Impacts of Physical Parameterization Schemes on Typhoon Doksuri (2023) Forecasting from the Perspective of Wind–Wave Coupling

Abstract

Tropical cyclones (TCs) form over warm ocean surfaces and are driven by complex air–sea interactions, posing significant challenges to their forecasting. Accurate parameterization of physical processes is crucial for enhancing the precision of TC predictions. In this study, we employed the Weather Research and Forecasting model coupled with the Simulating Waves Nearshore (WRF-SWAN) model to forecast Typhoon Doksuri (2023), which exhibited a secondary intensification process in the South China Sea (SCS). We also investigated its sensitivity to various atmospheric physical parameterization schemes (PPS). The findings indicate that improvements in microphysical and cumulus convection parameterizations have significantly enhanced the prediction accuracy of Typhoon Doksuri’s trajectory and intensity. The simulation of sea surface heat flux is primarily influenced by the microphysical scheme, while the cumulus convection scheme substantially affects the representation of the typhoon core’s size and shape. Variations in the wind field induce differences in wave height, potentially reaching up to 2–3 m at any given moment. This study provides valuable insights into the effective selection of physical parameterizations for improving typhoon forecasts.

1. Introduction

Tropical cyclones (TCs) are a form of extreme catastrophic weather on a mesoscale, exerting significant impacts on both human society and the environment. They result in substantial economic losses and human casualties annually in coastal regions [1]. TCs form and develop through intricate air–sea interactions, accompanied by extreme weather phenomena such as strong winds, heavy rain, and storm surges. The Northwestern Pacific Ocean is the world’s most active region for TC formation due to its expansive ocean surface spanning middle and low latitudes, which have relatively warm sea surface temperatures conducive to TC genesis. The South China Sea, interconnected with the Northwestern Pacific Ocean, features extensive coastal infrastructure and construction. Coupled with its distinctive coastal topography, this underscores the importance of studying tropical cyclones in the South China Sea [2,3,4].

A number of studies, based on the statistical analysis of historical data and forecasts for the future, have identified a trend of increasing TC intensity and a corresponding increase in the threat of strong TCs [5,6]. The influence of climate change on typhoon activity patterns plays a crucial role in this context. For example, the main region of occurrence for the El Niño Southern Oscillation (ENSO) is the Pacific Ocean, where ENSO impacts typhoons through its regulation of both the atmosphere and the ocean [7,8,9]. In the atmospheric domain, ENSO leads to changes in vertical wind shear, humidity distribution, low-level atmospheric vorticity, and the intensity and position of the subtropical high-pressure system [10,11,12]. In the oceanic domain, it is reflected in changes in sea surface temperature, as well as adjustments in the upper ocean’s heat content and structure [12,13]. Therefore, typhoon forecasting remains a long-standing challenge. Although the forecasting capability of numerical models has significantly improved in recent years through the development of techniques such as satellite observation data assimilation and parameterization of physical processes, forecasting TCs with rapid intensification, secondary intensification, and sudden path changes remains challenging [14,15,16]. The principal mechanism through which numerical models facilitate extreme weather forecasting is by parameterizing dynamic and thermal processes [16]. Therefore, for tropical cyclones subject to air–sea interactions, optimizing the physical parameterization scheme and incorporating oceanic parameters are crucial for enhancing forecast accuracy [17,18,19].

The intensity of TCs is closely linked to the state of the oceans. Numerous studies have confirmed the role of surface heat fluxes in influencing TC intensity [20]. Oceanic heat flux, which supplies energy to the atmosphere, is a critical source of energy for TCs. The transfer of heat flux resulting from elevated sea surface temperatures (SSTs) can modulate TC intensity, while localized SST decreases due to TCs can also impact oceanic conditions [21]. During TC formation, warm surface water releases sensible heat, transferring some of its energy to the overlying atmosphere. Simultaneously, moisture evaporates from the ocean surface, converting water into water vapor and storing latent heat [22]. Bryan improved numerical simulation parameterization to better characterize the thermal flux and momentum exchange between the ocean and the atmosphere and investigated its impact on hurricane structure [23].

