The genesis of a tropical cyclone (TC) in the western North Pacific (WNP) occurs under influences of a wide range of spatio-temporal variabilities. The summertime monsoon trough provides favorable background conditions for a TC precursor to develop through multi-scale interactions (Holland 1995). The Madden–Julian Oscillation (MJO; Madden and Julian 1971) or boreal summer intraseasonal variability (BSISO) (e.g., Wang and Rui 1990) has positive influences on TC genesis in the convectively active period (Liebmann et al. 1994; Maloney and Hartmann 2001; Schreck and Molinari 2011; Hsu et al. 2011; Yoshida et al. 2014). Easterly waves, mixed Rossby-gravity (MRG) waves, and tropical depression (TD)-type disturbances often serve as TC precursors (Yanai 1961; Frank and Roundy 2006; Ritchie and Holland 1999; Schreck et al. 2011; Dickinson and Molinari 2002; Zhou and Wang 2007). The TD-type disturbance is dynamically close to the synoptic-scale wave train (SWT) that often appears before and after TC genesis, with a zonal wavelength of 2500–3500 km, northeast–southwest oriented phase tilt, and northwestward phase propagation. Middle- to upper-tropospheric disturbances sometimes play an important role in TC genesis through the enhancement of upper-level divergence or intensification of middle tropospheric vorticity (Sadler 1976; Briegel and Frank 1997; Heta 1990; Ritchie and Holland 1997; Tam and Li 2006; Zhou and Wang 2007). On the other hand, the critical role of diabatic processes in TC genesis has been broadly recognized (e.g., review by Houze 2010), which often appears as “mesoscale convective systems (MCSs)” (Bister and Emanuel 1997; Ritchie and Holland 1999) or “hot towers” (Simpson et al. 1998; Hendricks et al. 2004). Based on numerical experiments with resolving mesoscale convection (MC), Yamasaki (1983) demonstrated the importance of successive formation of MC and its interaction with storm-scale fields, especially in the early formation stage of TCs.

Specific aspects of the above-mentioned factors have been well documented by previous studies (e.g., Fu et al. 2007; Kerns and Chen 2013; Maloney and Hartmann 2001; Nolan 2007; Ritchie and Holland 1999; Yoshida and Ishikawa 2013; Xu et al. 2013, 2014; Wu and Duan 2015). In the individual cases of TC genesis, however, multiple factors simultaneously have an influence, and may interact among themselves. To further deepen our understanding of this issue, it is useful to examine the combination of multiple factors by detailed case studies.

A global cloud-permitting simulation is one of useful methods for such an investigation. This approach has the merit of representing a broad range of scale interactions from mesoscale cloud systems to planetary scale variabilities in a seamless framework (Kodama et al. 2015; Nakano et al. 2015; Yamada et al. 2016). It requires fewer assumptions to represent moist convection than conventional general circulation models by explicit treatment and is also free from lateral boundary problems in regional models (Warner et al. 1997), although the model does not necessarily outperform convectional general circulation models or regional nonhydrostatic models in all aspects due to its own biases (Kodama et al. 2015).

In this study, we focused on Typhoon Fengshen (2008), which was formed during the onset of the WNP monsoon (Geng et al. 2011) and in an active phase of BSISO (Katsumata et al. 2013). Based on an examination of the European Center for Medium-Range Weather Forecasts (ECMWF) year of tropical convection (YOTC) operational analysis (Waliser et al. 2012; hereafter ECMWF YOTC), Xu et al. (2013) categorized the synoptic-scale precursor of Fengshen as the “SWT” type, whereas Yoshida and Ishikawa’s (2013) index, based on the 850-hPa flow pattern, classified Fengshen as the “monsoon confluence” type.

