Formation and Development of a Mountain-induced Secondary Center inside Typhoon Morakot (2009)

When a tropical cyclone (TC) passes over a mountain, significant variations in its track and intensity will occur owing to the structural modification associated with orographic impacts. This has been widely studied for TCs over the Central Mountain Range (CMR) in Taiwan (Wang, 1980; Wu and Kuo, 1999; Wu, 2001; Lin et al., 2002, 2006; Huang and Lin, 2008; Lee et al., 2008; Huang et al., 2011; Van Nguyen and Chen, 2011), over Luzon Island of the Philippines (Bender et al., 1987), and over the Sierra Madre of Mexico (Zehnder and Reeder, 1997).

The primary knowledge on the structure and intensity changes of TCs crossing the CMR was derived from analyses of track characteristics. For example, Wang (1980, 1989) found that strong and deep typhoons with maximum sustained wind of more than 50 m s^-1 and a vertical extent of more than 10.7 km tend to pass over the CMR with continuous tracks, whereas weak and shallow typhoons with maximum sustained wind of about 20-50 m s^-1 and a vertical extent of less than 6 km often produce discontinuous tracks over Taiwan. In the latter cases, the low-level center of the typhoon with a westbound track can be blocked and even filled by the CMR. Meanwhile, several vorticity centers (VCs, with closed circulation in the streamline field but lacking a low-pressure center) or secondary lows [SLs, with closed isobars in the pressure field but lacking a closed circulation, defined by (Lin et al., 1999)] form on the west side (leeside) of the CMR, one of which may develop into a well-defined secondary center (SC) with closed isobars, closed circulation and a warm core. When the SC couples vertically with the mid-level center of the primary typhoon passed over the CMR, a new typhoon is reestablished. Yeh and Elsberry (1993a, 1993b) identified that it is easier for a slow-moving typhoon crossing a mountain to produce an SC.

Figure 1.  The NCEP/GFS analyses field. 1000-hPa geopotential height (contour interval: 15 m) and horizontal wind (full barb is 10 m s^-1) at (a) 1200 UTC 7 August and (b) 1800 UTC 7 August 2009; (c) 500-hPa geopotential height and horizontal wind at 0000 UTC 6 August; (d) 850-hPa geopotential height and horizontal wind at 0000 UTC 6 August; and (e) sea level pressure (contour interval: 2 hPa) and 10-m wind at 0000 UTC 6 August 2009. "T", "SLn (n=1,2)" and "SC" denote the primary center, secondary low and secondary center, respectively.

(Lee et al., 2008) indicated that an SC's formation makes the track forecast more difficult, because the SC can modify the TC's structure or even replaces the primary center. Based on previous studies (Yeh and Elsberry, 1993a, 1993b; Lin et al., 2005, 2006; Jian et al., 2006; Lee et al., 2008), three categories of structural reorganization of TCs crossing a mountain can be summarized: (i) downward extension from the upper-level remnants of a storm passing over a mountain; (ii) upward growth of an SC produced on the leeside of a mountain; and (iii) vertical coupling between the SC and the remnant of the primary storm over a mountain.

In general, a pair of positive and negative vortices always appears on the leeside when a uniform flow is passing over a mountain, and three formation mechanisms for the vortex pair have been proposed: (i) the separation of the viscous boundary layer from the underlying surface (Hunt and Snyder, 1980); (ii) the titling of baroclinically generated horizontal vorticity produced by the downslope adiabatic warming (Smolarkiewicz and Rotunno, 1989); and (iii) the internal and boundary potential vorticity (PV) anomalies within the elongated wake or wave-breaking region on the leeside of a mountain, defined as PV generation (Smith, 1989; Schä r and Durran, 1997; Lin et al., 2006).