Extreme nearshore waves represent a significant manifestation of the threats posed by TCs to coastal regions. Waves and typhoons have a complex bidirectional feedback relationship [19,24]. Typhoons provide the energy for the generation and propagation of waves through strong winds and low pressure, which leads to the amplification of ocean surface waves, especially around the typhoon center and eyewall. In turn, wave breaking and spray effects alter the energy exchange at the air–sea interface, enhancing surface evaporation [25]. At the same time, the presence of waves modifies the roughness of the sea surface in the lower layers of the typhoon, further influencing the typhoon’s intensity and structure [26]. This bidirectional feedback mechanism makes the interaction between typhoons and waves dynamic and mutually dependent, with both factors jointly influencing the typhoon’s evolution. Therefore, investigating the relationships between strong winds and waves under TC conditions is crucial for improving the accuracy of TC intensity forecasts. Globally, current observations of TC-induced waves are not comprehensive [27], but numerical models can compensate for this lack of operational data to some extent. Numerous studies have enhanced wave simulations by refining wave parameterization in models [28,29]. Wave height, a critical wave parameter, is closely associated with storm tracks, frequency, and intensity [30]. This paper explores the effect of strong winds on wave height.

This study aims to enhance the forecasting of Typhoon Doksuri in 2023 by utilizing the Weather Research and Forecasting (WRF) atmospheric model, coupled with the Simulating Waves Nearshore (SWAN) wave model, to investigate the wind–wave interaction. Superior WRF simulation outcomes are critical for improved coupling performance. Numerous studies have demonstrated that variations in microphysics and cumulus parameterization schemes affect the size, intensity, and convective processes of tropical cyclones’ simulations [17,31,32,33]. Therefore, conducting sensitivity experiments to evaluate the impacts of different physical parameterization schemes (PPSs) is essential. This analysis focuses on the perspectives of typhoon track, intensity, and sea surface heat flux.

2. Materials and Methods

2.1. Model Configuration

The Coupled Ocean Atmosphere Wave Sediment Transport (COAWST) Modeling System (Version 3.8) is used to implement the models’ coupling in this paper (https://github.com/DOI-USGS/COAWST (accessed on 17 January 2025)), which incorporates multiple model components for atmosphere (WRF), ocean (Regional Ocean Modeling System, ROMS), waves (SWAN, WAVEWATCH III), and sediment transport. Different models exchange variables online under the control of the Model Coupling Toolkit (MCT) coupler, the spherical coordinate remapping interpolation package (SCRIP) is used to calculate interpolation weights between domains, and the coupling simultaneously supports mode grid refinement [34]. The COAWST system contributes to a better understanding of ocean–atmosphere–wave interaction processes; for this study, both WRF and SWAN are run based in COAWST.

2.1.1. Atmospheric Model

WRF is a mesoscale atmospheric model which has been widely used for forecasting as well as realistic and idealized research experiments for numerous weather phenomena [35]. To enhance the tracking of TC centers, we implemented the vortex-following option in WRF to establish moving nests. This study used a 24-h rolling forecast for typhoon prediction, where each 24-h forecast is based on the latest GFS meteorological forecast background field to reduce errors caused by inaccuracies in the initial field (same for coupling cases).

WRF is configured with triply nested grids at 27 km, 9 km, and 3 km, respectively. The innermost nest is the moving nest, which locks in the position of TC Doksuri (Figure 1a). From the outside to the inside, the number of grid points are 78 × 69, 187 × 130, and 67 × 64, respectively. In the vertical direction, the model contains a vertical structure of 45 layers with the model top at 50 hPa. The time step of the WRF is 90 s. In the output of the TC simulation results, key parameters such as the TC center’s position, the minimum sea level pressure (MSLP), and the maximum wind speed near the center (MWS) are outputted by the automatic vortex following mode.

In this study, we set up sensitivity tests of PPSs with microphysics and cumulus parameterization as variables to obtain better performing scheme combinations for model coupling (Section 2.1.3). The non-variable PPS used in the simulations is as follows: RRTMG for longwave and shortwave radiation, YSU for boundary layer, MM5 for surface layer, and Noah for land use.

