Yu Yang (Orcid ID: 0000-0002-5527-1540)
Chen Shu‐Hua (Orcid ID: 0000-0001-5929-1074)
This study uses a coupled atmosphere-ocean model with different numerical settings to investigatethe mean and eddy momentum transfer processes responsible for Typhoon Muifa’s (2011) early rapid intensification (RI).
Three experiments are conducted. Two use the coupled model with a horizontal resolution of either 1-km (HRL) or 3-km (LRL). The third (NoTCFB) is the same as LRL but excludes TC-induced SST cooling. HRL reasonably reproduces Muifa’s intensity during its rapid intensification and weakeningperiods.
The azimuthal mean tangential and radial momentum budgets are analyzed before the RI rates diverge between HRL and LRL. Results show that the dominant processes responsible for Muifa’s intensification are different in HRL and LRL. For HRL, the net eddy effect intensifies the storm’s circulation and contracts the eyewall during early RI, and it dominates the net mean-flow effect inside the radius of maximum wind (RMW), except near the surface and between 2 and 5 km close to the RMW. In contrast, the mean and eddy effects in LRL almost cancel inside the RMW, while the mean-flow effects dominate and intensify tangential winds outside. Without TC-induced SST cooling, Muifa in NoTCFB reaches a similarstorm intensity as in HRL but its rapid weakening rate is substantially underestimated.
The dominant mechanisms for tangential wind intensification in NoTCFB are similar to those in LRL, but their magnitudes are larger, implying a misrepresentation of the dominant momentum transferprocesses in NoTCFB during RI. For the radial momentum budget analysis, the dominant processes are similar among the three experiments except for some differences in their locations and strengths.
Tropical cyclones (TCs), commonly known as typhoons in the western North Pacific (WP) and hurricanes in the eastern North Pacific (EP) and North Atlantic (AL), are among the most destructive weather phenomena in the world. In the past three decades, TC track prediction has improved substantially, but improvements in TC intensity prediction have progressed more slowly (DeMaria et al. 2014, 2007). TC tracks are largely controlled by the large-scale environmental steering flow (Chan 2005). Continuous improvements in ensemble forecasts and data assimilation have enabled numerical weather prediction (NWP) systems to better represent the large-scale atmospheric flow steering TCs, resulting in improved TC track predictions (Franklin et al. 2003; Aberson 2001; Dong and Zhang 2016; Landsea and Cangialosi 2018). But, in addition to the influence of the large-scale environment, TCs’ intensity changes, and in particular rapid intensification (RI), involve cloud-scale processes, internal dynamics, and multi-scale interactions within TCs. These are still not well understood and potentially not well-resolved by current NWP model resolutions, lowering intensity prediction skill (Trabing and Bell 2020).
RI is defined as an increase in a TC’s maximum 10-m wind speed (Vmax) of at least 15.4 m/s (~30 knots) in 24 h (Kaplan and DeMaria 2003) or a decrease in the storm’s minimum sea-level pressure (MSLP) of at least 42 hPa in 24 h (Holliday and Thompson 1979). RI occurs at least once during most intense TCs’ lifetimes. On average, 37% of WP TCs (Wang and Zhou 2008), 42% of EP TCs (Wang and Jiang 2021), and 31% of AL TCs (Kaplan and DeMaria 2003) undergo RI. Though all category 4 and 5 hurricanes on the Saffir–Simpson hurricane scale (SSHS) (Simpson 1974) and 90% of super typhoons (category 5 typhoons on SSHS) undergo RI at least once (Wang and Zhou 2008), RI remains challenging to predict. For example, small environmental perturbations may change a TC’s structure during RI and result in large intensity changes (Zhang and Tao 2013; Judt and Chen 2016; Li et al. 2019).