In the past two decades meteorologists have paid attention to the formation and development of SCs. Lin et al. (1999, 2006) found that an anticyclonic vortex cannot be produced on the leeside, since the mountain is encircled by a large cyclonic circulation of the TC. Based on an analysis of the PV budget, they moreover identified that the inner-core PV of the primary center can be transported downstream around the northern tip of the CMR, and its strength may even abruptly increase after the import of new PV generated by wave breaking over the downslope of the CMR. Consequently, a high-PV zone emerges over central western Taiwan, where SCs tend to form. (Wu, 2001) and (Lee et al., 2008) also indicated that SCs can develop from an SL, which are produced by the downslope adiabatic warming of easterly wind passing over the northern CMR. (Jian et al., 2006) investigated the evolution of Typhoon Dot (1990) and found that an SL formed over the northwestern CMR, while a VC was produced by vortex stretching over the southwestern CMR. After that, the SL drifted southwestward and merged with the VC, leading to a well-defined SC over the southwestern coast. Therefore, an SC can be induced and develop via downslope adiabatic warming (Wu, 2001; Lee et al., 2008), vortex stretching (or convergence) (Chang, 1982; Yeh and Elsberry, 1993a, 1993b), the horizontal advection initiated from the primary center (Lin et al., 1999, 2006), and the cooperation between two or among three of the abovementioned mechanisms (Jian et al., 2006). In general, as a thermodynamic process, downslope adiabatic warming results in an SL, while as dynamic processes vortex stretching and horizontal advection produce VCs. Thus, the formation of a well-defined SC needs to consider the cooperation between thermodynamic and dynamic processes. However, some questions still remain: In addition to the abovementioned mechanisms, are there other mechanisms for the formation of an SC? If the evolution of an SC is associated with the cooperation of the abovementioned mechanisms, which is the most important factor, or which dominates during different stages?

Slow-moving, moderate-intensity Typhoon Morakot (2009) passed over the CMR in August 2009 (Hall et al., 2010; Peng et al., 2011). Based on the 1000-hPa geopotential height and horizontal wind vectors from the National Centers for Environmental Prediction (NCEP) operational analyses of the 0.5^° Global Forecast System (GFS), an SL (SL1 in Fig. 1a, without closed circulation in the wind field) appeared over the northwestern CMR when Morakot approached the eastern coast of Taiwan at 1200 UTC 7 August 2009. The northwesterly wind of the typhoon circulation experienced a northward deflection by the orographic blocking when it approached the southwestern CMR (Fig. 1a). As Morakot moved inland, in addition to its low-level center being gradually filled by the CMR, an SC (SC in Fig. 1b, with a closed isobar and closed circulation) formed over southwestern Taiwan where the airstream had a northward deflection 6 h previously (Fig. 1a). Another SL (SL2 in Fig. 1b) emerged offshore over southeastern Taiwan at 1800 UTC 7 August 2009. Unfortunately, the evolution of the SC inside Morakot cannot be revealed in detail by the operational analyses of NCEP/GFS at 6-h intervals, so it is worthy of study using a high-resolution numerical model.

In section 2, the synoptic background and life cycle of Morakot are briefly reviewed. Section 3 describes the numerical model and experimental design. Section 4 examines and discusses the formation of the SC and its influence on the evolution of Morakot's structure during its crossing of the CMR. Concluding remarks are given in section 5.

Typhoon Morakot (2009) developed from a tropical depression over the western North Pacific on 2 August 2009, and then slowly moved towards Taiwan. According to the official report of the China Meteorological Administration (CMA), Morakot reached peak intensity with an estimated maximum sustained wind of 40 m s^-1 on early 7 August. Morakot made landfall at Hualian County on late 7 August, and took 48 h to cross Taiwan Island and the Taiwan Strait. Consequently, it made landfall once again on the eastern coast of mainland China on late 9 August (Fig. 2b). Morakot was embedded in a monsoon gyre (Ge et al., 2010; Wu et al., 2011), leading to strong southwesterly flow during the period from the first landfall in Taiwan to the second landfall in mainland China (Figs. 1c and d). Meanwhile, the southwesterly monsoon carried large amounts of water vapor towards Taiwan (Lee et al., 2011; Lin et al., 2011). There is evidence that the slow movement of Morakot and the transportation of abundant water vapor were important factors causing the extreme rainfall in southern Taiwan (Hong et al., 2010; Chien and Kuo, 2011; Van Nguyen and Chen, 2011).