2.1.2. Wave Model

The SWAN model was used for nearshore wave simulation here. SWAN simulates the generation and propagation of wind waves in coastal waters and includes the processes of refraction, diffraction, shoaling, wave–wave interactions, and dissipation due to white capping, wave breaking, and bottom friction [36,37].

The wind field input for SWAN can be sourced from theoretical models, empirical formulae, reanalysis datasets, or mesoscale meteorological models. Among these options, mesoscale meteorological models provide high accuracy and are better suited for supplying the wind field conditions required by SWAN. In this study, the coupled wind and wave prediction was achieved through WRF-SWAN. Here, WRF furnishes the wind variables for SWAN, while SWAN generates outputs detailing wave parameters such as wave heights, wavelengths, and periods [34]. These wave parameters are input into the WRF model, affecting the calculation of sea surface roughness [38,39]:

$$z_0=1200\cdot Hgt{\left({\displaystyle \frac{Hgt}{Len}}\right)}^{4.5}+0.11\cdot {\displaystyle \frac{v}{u_{\ast }}}$$

(1)

where $z_0$ means the sea surface roughness, Hgt means the significant wave height, Len means the wave length, v represents the viscosity, and $u_{\ast }$ means the surface stress.

Specifically, an increase in wave height enhances surface wind stress, alters the drag coefficient, and increases surface friction, leading to higher wind speeds and turbulence, which in turn affects the stability of the atmospheric boundary layer. At the same time, changes in the waves modify the sea surface’s reflective properties, influencing the radiation balance of the atmospheric boundary layer, thereby impacting local weather processes.

In this study, SWAN adopted the same horizontal grid as the WRF inner domain (d02) of 186 × 129 grid points, with an approximate horizontal resolution of 9.8 × 9 km (Figure 1b). The time step of SWAN was set at 90s.

2.1.3. Experimental Designs

The National Center for Atmospheric Research (NCAR) has developed a physics suite named “TROPICAL” to simulate the track and intensity of TCs within the WRF model. This study conducts sensitivity tests based on the “TROPICAL” suite as the CTL group, modifying microphysics and cumulus parameterization schemes (

Table 1. Experiment design on the PPS of WRF (“√” represents that the option has been selected.).

Experiment			CTL	C1	C2	C3	C4	C5
WRF	Microphysics Parameterization	WSM6	√	√
		Goddard 4-ice			√	√
		Milbrandt 2-mom					√	√
	Cumulus Parameterization	New Tiedtke	√		√		√
		KIAPS SAS (KSAS)		√		√		√

). The test group exhibiting optimal performance will be chosen for integration with SWAN, and its results will be compared to those of the control group.

Cyclone hydrometeors (such as cloud water, rainwater, cloud ice, snow, and graupel) are closely associated with the cyclone’s diabatic heating and vertical velocity structure, making them effective indicators of storm intensity. The microphysics parameterization schemes in WRF focus on representing such processes. Based on previous studies and the approaches used by various schemes for calculating cyclone hydrometeors, this study selects the following schemes as the simulation variables. The WSM6 scheme has been developed by incorporating additional processes related to graupel into the WSM5 scheme. The prognostic variables of water substance in this scheme include the mixing ratios of water vapor (q_V), cloud water (q_C), cloud ice (q_I), snow (q_S), rain (q_R), and graupel (q_G) [40]. The Goddard 4-ice scheme developed the 4ICE (cloud ice, snow, graupel, and frozen drops/hail) scheme to expand the ability of the microphysics to include more intense convection; furthermore, hail and graupel can occur in real weather events simultaneously. Therefore, a 4ICE scheme with both graupel and hail is useful for numerical weather prediction [41]. The Milbrandt 2-mom scheme consists of six hydrometeor categories, and a closure formulation for calculating the source and sink terms of a third moment of the size distribution; the radar reflectivity is established and used to predict all precipitating hydrometeor categories [42].

