4340 MONTHLY WEATHER REVIEW VOLUME 142

Improvements to a Initialization Scheme and Impacts on Forecasts

HIEP VAN NGUYEN* AND YI-LENG CHEN Department of Atmospheric Sciences, University of Hawaii at Manoa, Honolulu, Hawaii

(Manuscript received 10 October 2013, in final form 11 August 2014)

ABSTRACT

This study makes improvements to the tropical cyclone (TC) initialization method introduced by Nguyen and Chen (i.e., the NC2011 scheme). The authors found that prescribing sea level pressure associated with the initial vortex using a modified Fujita formula has very little impact on the vortex structure and intensity during a series of 1-h model integration and relocation. On the other hand, inserting an artificial warm core makes the vortex spin up much faster. When a warm core is inserted during the initial spinup process, the computational time required for model initialization is reduced by 1/2–1/3. Because prescribed sea level pressure is not required to spin up the vortex, information on vortex size, such as radius of maximum wind, is no longer needed. The performance of the improved NC2011 scheme with an initial prescribed warm core during the initial spinup process is tested for that made over southern or in 2006. Before landfall, these storms were over the open ocean where conventional data were sparse, without special observations. Two sets of model runs, with (NC2011-CTRL) and without (CTRL) vortex initialization, are performed for comparison. The initial and time-dependent boundary conditions are from the NCEP Final Analyses (FNL). There are twelve 48-h simulations in each run set. Results show that the vortex initialization improves TC track and intensity simulations.

1. Introduction bogus data have been known to significantly improve TC track and intensity forecasts (Zou and Xiao 2000; Pu and Tropical cyclone (TC) initialization in numerical Braun 2001; Wu et al. 2006; Zhao et al. 2007; Zhang et al. models is a crucial factor affecting TC track and intensity 2012; and others). A TC initialization technique needs forecasts. Because TCs develop and spend most of their to generate a vortex with a structure similar to the real lifetime over the open ocean before making landfall, TC and remain compatible with the numerical model high-resolution observational data depicting detailed TC employed. structure in the core area are frequently lacking. Usually, Recently, Nguyen and Chen (2011) developed a new a bogus TC vortex is used to improve the TC structure in TC initialization technique (hereafter the NC2011 the model initial conditions. The bogus vortex can be scheme) to generate high-resolution initial conditions constructed with empirical functions of pressure and/or using the Advanced Research version of the Weather winds (Fujita 1952; Holland 1980; Davidson and Weber Research and Forecasting (WRF) Model (ARW; 2000; Davis and Low-Nam 2001; Kwon and Cheong 2010) Skamarock et al. 2005). The NC2011 scheme signifi- or by model integration (Kurihara et al. 1993; Liuetal. cantly improves model initial conditions with better de- 1997). Data assimilation of both observed and vortex piction of TC intensity and structure including minimum sea level pressure (SLP) (Pmin); maximum wind speed * Current affiliation: Institute of Meteorology, Hydrology and (Vmax); and 3D distributions of temperature, pressure, Climate Change, Hanoi, Vietnam. winds, water vapor, as well as the condensate mixing ratio and asymmetric structure of rainbands. Nguyen and Chen (2011) performed seven 48-h simulations for Corresponding author address: Yi-Leng Chen, Dept. of Mete- orology, School of Ocean and Earth Science and Technology, Morakot (2009). Another 48-h simulation was initialized University of Hawaii at Manoa, Honolulu, HI 96822. about 36 h before landfall over for both Typhoon E-mail: [email protected] Jangmi (2008) and Typhoon Kalmaegi (2008). For all the

DOI: 10.1175/MWR-D-13-00326.1

Ó 2014 American Meteorological Society Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4341

