Modeling Rapid Intensification of Typhoon Saomai (2006) with the Weather Research and Forecasting Model and Sensitivity to Cloud Microphysical Parameterizations

JournalOctober of2012 the Meteorological Society of Japan, Vol. 90,J. MING No. 5, etpp. al. 771−789, 2012 771 DOI: 10.2151/jmsj.2012-513

Jie MING, Shoujuan SHU, Yuan WANG, Jianping TANG, and Baojun CHEN

Key Laboratory of Mesoscale Severe Weather, Ministry of Education, School of Atmospheric Sciences, Nanjing University, Nanjing, China

(Manuscript received 28 June 2011, in final form 8 May 2012)

Abstract

Typhoon Saomai (2006) was one of the most severe typhoon landfalls in China from 1956 to 2010. The rapid intensification process of Typhoon Saomai is simulated with the advanced research version of the- WeatherRe search and Forecasting (ARW) modeling system using different cloud microphysical parameterization schemes. The horizontal spacing of the finest nested mesh is 1.5 km. The intensity, precipitation, and inner-core structures of the simulated typhoons are verified against the observations. The performances of various cloud microphysical parameterization schemes are compared. It is found that varying the microphysics scheme generates little sensitivity in track, but results in pronounced deviations in intensity and inner-core structures. The results indicate the condensation and depositional growth of graupel or snow of the suitable cloud microphysical parameterization scheme enhances the diabatic heat releasing in the inner core region. The released diabatic heating determines the intensity and inner-core structures of typhoon. Furthermore, a positive feedback associated with the diabatic heating plays an important role in the intensification of the simulated storm with a suitable cloud microphysical parameterization scheme.

It is generally well recognized that intensity change 1. Introduction results from complex interactions between a storm Tropical cyclones (TCs) are one of the most destruc- internal dynamics and its large-scale environment. It is tive natural disasters in the world. Owing to advance- often not well forecasted due to the lack of knowledge ments in observations and numerical modeling, much of either the storm structure or environmental condi- improvement has been made in TC track forecasts, but tions over the open ocean. Especially, rapid intensifi- the intensity forecasts has improved less than track cation (RI) is one of the most challenging and the least forecasts (DeMaria et al. 2005). As a result, intensity well understood problems in tropical cyclone predic- forecast still remains a challenging problem in both tion today. Furthermore, the most intense TCs in the operational and research communities (Bender and Northern Hemisphere often have a RI period during the Ginis 2000; Krishnamurti et al. 2005; Rogers et al. peak season, and account for nearly three-quarters of 2006). all the RI events (Frederick 2003; Kaplan and DeMaria 2003; Ventham and Wang 2007; Shu et al. 2012). RI Corresponding author: Jie Ming, School of Atmospheric also poses a significant threat to ships and a growing Sciences, Nanjing University, 22 Hankou Road, Nanjing, Jiangsu, 210093 China number of coastal communities. E-mail: [email protected] Intensity changes have been investigated in ©2012, Meteorological Society of Japan numerous previous studies (Frank 1977; Willoughby 772 Journal of the Meteorological Society of Japan Vol. 90, No. 5