Figure 1 shows the atmospheric conditions during the genesis of Fengshen in the ECMWF YOTC and the National Centers for Environmental Prediction (NCEP) final analysis (hereafter NCEP FNL). Both sets of data represent a low-level westerly over the equatorial western Pacific and a large-scale cyclonic gyre associated with the BSISO event (Fig. 1a, b), and moisture was accumulated within the gyre (Fig. 1c, d). Interestingly, differences are found between the analyses in moisture and flow fields that might affect convective behavior. Under these large-scale conditions, Yamada et al. (2012), who analyzed the in situ observation and cloud-permitting simulation data, revealed the critical role of the vertical superposition of mesoscale convective vortices for this case. Thus, Fengshen is a suitable target for an investigation of the multiple factors at different scales and the convective effects on the TC. In this study, we investigated these issues using a coordinated set of simulations.

Fig. 1
figure 1

a, b Zonal velocity (shading), relative vorticity (red contour; 1, 3, and 5 × 10−5 s−1), and wind vectors at 850 hPa, c, d column-integrated water vapor (shading) and 200-hPa wind vectors, at 0000 UTC on 15 June 2008, e, f height–longitude section of relative vorticity at 0000 UTC on 16 (blue), 17 (black), 18 (green), and 19 (purple) June 2008, along the latitude across the center of the incipient disturbance, in the ECMWF YOTC (a, c, e) and the NCEP FNL (b, d, f). Contour lines of 850-hPa relative vorticity averaged between 6 hours before and after 0000 UTC on 15 (red), 16 (blue), 17 (black), and 18 (green) June 2008 are also depicted in c, d (1, 5, and 10 × 10−5 s−1). Crosses in a, b, e, f indicate the location of the center of the incipient disturbance at 900 hPa, and triangles in e and f indicate the location of the center of the maximum vorticity in the free atmosphere


Numerical model

Nonhydrostatic Icosahedral Atmospheric Model (NICAM; Satoh et al. 2008, 2014) was used for the simulation in this study. The horizontal mesh size was 3.5 km (“cloud-permitting”) and 40 vertically uneven levels were taken. Moist processes were explicitly calculated with the NICAM Single-moment Water 6 scheme (NSW6; Tomita 2008), and no cumulus parameterization was used. The atmospheric turbulence and radiation processes were calculated using a modified version of the Mellor–Yamada–Nakanishi–Niino level 2 scheme (Mellor and Yamada 1982; Nakanishi and Niino 2006; Noda et al. 2010) and Model Simulation radiation TRaNsfer code (MSTRN) X (Sekiguchi and Nakajima 2008), respectively. The land surface process was represented by a bucket model. Sea surface temperature (SST) was predicted using a slab ocean model, and was nudged to the weekly mean observations from National Oceanic and Atmospheric Administration (NOAA) Optimum Interpolation Sea Surface Temperature (Reynolds et al. 2002) with a relaxation time of 5 days.

Experimental setup

In this study, four simulations were performed with initial data interpolated from the 0.5° gridded ECMWF YOTC, with varying cloud microphysical settings (Table 1) to determine their impacts on the evolution of the large-scale variability and the initial disturbance. In “E_mp1,” sedimentation of cloud ice was included, as was the case in the 3.5-km mesh simulation of Fengshen (Hashino et al. 2013, 2016; Yamada et al. 2016). In “E_mp2,” the default setting was used (the same as Nasuno 2013). In “E_mp3,” the upper-tropospheric clouds were tuned for cloud-radiative balance in the 14-km mesh multi-decadal simulations (Kodama et al. 2015). In “E_mp4,” the terminal velocity of precipitating condensates was reduced, which was tuned for 14-km mesh ensemble simulations of MJO events (similar to Miura et al. 2015). The impact of atmospheric conditions (Fig. 1; Additional file 1: Figure S1) was examined by the “N_mp2” run, where the 1.0° gridded NCEP FNL was used to create the initial data, with other settings the same as in “E_mp2." The thermal conditions were more stabilized in the NCEP FNL than in the ECMWF YOTC, with more humidity in the free atmosphere (Additional file 1: Figure S1a, b). In all runs, the land surface and the slab ocean model were initialized using the NCEP FNL. The simulation period was 6 days starting from 0000 UTC 15 June 2008, approximately 4 days prior to the declaration of Tropical Storm Fengshen by the Joint Typhoon Warning Center (JTWC).