Figure 2.  (a) Model domains for D01 (9-km grid), D02 (3-km grid) and D03 (1-km grid). (b) The best track (line with triangles) and the simulated track in D01 by CTRL (line with circles) every 6 h from 0000 UTC 6 to 1200 UTC 9 August, and by FLAT (line with squares) every 6 h from 0000 UTC 7 to 1200 UTC 9 August. (c) Observed (dashed line) and simulated intensity in D03 by CTRL (solid line) and FLAT (dotted line) every 6 h from 0600 UTC 7 to 0000 UTC 9 August.

The numerical simulations in this study were conducted and performed with the WRF model (version 3.7; Skamarock and Klemp, 2008). Figure 2a shows the three nested domains with 9-, 3- and 1-km horizontal resolutions. All domains had 35 vertical (η) levels, with the finest resolution near the boundary layer and the model top at 50 hPa. The terrain datasets and land-use information were obtained from the U. S. Geological Survey with a 30-s temporal resolution. The WRF single-moment six-class microphysics (Hong et al., 2004), rapid radiative transfer model for longwave radiation (Mlawer and Clough, 1997), Dudhia shortwave radiation scheme (Dudhia, 1989), and Yonsei University planetary boundary layer scheme (Noh et al., 2003; Hong et al., 2006) were adopted in all the domains. In addition, the modified version of the Kain-Fritsch convective parameterization scheme (Kain, 2004) was applied in the outermost domain, but no cumulus scheme was used in the inner domains.

There was independent integration in each domain because of the computational limitations: one-way interpolation was used between the domains of 9- (D01), 3- (D02) and 1-km grids (D03) (Fig. 2a). The bogus scheme was not used in each domain. Two experiments were designed as follows:

Control experiment (CTRL): (i) in the 9-km D01, the initial and boundary conditions were provided by the 0.5^° operational analyses of NCEP/GFS at 6-h intervals. Firstly, D01 ran freely from 0000 to 2000 UTC 6 August 2009, and then the horizontal wind retrieved from QuikSCAT data based on Ma and Tan (2010) were assimilated to obtain a new initial model field for the subsequent integration from 2000 UTC 6 to 1200 UTC 9 August. (ii) In the 3-km D02, the initial and boundary conditions were provided by the outputs of D01, and was run for 64 h from 2000 UTC 6 to 1200 UTC 9 August. (iii) The 1-km D03 was run for 42 h from 0600 UTC 7 to 0000 UTC 9 August, with the initial and boundary conditions from the outputs of D02.

Sensitivity experiment (FLAT): the experimental design was the same as CTRL, except that the overall terrain height on Taiwan Island was set to 1 m. The experiment adopted the methodology of (Yang et al., 2008) and (Tang et al., 2012) to deal with the vertical volume that was originally occupied by the terrain.

Note that all analyses and discussions in the following sections are based on the output of the 1-km grid (D03), except for the simulated tracks (Fig. 2b) and the calculation of steering flow from D01, and circulation centers and parcel trajectories from D02.

4.1.1. Track and intensity

The track of the simulated minimum sea level pressure (MSLP) center in CTRL is shown in Fig. 2b, as well as the best-track data issued by the CMA. In general, Morakot's track was reproduced well, especially regarding the two landfall sites——namely, on the eastern coast of Hualian County for the first landfall and on the northeastern coast of Fujian Province in China for the second landfall. A difference appeared over northwestern Taiwan in which the simulated storm presented a looping motion (Fig. 2b). This was responsible for the determination of the MSLP center, which may have been easily influenced by the downslope adiabatic warming and depression effect over the northwestern CMR, where SLs always appear. As for the typhoon intensity, the weakening process after landfall in Taiwan was also captured by the numerical simulation in the 1-km D03 (Fig. 2c).