Many studies have investigated the mechanisms that lead to RI, including the environmental conditions (Kaplan and DeMaria 2003; Kaplan et al. 2015) and TC multi-scale interactions (Judt and Chen 2016; Judt et al. 2016). For the large-scale environment, for example, the impacts of Sea Surface Temperature (SST) and vertical wind shear on RI have been widely studied (Chih and Wu 2020; Črnivec et al. 2016; Fudeyasu et al. 2018; Hendricks et al. 2010). SST influences the amount of energy available for TC development through latent and sensible heat fluxes (Malkus and Riehl 1960) and determines the maximum potential intensity that a TC can achieve (Emanuel 1999; Foltz et al. 2018; Zeng et al. 2007; Demaria and Kaplan 1994). Vertical wind shear influences TCs through its impact on vertical alignment of the vortex in the lower and middle levels and convective activity, as well as dry air intrusion and precipitation, which provide feedbacks to TC intensity (Nguyen and Molinari 2015; Tao and Jiang 2015; Zhang and Tao 2013; Tao and Zhang 2015). Besides the environmental influence, a TC’s internal dynamics may govern whether the storm undergoes RI. RI is usually associated with intense inner-core (eyewall) convective bursts (Steranka et al. 1986; Sanger et al. 2014) and asymmetric convective structures (Zhang et al. 2001; Montgomery et al. 2006; Van Sang et al. 2008; Fang and Zhang 2011; Persing et al. 2013; Kilroy and Smith 2016). The asymmetric eddy processes can change a TC’s intensity through the barotropic and baroclinic energy cascade (Bhalachandran et al. 2020). Agradient wind forcing in the friction layer contributes to the spin-up of RI (Montgomery et al. 2020).
As computational resources have significantly increased, state-of-the-art NWP systems have increasingly used cloud-resolving resolutions (usually higher than 5 km) to capture TC internal dynamics (Judt et al. 2016; Li et al. 2019). Some studies have shown that high-resolution cloud-permitting NWP models have improved representations of TC structures such as spiral rainbands and secondary circulations (Gopalakrishnan et al. 2011, 2012; Rogers 2010; Jin et al. 2014). However, most of these studies utilized atmospheric models alone. Increasing model resolution in a NWP system that only has an atmospheric component can better capture a storm’s intensity but may cause other problems. For example, the Taiwan Central Weather Bureau recently improved its Typhoon Weather Research and Forecast (TWRF) model by increasing its spatial resolution from 15 km to 3 km. The cloud-resolving (3 km) TWRF model can better predict RI and has lowered the intensity forecast errors for lead times of 0- 48 hours. However, the model overestimates TC intensity when the forecast lead time is greater than 48 hours, particularly for slow-moving TCs (Leroux et al. 2018; Hsiao et al. 2020). This is due in part to the absence of TC-ocean interaction in the TWRF model.
TCs often have limited impacts on SST cooling before they intensify (Dare and Mcbride 2011; Sandery et al. 2010; Lin et al. 2008, 2009). However, once a storm intensifies, TC-induced SST cooling cannot be ignored because it decreases surface latent and sensible heat fluxes, providing a negative feedback to TC intensification (Schade and Emanuel 1999; Chan et al. 2001; Sandery et al. 2010; Vincent et al. 2012). Despite progress, challenges remain on how SST feedback affects RI in a high-resolution atmosphere-ocean coupled system (Kanada et al. 2017; Li et al. 2019). While an atmosphere-ocean coupled model may reduce atmospheric models’ overestimates ofstorms’ maximum intensities (Leroux et al. 2018; Hsiao et al. 2020), it can produce weaker TCs during the earlier stage of a forecast (e.g., forecast length less than 48 hours), which reduces RI predictability. Thus an intuitive question to ask is “Is the commonly accepted 3- or 4-km cloud-resolving resolution capable of forecasting RI when the TC-induced SST cooling feedback is considered?” If not, what are the deficiencies compared to a model with even finer resolution, such as 1 km or higher?
In an idealized study, Montgomery et al. (2020) conducted a TC simulation with a 1-km resolution and applied a momentum budget analysis to investigate the contribution of mean and eddy momentum transfer during RI. One important finding in their study is that the asymmetric (eddy) wind component and unbalanced mean gradient wind flow contribute substantially to TC intensification within the radius of maximum wind (RMW). Our study is different from that ofMontgomery et al. (2020) in that it explores how mean and eddy momentum transfer during RI may change under different model resolutions and TC-induced SST feedback, using super Typhoon Muifa (2011) as a case study. More specifically, we use an atmosphere-ocean coupled model for numerical simulations, while Montgomery et al. used an atmospheric model alone. In addition, ours is a real case study with the use of a sophisticated microphysics scheme, while theirs was an idealized case study using a simple warm microphysics scheme. They also excluded radiation parameterization in their simulation. However, our momentum budget analysis mainly follows Montgomery et al. (2020).