The energy released from condensation and precipitation in convective clouds plays a crucial role in the development of a typhoon. Cumulus parameterization schemes redistribute air in gridded columns to account for vertical convective fluxes, which has also been selected as a variable to evaluate the effectiveness of typhoon simulation. The New Tiedtke scheme can largely improve the simulations on tropical variations as well as the diurnal cycle of precipitation; this scheme is improved in terms of new trigger functions for deep and shallow convection initiations, new convection closures for shallow and deep convection, etc. [43]. The KSAS scheme improves the performance of the cumulus parameterization across the gray-zone resolutions of the numerical model (1 km to 10 km) through the combination of two different updraft fractions [44].

2.1.4. Model Verification

The model’s performance is validated based on the track and intensity of TCs.

For track, Bias is used to calculate the deviation between the forecast result and the observation result:

$$Bias={\displaystyle \frac{1}{N}}\sum _{i=1}^n\left(S_i-O_i\right)$$

(2)

where N is the total number of TC locations, S_i is the simulated location of the TC center, O_i is the observed location of the TC center, and the subtraction of S_i and O_i indicates the calculation of the geographical distance between the two latitude and longitude locations.

For intensity, the Bias, Pearson product-moment correlation coefficients (R), root mean squared error (RMSE), mean absolute error (MAE), standard deviation (SD), and Nash–Sutcliffe efficiency (NSE) are used to evaluate the deviation between the forecast result and the observation result:

$$R={\displaystyle \frac{{\sum }_{i=1}^n\left(S_i-\overline{S}\right)\left(O_i-\overline{O}\right)}{\sqrt{{{\sum }_{i=1}^n(S_i-\overline{S})}^2}\sqrt{{\sum }_{i=1}^n{\left(O_i-\overline{O}\right)}^2}}}$$

(3)

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{\left(S_i-O_i\right)}^2}$$

(4)

$$MAE={\displaystyle \frac{1}{n}}\sum _{i=1}^n\left|S_i-O_i\right|$$

(5)

$$SD=\sqrt{{\displaystyle \frac{1}{n}}\sum _{i=1}^n{\left(x_i-\overline{x}\right)}^2}$$

(6)

$$NSE=1-{\displaystyle \frac{{\sum }_{i=1}^n{\left(O_i-S_i\right)}^2}{{\sum }_{i=1}^n{\left(O_i-\overline{O}\right)}^2}}$$

(7)

where n is the total number of data, $\overline{S}$ is the mean of the simulation values, $\overline{O}$ is the mean of the observation values, x_i is the value of each set of data, and $\overline{x}$ is the mean value of each set of data.

In order to provide a more comprehensive evaluation of the simulation performance of the test groups, a normalized statistical score, rate, is employed to assess the simulation performance. A rate closer to 0 indicates a superior simulation, and the optimal group will be utilized for subsequent wind–wave coupling.

$${rate}_{RMSE}={\displaystyle \frac{{RMSE}_i-{RMSE}_{min}}{{RMSE}_{max}-{RMSE}_{min}}}\times 100\%$$

(8)

$${rate}_{SD}={\displaystyle \frac{\left|{SD}_i-{SD}_{obs}\right|}{max\left(\left|{SD}_{max}-{SD}_{obs}\right|,\left|{SD}_{min}-{SD}_{obs}\right|\right)}}\times 100\%$$

(9)

$${rate}_R={\displaystyle \frac{1-R_i}{1-R_{min}}}\times 100\%$$

(10)

$$rate=\left({rate}_{RMSE}+{rate}_{SD}+{rate}_R\right)\times {\displaystyle \frac{1}{3}}$$

(11)

where i is the number of locating TC (i = 1, 2, 3…n), max is the maximum value of the variable, and min is the minimum value of the variable.

2.2. Data

The initial and boundary conditions and the time-varying sea surface temperature (SST) data supplied to WRF were driven by real-time global forecast data from the Global Forecast System (GFS) of the National Centers for Environmental Prediction (NCEP), with 0.5° of spatial resolution (available online at https://www.ncei.noaa.gov/products/weather-climate-models/global-forecast (accessed on 17 January 2025)).