4.1.2. Accumulative rainfall

As identified by (Wang et al., 2012), Morakot's key feature was its asymmetric structure, with an intense and long-lasting rainband over the southern Taiwan Strait, leading to the heavy rainfall in southern Taiwan. Therefore, good reproduction of this rainband is an important evaluation standard in terms of a successful numerical simulation. Figures 3a and b compare the 20-h accumulative rainfall simulated by CTRL and observations from Taiwan Central Weather Bureau (CWB) during Morakot's crossing over Taiwan from 1600 UTC 7 to 1200 UTC 8 August. In general, the model reproduced the heavy rainfall over southern Taiwan well, particularly with respect to the rainfall maxima observed at the border of Kaohsiung and Pingtung (near 23.0^°N), and three poor rainfall areas in Taipei, Yilan and Changhua (Figs. 3a and b). However, the rainfall amount was a little overpredicted, e.g., the simulated rainfall maximum of 1338 mm versus the observed maximum of 1064 mm. Note that the simulated rainfall over the southern tip was greatly overestimated, as were the simulations reported by (Van Nguyen and Chen, 2011) and (Wang et al., 2012). Although this overestimated rainfall cannot be verified, due to the few rain gauges in these remote mountainous areas, the rainfall estimations from the Tropical Rainfall Measuring Mission satellite (Wang et al., 2012, Fig. 4) and operational radar in Taiwan (Figs. 4f-h) partly support that there should have been heavy rainfall over the southern tip. In general, there were two principal rain bands contributing to the heavy rainfall over southern Taiwan: (i) a high-reflectivity rainband, as displayed by Line1 in Fig. 4a, corresponding to a southwest-northeast-oriented rainfall area during 1200 UTC 7 to 0000 UTC 8 August (Fig. 3c); and (ii) Line2 in Figs. 4b and c, associated with a quasi-east-west-oriented rainfall area near 23.0^°N during 0000-1200 UTC 8 August (Fig. 3d).

Figure 3.  (a, b) 20-h accumulative rainfall (units: mm) from 1600 UTC 7 to 1200 UTC 8 August (a) simulated by CTRL and (b) observed by Taiwan CWB; (c) 8-h rainfall from 1200 UTC 7 to 0000 UTC 8 August; and (d) 12-h rainfall from 0000 to 1200 UTC 8 August. The locations of the rain gauges at Yuyao shan (23.0^°N, 120.7^°E), Majia (22.7^°N, 120.65^°E) and Fangliao (22.4^°N, 120.6^°E) are marked in (a).

Figure 4.  (a-d) 850-hPa wind vectors (units: m s^-1) and radar reflectivity (color shading; units: dBZ) simulated by CTRL and (e-h) corresponding radar reflectivity observations at 1200 UTC 7, 1800 UTC 7, 0000 UTC 8 and 1200 UTC 8 August. "Line1" and "Line2" denote two intense rain bands.