TABLE 1. Abbreviations of sensitivity tests and their descriptions. TABLE 2. Abbreviations and descriptions for sensitivity tests with a warm core. Abbreviation Description Cumulus NSLP WRF single-moment 6-class microphysics scheme Abbreviation Microphysics parameterization and Grell–Devenyi cumulus parameterization without prescribed SLP Eta-BM Eta (Rogers Betts–Miller–Janjic WSLP As in NSLP, but SLP is prescribed during cycle run et al. 2001) (Janjic 1994, 2000) for with observed minimum center SLP domain 1 only LSLP As in NSLP, but SLP is prescribed during cycle run Lin-KF Lin et al. (1983) Kain–Fritsch (Kain 2004) with minimum center SLP of 870 hPa for domain 1 only WRF-GD WRF single-moment Grell–Devenyi (Grell and 6-class (Hong et al. Devenyi 2002) for cases considered, using NC2011 to construct the initial 2004; Hong and Lim domain 1 only TC vortex leads to significant improvement in the simu- 2006) lations of track, intensity, radar reflectivity patterns, and Eta-GD Eta Grell–Devenyi for domain TC circulations. Nguyen and Chen (2011) also showed 1 only that for (2009), using the NC2011 initialization scheme results in better prediction of rain- factors affecting the TC initialization process are dis- fall distribution before and after landfall because the cussed in section 3. Some preliminary results on verifi- scheme is able to depict better storm structure. cations of TC intensity and track simulations with the In this study, the impacts of prescribed SLP and pre- improved NC2011 scheme for four landfalling typhoons scribed warm core structure on spinning up a TC vortex in 2006 are discussed in section 4. A summary and dis- in the NC2011 scheme are further investigated. One of cussion are given in section 5. the main challenges in TC research is forecasting the intensity change (Kucas 2010) and improving typhoon 2. Data and model descriptions intensity at the model initial time in high-resolution models can help intensity forecast. In this work, this This research uses version 3.1 of the ARW Model, problem is investigated for typhoons that occurred in which is a nonhydrostatic, three-dimensional primitive a relatively data-sparse region to determine the extent equation model. Two nested domains with horizontal to which large-scale settings affect the TC spinup pro- grids of 18 km (287 3 226) and 6 km (379 3 316), re- cess and subsequent model simulations. Nguyen and spectively, and 38 vertical levels from the surface to the Chen (2011) focused on typhoons that occurred over the 50-hPa level are employed in the model with two-way Taiwan area where the observational network is dense nesting. The full sigma levels are as follows: 1.00, 0.97, with supplementary dropsonde observations over the 0.95, 0.92, 0.89, 0.86, 0.84, 0.81, 0.78, 0.76; 0.73, 0.70, adjacent oceans (Wu et al. 2005). 0.68, 0.65, 0.62, 0.59, 0.57, 0.54, 0.51, 0.49, 0.46, 0.43, The case used in this study is Chanchu (2006), which 0.41, 0.38, 0.35, 0.32, 0.30, 0.27, 0.24, 0.22, 0.19, 0.16, formed as a tropical depression on 9 May 2006 around 0.14, 0.11, 0.08, 0.05, 0.03, and 0.00. The model con- 8.38N, 133.48E. It made landfall over the on figurations include the following: the WRF single- late 11 May 2006 and weakened. However, after crossing moment 6-class microphysics scheme (WSM6; Hong the Philippines, it intensified as a super typhoon in a data- et al. 2004; Hong and Lim 2006), the Grell–Devenyi sparse region between Vietnam and the Philippines. It cumulus parameterization scheme (Grell and Devenyi made landfall again over the southern China coast on 18 2002) for the 18-km domain only, a Rapid Radiative May 2008. At 0000 UTC 15 May 2006, Transfer Model scheme (Mlawer et al. 1997)forlong- (2006) had a minimum SLP (Pmin) of ;916 hPa and wave radiation, the Dudhia scheme (Dudhia, 1989)for 2 a maximum wind speed (Vmax) of ;64 m s 1 based on shortwave radiation, the Monin–Obukhov similarity the Joint Typhoon Warning Center (JTWC) best track scheme (Monin and Obukhov 1954) for surface layer data (http://www.usno.navy.mil/JTWC). We investigate physics, and the Yonsei University scheme (Noh et al. the TC vortex associated with Typhoon Chanchu (2006) 2003) for planetary boundary layer physics. The pre- initialized at 0000 UTC 15 May. In addition, three more cipitation schemes used are the same as Wu et al. landfalling typhoons (, Prapiroon, and Xangsane) (2012). The initial and lateral boundary conditions are occurred in the same region during 2006. These storms from the National Centers for Environmental Pre- are used to assess the impact of the new TC initialization diction (NCEP) Final Analyses (FNL) with a 18 hori- technique on the intensity and track simulations. Data zontal resolution. Two separate simulations with and and model descriptions are discussed in section 2.Abrief without (CTRL) the NC2011 scheme are performed for description of the NC2011 scheme and sensitivity tests on comparison.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4342 MONTHLY WEATHER REVIEW VOLUME 142

TABLE 3. The name, number of run cases, initial time, minimum central SLP, and maximum temperature anomalies (K) for the four TCs in 2006.