1988; Frank and Ritchie 1999; Montgomery et al. These results also suggested that without evaporative 2006). It is known that the large-scale environmental cooling and melting of snow and graupel, the down- conditions such as vertical wind shear and upper level drafts become much weaker, which was favorable trough or cold low dynamics and thermodynamics in for intensification. Zhu and Zhang (2006) studied the the inner core region, and air–sea interactions all play effect of various cloud microphysical processes on important roles in determining the intensity change intensity and inner core structure of Hurricane Bonnie (Davis and Emanuel 1988; Merrill and Velden 1996; (1998) with the fifth-generation Pennsylvania State Willoughby and Black 1996; Bosart et al. 2000; University–National Center for Atmospheric Research Nolan et al. 2007; Shay et al. 2000). Among all these (PSU–NCAR) Mesoscale Model (MM5) model. The factors, the processes occurring in the inner core are results indicated that a weak storm can be produced by closely associated with the parameterization schemes removing all ice particles from the cloud microphysical in numerical models. Different cumulus, planetary processes due to greatly reduced latent heat release and boundary layer (PBL) and cloud microphysical param- much slower autoconversion and accretion processes. eterization schemes in numerical models can signifi- The cooling of melting ice particles and evaporation of cantly influence the intensity change of TCs (Karyam- rainwater had a breaking effect on the development of pudi et al. 1998; Braun and Tao 2000; Zhu and Zhang the hurricane. In the six experiments of Zhu and Zhang 2006). (2006), the most rapid intensification of the storm was In recent years, numerical models can be run at very produced when evaporation processes were removed. high resolutions using sophisticated physical param- McFarquhar et al. (2006) also compared the roles of eterization, giving three-dimensional simulations of the PBL parameterization and cloud microphysical tropical cyclones thanks to the advances in computing processes in the simulation of Hurricane Erin (2001) capacity,(e.g., Liu et al. 1997; Zhang et al. 2000; Braun using the MM5 model. They showed that the conden- and Tao 2000; Rogers et al. 2003; Ming et al. 2009). sation process in the cloud microphysical scheme had While cumulus convective parameterization is not a major impact on the forecast of Erin’s final inten- applied at this resolution, the explicit representation of sity. Recently, Li and Pu (2008) conducted a series of cloud microphysical processes is a key component in numerical simulations to examine the sensitivity of the three-dimensional numerical models. simulation to available cloud microphysical and PBL Previous studies indicated that hurricane intensity parameterization schemes. The results indicated that and structure were greatly influenced by the cloud the numerical simulations of the early rapid intensi- microphysics in numerical models. Willoughby et al. fication of Hurricane Emily are very sensitive to the (1984) and Lord et al. (1984) showed the sensitivity choice of cloud microphysical and PBL schemes in the of simulated tropical cyclone structure and inten- ARW model. Specifically, with different cloud micro- sity to the cloud microphysics in a two-dimensional physical schemes, the simulated minimum central sea axis-symmetric non-hydrostatic model with 2km hori- level pressure (MSLP) varies by up to 29 hPa, and the zontal resolution. The results showed that the ice-phase use of various PBL schemes has resulted in differences cloud microphysical scheme could produce a lower in the simulated MSLP of up to 19 hPa during the 30-h minimum surface level pressure, but the warm-rain- forecast period. only cloud microphysical scheme produced a rapid Cloud microphysical processes have been shown intensification while the ice phase cloud microphys- to be critical to the realistic simulation of TCs by ical scheme generate a slowly developing storm. Their numerical models. Unlike the many observational results also showed that more realistic downdrafts and and modeling studies of hurricanes in Atlantic, few convective rings were produced with the ice-phase studies have been conducted to investigate the effects cloud microphysical scheme. With a three-dimensional of different cloud microphysical parameterization hydrostatic primitive equation model, Wang (2002) schemes on the simulation of typhoon intensity and conducted five numerical experiments to test the effects inner-core structures in western North Pacific. Further- of variations in cloud microphysical parameterization more, it is as yet unclear whether and to what degree schemes on the intensification, structure, and intensity the simulated typhoon intensity and inner-core convec- of an idealized hurricane. The results demonstrated tive structures can be affected by using different cloud that the use of warm-rain cloud microphysical scheme microphysical parameterization schemes. This study causes a faster intensification rate of the storm than the aims to gain better understanding of how the cloud mixed-phase scheme did, partly due to the stronger microphysical parameterizations are responsible for condensational heating in the warm-rain processes. intensity and structure change of typhoon by analyzing October 2012 J. MING et al. 773

Nest domain Tracks of Saomai (2006)

34N Best track (a) (b) LIN WSM6 GODDARD 30 THOMPSON MORRISON 30N ZJ HB 8/10 JX FJ 26N 26 D2 8/09 Latitude Latitude 22N