The real-time release system of TCs from China Meteorological Administration (CMA) provides information on TCs’ track and intensity, which is used for the comparison of forecast results (available online at https://tcdata.typhoon.org.cn/zjljsjj.html (accessed on 17 January 2025)).

The marine topographic data used for SWAN are from the General bathymetric Chart of the Oceans (GEBCO) in 2023 (available online at https://download.gebco.net/ (accessed on 17 January 2025)) (Figure 1c).

2.3. Case Description

Doksuri was the fifth named storm of the 2023 Northwest Pacific cyclone season, forming on the morning of 21 July 2023. It moved northwestward and developed into a super TC, reaching peak intensity on 25 July and then landfall in the Philippines on the next day, intensifying into a super TC again on the evening of 27 July, and making landfall a second time off the coast of Fujian Province in China. Doksuri weakened to a tropical depression on 29 July and ceased to be numbered, but its residual circulation continued to move northward and penetrate deep into inland China. Doksuri, characterized by high intensity, long duration, and secondary intensification, had a severe impact on both China and the Philippines; it caused extreme rainfall and disasters. In this paper, we will use TC Doksuri as a case study. The forecast period for Doksuri was from 26 July at 0000 UTC to 28 July, with a forecast interval of 24 h.

3. Results

3.1. Comparison of PPS Sensitivity Tests

3.1.1. TC Track and Intensity

Based on the TC data provided by the CMA, Doksuri, which had already developed into a super TC in the Western North Pacific, weakened from 26 July to 27 July; this period is referred to as the weakening phase here. Then, it witnessed a secondary phase of intensification from 27 July to 28 July, developing into a super TC once again, and we refer to this period as the secondary enhancement phase.

Figure 2 shows the track results of different test groups of PPSs. Prior to 1200 UTC on 27 July, the simulated paths exhibited minimal divergence. However, after this time, the test groups, except for CTL, displayed noticeable shifts, generally aligning more closely with the path provided by the CMA. A comparison with the CTL revealed an improvement in Doksuri’s track bias across all test groups (

Table 2. Track deviations of Doksuri.

	CTL	C1	C2	C3	C4	C5
Track bias (km)	83.55	70.92	58.89	62.02	58.80	65.93

). Among these, C4 had the smallest path deviation of 58.80 km, followed by C2 with a deviation of 58.89 km. Notably, there are significant differences in the microphysical schemes utilized by C2, C4 and CTL. C2 employed the Goddard 4-ice scheme, C4 used the Milbrandt 2-mom scheme, while CTL employed the WSM6 scheme. In terms of landing site capture, there was no discernible difference in performance between the experimental groups, with CTL and C1 being the closest to the landing position provided by the CMA.

As illustrated in Figure 2, multiple variable groups exhibited a sudden path deviation in the afternoon of July 27th. This period marked the critical phase when Doksuri underwent its secondary intensification into a super typhoon, characterized by significant short-term intensification and dramatic changes in the atmospheric environment. Given that this study utilized a rolling forecast approach with daily updates to the meteorological background fields, the path deviation is speculated to result from substantial differences in the background field before and after the update. The figure shows that, after the deviation, the paths of all groups align more closely with the observed results. This not only highlights the necessity of the rolling forecast approach but also reflects the challenges of using the initial field to capture rapid atmospheric changes [6,45].

Comparison of the MSLP and the MWS between the CMA results and simulation results are shown in Figure 3. In contrast to the CMA, all test groups demonstrated a tendency to overestimate the weakening phase of Doksuri and underestimate the secondary enhancement phase. However, the different settings of PPSs reflected varying degrees of improvement.

Table 3. Model deviations of Doksuri’s MSLP and MWS. The bold numbers are the best performance values in each group.