Name Initial time Pmin (hPa) dT (K) Chanchu 0000 UTC 14 May 960 4 Chanchu 1200 UTC 14 May 940 6 Chanchu 0000 UTC 15 May 930 7 Chanchu 1200 UTC 15 May 930 7 Chanchu 0000 UTC 16 May 930 7 Durian 0000 UTC 2 Dec 965 4 Durian 1200 UTC 2 Dec 965 4 Durian 0000 UTC 3 Dec 955 5 Prapiroon 0000 UTC 2 Aug 980 3 Prapiroon 1200 UTC 2 Aug 970 3 Xangsane 0000 UTC 29 Sep 960 4 Xangsane 1200 UTC 29 Sep 955 5

integrated vortex are close to the observed values. Nor- mally, the required number of cycles is about 80. The first important assumption used in the NC2011 initialization method is that within each 1-h cycle run the TC may move; however, it is still embedded in almost the same environment without many changes. The second as- sumption is that the TC structure is mainly controlled by

21 the environmental conditions in which it is embedded. To FIG. 1. The changes in (a) maximum wind speed (m s ) and (b) minimum SLP (hPa) with number of cycle runs for NSLP satisfy the first assumption, it is desirable for the duration (blue), WSLP (red), and LSLP (green) sensitivity test runs. of each cycle run to be relatively short (,3h)toensure that the environment in which it is embedded does not change significantly. 3. Factors affecting the spinup of a TC vortex in the In the rest of this section, sensitivity tests (Tables 1 NC2011 scheme and 2) are performed for Typhoon Chanchu (2006) and initialized at 0000 UTC 15 May 2006 to investigate the a. Review of the NC2011 scheme physical processes affecting the spinning up of an intense In the NC2011 scheme, model cycle runs were used to TC in the NC2011 scheme. At this time, the global construct a TC vortex with the environment conditions analysis from NCEP FNL with a 18 horizontal resolution almost unchanged. In each 1-h model integration, the significantly underestimates the TC intensity because of initial SLP distributions associated with the TC vortex the coarse resolution and lack of in situ data in the storm within R , 400 km were constructed by a modified Fujita core region. (1952) formula using the observed minimum SLP (Pmin) b. Impact of prescribed SLP on the spinup process and storm size (Rmax) from the best track data. The model is then integrated for a period of an hour. During The first three sensitivity tests (Table 1) are designed to the model integration within a 1-h cycle run, the TC determine the need to prescribe SLP in the NC2011 vortex is free to develop under the given large-scale scheme. In the NSLP run, SLP is not prescribed during the conditions from the global data. At the end of the 1-h cycle runs. In the WSLP run, the SLP is prescribed by model integration of each cycle, the TC structure is sep- amodifiedFujita (1952) formula (Nguyen and Chen 2011) arated from the large-scale environment using a modifi- using the best track minimum surface pressure (Pmin) of cation to Kurihara et al. (1993). The TC vortex structure 916 hPa and radius of maximum wind (Rmax) of 25 km. at the end of the current 1-h cycle run is then relocated to Beyond the radius of R 5 400 km, the values of all vari- the observed location of the actual TC and used as the ables are the same as in the global analysis. In the LSLP initial vortex for the next 1-h cycle run with the large- run, the prescribed SLP is constructed with the observed scale environment provided by global analysis at the record Pmin ;870 hPa of Typhoon Tip (1979) (Dunnavan model initial time. The above cycle run is repeated until and Diercks 1980) with the Rmax of 25 km. We per- the minimum SLP and the maximum wind speed of the formed 90 cycle runs for each experiment.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4343

21 FIG. 2. The changes in maximum wind speed (contour, m s ), vertical wind (vector), and temperature anomalies (K) for CTRL and different cycle runs for the Typhoon Chanchu case initialized at 0000 UTC 15 May 2006.