22 8/08 D2 18N D1

110E 120E 130E 140E 150E 18 Longitude 115 120 125 130 135 140 Longitude

Fig. 1. (a) Locations of the model domains for the numerical simulations of Typhoon Saomai (2006). D1 is the 4.5-km grid and D2 is the nested 1.5-km grid. FJ, ZJ, JX, HB is stand for Zhejiang, Fujian, Jiangxi and Hubei provinces. (b) The tracks of Typhoon Saomai from the best track analysis (every 6-h) by JMA (2006) and the model simulations (every 6-h) from 0000 UTC 08 August to 1200 UTC 10 August 2006. the numerical simulations of typhoon Saomai (2006). moved westward and passed through Fujian province. In this study, the rapid intensification of Typhoon It reduced to a tropical storm within Geyang of Jiangxi Saomai (2006) and sensitivity to various cloud micro- province 1 day later. Finally, the intensity weakened to physical parameterization schemes are investigated. a depression in Hubei province. The performance of various cloud microphysical When Saomai made landfall at Zhejiang province, parameterization schemes is evaluated. Furthermore it attained intensity with a minimum central pressure the possible effects on the intensification of Typhoon of 920 hPa and maximum surface wind speed of 60 Saomai (2006) are elucidated. The paper is organized m/s. Saomai was the strongest typhoon that ever as follows. Section 2 includes a brief overview of the occurred over China’s offshore region, and the most Typhoon, and Section 3 includes a description of the powerful typhoon ever to have made landfall over numerical model and experimental design. Numerical mainland China. Saomai ripped through Zhejiang, results are verified to various observations in Section 4. Fujian, Jiangxi and Hubei province of Southeast China The sensitivity of various cloud microphysical param- (Fig. 1a), it battered provinces of Zhejiang and Fujian eterization schemes to simulate typhoon intensity and with powerful winds and heavy rain as it made land- structures are examined in Section 5. Concluding fall, destroyed tens of thousands of buildings, sank remarks and discussion are drawn in Section 6. more than 1,000 boats and downed power lines which suspended the electricity supply in six cities. Zhejiang 2. A brief overview of Typhoon Saomai (2006) and Fujian were devastated by the storm and the State Typhoon Saomai (2006) originated from a trop- Flood Control and Drought Relief Headquarters said ical disturbance near the east of Chuuk on July 31. economic losses in both provinces reached 11.3 billion It gradually increased in organization over the next yuan (U. S. $1.4 billion). 54,000 homes were destroyed several days as it moved northwestward. The system and 122,700 hectares (303,000 acres) of farmland developed to a tropical depression near the southeast ruined by the strong winds and floods. Saomai affected of Guam on 4 August 2006. Saomai moved northwest- around 6 million people and displaced 1.7 million ward, intensified quickly and reached typhoon inten- residents. Several ports were forced to close and the sity by 0600 UTC 7 August with a central location of typhoon disrupted all forms of transport. At least 441 18.8oN, 138.2oE. The typhoon continued to intensify people were killed by the storm in China. Because of rapidly, and became a super typhoon with maximum the huge losses, the name “Saomai” was retired from wind speed of 51.4 m/s on 0600 UTC 9 August. western North Pacific typhoon name list and became Subsequently, Saomai made a landfall at Cangnan the specified name of the No. 8 typhoon of 2006. of Zhejiang Province at 11 UTC 10 August and then 774 Journal of the Meteorological Society of Japan Vol. 90, No. 5