Table 1 Parameter settings of cloud microphysics

Analysis method

The 1.0° gridded (3-hourly) data were created for comparison with the objective analyses. The center of the simulated incipient vortex was identified by the following procedure: (1) the 1.0° gridded horizontal winds were smoothed twice with a 9-point weighted-mean filterFootnote 1 to calculate relative vorticity, and the resulting relative vorticity was smoothed in the same way; (2) the location of the maximum relative vorticity was detected as the first guess; (3) horizontal winds within a 5.0° square domain surrounding the first guess were smoothed by operating the 9-point weighted-mean filter 300 times to calculate relative vorticity; and (4) the location of the maximum relative vorticity at the 1-km altitude was defined as the center of the disturbance.Footnote 2 The center of the incipient vortex in the analyses was defined by the maximum location of the 900-hPa relative vorticity. For evaluation of the simulated track and central SLP, the JTWC best track data were used. The 0.25° gridded Tropical Rainfall Measuring Mission (TRMM) 3B42v7 (Huffman et al. 2007) at 3-h intervals was used to quantify the observed convective activity.

Results and discussion

Large-scale fields

Figure 1 presents the large-scale fields in the analyses. On 15 June 2008, the incipient disturbance of Fengshen was located at the eastern edge of the large-scale (~2000 km) gyre to the north of the equator at 850 hPa, where zonal convergence was pronounced (Fig. 1a, b). This is one of the typical types of favorable environment for tropical cyclogenesis (Yoshida and Ishikawa 2013). The precursor of Fengshen (outlined by thick contour lines) moved northwestward with increasing vorticity in the east to northeast of the large-scale gyre (outlined by thin contour lines in Fig. 1c, d). In the vertical, the vorticity maximum was rather limited in the lower troposphere on 16 June, and the distance between the lower and middle tropospheric maxima was 1~2° until 19 June (Fig. 1e, f), suggesting that they had different origins. The vertical penetration was slower in the NCEP FNL than in the ECMWF YOTC. In the upper troposphere, a westerly was dominant in the northern hemisphere, indicating a pronounced vertical shear of horizontal wind (Fig. 1c, d).

Track and intensity of Fengshen

In the macroscopic evolution described above, Fengshen was first detected as tropical depression by JTWC at 0600 UTC on 17 June 2008. The observed and simulated track and central SLP are plotted in Fig. 2. In the best track, the central SLP started decreasing at around 0000 UTC on 18 June (Fig. 2b), and recorded its minimum pressure at 0000 UTC on 21 June. A tendency for a northward track bias was found in the simulations (Fig. 2a),Footnote 3 with earlier onset of pressure reduction in E_mp2 and E_mp3, and an eventual overdevelopment in E_mp2 (Fig. 2b). In contrast, N_mp2 showed scarce deepening of the initial disturbance. Delayed evolution was found in E_mp4, where the latent heating rate was systematically reduced, especially in the lower troposphere (Table 1; Additional file 1: Figure S1c).

Fig. 2
figure 2

a Track of Fengshen and b time series of central sea level pressure (SLP) in the best track (black) and simulations; E_mp1 (red), E_mp2 (blue), E_mp3 (yellow), E_mp4 (green), and N_mp2 (purple). Tracks before 0600 UTC on 17 June and after 0000 UTC on 21 June 2008 are plotted by thin lines and crosses

Simulated large-scale fields

In the following section, we discuss the difference in the simulation results in terms of the origin of the middle tropospheric vorticity disturbance. Figure 3 shows the 400-hPa vorticity fields at 1-day intervals in the analyses (Fig. 3a–d) and simulations (Fig. 3e–g). The lower tropospheric wind fields in the same period are presented in Additional file 2: Figure S2. To extract large-scale disturbances, the 1.0° gridded horizontal winds were smoothed ten times with a 9-point weighted-mean filter to calculate relative vorticity, and the resulting vorticity was smoothed in the same way. On 15 June 2008 (initial date of simulation), positive vorticity anomalies (troughs) developed around 130°E and 150°E in the subtropical middle to upper troposphere (Fig. 3a, b), and were extending southeastward and southwestward, respectively. They overlapped the lower tropospheric gyre. In the subsequent three days, the middle tropospheric trough intensified and migrated northwestward with a vorticity maximum at 130°–140°E (Fig. 3c, d). The center of the low-level incipient vortex (crosses in Fig. 3) was located on the northeastern rim of the vorticity maximum during the migration. Prior to the intensification of the middle tropospheric vorticity, convective activity was enhanced in the broad domain corresponding to the lower tropospheric large-scale gyre (shading in Fig. 3; Additional file 2: Figure S2). The growth of middle tropospheric trough was slower in NCEP FNL than in the ECMWF YOTC (Fig. 3c, d).