This paper is organized as follows. Section 2 introduces the model and numerical experiments used in this study. Section 3 presents model results and the dominant processes responsible for early RI changes when applying different atmospheric resolutions and TC-ocean coupling feedback. Summary and conclusions are provided in section 4.
2 Numerical Model and Experiment Design
2.1 Super Typhoon Muifa (2011)
Typhoon Muifa (2011) was the ninth named TC and the second super typhoon in the 2011 WP typhoon season. Muifa became a tropical depression on July 27 and a named tropical storm on July 28. At 00Z July 30, Muifa intensified into a category 3 typhoon on SSHS and underwent RI during 00Z to 18Z July 30. Muifa slowly drifted northward during its RI. Based on the China Meteorological Agency (CMA) best track data (Ying et al. 2014), the storm strengthened into a category 5 super typhoon in less than 24 hours, reaching its peak intensity with a Vmax of 65 m/s and MSLP of 915 hPa at 1800 UTC July 30. Muifa went through rapid weakening (RW) after RI and became a category 4 typhoon on July 31. After RW, the storm continued to move northward with relatively steady intensity and then entered the East China Sea on August 6.
Our period of interest for this study is Muifa’s RI phase, while the RW period will be the focus of a follow-up study. Typhoon Muifa is chosen for the study because the storm underwent both RI and RW over the open ocean in the western North Pacific, making it an ideal candidate for exploring the impact of TC-ocean coupling on its intensity.
2.2 Model Description
A high-resolution atmosphere-ocean model is used to conduct numerical experiments. The coupled model used in this study is based on the Coupled Ocean-Atmosphere-Wave-Sediment Transport (COAWST) model (Warner et al. 2010). The COAWST version 3.6 model consists of several components including the non-hydrostatic Weather Research and Forecasting (WRF) model (version 4.1.2; Skamarock et al. 2019) for the atmosphere, the hydrostatic Regional Ocean Modeling System (ROMS) model (version 3.9; Shchepetkin and McWilliams 2005) for the ocean, and the Simulating Waves Nearshore model (version 41.31; Booij et al. 1999) for surface waves. All model components have been successfully used in a broad range of applications. In this study, we use only the WRF and ROMS (WRF-ROMS) components in order to include ocean-atmosphere coupled feedback in Muifa’s intensity prediction.
2.3 Numerical Setting
Figure 1 shows the domains of WRF (black boxes) and ROMS (the red box) that are used in this study. WRF uses four nested domains with horizontal resolutions of 27, 9, 3, and 1 km and domain sizes of 280×280, 541×541, 403×403, and 802×802 grid points, respectively, referred to as d01 to d04. The outer two WRF domains (d01 and d02) are fixed in location, while the inner two domains (d03 and d04) are moving nests that follow the TC center, defined by the minimum geopotential height at 850 hPa. In the vertical direction, the WRF model has 45 levels in a stretched grid, with a higher resolution near the surface. The model top is at 50 hPa. The WRF parameterizations include the unified Noah land surface model (Tewari et al. 2004), the revised fifth-generation Mesoscale Model (MM5) Monin-Obukhov surface layer scheme (Jiménez et al. 2012), the Shin-Hong scale-aware Planetary Boundary Layer (PBL) scheme (Shin and Hong 2015), the Rapid Radiative Transfer Model for General circulation models (RRTMG) for shortwave and longwave radiation (Iacono et al. 2008), and the Purdue Lin microphysics scheme for resolved cloud processes (Chen and Sun 2002). The Kain-Fritsch (KF) cumulus scheme (Kain 2004) is applied to d01 and d02 only. Both the initial and lateral boundary conditions of the WRF model are extracted from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) Final Analysis (FNL) with a horizontal resolution of 1° (~108 km) and a temporal resolution of 6 hours. As SST feedback is important for our study and SST varies significantly in space, differences in the experiments’ forecasted TC tracks can influence the results and potentially bias our conclusions. Thus, to improve TC track simulations, meteorological conditions (e.g., winds, temperature, and moisture) above the PBL in d01 and d02 are nudged to FNL data using different nudging techniques. Four-dimensional grid nudging (FDDA; Liu et al. 2008; Stauffer and Seaman 1994) is applied to d01 to ease the transition from FNL global model data to WRF model simulations. This is necessary due to the use of different physics parameterizations and numerical methods in these two models. Spectral nudging (Miguez-Macho et al. 2004) with wavelengths greater than 700 km is applied to d02. This is to ensure that the large-scale steering flow is well-simulated and produces a realistic TC track while still allowing smaller-scale features to develop in d02 and propagate into d03 through boundary conditions, aiding the development of TC intensity in d03. There is no nudging employed in d03 and d04 so that the TC can freely develop in those domains.