Figure 5 shows the streamline field at 925, 700 and 500 hPa during Morakot's crossing of Taiwan. Prior to landfall, the storm centers were separated in the vertical direction. When the low-level center (below 4 km, lower than the highest altitude of the CMR) approached the central eastern coast of Taiwan, the mid-level center (above 4 km, denoted at 500 hPa herein) was located more than 100 km toward the southwestern coast (Figs. 5a and b). The dislocation between the low-level center and the mid-level center was associated with the strong southwesterly wind (Hong et al., 2010; Van Nguyen and Chen, 2011). The steering flows, defined by the area-averaged wind inside an annulus between 300 and 700 km from the simulated MSLP centers at 850, 700 and 500 hPa, showed that the low-level wind (850 and 700 hPa) changed from southeasterly to southwesterly, while a quasi-easterly and gradually weakening wind prevailed at 500 hPa during 2000 UTC 6 to 0600 UTC 7 August (Fig. 6a), which closely agrees with in the findings of Chien and Kuo (2011, see their Fig. 11). As a result, the initially overlapped storm centers had completely separated in the vertical direction by 0600 UTC 7 August (Fig. 6b). Similar to the methods used in (Jian et al., 2006) and (Van Sang et al., 2008), the circulation centers of the simulated storm in Fig. 6b were defined by the weighted-average VCs within a radius of 50 km from the streamline centers at 850 hPa and 500 hPa. Moreover, a strong northeasterly wind shear, derived from the steering flow at 500 hPa and 850 hPa (Fig. 6a), made the mid-level center of Morakot above the CMR tilt to the southwest relative to the low-level center (Fig. 5a and Fig. 6b), similar to that in previous studies (Wang and Holland, 1996; Reasor et al., 2004, Wu et al., 2011). Consequently, when the low-level center made landfall on the central eastern coast of Taiwan, the mid-level center was far away the southeastern coast (Fig. 5a). Then, the low-level center was filled over the eastern coast of Taiwan (Figs. 5a-c and 6b), whereas the low-level northerly wind of the storm flowed around the northern tip of the CMR and met the southwesterly monsoon to result in a confluent flow over the southern Taiwan Strait (denoted by CZ1 in Figs. 5a and b), corresponding to Line1 in Figs. 4a and b. The backward trajectories of air parcels released within CZ1 at 500 m above ground level (AGL), calculated by the Flexible Particle Dispersion Model with WRF (Stohl et al., 2005), demonstrate that this confluent flow was caused by Morakot's northerly wind and the southwesterly monsoon (Fig. 7a). By about 1400 UTC 7 August, a VC had emerged from the northern edge of CZ1, when the confluent flow was blocked by the southern CMR (Fig. 5b). After that, the VC moved northward along the western CMR and approached central western Taiwan (Fig. 6b), before finally coupling with the primary mid-level center of the storm passing over the CMR to reestablish a new and vertically stacked typhoon at 0000 UTC 8 August (Figs. 5c-e).

Figure 5.  Time series of simulated streamline at 925, 700 and 500 hPa in CTRL at (a) 1000 UTC 7, (b) 1400 UTC 7, (c) 1600 UTC 7, (d) 2000 UTC 7, (e) 0000 UTC 8, and (f) 0600 UTC 8 August. The black stars denote the primary centers; the solid (dashed) line indicates the vertical extent of the typhoon (secondary) center; CZ1 denotes a convergence zone (or confluent flow) between the northerly wind of the storm and the southwesterly wind; and VC indicates the location of the vorticity center.

Figure 6.  (a) Environmental steering flow (units: m s^-1) simulated in D01 by CTRL, defined by the area-averaged wind inside an annulus between 300 and 700 km from the circulation centers at three levels (850, 700 and 500 hPa) from 2000 UTC 6 to 0600 UTC 7 August. (b) Tracks of hourly circulation centers at 850 hPa (line with circles) and 500 hPa (line with squares) in CTRL from 2000 UTC 6 to 1200 UTC 8 August, with triangles denoting the tracks of the SCs at 850 hPa. The circulation centers are defined by the weighted-average vertical VC within a 50-km radius away from the streamline center, and the terrain height is indicated by shading with an interval of 500 m.

Figure 7.  The 8-h (a) backward and (b) forward trajectories (time interval: 1 h) for 10 air parcels inside D02 in CTRL at 1400 UTC 7 August. The positions of parcels were chosen at 500 m AGL within CZ1 (displayed in Fig. 5b).

4.3.1. Formation

The evolution of the simulated MSLP and 10-m wind vectors are shown in Fig. 8. Figures 8a-c show that the typhoon center was gradually filled by the land after landfall at Hualien, and two thermodynamic SLs meanwhile emerged on the leeside——namely, SL1 in the northwest, relative to the easterly wind passing over the northern CMR; and SL2 at the southeastern coast, relative to the westerly wind passing over the southern CMR (Figs. 8a-c)——consistent with the analysis of NCEP/GFS (Fig. 1a), which were caused by the downslope adiabatic warming and depression effect. In addition, a dynamic VC was produced over southwestern Taiwan, characterized by a low-pressure center and closed isobars (Figs. 8a-c). The VC was produced due to the interaction between the confluent flow and the CMR. Because the VC was initiated on the windslope relative to the westerly wind, there was an absence of a warm core inside it until 1500 UTC 7 August (Fig. 9a). Therefore, the VC was not a well-defined SC in this formation stage.