Figure 1 shows that during the spinup cycles, there are structure will be mainly determined by the given large- no significant differences in Vmax (Fig. 1a)andPmin scale conditions and the model employed. (Fig. 1b) among the NSLP (pink), WSLP (blue), and c. A sensitivity test with an initial warm core LSLP (green) runs. All three runs show that Pmin (Vmax) significantly decreases (increases) during the first 40 cycle Because observational studies have shown that the runs. Both Pmin and Vmax are close to best track values maximum temperature anomaly within the eyewall of at about cycle 70. From cycle 70 to 90, both the Pmin and hurricanes occurs in the upper levels (e.g., Hawkins and Vmax of the three runs do not change significantly. It is Rubsam 1968), model sensitivity tests with a prescribed very striking that the prescribed SLP associated with the warm core using NC2011 (WRF-GD) (Table 2) are per- TC does not affect the spinup process in the NC2011 formed. The warm core test run is configured the same as scheme. In addition, even without prescribing the hori- the NSLP run except with a prescribed symmetric warm zontal distribution of SLP the storm is able to reach the core with a maximum (dT)max at the 250-hPa level. The observed intensity during the spinup process. These re- magnitude of the maximum temperature anomaly is de- sults attest to the fact that the storm environment has pendent on the best track minimum surface pressure a significant impact on the observed typhoon intensity. (Pmin) using a linear relationship for typhoons over the Because prescribed sea level pressure is not required to northwestern Pacific suggested by Velden et al. (1991) spin up the vortex, the information on vortex size, such as [(dT)max 5 (1013. 2 Pmin 2 2.)/11.8]. The warm core Rmax, is no longer needed. Thus, the storm size and has a radius of 60 km with a Gaussian weighting function

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4344 MONTHLY WEATHER REVIEW VOLUME 142

constructed by satellite observations will reduce the number of cycles in the initialization of TC vortex using the NC2011 method. d. Sensitivity tests of different precipitation schemes with an initial warm core Because latent heat release is essential for the spin up of the TC vortex, model sensitivity tests are performed with three different combinations of cumulus parame- terization schemes and microphysics schemes (Table 2). The microphysics and cumulus schemes are well tested by NCEP and recent publications related to TC simu- lations. Specifically, the Eta microphysics (Rogers et al. 2001) and Betts–Miller–Janjic cumulus parameteriza- tion (Betts and Miller 1993; Janjic 1994) schemes are used by NCEP and Nguyen and Chen (2011). The combination of Lin’s microphysics (Lin et al. 1983) and Kain–Fritch’s cumulus parameterization schemes (Kain and Fritsch 1993; Kain 2004) are used by Zhang et al. (2012). For all the schemes tested, a warm core structure is inserted during the first ten 1-h cycles. The inserted warm core helps the rapid spinup of the TC vortex re-

21 gardless of the precipitation schemes employed (Fig. 3). FIG. 3. The changes in (a) maximum wind speed (m s )and From cycle 11, no prescribed warm core is inserted, (b) minimum SLP (hPa) with number of cycle runs for WRF- GD (red), Eta-BM (black), Lin-KF (blue), and Eta-GD (green) which allows the TC structure to adjust to the given sensitivity test runs. large-scale environmental conditions. The adjustment process initially reduces the intensity of the spinup (Kidder et al. 2000; Kwon and Cheong 2010)inboththe vortex for the first ten 1-h cycles or so (Fig. 3) as the TC horizontal and vertical direction. The maximum tem- warm core structure adjusts to environmental condi- perature anomalies [(dT)max] for different typhoon in- tions. After that, the spinup process of the TC vortex tensities are shown in Table 3. The warm core is inserted continues. There is about 10-hPa difference in Pmin and 2 into the initial temperature fields during the first ten 1-h 5ms 1 difference in Vmax between the different cycles. The vertical cross sections of the inserted warm schemes. In this case, the combination of Bells–Miller– core and the development of the warm core during the Janjic cumulus parameterization and Eta microphysics cycle run are shown in Fig. 2. With the prescribed warm tends to underestimate the storm intensity as compared core, the Pmin associated with the TC vortex drops with the best track data. Nevertheless, it is apparent that quickly toward the best track data. Because the warm inserting an artificial warm core during the first ten 1-h core structure is prescribed as an external forcing, the cycles or so in the modified NC2011 scheme greatly re- Vmax also adjusts rapidly toward the best track value duces the number of cycles and computational resources (Fig. 3, red). Once the prescribed warm core is removed required to spin up the TC vortex for all four schemes after cycle 10, the TC vortex starts to adjust to the given (Fig. 3) tested. large-scale environment in the following cycling runs. In the WRF-GD run, the vortex can reach the best 4. Verifications of TC track and intensity track Pmin and Vmax after about 40 cycles as com- simulations pared to about 70 cycles in the original NC2011 scheme (Fig. 1,red). Chen et al. (2014) initialized 18 TCs that occurred In most cases, the TC warm cores are not well resolved during 2003–13 over the western Pacific using WRF- by global analysis in the initial conditions. Three- GD precipitation schemes with the modified NC2011 dimensional structures of TC warm cores can be con- method (NC2011-CTRL) (e.g., with a warm core and structed from satellite observations with an Advanced without prescribed SLP as described in section 3d). Microwave Sounding Unit (Kidder et al. 2000). It is They demonstrate that the NC2011-CTRL TC initial- apparent that 3D data assimilation of temperature data ization scheme is capable of reproducing the observed from satellite data or inserting a warm core structure TC structure as compared with satellite microwave