an extension of the WRF single-moment five-class 3. Model description and experimental design scheme (Hong et al. 2004) with graupel included. The For numerical simulations, the ARW modeling WSM6 scheme improved the number concentration, system version 3.1 (Skamarock et al. 2005) is used. accretion, and ice nucleation in the cloud ice formation Three-dimensional simulations are performed using processes (Hong and Lim 2006). The Goddard micro- two-way interactive nesting in two domains with hori- physical scheme is coming from Goddard Cumulus zontal grid spacing of 4.5 km (D1) and 1.5 km (D2). Ensemble (GCE) model’s (Tao and Simpson 1993) Figure 1a shows the location of the domains, the one-moment bulk microphysical scheme. It is mainly center of D1 is at 24.1oN and 127.8oE. The numbers based on Lin et al. (1983) with additional processes of horizontal grid points are 500 × 300 and 421 × 421 from Rutledge and Hobbs (1984). The Goddard micro- for D1 and D2 respectively. The outer domain D1 is physical schemes have several modifications. There is integrated from 1200 UTC 7 August to 1200 UTC an option to choose either graupel or hail as the third 10 August 2006, and the inner domain starts at 0000 class of new ice saturation techniques (Tao et al. 1989, UTC 8 August. D2 is an automatic vortex-following 2003), here the graupel is chosen. All microphysical moving nest grid so that the center of the domain is processes that do not involve melting, evaporation always located at the center of the typhoon. The model or sublimation (i.e., transfer rates from one type of tracks the vortex center every 20 minutes and then the hydrometeor to another) are calculated based on one inner domain moves if necessary. The vortex center thermodynamic state. This ensures that all of these is determined by finding the minimum geopotential processes are treated equally. Thompson et al. (2004) height at 500 hPa. The advantage of using a moving scheme is a new bulk microphysical parameterization frame is that the typhoon does not leave the domain (BMP) that has been developed for using with WRF during long-period simulations with a limited domain or other mesoscale models. Unlike any other BMP, the size. Forty-seven σ levels are used from the surface to assumed snow size distribution depends on both ice the top at 50 hPa. water content and temperature, and is represented as a The initial and boundary conditions for the ARW sum of exponential and gamma distributions. Further- model simulations are derived from the Japan Mete- more, snow assumes a non-spherical shape with a bulk orological Agency (JMA) 6-hourly gridded regional density that varies inversely with diameter as found in analyses at 20 km × 20 km horizontal resolution with observations and in contrast to nearly all other BMPs 20 pressure levels. The JMA analyses were produced that assume spherical snow with constant density. The using a multivariate three dimensional optimum inter- Morrison et al. (2009) scheme is based on the two-mo- polation (OI) method to combine first-guess fields from ment bulk microphysical scheme of Morrison et al. JMA’s regional spectral model (RSM) with observa- (2005) and Morrison and Pinto (2006). Prognostic tions from a variety of platforms (JMA 2002; Hosomi variables include number concentrations and mixing 2005). The model physics options are the same for the ratios of cloud ice, rain, snow, and graupel/hail, and two domains. The Rapid Radiative Transfer Model mixing ratios of cloud droplets and water vapor (total (RRTM) longwave radiation (Mlawer et al. 1997) and of 10 variables). The prediction of two-moments (i.e., Dudhia shortwave radiation schemes (Dudhia 1989) both number concentration and mixing ratio) allows are adopted. For the parameterization of turbulence for a more robust treatment of the particle size distribu- in the PBL, the Mellor-Yamada-Janjic PBL scheme tions, which are crucial for calculating the microphys- (Mellor and Yamada 1982; Janjic 2002) is used. There ical process rates and cloud/precipitation evolution. are no cumulus parameterization schemes for the two The present study focuses on the period from 0000 domains. UTC 8 to 1200 UTC 10 August 2006, which covers The five sensitivity experiments (see Table 1) are nearly the entire lifecycle of the typhoon from the designed to examine the effects of different cloud developing stage to the decaying stage. All simula- microphysical parameterizations on the typhoon inten- tions begin at the same initial conditions and the differ- sity and structure changes. All the five schemes include ences in the simulations will solely rely on the physics six categories of hydrometeors: vapor, cloud water, rain, options used in the ARW model. cloud ice, snow, and graupel. Specifically, the Purdue 4. Model verification Lin scheme (LIN) is a relatively sophisticated scheme, and is based on Lin et al. (1983) with some modifica- 4.1 Track and intensity tions (Chen and Sun 2002). The WRF single moment Figure 1b compares the simulated tracks from six-class (WSM6) scheme (Hong and Lim 2006) is different experiments with the JMA best track from October 2012 J. MING et al. 775

Table 1. The Lists of the cloud microphysical parameterization scheme sensitivity experiments and their physical options. Experiments Cloud microphysics option Features

LIN Purdue Lin scheme Exponential size distributions of rain, snow and graupel and Including ice sedimentation

WSM6 WSM 6-class graupel scheme New method for mixed-phase particle (snow and graupel) fall speed

GODDARD Goddard GCE scheme Option of choosing either graupel or hail and new saturation techniques

THOMPSON Thompson scheme Assume snow size distribution depend on ice and temperature and non-spherical shape with a bulk density

MORRISON Morrison 2-Moment scheme Allow more robust treatment of the particle size distribution and mixed-phase process

Intensity of Typhoon Saomai (2006) Intensity of Typhoon Saomai (2006) 1000 (a) 120 (b)

980 100

960 80

60 940

40 Minimum Pressure (hPa ) Best track Best track 920 LIN Maximum Surface Wind (kts ) LIN WSM6 20 WSM6 GODDARD GODDARD THOMPSON THOMPSON MORRISON MORRISON

900 0 8/08 8/09 8/10 8/08 8/09 8/10 Date Date

Fig. 2. Time series of simulated (a) minimum central pressure (hPa) and (b) the maximum surface wind speed (kts) of Saomai from 0000 UTC 08 August to 1200 UTC 10 August 2006 for all the experiments and the corresponding best analysis by JMA.