In this study, the ROMS model uses a single fixed domain (red box in Figure 1) with a horizontal resolution of 1/12° (~9km) covering 110° to 150°E and 5° to 45°N. The ROMS model has 40 layers in the vertical direction, with higher resolutions near the sea surface generated fromShchepetkin’s double stretching function (Shchepetkin and Mcwilliams 2009). The ROMS parameterizations are the same as those used in Yu et al. (2017, 2020), including Mellor–Yamada level-2.5 scheme (Mellor and Yamada 1982) for vertical mixing and Smagorinsky scheme (Smagorinsky 1963) for horizontal diffusion. Shortwave radiation penetration in the ocean is modeled using a double exponential irradiance absorption function (Paulson and Simpson 1977)with Jerlov water type I parameters over the open ocean (deeper than 200 m) and Jerlov water type II parameters in the marginal seas (water depth shallower than 200 m). The wind stress and latent and sensible heat fluxes are calculated using the COARE 3.0 bulk formula (Fairall et al. 2003, 1996b) with an effective wind speed correction method (Bye and Wolff 1999). Note that the wind stress accounts for a leveling off of the drag coefficient for winds greater than 30-40 m/s (Powell et al. 2003). The “cool skin” effect on surface heat fluxes is also included in the model using an empirical correction scheme (Fairall et al. 1996a). The lateral boundary conditions of the ROMS model are derived from the daily frequency 1/12° (~9 km) resolution HYbrid Coordinate Ocean Model (HYCOM) Global Ocean Forecasting System (GOFS) 3.1 reanalysis. The initial conditionsof the ROMS model are obtained from a spin-up run, which is a ROMS alone simulation, initialized with the HYCOM data on 1 July 2011 and forced with HYCOM and NCEP National Center for Atmospheric Research (NCEP/NCAR) reanalysis as its lateral boundary conditions and atmospheric forcing, respectively. Harmonic constants of ten principal tidal constituents (M2, S2, N2, K2, K1, O1, P1, Q1, Mf, and Mm) derived from the global inverse tidal model of TPXO7.0 (Egbert et al. 1995; Egbert and Erofeeva 2002) are specified at the lateral boundaries of the ROMS model as the tidal forcing.
WRF and ROMS communicate with each other every 15 minutes using Model Coupling Toolkit (MCT) (Larson et al. 2005; Jacob et al. 2005; Warner et al. 2008). WRF provides sea level pressure, 10-m wind, 2-m temperature, 2-m relative humidity, surface longwave and shortwave radiation fluxes, and surface evaporation and precipitation rates to ROMS, which in turn provides SST to WRF. The interpolation weights between the different WRF and ROMS grids are generated by the Spherical Coordinate Remapping Interpolation Package (SCRIP; Jones 1999).
2.4 Experimental Design
To examine the impact of model resolution on Muifa’s RI prediction, two numerical experiments with different atmospheric resolutions are conducted (Table 1). The first experiment, our control run with a higher model resolution (HRL), uses the WRF-ROMS coupled model with the numerical setting described in Section 2.2. This run includes four WRF domains (d01-04), with the highest resolution of 1 km in d04, and one ROMS domain (Figure 1). The second experiment (LRL) is the same as the HRL run, except that only three domains (d01-03) are used in the WRF model, with the highest resolution of 3 km in d03. In addition to these two WRF-ROMS simulations, we also carry out a numerical experiment, called NoTCFB (Table 1), to explore the importance of TC-induced SST cooling for Muifa’s RI and RW prediction. NoTCFB uses the WRF model only. The WRF setting is the same as that in LRL, and SST is updated every 15 minutes using prescribed values. The prescribed SSTs in NoTCFB are produced by another WRF-ROMS coupled simulation, which uses the same numerical setting as in LRL except Muifa’s vortex in the initial conditions was removed by applying a TC vortex removal scheme (Davis and Low-Nam 2001). Thus there is no TC-induced SST cooling in the lower boundary condition. In all experiments, the model (either WRF-ROMS or WRF alone) is integrated for 9 days from 00Z July 28 to 00Z August 6, 2011, and WRF data is saved every 15 min for analysis.