Fig. 3
figure 3

Horizontally smoothed 400-hPa relative vorticity (blue), 850-hPa wind vectors, and relative vorticity (red) at 0000 UTC on 15 June 2008, in the a ECMWF YOTC and b NCEP FNL, with the precipitation rate averaged between 1200 UTC on 13 June and 0000 UTC on 15 June 2008, in TRMM 3B42v7 (shading). Horizontally smoothed 400-hPa relative vorticity at 0000 UTC on 16 (blue), 17 (black), and 18 (green) June 2008, in the c ECMWF YOTC and d NCEP FNL with the precipitation rate averaged between 1200 UTC on 15 June and 0000 UTC on 17 June 2008 in TRMM 3B42v7 (shading). Same as c, d, except for the simulation case e E_mp2, f N_mp2, and g E_mp4 at the 7.6-km altitude (approximately corresponding to 400 hPa), with the vertically integrated latent heating rate (shading). In all panels, the unit of relative vorticity is 10−5 s−1 and only positive values are shown. Crosses indicate the center of the incipient disturbance on 15 (red), 16 (blue), 17 (black), 18 (green), 19 (purple), and 20 (orange) June 2008. Black crosses in a and b indicate the center of the large-scale gyre

The evolutions described above were reasonably simulated in E_mp2 (Fig. 3e; Additional file 2: Figure S2c), E_mp1, and E_mp3 (not shown). In N_mp2, the results were substantially different, with weak development of the middle tropospheric trough, which was less coherent to the latent heat release and lower tropospheric gyre (Fig. 3f; Additional file 2: Figure S2d). In E_mp4, a weak trough was formed after 16 June, with smaller latent heat release over the lower tropospheric gyre than in E_mp2 (Fig. 3g; Additional file 2: Figure S2e). These results indicate that the large-scale latent release associated with the lower tropospheric conditions (e.g., the active period of BSISO) was important to the growth of the middle tropospheric trough and subsequent development of TC within the vertically coherent large-scale positive vorticity anomaly. In the lower troposphere, a wave train-like structure was confirmed in the analyses and the TC developing simulation, with northeast-southwest phase tilt in parallel to the middle tropospheric trough (Additional file 2: Figure S2a–c). The structure was clearer in E_mp2 than in the analyses, while it was less evident in E_mp4 and N_mp2.

In order to interpret the differences in evolution among the simulations (Fig. 4e–g), the static stability \(N^{ 2} = g/\theta_{0} \partial \theta / \partial z\) and intertial stability \(I^{2} = \left( {f + \partial rv_{\text{T}} /r \partial r} \right) \left( {f + 2 v_{\text{T}} /r} \right)\) were examined, where θ is potential temperature, θ 0 (=300 K) is the reference potential temperature, r is radius, g (= 9.8 m s−2) is the gravitational acceleration, f is the Coriolis parameter, and v T is the tangential wind. According to the Eliassen’s (1952) balanced model, N 2 and I 2 have been used to diagnose the vertical and horizontal extents of the thermally forced transverse circulation, respectively (Schubert and Hack 1982; Shapiro and Willoughby 1982). Figure 4 shows the N 2 , I 2, and \(\varGamma = \left( {{{N^{2} } \mathord{\left/ {\vphantom {{N^{2} } {I^{2} }}} \right. \kern-0pt} {I^{2} }}} \right)^{{{1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-0pt} 2}}}\), averaged within the 600-km radius of the large-scale vortex in the simulations.Footnote 4