Figure 8.  Simulated MSLP (contour interval: 2 hPa) and 10-m wind vectors (units: m s^-1) in CTRL at (a) 1400 UTC 7, (b) 1500 UTC 7, (c) 1600 UTC 7, (d) 2000 UTC 7, and (e) 0000 UTC 8 August. "TC", "SLn (n=1,2,3)," and "VC" denote the primary typhoon center, secondary lows, and vorticity center, respectively. Lines SN and EW (S'N' and E'W') cross the VC at 1500 UTC 7 August (0000 UTC 8 August).

Figure 9.  Simulated horizontal wind vectors (units: m s^-1) and temperature (color shading; units: ^°C) at the lowest model level, superposed with the terrain height (contour interval: 500 m) at (a) 1500 UTC and (b) 2100 UTC 7 August. The black and green lines denote the hourly backward trajectories released at 500 m AGL inside the SC.

To highlight the formation process, the hourly horizontal wind vectors and absolute vorticity field at the lowest model level are given in Fig. 10. At 1200 UTC 7 August, a confluent flow appeared between the northerly wind of the storm and the southwesterly wind, resulting in a quasi-westerly wind near 22.8^°N (Fig. 10a). When this westerly wind was blocked by the southern CMR, a part of the airflow deflected northward to become southerly wind along the western slope (Figs. 10b and c). This southerly wind then deflected westward again when it encountered the wider and higher orographic barriers near 23.2^°N, before finally merging with the northerly wind to produce a closed circulation (Fig. 10d). The forward trajectories of air parcels released within the confluent flow confirm this closed circulation (Fig. 7b). Figure 11 shows vertical cross sections of temperature and horizontal wind crossing the SC (along lines EW and SN in Fig. 8b) at 1500 UTC 7 August. The results indicate that a strong westerly flow was located near 22.7^°N (to the south of the VC). As a result, a positive shear vorticity -∂ u/∂ y>0 would have been produced at the northern edge of the strong westerly flow (Fig. 11a). Moreover, when the westerly wind was blocked by the southern CMR to result in southerly flow along the western slope of the CMR, another component of positive shear vorticity ∂ v/∂ x>0 could be produced (Fig. 11b). Consequently, there was considerable positive vertical vorticity ∂ v/∂ x-∂ u/∂ y>0 over southwestern Taiwan, even though it looked partially asymmetric at this time. The vertical cross sections of zonal wind and meridional wind at line AA' (displayed in Fig. 10d) show that there was reversed flow (easterly wind) to the north of the VC (Fig. 12), which also reconfirmed that closed circulation appeared over southwestern Taiwan at 1500 UTC 7 August.

Figure 10.  Simulated horizontal wind vectors (full barb is 10 m s^-1) and absolute vorticity (shading; units: 10-5 s-1) at the lowest model level from 1200 UTC to 15 UTC 7 August. The blue contours denote the terrain height (contour interval: 500 m), and green line AA’ is located at 23.2^°N.

Figure 11.  Vertical cross sections of temperature (shading; units: ^°C) and (a) zonal wind (contour interval: 4 m s-1) along line SN, and (b) meridional wind along line EW (displayed in Fig. 8b) at 1500 UTC 7 August; (c) zonal wind along line S'N', and (d) meridional wind along line E'W' (displayed in Fig. 8e) at 0000 UTC 8 August. The black dots below the horizontal coordinate denote the location of the SC.

Figure 12.  Vertical cross sections of zonal wind vectors (shading; <0 m s^-1) and meridional wind (contour interval: 4 m s^-1) along line AA' (displayed in Fig. 10d) in CTRL at (a) 1300 UTC 7 and (b) 1500 UTC 7 August.