3 Model Results and Discussion
3.1 Model Results against observations
a. TC Intensity prediction
As mentioned earlier, SST differences caused by different TC track forecasts can have a great impact on TC development. Therefore, FDDA and spectral nudging were applied to WRF d01 and d02, respectively, to ensure reasonable and comparable track simulations in the experiments. As shown in Figure 2a, all experiments closely reproduce Muifa’s observed track, except during the early simulations when the storm’s center is difficult to identify due to its weak intensity. These well-simulated Muifa tracks are obtained because the use of nudging ensures that the model’s flow field effectively resembles the FNL large-scale flow, which constrains Muifa’s movement. These reasonable TC track simulations allow us to more precisely investigate the impacts of model resolution and TC-induced SST cooling on Muifa’s RI and RW forecasts.
Figures 2b and 2c show intensity predictions in terms of Vmax and MSLP. The simulated RI in all experiments is delayed about 12 hours compared to the CMA best track data (i.e., from 0000 UTC to 1800 UTC July 30 in observations and 1200 UTC July 30 to 1200 UTC July 31 in the model). The delay in the predicted TC’s intensification is not unusual. In addition to model errors (e.g., physics parameterizations), one of the reasons is related to issues with the TC’s initial conditions (Chang et al. 2020). An example is the incorrect position of the initial storm, which can be seen in Figure 2a, where the FNL TC location is a few hundred kilometers away from the CMA best track location. Another possible reason is the model spin-up problem, since cloud-scale features are not present in the initial conditions. Despite the delayed start of RI in the model, the HRL experiment is able to reproduce Muifa’s observed maximum intensity after completion of RI, though overall the model has a slightly weaker intensification rate. The HRL storm reaches a minimum MSLP of 920 hPa (Figure 2b) and highest Vmax of 61 m/s (Figure 2c) at 1200 July 31, compared to the observed 915 hPa and 65 m/s. With a lower resolution in LRL, the storm’s maximum intensity is 935 hPa in MSLP (blue line in Figure 2b) and 50 m/s in Vmax (blue line in Figure 2c). LRL underestimates Muifa’s maximum intensity by 15 hPa and 11 m/s, suggesting that in the coupled WRF-ROMS, the commonly accepted cloud-resolving resolution of 3 km cannot reproduce the observed RI and the maximum storm intensity. However, when the TC-ocean coupled feedback is turned off (i.e., NoTCFB), the WRF model with a 3-km resolution is capable of simulating Muifa’s RI, though it greatly underestimates the RW rate after RI (magenta lines in Figures 2b and 2c). This result is consistent with that found in the TWRF forecast (discussed in the introduction), which also uses the WRF model alone with a 3-km resolution. While a high spatial resolution (e.g., 1 km) is required for the WRF model to capture convective-scale mechanisms that are important for the storm’s RI, the TC-induced SST cooling feedback is critical for the TC’s weakening after RI, emphasizing the importance of TC-ocean coupling in TC intensity prediction.
b. Storm morphology during RI
We further evaluate the model forecast by comparing the simulated TC’s cloud top temperature (CTT) with observations from Multifunctional Transport Satellites series 2 (MTSAT-2; also known as Himawari 7). MTSAT-2 is a geostationary satellite that was launched in February 2006 and was operated by the Japan Meteorological Agency (JMA) until it was decommissioned in July 2015. MTSAT-2 provided imagery in five wavelength bands including one visible channel with a resolution of 1 km and four infrared channels with a resolution of 4 km. In this study the MTSAT-2 brightness temperature (BT) in the 10.3-11.3 μm infrared channel (IR1) is used.