0000 UTC 08 August to 1200 UTC 10 August. wind speed from all the experiments to the best track Typhoon Saomai (2006) kept moving northwestward from JMA. Because model initialized with JMA RSM during its lifecycle. Although all the simulated tracks analysis field is at 20 km × 20 km resolution, there are are shifting a slightly northward, the track forecast in no differences at the beginning of all the experiments, different experiments are quite similar. All simulations but eventually the intensity forecasts are significantly reproduce the observed west-northwestward storm different in the five experiments. Pronounced differ- movement, which means steering flow produced by ences in intensity are evident between 12 and 54 h, the background circulation is similar in all the experi- with the extreme amplitude of differences at 39 hPa. ments. Overall, the track forecast of Typhoon Saomai is Overall, small differences in intensities are found not very sensitive to the cloud microphysical schemes among all the simulations in the first 12 hours. After- in the ARW model. ward, the simulations with the Lin and Thompson Figure 2 compares the time series of the simulated schemes produce the quickest intensification, and have minimum sea level pressure and maximum surface similar trends with the observation. After 24 h, the 776 Journal of the Meteorological Society of Japan Vol. 90, No. 5

(a) 26N (b) 26N (c) 100 60 24N 24N

Latitude 40 Latitude

22N 22N 20

128E 130E 132E 128E 130E 132E Longitude Longitude 10

(d) 26N (e) 26N (f) 26N 5

24N 24N Latitude Latitude Latitude 24N 2

22N 22N 22N 1 128E 130E 132E 128E 130E 132E 128E 130E 132E Longitude Longitude Longitude

Fig. 3. (a) The AMSR-E 89-GHz composite microwave imagery at 0015 UTC 09 August 2006 and simulated hourly rainrate (mm/h) at 0300 UTC 09 August 2006 from all five experiments: (b) LIN, (c) WSM6, (d) GODDARD, (e) THOMPSON, (f) MORRISON.

simulation with Lin scheme has a larger deepening rate reflectivity. Figure 3 compare the AMSR-E 89-GHz and over deepens with a magnitude of 10 hPa at 0000 composite microwave imagery and simulated hourly UTC 10 August. In contrast, other simulations could rainrate at 0300 UTC 09 August. The experiments with not capture this characteristic. In particular, the simu- Lin and Thompson schemes, which produce stronger lations with WSM6 and Morrison schemes generate a storms, have more compact eyewalls with heavier weaker storm and a slower deepening rate. Simulation precipitation and a small eye, similar to the observa- with Goddard scheme generates a similar intensity with tion. The distribution of rainfall is more symmetric Thompson scheme at 54-h of simulation, but intensi- and the size of storm is relatively small. Especially, the fies slowly. All experiments have the same weakening storm with Lin scheme produces the smallest storm but trend in intensity after landfall. Moreover, the differ- has a broader heavy rainfall area. In contrast, the other ences in the maximum surface wind speed are relative three storms have large eyes and asymmetric eyewalls small. Especially, the simulation with Lin scheme is with heavy rainfall occurring in the south part of the overestimated after 12 h, and the difference between eyewalls. On the other hand, the observed (Zhao et al. the maximum surface wind speed of simulation with 2008) and simulated radar reflectivity are compared. Thompson scheme and observation reduces to zero The horizontal resolution of observed radar reflectivity during the entire day of 09 August. All the storms reach is 1 km. Figure 4 depicts the observed and simulated the same speed at 54-h of simulation. In other words, radar reflectivity images captured at 1km valid at 1602 the Lin and Thompson schemes are able to reproduce a UTC and 1600 UTC 09 August respectively. All the similar deepening trend to the observation. experiments have relatively large eyes compared to the observation. The reason for the large eyes is probably 4.2 Precipitation and radar reflectivity due to the initial conditions. The storms with WSM6, Now let us shift our attention to the verification of the Goddard and Morrison schemes, which are the weaker system scale features, such as precipitation and radar storms, have larger eyes and looser eyewalls. The high October 2012 J. MING et al. 777

150 150 150 100 (a) 100 (b) 100 (c) 60

) 50 50 50 0 0 0 50

Distance(km −50 −50 −50 Distance (km) Distance (km) −100 −100 −100 40

−150 −150 −150 −150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150 Distance (km) Distance (km) Distance (km) 30 150 150 150