4.3.2. Development

Although the VC featured closed circulation with high vertical vorticity at 1500 UTC 7 August, the absence of a warm core and symmetric structure meant that it was still not a well-defined SC at this time. As the VC drifted northward along the western slope of the CMR, due to the steering of the southerly flow and complete departure from the confluent flow, its asymmetry gradually disappeared, ultimately becoming rather symmetric in structure over central western Taiwan at 0000 UTC 8 August (Figs. 11c and d). Similarly, the temperature inside the VC gradually increased as it moved northward during 1400 UTC 7 to 0000 UTC 8 August (Fig. 13a). Consequently, the VC has a warm core (Fig. 13b). The backward trajectories show that the warm air, produced by the adiabatic warming on the downslope of the northern CMR during the passage of the easterly wind across northern Taiwan, was continuously advected into the VC (Fig. 9b), contributing to the increase in temperature inside the VC (Fig. 13). So, by 0000 UTC 8 August, the VC had developed to become a well-defined SC with a near-symmetric, closed-circulation and warm-core structure.

Figure 13.  (a) Time-height (AGL) section of the area-averaged temperature (shading; units: ^°C) within 50 km from the circulation center of the SC (shown in Fig. 6b) and (b) the evolution of surface temperature difference between being averaged within 50 km and inside an annulus between 50 and 100 km from the circulation center of the SC in CTRL.

A sensitivity experiment with the CMR removed was performed to examine the effect of orography on the structural evolution of Morakot during its crossing of the CMR. As shown by Fig. 2b, prior to landfall, the simulated track in FLAT looked similar to that in CTRL, apart from a more northward track closely before landfall because of the absence of topographic channel impact, as proposed by Yeh and Elsberry (1993a, 1993b). After landfall, FLAT also simulated a more northward track than that observed and in CTRL. This is because the lower mountain height led to a more northward steering flow and track, in agreement with previous studies (Yang et al., 2008; Liang et al., 2011; Wang et al., 2012).

Figure 14 shows the streamline field of the simulated storm in FLAT, every 6 h from 1200 UTC 7 to 1200 UTC 8 August. Similar to the simulation in CTRL, the storm centers also inclined southward with altitude prior to landfall, and the low-level northerly wind of the storm also coupled with the southwesterly wind to result in a confluent flow over the southern Taiwan Strait, but no SC formed over southwestern Taiwan (Fig. 14a), because of the absence of closed circulation due to the orographic blocking.

Figure 14.  As in Fig. 5 but for FLAT every 6 h from 1200 UTC 7 to 12 UTC 8 August.

Based on the above analysis, the evolution of the SC inside Morakot can be divided into two stages: formation and development. As the primary center moved toward the eastern coast of Taiwan, its cyclonic wind had flowed around the northern tip of Taiwan and met the strong southwesterly wind to result in a confluent flow (CZ1 in Fig. 5b). The westerly wind within the CZ1 would deflect northward to induce a southerly wind along the western slope of the CMR, due to being blocked by the CMR. As a result, a VC with high vertical vorticity and closed circulation was produced over southwestern Taiwan (Fig. 10d and Figs. 11a and b). Because the VC initially emerged in the relatively cold area on the windslope relative to the westerly wind, it had no warm core in the initial formation stage (Fig. 9a). As the VC departed from its formation region and drifted northward along the western coast of Taiwan, similar to the findings of (Lee et al., 2008) and Lin et al. (1999, 2006), warm air produced within the wake flow of an easterly wind passing over the northern CMR was continuously transported into the VC to produce a warm core (Fig. 9b). Consequently, the VC became a well-defined SC with a warm-core and near-symmetric structure. In summary, although the SC inside Morakot was initially produced by a dynamic process, its development was closely associated with the thermodynamic process documented by previous studies.

But what caused the uniqueness of the formation of the SC inside Morakot? Comparing the wind fields of typhoons Morakot (2009), Mindulle (2004) and Dot (1990), the largest difference is that for Morakot there was a quasi-east-west-oriented, strong and long-lasting confluent flow sustained over the southern Taiwan Strait (Fig. 10; Jian et al., 2006, Fig. 5; Lee et al., 2008, Fig. 12), which was likely a very important factor in producing the VC inside. Moreover, as identified by (Chien and Kuo, 2011), the maintenance of the confluent flow was because of two unique features of Morakot: the strong southwesterly flow and the slow translation speed of the TC. Firstly, the southwesterly wind was so strong that it encountered the northerly wind of the storm to result in a confluent flow over the southern Taiwan Strait prior to landfall. Secondly, the translation of Morakot crossing Taiwan was sufficiently slow that there was abundant time for interaction between the confluent flow and the southern CMR to produce a VC.

