latter was obtained following the procedure described in Section 3.1. The maximum wind speed near the center was 34.8 m s⁻¹ in the GFS analysis and 35.4 m s⁻¹ in the best track. A major difference between the bogus vortex and the GFS analysis vortex is the size of the eyewall, characterized by the ring of maximum 10 m wind speed. The bogus vortex has a larger radius of the eyewall than the GFS analysis vortex. Two additional differences between the bogus vortex and GFS analysis vortex are that the bogus vortex has a broader radial profile of sea-level pressure than the analysis vortex, and the bogus vortex has an axisymmetric wind distribution, while the analysis vortex has an asymmetric wind distribution.

By replacing the analysis vortex with the bogus vortex using Equation (11), we obtain a new field to serve as the initial condition for the forecast experiment Exp_BVI, whose typhoon center location is closer to the best track through repositioning, and minimum sea-level pressure is well aligned with typhoon center. The forecast differences between the Exp_BVI and CTRL experiments are compared in the following section.

4.3. Forecast Results

4.3.1. BVI Results

The first pair (CTRL and Exp_BVI) of 48 h forecast experiments was initialized at 1800 UTC 23 July 2023. The outer domain (9 km resolution) is fixed, and the inner domain (3 km resolution) is movable at 15 min intervals. Figure 8a presents the outer domain and two inner domains at 1800 UTC 23 July and 1800 UTC 26 July 2023, along with the CTRL and Exp_BVI, the sea-level pressure and 500-hPa geopotential height fields in the GFS analysis. The vertical distribution of the 115 model levels with respect to pressure is provided in Figure 8b. Typhoon Doksuri was located around 15°N, characterized by an isolated vortex of low sea-level pressure and low geopotential at 500 hPa. The minimum geopotential height at the vortex center is 5561 gpm. Track differences are clearly evident between the CTRL and Exp_BVI simulations.

Figure 9a shows the best track, the NCEP GFS analysis 10 m sea surface wind-determined track, and the minimum sea-level pressure-determined track of Typhoon Doksuri from 1200 UTC 21 to 0600 UTC 28 July, at 6 h intervals. Distances between the best track and the 10 m sea surface wind-determined track, and between the best track and the minimum sea-level pressure-determined track are provided in Figure 9b. Intensity categories are indicated by hollow symbols except at 0000 UTC, when they are indicated by solid symbols. For convenience, we added the evolution of the maximum sustained wind in Figure 9b. Both the 10 m wind circulation center and the center of minimum sea-level pressure closely followed the best track (Figure 9a). However, track errors were larger before 1200 UTC 22 July, varying from 60 to 130 km. This was the period when the cyclone was relatively weak, with the maximum sustained wind below 34 kt. As the storm intensified further, track errors decreased to about 20–30 km. The track differences between different numerical weather prediction (NWP) systems ranged from 20 to 30 km for all tropical cyclones over the western North Pacific during 2022–2023 [66]. Therefore, the positioning of Typhoon Doksuri in the GFS analysis was reasonably good.

Since the bogus vortex is axisymmetric, we assess the axisymmetry of Typhoon Doksuri to determine the appropriate times to deploy the axisymmetric bogus vortex. Figure 10a presents the temporal evolution of the 34 kt wind radii in the northwest, northeast, southwest, and southeast quadrants from the best track of Typhoon Doksuri during 1200 UTC 21–0600 UTC 28 July 2023, at 6 h intervals; Figure 10b shows the temporal evolution of the wavenumber-0 amplitude of 500-hPa relative humidity from the GFS analysis. During the intensifying period, differences in the 34 kt wind radii among the northwest, northeast, southwest, and southeast quadrants were relatively small at 1800 UTC 23 and 0000 UTC 24 July. The symmetry (i.e., wavenumber-0 component) of 500-hPa relative humidity at these two times was stronger and extended to larger radial distances (~300 km) than at other times. These were the reasons for carrying out the four forecasts described in Section 3.4: two initialized at 1800 UTC 23 July with (Exp_BVI) and without (CTRL) BVI and the other two initialized at 0000 UTC 24 July with (Exp_BVI) and without (CTRL).