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4345

FIG. 4. Best track (circle, black) and 48-h simulated tracks (triangle, red) for five 48-h runs for Chanchu initialized every 12 h from 0000 UTC 14 May to 0000 UTC 16 May 2006 for (a) the CTRL and (b) NC2011-CTRL runs. The TC tracks are marked every 6 h. observations, including asymmetric spiral rainbands over southern China or Vietnam. The model results are and eyewall structure (single vs double), with an intensity used to verify the performance of the NC2011-CTRL very close to the best track data. The mean absolute error runs, in terms of TC intensity and track simulations. 2 (MAE) in Vmax and Pmin are 3.3 m s 1 and 2.6 hPa, These four storms are all 2006 typhoons that crossed the respectively. Furthermore, the storm size in the model Philippines and made landfall either over southern China initial conditions also compares favorably with satellite or Vietnam. Two types of initial conditions, one with the observations. In this section, the NC2011-CTRL scheme NC2011-CTRL TC initialization method and the other is used to simulate Chanchu and three other typhoons from the global analysis (CTRL), are used to initialize (Table 3) in the same region in 2006 that made landfall the ARW model every 12 h after the TC crosses the

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4346 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 5. Time series of (a) maximum wind speed and (b) minimum SLP for observations (black), NC2011-CTRL (red), and CTRL (green) for five 48-h runs for Chanchu initial- ized every 12 h from 0000 UTC 14 May to 0000 UTC 16 May 2006. Values are marked every 6 h.

Philippines and reaches 1178E about 36 h prior to landfall are performed for each type of model initialization over Vietnam or China. For each type of model initiali- scheme (Table 3). zation, a total of twelve 48-h simulations were performed. For the case of Chanchu (2006), both the CTRL The mean error statistics of track, maximum wind speed, (Fig. 4a) and the NC2011-CTRL (Fig. 4b) runs capture and minimum SLP are computed for comparison and the almost 908 directional change from westward to verifications. The model initial time for these runs is northward in late 14 May 2006. Both runs show simulated shown in Table 3. track bias to the left of the best track. For intensity sim- ulations, because the TC vortex in CTRL is much weaker a. Typhoon Chanchu (2006) than observed, Vmax (Pmin) in the CTRL run increases Chanchu was a supertyphoon over the northwest (decreases) rapidly within the 48-h simulated period Pacific in 2006. The typhoon affected the Philippines, (Fig. 5, green). For five CTRL runs in the Chanchu case, Vietnam, China, Taiwan, and . There were a total after 48 h of simulation, the simulated intensity is still of almost 300 casualties, with losses totaling more than weaker than observed (Fig. 5, green). For the NC2011- $1 billion U.S. dollars (USD). For Chanchu, five 48-h CTRL runs, both Vmax and Pmin (Fig. 5, red) are close runs, initialized every 12 h from 0000 UTC 14 May 2006 to best track values at the model initial time without