100 (d) 100 (e) 100 (f) 20 50 50 50

0 0 0 10 −50 −50 −50 Distance (km) Distance (km) Distance (km)

−100 −100 −100

−150 −150 −150 −150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150 −150 −100 −50 0 50 100 150 Distance (km) Distance (km) Distance (km)

Fig. 4. (a) 1-km height observed radar reflectivity (dBZ) at 1602 UTC 09 August, and 1-km height simulated radar reflectivity (dBZ) at 1600 UTC 09 August 2006 from all five experiments: (b) LIN, (c) WSM6, (d) GODDARD, (e) THOMPSON, (f) MORRISON.

echo region is not closed and the magnitude of radar concentrated. The storm with Thompson scheme has a reflectivity is smaller. The structures of storm with Lin similar structures compared to the observation. There scheme are consistent with the observation and seem is a region of enhanced reflectivity above the melting to have the concentric eyewall. But the outer rainband layer (around 6-km altitude). This region, clearly seen is not closed. Because the eye of simulated storm is in the 1%–1.6% contour, is likely due to have enhanced too large, the inner eyewall does not contract. The precipitated ice supported by the stronger low-level storms with Lin and Thompson schemes have closed updraft. The storm with Lin scheme also has a similar and more organized eyewalls with embedded cores of structure in the upper level except a high percentage high reflectivity. However, the storm with Lin scheme of points at weak reflectivity, but the magnitude of has broader high echo region and that with Thompson contour is small due to the less total points. In the scheme has narrow one. The magnitude of peak reflec- lower level, it has two bands with higher percentage of tivity is higher in the simulations than observations by points between 30 dBZ and 50 dBZ, which is related to 10-15 dBZ. This feature is similar to other modeling the formation of concentric eyewalls. The storms with study (Rogers et al. 2007). WSM6 and Morrison schemes have the same charac- Figure 5 provides a more detailed look at the three teristics, but a lower percentage of points in the middle dimensional statistical properties of inner core structures to upper level compared to the observation. The storm by showing contoured frequency by altitude diagrams with Goddard scheme has a high percentage of points (CFADs, Yuter and Houze 1995) of radar reflectivity between 30 dBZ and 40 dBZ, and two centers at 4 km with the bin size of 2 dBZ. The inner core region is and 7 km. an area of 225 km × 225 km centered at the typhoon’s surface minimum pressure. The storm with Lin scheme 4.3 Vertical distributions of hydrometeors has the least amount of total points, it means the storm Figure 6 compares the normalized vertical profiles produced by Lin scheme has small size and is more of cloud water, rainwater, cloud ice, precipitated ice 778 Journal of the Meteorological Society of Japan Vol. 90, No. 5

10 (a) 9 (b) 9 (c) 8

6 6 6 1.4 Height (km) 4 Height (km) Height (km) 1.2 3 3 2 1 10 20 30 40 50 10 20 30 40 50 10 20 30 40 50 Radar Reflectivity (dBZ) Radar Reflectivity (dBZ) Radar Reflectivity (dBZ) 0.8

9 9 9 0.6 (d) (e) (f) 0.4 6 6 6 0.2 Height (km) Height (km) Height (km)

3 3 3

10 20 30 40 50 10 20 30 40 50 10 20 30 40 50 Radar Reflectivity (dBZ) Radar Reflectivity (dBZ) Radar Reflectivity (dBZ)

Fig. 5. Contoured frequency by altitude diagrams of (a) observed radar reflectivity (dBZ) at 1602 UTC 09 Au- gust, and simulated radar reflectivity (dBZ) at 1600 UTC 09 August 2006 from all five experiments: (b) LIN, (c) WSM6, (d) GODDARD, (e) THOMPSON, (f) MORRISON.