Fig. 4
figure 4

Vertical profiles of a, d) static stability, b, e inertial stability, c, f Γ, and g latent heating rate averaged within the 600-km radius of the large-scale lower tropospheric vortex ac at 0003 UTC on 15 June, df at 0000 UTC on 17 June, and g the average between 1800 UTC 15 June and 0000 UTC 17 June 2008, in E_mp2 (blue), E_mp4 (green), and N_mp2 (purple)

On 15 June, I 2 was small above the 6-km altitude, which led to a large Γ in both analyses (Fig. 4a–c). On 17 June, convective activity induced thermal stabilization and enhancement of the vortex (Fig. 4d, e, g). In N_mp2 (E_mp2 and E_mp4), the increase in I 2 was limited below (extended above) the 6-km altitude, reflecting the different strengths of diabatic forcing. As a result, Γ in the middle troposphere was smaller in E_mp2 and E_mp4 than in N_mp2 (Fig. 4f), and the Rossby depth (=1/(Γk); k is the radial wave number), which represents the vertical size of the forced circulation (Schubert and McNoldy 2010), in E_mp2 was approximately 1.6 times that in N_mp2. This elucidates the subsequent coupling of the large-scale lower and middle tropospheric vortices in E_mp2, which was substantially weaker in N_mp2.

Simulated evolution of the TC inner core

Given the spin-up of the large-scale circulation in both the lower and the middle troposphere loosely coupled by enhanced convection, what determines the onset of the pressure deepening and subsequent intensification? Figure 5 presents the time evolution of the inner core of the simulated storm. Because of the ambiguous structure in N_mp2, only the averages within the 50- and 100-km radius are presented for this case (Fig. 5e).

Fig. 5
figure 5

(Upper) Height–time sections of the latent heating rate (shading; 10−3 K s−1) and tangential velocity (contour; m s−1) averaged within the 50-km and 100-km radius, respectively, (lower) time series of the vertically integrated latent heating rate averaged within the 22-km (red) and 50-km (green) radius, the 7.6-km altitude tangential wind averaged within the 100-km radius (purple), and distance between the center of the vortex at the 1- and 7.6-km altitude (black) in the simulation cases a E_mp1, b E_mp2, c E_mp3, d E_mp4, and e N_mp2. f Time series of the vertically integrated latent heating rate within the 100-km radius for cases E_mp1 (red), E_mp2 (blue), E_mp3 (yellow), E_mp4 (green), and N_mp2 (purple)

Deep latent heating associated with convective events within the 50-km radius occurred at 12 h or at shorter intervals during 16–17 June in the TC developing cases (Fig. 5a–c). An increase in the azimuthal mean tangential wind in the middle troposphere occurred simultaneously or several hours after the deep convective events (upper panels). An increase in the tangential wind was less clear in E_mp4 and N_mp2 in this period (Fig. 5d, e). This difference was attributable to the weakness of the large-scale vorticity that preconditioned the intensification of the TC inner core in these runs. In the TC developing runs, an abrupt increase in the column-integrated latent heat release occurred over a 1~1.5-day period around 18 June (Fig. 5a–c, f). As a measure of the vertical tilt of the inner core, the distance between the locations of the vorticity maximum at the 1- and 7.6-km altitudes (approximately corresponding to 900 and 400 hPa, respectively) was calculated (black lines in the lower panels). Concurrent with the outbreak of the latent heat release, the distance sharply decreased, and the acceleration rate of the middle tropospheric tangential wind (purple lines in the lower panels) abruptly rose. This indicates the establishment of an upright deep TC inner core within the southwestward slanted large-scale vorticity fields (Fig. 3). In E_mp2 and E_mp3, the convective enhancement was more vigorous and was rooted below the 2-km altitude, where it was sustained for ~10 h (Fig. 5b, c, f) around 0000 UTC on 18 June. These elucidate the earlier onset of pressure deepening in these runs (Fig. 2b). In E_mp1 and E_mp2, a prominent increase in latent heat release was confined very close to the storm center (r < 22 km) during the rapid change (late on 18 June; Fig. 5a, b). In these runs, a very strong single convective event made a substantial contribution to the formation of the TC inner core. Figure 6 presents zoomed plots of the relative vorticity before (Fig. 6a, c) and during the explosive latent heat release (Fig. 6b, d) in E_mp2. Before the event, the low-level incipient vortex consisted of small vorticity peaks, with relatively short lifetimes. The flow fields in the middle troposphere were elongated in the southwestward direction, and the centers of both two levels did not coincide. In Fig. 6b, d, a long-lasting deep convective event can be seen close to the center, and a vertically coherent vorticity peak, or a TC inner core, was established in this region. Not all the deep convective events within the 22-km radius led to significant intensification of axisymmetric tangential wind (Fig. 5). In some cases, the merger of the small vorticity peaks accounted for the development of a TC inner core, rather than the occurrence of a few prominent convective events (e.g., E_mp4). In summary, the deep explosive convective event, analogous to a “hot tower," does not appear to be a necessary nor sufficient condition for the establishment of a TC inner core or the onset of pressure reduction. However, it certainly played a crucial role when it occurred with the right timing and location in association with an environment that was preconditioned by the superposition of multiple (dynamical and convective) large-scale factors with different origins.