	MSLP							MWS
	R	Bias	MAE	RMSE	SD	NSE	Rate	R	Bias	MAE	RMSE	SD	NSE	Rate
CTL	0.54	6.95	10.5	13.33	6.38	0.02	96.07	0.51	−0.18	4.85	6.14	5.66	0.15	50.26
C1	0.62	0.03	9.43	10.58	7.56	0.38	52.07	0.47	1.28	5.63	6.89	6.45	−0.08	49.79
C2	0.66	4.35	9.13	11.36	6.10	0.29	65.45	0.67	3.42	5.54	6.46	6.83	0.05	29.94
C3	0.59	−1.25	9.36	10.99	7.67	0.34	58.55	0.62	3.73	5.13	6.56	5.59	0.02	51.84
C4	0.61	4.42	9.03	11.67	6.75	0.25	69.61	0.67	5.10	6.43	8.11	8.32	−0.49	87.42
C5	0.50	2.76	9.99	11.97	7.26	0.21	78.28	0.56	2.23	5.48	6.21	5.45	0.13	52.47

presents the simulated deviation results for each group regarding MSLP and MWS. Firstly, in terms of MSLP, the NSE index of the C1 is closest to 1, with the lowest rate indicating the best overall performance. Additionally, the rate and NSE indices of all variable groups outperform those of the CTL group. Secondly, in terms of MWS, the NSE index of the CTL is closest to 1, while the C2 has the lowest rate. Furthermore, Table 3 shows varying degrees of improvement in aspects such as correlation and bias simulation for the variable groups compared to the control group. For example, C2 and C4 show higher simulation correlations for both MSLP and MWS than the CTL, and all variable groups exhibit lower bias for MSLP than the CTL. The results of the NSE index also indicate that all simulation results are close to the average level of the observed values.

It was noted that CTL, C1, and C5 displayed significantly different performances between simulating MSLP and MWS. For example, CTL had the worst simulation of MSLP among all the groups but had the best performance for the simulation of MWS, which suggests that the CTL did not accurately simulate the relationship between pressure and wind during Doksuri, a limitation that has also been observed in previous studies [17,32]. The relationship between TC wind and pressure is influenced by various factors, including the TC’s location, size, and environmental pressure. The effectiveness of the model in forecasting TCs relies on its ability to quantify and calculate these influencing processes. Enhancing the quantification of the wind–pressure relationship may therefore improve forecast accuracy.

The analysis of rates for each group regarding MSLP and MWS reveals that C1 showcases the most favorable simulation performance at MSLP, whereas C2 demonstrates the most favorable simulation performance at MWS. The average rates of all groups indicate that C2 exhibits the best overall simulation performance. Combining the bias of track simulation, C2 was chosen as the group with the best results in terms of Doksuri’s path and intensity, which is used for the subsequent WRF-SWAN coupling.

The mesoscale wind field plays a crucial role in the intensity of TCs and interaction between wind and waves in the later stages. The time chosen for comparison is 27 July at 0000 TC. Figure 4 shows the wind field and the sea level pressure (SLP) field of each test group. In terms of TC structure, the windless or still-wind area at the center can be identified as the TC eye. Surrounding the eye is the eyewall area characterized by strong convective motion, where the strongest gusts of wind and waves of 10 m or more are mainly found. Firstly, the structure of the TC eye is initially analyzed in terms of SLP and wind speed. C1 exhibited the smallest eye size, followed by CTL, C3, and C5, with no significant difference, while C2 and C4 have the largest eye sizes. Secondly, the SLP within the TC eye area also vaied significantly across groups. In particular, the SLP values of C1, C3, and C5 are considerably lower compared to the other groups. Furthermore, the wind distribution reveals the asymmetry of the intensity of the TC eye wall in each group; the high wind speed areas of CTL, C2, C4, and C5 are located to the upper left of the TC eye, while those of C1 and C3 are distributed to the lower right of the TC eye.