4.3.2. Impact of BVI on Intensity and Track Forecasts of Typhoon Doksuri

The track forecasts from the first pair of Exp_BVI and CTRL experiments are shown in Figure 11. Specifically, Figure 11a shows the moving tracks of Typhoon Doksuri from the best track (black), the CTRL (red), and Exp_BVI (blue) experiments during 1800 UTC 23 July–1800 UTC 26 July 2023, at 6 h intervals. All tracks show the northwestward motion of Doksuri, resulting from the dominant β-effect. The CTRL track is closer to the best track than the Exp_BVI. The latter has a northwestward bias, due most likely to its larger steering flow than the CTRL. Track errors of the CTRL forecast and the Exp_BVI forecast, shown in Figure 11b, are below 60 km during the 24 h lead time for both CTRL and Exp_BVI, with the mean 24 h track error being 8.45 km. Track errors increase with time during 24–48 h forecasts, reaching 130 and 221.56 km at 48 h in the CTRL and Exp_BVI experiments, respectively. The BVI had a limited impact on track forecasts during the first 24 h because only the inner-core vortex was modified [67]. It had a more significant impact on track forecasts during the 24–48 h as a result of an interaction between the vortex and the environment. Verification of forecast track against scatterometer data is also performed. Compared with the forecast track errors verified with the best track, the forecast track errors from both CTRL and Exp_BVI experiments are smaller in the 36 h and comparable afterward (36–72 h).

A detailed comparison of the steering flow between the CTRL and Exp_BVI is provided in Figure 12, for the 54 h forecasts initialized at 1800 UTC 23 July 2023. The 54 h forecasts of deep-layer geopotential height and wind vectors from the CTRL (Figure 12a) and Exp_BVI (Figure 12b) are decomposed into the forecast vortex (Figure 12c,d) and the environment geopotential and wind vector (Figure 12e,f). The steering flow calculated from the environment flow is shown in Figure 12e,f. The steering flow of 3.09 m s⁻¹ in the CTRL was directed northeastward, while the steering flow of 3.04 m s⁻¹ in the Exp_BVI was directed northwestward due to stronger southeasterly flows northeast of Duksuri’s center. The forecast vortices were compact, with a deep low in the deep-layer geopotential field and cyclonic circulation (Figure 12c,d). The Exp_BVI forecasted a stronger, larger vortex positioned further north than the CTRL. Results in Figure 8 and Figure 9 highlight that while large-scale environments control the overall track, the temporal evolution of the initial bogus vortex and its interaction with the environment could modulate the track forecast beyond about 24 h lead time.

The impact of BVI on the intensity forecast of Typhoon Duksuri is shown in Figure 13. Figure 13a compares the temporal evolution of maximum sustained wind speed among the CTRL forecast, the Exp_BVI forecast, and the best track, as well as the differences in maximum sustained wind speed between the CTRL forecast and the best track, and between the Exp_BVI forecast and the best track. Figure 13b is the same as Figure 13a, except it is for the minimum sea-level pressure. The best track data indicates a period of rapid intensification (RI, +39 kt) from 1800 UTC 23 July to 1800 UTC 24 July, followed by rapid weakening (RW, –30 kt) during 25–26 July. The Exp_BVI forecasted the RI and RW reasonably well, producing an intensification of +49.38 kt during the RI period and an intensity weakening of –17.91 kt during the RW period. In contrast, the CTRL experiment gave a weaker RI (+31.97 kt) and a much weaker RW (−1.42 kt). The maximum simulated wind speed reached 118.91 kt, closer to the best track value of 128 kt than the CTRL (104.75 kt). The mean absolute error (MAE) of maximum simulated wind speed was 18.40 kt and 8.77 kt for the CTRL and Exp_BVI, respectively. The temporal evolution of minimum sea-level pressure in the Exp_BVI forecast also follows the best track data better. Based on the best track data, a reduction of −40 hPa occurred in the minimum sea-level pressure during the RI period from 1800 UTC 23 July to 1800 UTC 24 July. The reduction in minimum sea-level pressure was −36.46 hPa in the Exp_BVI and −25.49 hPa in the CTRL forecasts. The MAE of minimum sea-level pressure was reduced from 9.80 hPa (CTRL) to 4.73 hPa (Exp_BVI) by implementing the BVI. These results clearly demonstrate the improvement in intensity forecasts by scatterometer-based BVI. It is noted that the BVI track exhibits a right-of-track bias and lies farther from the island than the CTRL and best tracks (Figure 12). This displacement could partly favor stronger typhoon intensity by reduced terrain-induced frictional dissipation and structural disruption [68].

Results for the two forecasts initialized at 0000 UTC 24 July with and without BVI are presented in Figure 14 and Figure 15, which are similar to Figure 11 and Figure 13, respectively. Compared with the best track, the Exp_BVI initialized at 0000 UTC 24 July produced a better track forecast of Typhoon Doksuri (Figure 14) than the Exp_BVI initialized at 1800 UTC 23 July (Figure 12). The track errors in the Exp_BVI forecast are similar to the CTRL within 0000 UTC 24 July–0000 UTC 26 July 2023, and slightly larger than the CTRL within 0000 UTC 26 July–0000 UTC 26 July 2023. Track errors of the CTRL and Exp_BVI forecasts are below about 60 km during the 48 h lead time and increase to 170 km by the 72 h lead time. The intensity evolution (Figure 15) further confirms that BVI reduces intensity errors in the 72 h forecasts initialized at 0000 UTC 24 July 2023.