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4347

FIG. 6. Best track (black) and 48-h simulated tracks (red) for three 48-h runs for Durian initialized every 12 h from 0000 UTC 2 Dec to 0000 UTC 3 Dec 2006 for (a) the CTRL and (b) NC2011-CTRL runs. The TC tracks are marked every 6 h. significant adjustments in Vmax and Pmin during the southern coast of Vietnam, weakened, and finally dis- NC2011-CTRL simulations (Fig. 5,green).TheNC2011- appeared on late 5 December. Durian caused severe CTRL runs tend to underestimate Vmax and have a damage in the Philippines and Vietnam. According to slightly lower Pmin as compared with the best track data a report from ReliefWeb (http://www.reliefweb.int), the (Fig. 5,red). damage to the Philippines and Vietnam caused by Supertyphoon Durian was more than $10 billion USD b. (2006) with more than 1000 casualties. For Durian, three 48-h Durian is a rare supertyphoon that formed in December runs initialized every 12 h at 0000 UTC 2 December, 2006 over the northwest Pacific and affected Vietnam and 1200 UTC 2 December, and 0000 UTC 3 December the Philippines. Durian formed as a tropical depression on 2006 are performed for each type of model initializa- 25 November 2006 at about 9.68N, 147.28E. It became tion scheme (Table 3). a tropical storm on 26 November 2006 and a supertyphoon For track simulations, both the CTRL and NC2011- on 29 November, about 24 h before it made landfall over CTRL runs capture the southwestward movement of the Philippines. Typhoon Durian maintained its strong Durian (Fig. 6). For the intensity simulations, the CTRL 2 intensity for almost five days. It eventually moved to the runs simulate neither Vmax larger than 38 m s 1 nor Pmin

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4348 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 7. Time series of (a) maximum wind speed and (b) minimum SLP for observations (black), NC2011-CTRL (red), and CTRL (green) for three 48-h runs for Durian initialized every 12 h from 0000 UTC 2 Dec to 0000 UTC 3 Dec 2006. Values are marked every 6 h. less than 990 hPa (Fig. 7, green). The NC2011-CTRL was more than $1 billion (USD) with a total of more than runs (Fig. 7, red) show much improvement in Vmax and 100 casualties in China and the Philippines. For Prapiroon, Pmin simulations in comparison with the CTRL runs, two 48-h runs initialized at 0000 and 1200 UTC 2 August but the simulated intensity is weaker than in the best 2006 are performed for each type of model initializa- track data. tion schemes (Table 3). For track simulation, the NC2011-CTRL runs sig- c. (2006) nificantly improve the track simulation during the Prapiroon developed as a tropical depression on 29 early hours of simulations as compared with the CTRL July 2006 at about 13.58N, 129.28E, east of the Philippines. runs (Fig. 8). Both the CTRL and NC2011-CTRL runs It made its first landfall over the Philippines on 31 July tend to simulate track to the left of the best track. For 2006. After it crossed the Philippines, it intensified with the intensity simulations, the CTRL runs simulate 2 maximum wind speeds of ;18 m s 1 early on 1 August a weak TC vortex with no significant changes in Vmax 2006. Prapiroon reached its peak intensity early on 3 and Pmin within the 48-h simulations (Fig. 9, green). 2 August 2006 with maximum wind speed of ;33.5 m s 1 The NCB runs slightly overestimate vortex intensity about 6 h before it made landfall over China around (Fig. 9, red). The overestimation may be related to the 1200 UTC 3 August 2006. The typhoon weakened on simulated track being closer to the open ocean than the 4 August and disappeared on early 5 August 2006. best track data. This results in a decrease in the fric- Although Typhoon Prapiroon was relatively weak, with tional effect of landmasses and more fluxes from the 2 Vmax not exceeding 35 m s 1, the total reported damage ocean.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4349

FIG. 8. Best track (black) and 48-h simulated tracks (red) for two 48-h runs for Prapiroon initialized at 0000 and 1200 UTC 2 Aug 2006 for (a) the CTRL and (b) NC2011-CTRL runs. The TC tracks are marked every 6 h. d. (2006) just before it made first landfall over the Philippines on 27 September. After landfall, Xangsane weakened 2 Typhoon Xangsane formed on 25 September 2006 with maximum wind speeds of ;31 m s 1 early on 28 as a tropical depression centered at 11.88N, 129.18E. September. It then reintensified, reaching a maxi- 2 Xangsane reached its peak intensity, with a maximum mum wind speed of ;41 m s 1 on 30 September 2006, 2 wind speed of ;44 m s 1 and a minimum SLP ; 940 hPa, about a day before it made landfall over Vietnam

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4350 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 9. Time series of (a) maximum wind speed and (b) minimum SLP for observation (black), NC2011-CTRL (red), and CTRL (green) for two 48-h runs for Prapiroon initialized at 0000 and 1200 UTC 2 Aug 2006. Values are marked every 6 h.