3.1.2. Surface Heat Flux

Heat flux is regarded as an important source of energy for TCs. In order to gain further insight into the simulation of the TC’s intensity, in addition to the analysis of air pressure and wind speed at the center of Doksuri, latent heat flux (LHF) and sensible heat flux (SHF) are also incorporated here. Taking the secondary enhancement period as an example, the 24-h average LHF distribution for each group in the innermost domain of the WRF is presented in Figure 5, while the 24-h average SHF distribution is shown in Figure 6. These figures highlight the variations in Doksuri’s inner core structure and heat flux intensity among the groups. The thermal flux distribution patterns for each group are unique, with notable differences in their intensities. Upon examining both LHF and SHF, clear asymmetrical distribution patterns are observed. C1, C3, and C5 show a left-side bias, while CTL, C2, and C4 tend towards a right-side bias. In terms of LHF intensity, C3 exhibits the most significant average peak value at 961.99 W/m², closely followed by C4 at 936.19 W/m². For SHF, C1 records the highest average peak value at 190.10 W/m², followed by C4 at 176.38 W/m². Additionally, C5 consistently shows lower maximum values for both latent and sensible heat flux compared to the other groups.

Figure 5 and Figure 6 demonstrate that microphysical processes and cumulus convection are crucial in the representation of surface fluxes. The microphysical scheme primarily focuses on analyzing latent heat processes within the atmosphere. Within convective cloud clusters of a typhoon, water vapor rapidly condenses into water droplets or ice crystals, thereby releasing latent heat. This process aids in heating ascending air currents, facilitating convective development. Within the eyewall of a typhoon, strong winds induce seawater evaporation, generating water vapor. These vapors ascend to higher altitudes within updrafts, releasing latent heat and absorbing heat from the surrounding environment. All these processes collectively influence the intensity and structure of a typhoon. Different cumulus convection schemes treat the initiation, intensity, and distribution of cloud water differently. A strong convection scheme enhances the typhoon’s upward motion, leading to greater heat transfer through convection into the typhoon’s core, thus increasing latent and sensible heat flux. On the other hand, a weak convection scheme limits convection, resulting in relatively smaller heat flux.

3.2. Comparison of Wind–Wave Coupling

3.2.1. Wind Fields

The synthesis in Section 3.1 suggests that C2 emerges as the most effective group among all PPS sensitivity tests. To investigate the impact of PPS optimization on the wind–wave coupling TCs’ prediction, the coupled WRF-SWAN simulations were conducted in CTL (CTL_coupling) and C2 (C2_coupling), respectively.

Wind field at a height of 10 m and a grid resolution of 9 km for three time points are presented (Figure 7), along with the differences in wind speed and direction (Figure 8), which have direct impacts on wave generation and propagation.

For the wind speed, at 0000 UTC on 27 July, the central wind speed in the TC eye area of C2_coupling is demonstrably lower than that of CTL_coupling, with a maximum difference of 25.74 m/s. Conversely, in the TC eyewall area, the wind speed of C2_coupling surpasses that of CTL_coupling significantly, with a maximum difference of 17.65 m/s. The difference is mainly observed in the upper left part of the TC eye. At 1200 UTC on 27 July, the C2_coupling is higher than CTL_coupling, with the highest value of the difference occurring in the upper part of the TC center, at the wind speed of 23.14 m/s. At 0000 UTC on 28 July, when Doksuri was about to make landfall, the range of sea surface wind speed differences between the two groups was significantly reduced. However, the difference remained considerable, as evidenced by the inner wall of the TC. The highest difference in wind speed was observed between C2_coupling and CTL_coupling, with a difference of 29.93 m/s. Simultaneously, CTL_coupling exhibited a wind speed of 28.52 m/s higher than C2_coupling. In terms of wind direction, the closer to the TC center, the more pronounced the difference.

3.2.2. Wave Fields

Corresponding to the wind field, Figure 9 shows the performance and differences between the two groups on the wave height field. Comparing the differences in wind speed and wave height, we can see a striking similarity in their spatial and temporal distributions. This confirms that wave height is directly influenced by wind speed. At 0000 UTC on 27 July, the sea surface wave heights of CTL_coupling are primarily located on the right side of the TC eyewall, while those of C2_coupling are distributed on the upper right side of the TC eyewall. The difference at the TC eye wall is extremely significant; the wave heights of C2_coupling are, on average, higher than those of CTL_coupling, with a maximum difference of 3.34 m. On July 27th at 1200 UTC, as Doksuri approached land, the high-value area of waves narrowed due to the topography of the strait. Both groups exhibited higher wave heights on the right eyewall, with a maximum difference of 2.03 m between C2_coupling and CTL_coupling. At 0000 UTC 28 July, the wave heights further decreased, with C2_coupling still higher than CTL_coupling, with a maximum difference of 1.19 m between the two groups.