Fig. 6
figure 6

Relative vorticity and wind vectors in E_mp2 at a, c 1800 UTC on 17 June 2008 and at b, d 1200 UTC on 18 June 2008 (before and after the onset of intensification) at the a, b 7.6-km and c, d 1.5-km altitude in the coordinates with the origin located at the storm center


Large-scale fields and inner-core processes relevant to the formation and intensification of Typhoon Fengshen (2008) were investigated by a set of global cloud-permitting (3.5-km mesh) simulations. Fengshen was formed in mid-June during the onset of the WNP monsoon and the active phase of a BSISO event.

The TC genesis was preceded by the formation of a lower tropospheric large-scale gyre associated with enhanced equatorial westerlies and a middle tropospheric wave trough intruding above the gyre in both the ECMWF YOTC and the NCEP FNL. The incipient disturbance of Fengshen was located on the northeastern rim of the large-scale gyre, where moisture convergence was pronounced. Differences between the analysis datasets were found in the development of the trough and the horizontal distribution of moisture.

To understand the impact of atmospheric conditions and the convective effects on the TC environment and precursor disturbance, five simulations were performed by varying the initial data (ECMWF YOTC, NCEP FNL) and cloud microphysics settings (mp1–4).

In three simulations initialized using the ECMWF YOTC, a middle tropospheric trough developed over a few days following a large-scale latent heat release above the lower tropospheric gyre, allowing for the successive occurrence of deep convective events within the 50-km radius of the incipient disturbance, and a deep upright TC inner core was established.

In the run with reduced latent heating (mp4), the trough was weak and TC formation was significantly delayed, suggesting the importance of the latent heat release over the lower tropospheric gyre to the intensification of the middle tropospheric trough and vertical coupling of the large-scale vorticity, which facilitated the acceleration of middle tropospheric tangential wind around the incipient low-level vortex (<50–100 km).

In the run initialized using the NCEP FNL, where the latent heat release averaged over the entire simulation was the same as its ECMWF YOTC counterpart, neither the trough nor a TC developed, lacking the collocation between the latent heat release and large-scale gyre in the early period. Under the vertical wind shear increasing poleward, it was critical for the incipient vortex to experience upward penetration before it moved farther northward.

The wave train (SWT) is a dominant unstable mode under the WNP summertime background state (Aiyyer and Molinari 2003; Kuo et al. 2001; Lau and Lau 1990; Li 2006) and easily manifests from various disturbances (Xu et al. 2014; Wu and Duan 2015). In the case of Fengshen, our results suggest that the superposition of large-scale disturbances with different origins, namely a middle tropospheric trough and lower tropospheric gyre associated with the active period of a BSISO event, formed the wave train. The earlier and stronger development of a wave train-like structure down to the lower troposphere in the TC developing simulations was considered to be responsible for the northward displacement, as well as the tendency toward over-intensification.

It is a very challenging task to simulate/predict the behavior of multiple disturbances that may excite the wave train accurately, but that will improve the predictability of TC genesis in the WNP. A high-resolution global model may serve as a useful tool to achieve this.