The 6-hourly snapshots of observed IR1 BT by MTSAT-2 and simulated CTT in both HRL and LRL during 00Z July 30 to 00Z July 31 are shown in Figure 3. Compared to the MTSAT-2 observations (Figures 3a, d, g, j, and m), HRL (Figures 3b, e, h, k, and n) reasonably reproduces Muifa’s morphology during RI, including the eyewall (12-h shift in time due to the delay of model RI) and rain bands. Although the simulated TC CTTs in HRL and LRL are similar, there are important differences. For example, at 1200 UTC July 30, deep convective storms are embedded within an organized eyewall and the surrounding region. This is more accurately simulated in HRLthan in LRL (i.e., the cyan color region near the storm center), when compared to observations at 0000 UTC July 30 (12-h shift). HRL better reproduces Muifa’s eyewall changes during RI, especially its closed cloud structure at 1800 UTC July 30 (Figure 3k versus 3l), when the storm intensities between these two experiments start diverging. At 0000 UTC July 31, although eyewall features look similar between HRL and LRL, Muifa’s eye is more concentrated in HRL, which is closer to observations.
3.2 Wind and Kinetic Energy Analysis
Wind circulation is an important indicator of TC development. In this subsection, the azimuthally-averaged wind circulation with respect to the storm center axis is examined. The simulated TC center locations are estimated using the pressure centroid method described in Nguyen et al. (2014). However, instead of using the pressure field at a 2-km height in Nguyen et al. (2014), we use the sea level pressure (SLP) with a radius of 100 km and the first-guess location at the minimum SLP to estimate the TC’s centroid location at each model output time (every 15 minutes). SLP is chosen since it is a direct model output field.
All azimuthally-averaged data are from domain 3 (3-km resolution) and time-averaged data are from 15-min WRF output. Figure 4 presents the radius-height cross sections of the storm’s mean tangential, radial, and vertical winds from HRL and LRL. The data are averaged from 0600 UTC to 1800 UTC July 30 during the time before the storm intensity diverges between the simulations(see Figures 2b and 2c). Overall, the mean wind patterns from HRL and LRL are similar, as expected. Both experiments have strong inward radial winds, with maxima greater than 10 m/s within the boundary layer, and strong outward radial winds with maxima greater than 15 m/s at a height of about 16 km (Figures 4a and 4b). The maximum tangential wind greater than 35 m/s appears at the top of the boundary layer and is about 30-40 km away from the storm’s center (Figures 4d and 4e). The simulated eyewalls in HRL and LRL, identified by the strongest vertical motion, are located within a 20-50 km radius of the storm’s center axis and slope outward with height, with the maximum time-averaged vertical velocity of ~1-1.5 m/s between 10-15 km height (Figures 4g and 4h). While their patterns are similar, there are differences between the two experiments. HRL produces stronger wind circulation, consistent with a more intense storm, as shown in Figures 2a and 2b. The dipole patterns in wind difference between these two experiments (Figures 4c, 4f, and 4i) are mainly due to stronger winds within (outside) the eyewall in HRL (LRL), particularly the tangential component. Upward (~1 km) and inward (~5-10 km) shifts of the convective core (and less sloped) can also be seen in HRL. Compared to LRL, the maximum mean vertical velocity in the eyewall region in HRL is about 0.4 m/s stronger (Figure 4g versus 4h). The difference in mean tangential wind induces a stronger vertical component of relative vorticity inside the RMW for HRL but outside for LRL (figure not shown). The discussion above indicates that the application of the more expensive 1-km resolution in HRL makes the simulated eyewall circulation and convection not only stronger but also more concentrated (i.e., a smaller RMW) compared to LRL, consistent with other studies (Montgomery et al. 2020).
We further calculate the azimuthally-averaged symmetric (mean; Km) and asymmetric (eddy; Ke) kinetic energy for HRL and LRL. Their differences are presented in Hovmöller diagrams. Consistent with the dipole-pattern of wind differences between the two experiments, before 1800 UTC 30 July HRL has 10 J/m2 higher Km within a 50 km radius and 5 J/m2 lower Km outside (Figure 5a). For the eddy kinetic energy, HRL also has a higher Ke (10 J/m2 more) in Muifa’s inner core compared to LRL, but their difference is almost zero outside (Figure 5b). Moreover, the positive Ke difference is located within the radius of maximum positive Km differences (black dashed lines in Figure 5). The time-series results show that the difference in eddy kinetic energybetween HRL and LRL in the inner core region during the earlier RI period might play an important role in the RI rate separation between the two experiments at 1800 UTC July 30. This possibility is examined next.