4. Discussion

In the context of ongoing advancements in typhoon forecasting techniques, accurately predicting typhoons with varying characteristics remains a formidable challenge. Extensive research has underscored that enhancing our understanding of physical processes and refining parameterization schemes are pivotal avenues for bolstering accuracy [46]. A challenge also lies in the selective evaluation of existing PPS. Furthermore, ocean waves are not only an important medium for energy transfer from the upper ocean to the typhoon but also a form of extreme maritime disaster during a typhoon event. Therefore, they should be considered in the typhoon forecasting process.

In Section 3.1 of this study, the impact of different physical parameterization schemes on typhoon simulations was tested using the standalone WRF model. Then, in Section 3.2, the coupled WRF-SWAN model was used to compare the wind–wave simulation results of the CTL and C2 configurations. From Table 2, it is evident that the simulations of Doksuri’s path improved after adjusting the microphysics and cumulus parameterization schemes. In Table 3, variations in simulating the Doksuri’s central intensity are notable across different schemes. Additionally, it was found that under different schemes, there were obvious differences in the upward motion, size, and shape of the typhoon core. The shape of the inner core, for example, was eye-shaped, egg-shaped, and oval-shaped (Section 3.1.2). Upon coupling with SWAN, we quantified the impact of wind on wave height, with variations of up to 3.34 m on July 27th at 0000 UTC (Section 3.2.2). This underscores the importance of coupling effects. Hence, adjusting PPS aids in enhancing the accuracy of typhoon forecasting and directly influences the coupling effect of oceanic and atmospheric dynamics.

This study validated the sensitivity of wave height to strong wind intensity during a typhoon, highlighting the forecasting need for the wind–wave relationship in disaster prevention efforts. However, the feedback of waves on strong winds was not well simulated. Here, the comparison of coupled vs. uncoupled simulations under the same scheme is not presented. Theoretically, during a typhoon event, waves can influence the typhoon’s intensity, structure, and other characteristics through energy transfer and changes in surface roughness. However, in this study, the subtle differences between the coupled and uncoupled simulations were not significant enough to be visually compared. This may be because, under the specific setup of this case study, the influence of the variables provided by SWAN does not lead to a noticeable difference from the default ocean surface considerations already included in the WRF model. The standalone WRF model also incorporates time-varying sea surface temperatures and surface roughness. Therefore, the parameterization of the wave model needs to be more thoroughly discussed and adjusted in future work.

Moreover, it should also be noted that this study exclusively focused on Typhoon Doksuri as a case study and conducted sensitivity tests using a limited number of schemes. The PBL scheme, which involves the exchange of matter and energy between the atmosphere and the sea surface, is also of significant importance to the theme here. Future research should involve a broader range of cases and schemes for comparative analysis. This study focuses solely on the analysis of waves as an oceanic factor. However, the state of the upper ocean involves complex physical processes, such as sea level variations, mixed layer disturbances, and extreme storm surge, which have both direct and indirect relationships with typhoons [47,48]. In future work, ocean models (such as ROMS) should be considered in the coupling process to more comprehensively analyze the effects of physical parameterization in the ocean–atmosphere interaction.

5. Conclusions

The coupling of WRF-SWAN was employed for TC forecasting, while sensitivity tests were conducted on the microphysics and cumulus parameterization schemes of WRF to investigate their impact on forecast performance. The main conclusions are as follows: First, adjustments to the microphysics scheme and the cumulus scheme significantly improved the path and intensity of Doksuri, with the former playing a more prominent role relative to the latter. Second, the consideration of latent heat processes in the microphysics scheme significantly affects the intensity of the surface heat flux. Third, the wind fields show notable variability in response to varying schemes, resulting in a 2 to 3 m difference in wave height between groups at the same time.