The track forecast error of Exp_BVI is larger than that of the CTRL experiment, implying that the BVI-related intensity change induces subtle variations in the interaction between the adjusted vortex and the ambient flow.

The cloud and rain structures within and around Typhoon Doksuri can be examined by the spatial distributions of AHI infrared brightness temperatures at channel 13 (10.45 μm), which are sensitive to radiation from both the ocean surface and atmosphere. In convective regions containing ice particles, brightness temperatures decrease because of scattering. With a spatial resolution of 2 km and a 10 min full-disk refresh rate, the spatial distributions of low brightness temperatures effectively reflect clouds and rain structures in tropical cyclones. Figure 16 shows the spatial distributions of AHI channel-13 brightness temperatures from Himawari-9 observations at 1200 UTC 25 July 2023 (Figure 16a), and all-sky simulations from the CTRL and Exp_BVI forecasts valid at the same time (Figure 16b–e). Also shown in Figure 16 are the 10 m wind vectors from the GFS analysis at 1200 UTC 25 July 2023 (Figure 16a), and the 10 m wind vectors from the CTRL and Exp_BVI valid at the same time (Figure 16b–e). Specifically, Figure 16b,c are the 42 h model forecasts initialized at 1800 UTC 23 July, and Figure 16d,e are the 36 h forecasts initialized at 0000 UTC 24 July 2023. Results from the CTRL and Exp_BVI forecasts are presented in the left and right panels, respectively. The cloud and rain structures of Typhoon Doksuri are captured by the model forecasts, showing a warm eye of similar magnitudes of brightness temperatures to its surroundings (~290 K), a cold eyewall (~200 K), and an orientation of cloud and rain distributions from the southwest to northeast colder than the eye [69]. The 10 m winds show a cyclonic circulation, converging toward the center. Stronger winds are located in convective clouds and rain clusters revealed by the low brightness temperatures, reflecting the dynamical–thermodynamical consistency of model forecasts. Compared with the GFS analysis, the model-predicted wind fields exhibit stronger intensity. We note clear differences in eye size among the Himawari-9 AHI infrared imagery (Figure 16a) and the forecasts from CTRL (Figure 16b,d) and Exp_BVI (Figure 16c,e). The AHI infrared imagery shows the smallest eye, owing to its higher spatial resolution (2 km) compared to the model forecasts (3 km). The CTRL run produced a more realistic representation of the smaller eye than Exp_BVI. The reasons for the differences between CTRL and Exp_BVI are not yet fully understood and will be investigated in future work.

4.3.3. Impact of BVI on Intensity and Track Forecasts of Typhoon Gaemi

Forecast Impacts of BVI using scatterometer-derived 34 kt wind radii and center locations, as well as the resulting track errors of Typhoons Gaemi (2024), are shown in Figure 17. A noticeable feature is a southwestward track bias that appears in the 48–72 h CTRL forecast (Figure 17a). Track errors for both CTRL and Exp_BVI remain mostly below 100 km during the 72 h forecasts (Figure 17b). The CTRL track error oscillates with time; whereas, the Exp_BVI track error generally increases with time. Exp_BVI errors are smaller than CTRL during the first 18 h, become larger between 18 and 24 h, and are comparable to CTRL thereafter. Similarly to the results for Typhoon Doksuri, the intensification of Typhoon Gaemi is better captured by Exp_BVI (Figure 17c,d). However, the weakening in Exp_BVI is delayed by approximately six hours relative to the best track. At the 52 h forecast time, when the tropical cyclone track crosses Taiwan, the best track suggests the storm center (black square symbol) is located over the land; whereas, the storm center from the CTRL (red square symbol) and Exp_BVI (blue square symbol) is located to the southwest and southeast of the best track, respectively. This makes the Exp_BVI track almost overlap with the best track, but the CTRL track does not. The similarities between Typhoon Gaemi (2024) and Typhoon Doksuri (2023)—in terms of their genesis location, track, intensity evolution, and landfall site—make this pair of cases an ideal testbed for examining the cross-case applicability of research methodologies.

5. Discussion

This study validates an objective bogus vortex initialization (BVI) scheme that leverages satellite microwave scatterometer winds to improve the initial fields and forecasts of Typhoon Doksuri (2023).