4 Summary and Conclusions
While TCintensityforecasts have improved in the past two decades, challenges remain, especiallyfor RI. Even when TC RI is reasonably forecasted, are the dominant processes responsible for intensification correctly captured by numerical models? Without reasonably representing these processes, it is difficult to accurately predict the intensity of a storm that experiences both rapid intensification and weakening during its lifecycle.
This study aims to determine the dominant mean and eddy momentum transfer processes that are responsible for RI and investigate how those processes may change with model spatial resolution and TC-induced SST cooling feedback. Super typhoon Muifa, which occurred over the western North Pacific from late July to early August in 2011, is used as a case study. A fully coupledatmosphere-ocean WRF-ROMS model is used for simulations. Two numerical experiments are conducted with different horizontal grid spacing in WRF: 1 km (HRL) versus 3 km (LRL). A third experiment is the same as LRL but without TC-induced SST cooling feedback (NoTCFB).
The HRL experiment with a 1-km resolution can better resolve convection, allowing it to reasonably reproduce Muifa’s rapid intensification and weakening and its morphology during rapid intensification. When the resolution is reduced to 3 km, LRL still captures Muifa’s lifecycle but underestimates its maximum intensity by 15 hPa in MSLP and 11 m/s in the maximum 10-m wind, meaning weaker rapid intensification and weakening rates. After further removing the TC-induced SST cooling effect, NoTCFB (with a 3 km resolution) is capable of reproducing Muifa’s rapid intensification, but the rapid weakening becomes too weak compared to HRL, suggesting inappropriate dynamics caused by wrongly simulated surface energy fluxes.
The azimuthally-averaged wind circulation from 0600 UTC to 1800 UTC July 30 during early RI is compared between HRL and LRL. Their wind patterns are similar, including the maximum cyclonic circulation at the top of the boundary layer, strong inward radial winds within the boundary layer, and strong outward radial winds at upper levels. The dipole patterns in their wind differences result from the stronger intensity in the inner core (eyewall) region in HRL but greaterintensity outside the eyewall in LRL, as well as the upward and inward shifts of the eyewall in HRL.
The differences in azimuthally-averaged Km and Ke (symmetric and asymmetric kinetic energy, respectively) between HRL and LRL show that before 1800 UTC 30 July, HRL has 10 J/m2 higher Km within a 50 km radius and 5 J/m2 lower Km outside. For the eddy kinetic energy, HRL has 10 J/m2 higher Ke in Muifa’s inner core, which is located within the radius of the maximum positive Km differences.
Azimuthally averaged tangential and radial momentum budget analyses are performed to determine the contributions of mean and eddy momentum transfer to Muifa’s early RI before the TC intensity diverges between HRL and LRL. For the mean tangential wind, the maximum local rates of change (i.e., Vt in (1)) among the three experiments are distinct. HRL has the maximum positive tendency located inside the RMW, while LRL and NoTCFB have the maximum tendencies outside the RMW. Additionally, the maximum positive tendency has a larger area and greater magnitude in NoTCFB than in LRL. For HRL, the net eddy effect dominates the net mean flow effect in driving the acceleration of the mean circulation inside the RMW, and this is favorable for TC intensification and eyewall contraction. The net eddy effect is mainly driven by vertical advection of eddy tangential momentum at middle to upper levels and eddy radial vorticity flux at low levels. In contrast, over the same region inside the RMW, the mean and eddy effects almost cancel each other in both LRL and NoTCFB, except near the surface. Outside the RMW, the net mean flow effect dominates the net eddy effect for all three experiments, in which HRLhas the weakest magnitude and NoTCFB has the strongest. The primary contributor to the net mean flow effect outside the RMW is the vertical advection of mean tangential momentum. For the mean radial momentum analysis, although the patterns of local rates of change show some discrepancies among the three experiments, their dominant processes in general agree well with each other and are mainly driven by the combined effect of the mean gradient wind imbalance andthe horizontal and vertical mean flow advection.