The structural evolution of Typhoon Morakot (2009) during its passage across Taiwan was investigated in this study, including the vertical separation of the storm's centers prior to landfall in Taiwan, the formation and development of an SC, and the vertical coupling between the SC and the mid-level center of the primary storm. Because of different steering winds in the vertical levels, the centers of the primary storm exhibited vertical separation prior to landfall. Due to the orographic effect, the low-level center was filled on the east side of the CMR after landfall, while an SC was produced on the west side. Until the mid-level center passed over the CMR and approached central-western Taiwan, it coupled with the SC to result in a new vertically stacked typhoon.

This study provides new insight into the initial formation of the SC inside Morakot. When Morakot was approaching Taiwan, its cyclonic wind flowed around the northern CMR and met the southwesterly monsoon to result in a confluent flow over the southern Taiwan Strait. After the confluent flow was blocked by the CMR, a VC with high vertical vorticity and closed circulation was initially produced over southwestern Taiwan. At this formation stage, the VC exhibited an asymmetric structure and was lacking a warm core, because it emerged in a strong-wind-shear and cool area. As the VC drifted northward along the west slope of the CMR, the warm air produced on the downslope of the northern CMR was continuously transported into the VC to result in a warm core. Consequently, when the VC approached central western Taiwan, it developed to become a well-defined SC. Therefore, the SC inside Morakot was initiated from a VC, associated with dynamic interaction among the TC's cyclonic wind, southwesterly wind and orographic effects of the CMR (Fig. 15b), rather than initiated from the thermodynamic mechanism associated with the downslope adiabatic warming effect documented by previous studies (Fig. 15a). Note, however, that the subsequent development of the SC in this study was very similar to the evolution of the SL associated with the thermodynamic process proposed by (Lee et al., 2008) and Lin et al. (1999, 2006). In general, the evolution of the SC in our study does not contradict the studies of (Wu, 2001) and (Lee et al., 2008), but is just a complement, especially in the initial formation stage.

Figure 15.  Schematic diagram of the (a) thermodynamic and (b) dynamic mechanism of the SC in the initial formation stage. (a) When the TC closely approaches the eastern coast of Taiwan, the inner-core wind is strong enough to climb over the northern CMR to generate an SL over northwestern Taiwan due to the downslope adiabatic warming effect. (b) If there is a strong southwesterly wind, as the TC moves toward Taiwan, the outer flow moves around the southern tip of the CMR and meets the southwesterly wind to result in a confluent flow, from which a VC is induced over southwestern Taiwan due to the CMR's blocking.

The sensitivity experiment without the CMR confirmed the importance of the orographic blocking effect to result in the SC. Without orographic blocking, firstly, the low-level primary center could pass directly over Taiwan, instead of being blocked and filled. Secondly, although the easterly wind of the low-level primary center flowed over Taiwan, and still met the southwesterly monsoon to produce a confluent flow, there was no closed vortex produced over southwestern Taiwan. As a result, the primary storm passed over Taiwan without any structural reestablishment associated with the SC.

In addition, we also found that there should have been two main convective rain bands that contributed to the heavy rainfall over southern Taiwan during 7-9 August 2009——namely, one caused by the convergence between the northerly wind of Morakot and the southwesterly monsoon, and another closely associated with the SC. In combination with previous studies on the evolution of rainfall over southern Taiwan, the heavy precipitation near Shiao Lin mainly occurred on 8 August 2009, and looks to have been more associated with the intense convective rainband around the southern flank of the SC. Thus, further work is needed to examine the contribution of the SC to the heavy rainfall in southern Taiwan.