Evaluation of this pre-existing BVI scheme [4] in a real typhoon-forecasting scenario demonstrates several notable advantages:

• Improved initial vortex representation. The method substantially enhances the accuracy of the initial vortex center position and minimum sea-level pressure, thereby providing a more reliable starting point for forecasts.
• Significant gains in intensity prediction. While track forecasts show limited improvement, intensity forecasts are greatly enhanced—particularly during rapid intensification (RI) and rapid weakening (RW). The BVI scheme reduces forecast biases in both maximum sustained winds and minimum sea-level pressure and reproduces peak intensity more faithfully than the CTRL experiment. These findings reinforce the view that realistic initialization of the inner-core vortex is crucial for skillful intensity forecasting.
• Enhanced depiction of storm structure. Comparisons between simulated all-sky brightness temperatures (via RTTOV) and Himawari-9 AHI observations indicate that BVI produces more realistic inner-core structures, including a clear eye and a cloudy eyewall. The simulated dynamical and microphysical fields align quite closely with observations, suggesting that the WRF model configuration used in this study represents eyewall convection and cloud–rainband evolution reasonably well.

Overall, the results confirm that scatterometer winds provide indispensable value for typhoon monitoring and vortex initialization. The proposed objective BVI scheme effectively integrates these data to improve intensity forecasts and structural realism, highlighting both scientific importance and operational potential. However, we note that BVI produces slightly worse track forecasts while improving intensity forecasts. We also acknowledge that the results are based on only two cases and therefore are not statistically robust.

Looking forward, the scheme could be further advanced by incorporating additional satellite datasets—such as microwave and infrared brightness temperatures—within three-dimensional and four-dimensional variational assimilation (3D/4D-Var) frameworks. Such extensions may help bridge the gap between environmental-flow analyses and inner-core dynamics, thereby further improving intensity forecasts without degrading track forecasts across a broader range of tropical cyclones.

Two key parameters—R_34kt and P_c—are used in Fujita’s bogus vortex formula. The method developed in this study incorporates R_34kt from scatterometer data but uses P_c from the best track dataset. Because best-track data are typically released only after the typhoon’s lifecycle has ended, sometimes months later, a follow-on study is planned to derive P_c directly from satellite observations by combining scatterometers with microwave temperature-sounding data. Tian and Zou [70] demonstrated a strong correlation between P_c and microwave-temperature-sounder-derived warm-core anomalies at 250 hPa (see Figure 6), providing a promising pathway toward such an approach.

6. Conclusions

This study demonstrates the practical utility of an objective bogus vortex initialization (BVI) scheme that incorporates satellite microwave scatterometer winds to improve forecasts of Typhoon Doksuri (2023). The results confirm that scatterometer-derived winds provide valuable constraints for tropical cyclone monitoring and vortex initialization, and that their objective integration can yield measurable forecast benefits. Specifically, BVI markedly enhances the initial vortex depiction by improving the analyzed vortex center location and minimum sea-level pressure, thereby providing a more realistic and dynamically consistent initial condition for numerical prediction. While improvements in track forecasts are limited, intensity prediction is substantially strengthened—most notably during rapid intensification (RI) and rapid weakening (RW)—as evidenced by reduced systematic biases in maximum sustained winds and minimum sea-level pressure and a more faithful reproduction of peak intensity relative to the CTRL experiment. In addition, comparisons between simulated all-sky brightness temperatures (via RTTOV) and Himawari-9 AHI observations indicate improved storm structure realism under BVI, including a clearer eye and a more coherent cloudy eyewall, suggesting better representation of inner-core convection and cloud–rainband evolution. Nevertheless, BVI slightly degrades track skill in this study, implying that improvements to inner-core initialization alone may not translate into concurrent enhancement of the large-scale environmental flow.

The ultimate goal is to make the BVI method operationally practical for real-time forecasting. Another issue is the symmetric assumption imposed on the bogus vortex in this study. Incorporating an asymmetric structure is nontrivial. Because planetary-vorticity advection by the symmetric flow strongly influences the development of asymmetric TC structure, an asymmetric wind component can be obtained by integrating the barotropic vorticity equation using a symmetric vortex as its initial condition [1,6]. By adding this asymmetric component to a symmetric bogus vortex, Zou et al. [71] showed that assimilating GOES-13 and GOES-15 imager radiances in HWRF produced substantially larger positive impacts on both track and intensity forecasts of Tropical Storm Debby in the Gulf of Mexico compared to using a purely symmetric BVI. Building on these results, we plan to construct a realistic asymmetric bogus vortex informed by satellite imager-derived TC asymmetries.