mass content averaged over the storm inner core region The storm with Thompson scheme produces the largest at 2100 UTC 08 August 2006 compared to the obser- amounts of precipitated ices at upper levels attributed vation from the TRMM (Tropical Rainfall Measuring to the generation of snow, the shape is also consistent Mission) 2A12 product at 2244 UTC 08 August 2006. with observations in the upper levels except the peak Because TRMM scans the half of inner core region, is 2 km higher. The storm with Lin scheme produces the normalized vertical profiles are computed in the smallest amounts of precipitated ice, and the shape is same area covered by TRMM and used to compare coincident with the observation at the middle level. the mean features of different heights. The shapes of 5. Inner-core evolution and structures cloud water profiles are consistent with the observations. All the experiments have two peaks of cloud Furthermore, the simulation with WSM6 scheme, water at 1 km and 5 km level. For rainwater, the storm which is the weakest, and the simulations with Lin with Lin scheme produces larger amount of rainwater and Thompson schemes, which have rapid intensifi- at lower levels than the others do, which may be the cation, are chosen to analyze the inner-core structures reason for the stronger rainfall in the experiment with and mechanisms of rapid intensification. Time-height Lin scheme. Notable differences in hydrometeor distri- series of averaged hydrometeors, diabatic heating, butions are ice and precipitated ice. On the one hand, vertical motion and temperature perturbation are the peak of ice in every scheme is at different levels shown in Figs. 7, 8. (The average is computed within and the amount is less than other hydrometeors. The 225 km × 225 km centered at the typhoon’s surface storm with Thompson scheme produces the smallest minimum pressure.) The amounts of the mean cloud amounts of ice. On the other hand, the storms with water in all three experiments are close to each other. WSM6 and Morrison schemes have similar profiles of There are two peaks, which are the same as the profiles precipitated ice with the peak at 7km, the same as the in Fig. 6. One peak is at the level of 1 km and the other observation, resulting from the production of graupel. is at 5 km. The storm with Lin scheme produces much October 2012 J. MING et al. 779

16 16 16

12 12 12

8 8 8 Height (km) Height (km) Height (km) 4 4 4 (a) (b) (c) 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

16 16 16 cloud rain ice 12 12 12 snow+graup

8 8 8 Height(km) Height (km) Height (km)

4 4 4 (d) (e) (f)

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Fig. 6. Mean normalized vertical profiles of various hydrometeors at 2100 UTC 08 August 2006, for five experiments: (a) LIN, (b) WSM6, (c) GODDARD, (d) THOMPSON, (e) MORRISON, compared to (f) observation from TRMM 2A12 product at 2244 UTC 08 August 2006. The average is computed in the same area covered by TRMM.

more rainwater compared to other two experiments, layer corresponding to the maxima of heating in the and the storm with Thompson scheme generates the 8–10-km layer after RI has begun. However, the storm most precipitated ice, but the least ice among all three with WSM6 scheme produces less heating and precipi- experiments. The different distributions of hydrome- tated ice, as well as weaker upward motion. It does not teors induce the different structures and intensity of have RI and the difference of intensity in this exper- storm. Then the evolution of diabatic heating rate is iment is relative large compared to other two experi- examined (Fig. 8). Because the average is computed ments. within the inner core region, the main feature is the Figure 8 also displays the time-height evolution heating produced by the convections. The magnitude of thermodynamic properties over the inner core of cooling is smaller than the heating. So the analysis region. The temperature anomalies of storms with is focused on the change in heating. Corresponding Lin and Thompson schemes have a maximum in the to the intensity, the storms with Lin and Thompson upper troposphere. The magnitude of warm anomaly schemes produce prominent convective heating increases from 1 to 5oC as the storm deepens while the during 0600 UTC August 8 prior to the onset of RI. layer of anomaly increases from 9 to 15 km during the Then they produce persistent heating in the middle to period from 1200UTC 08 to 1200 UTC 09. In contrast, upper levels, and the intensity shows similar trends the storm with WSM6 scheme has a weaker warm with the observation. The heating induces deep layer core, and the top of warm core is lower than other two. updraft from about 1km to 16 km in these two exper- The warm temperature anomaly in the eye is due to iments, and their deeper updrafts are likely tied to the adiabatic warming associated with subsidence. There establishment of a well-defined secondary circulation. is positive feedback between the low pressure and Periods of strong updrafts occur both before and during the warm core. The warm core induced by adiabatic RI. They show intermittent maxima in the 10–12-km warming can produce a low pressure perturbation. The 780 Journal of the Meteorological Society of Japan Vol. 90, No. 5

16 16 16 14 14 14 12 (a) 0.2 12 (b) 0.2 12 (c) 0.2

10 0.15 10 0.15 10 0.15 8 8 8 0.1 0.1 0.1 6 6 6 Height (km) Height (km) Height (km) 4 0.05 4 0.05 4 0.05 2 2 2