2 on 1 October 2006. Typhoon Xangsane caused al- ;60 hPa and 35 m s 1 for Pmin and Vmax (Fig. 11, most 300 casualties and damages totaling about $700 green), respectively. In comparison, the simulated million (USD) in the Philippines and Vietnam. For errors in Vmax and Pmin in the NC2011-CTRL runs 2 Typhoon Xangsane, two 48-h runs, initialized at are about 15 hPa and 10 m s 1, respectively (Fig. 11, 0000 and 1200 UTC 29 September 2006 are per- red). formed for each type of model initialization schemes e. Mean error statistics for all four typhoons (Table 3). Figure 10 shows that NC2011-CTRL runs have sig- The mean error statistics of the four typhoons are nificant improvement in track simulations especially at computed from all 12 runs to investigate the performance early hours of simulations. The observed landfall point of the NC2011-CTRL runs in TC intensity and track is ;15.98N. The simulated landfall location in the simulations. The mean track errors in the NC2011-CTRL NC2011-CTRL runs is ;15.58N,whichisbetterthanin runs (Fig. 12a, red) are slightly smaller than those of the the CTRL run (;158N).ComparedtotheCTRLsim- CTRL runs (Fig. 12a, green) for all simulation periods. ulation, the NC2011-CTRL runs have better simula- The NC2011-CTRL runs show significant improvement tions for both Vmax (Fig. 11a)andPmin(Fig. 11b). At in intensity simulations. The mean absolute errors of 0000 UTC 30 September, Xangsane reached its second Vmax in the NC2011-CTRL runs (Fig. 12b, red) are al- peak intensity. Errors for CTRL runs at that time are ways smaller than in the CTRL runs (Fig. 12b, green) for

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4351

FIG. 10. Best track (black) and 48-h simulated tracks (red) for two 48-h runs for Xangsane initialized at 0000 and 1200 UTC 29 Sep 2006 for the (a) CTRL and (b) NC2011-CTRL runs. The TC tracks are marked every 6 h.

all simulated periods. For the simulated period up to 24 h, runs (Fig. 12c). Some of those runs occur after the storms 2 the errors in the NC2011-CTRL runs are about 6 m s 1 made landfall at 48 h of simulation. More cases with (Fig. 12b, red), which is significantly less than the results storms at various stages of their life cycles are needed to 2 for the CTRL runs (larger than 20 m s 1)(Fig. 12b, green) refine the model. for the same period. For mean absolute errors of Pmin, Model parameterization schemes introduce uncer- the NC2011-CTRL runs (Fig. 12c, red) are reduced by tainty in TC forecasting (Bao et al. 2012); therefore, we more than 60% when compared with the errors in the also investigated the effects of different precipitation CTRL runs (Fig. 12c, green) for simulated periods up to schemes on typhoon track and intensity forecasts ini- 24 h (Fig. 12c). For simulations from 30 to 48 h, the Pmin tialized by the modified NC2011 scheme. First we errors in the NC2011-CTRL runs (Fig. 12c,red)are performed an additional set of runs for these four smaller than those in the CTRL runs (Fig. 12c, green). storms using the same Grell–Devenyi cumulus pa- The exception is at 48 h where the Pmin errors in the rameterization scheme for the 18-km domain only but NC2011-CTRL runs are comparable to those in CTRL with the Eta microphysics option (Eta-GD). Though,

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4352 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 11. Time series of (a) maximum wind speed and (b) minimum SLP for observation (black), NC2011-CTRL (red), and CTRL (green) for two 48-h runs for Xangsane initialized at 0000 and 1200 UTC 29 Sep 2006. Values are marked every 6 h.

cumulus parameterization is not needed for high- the cumulus parameterization scheme is used for both resolution grids, our 6-km grid may not be adequate domains with Eta-microphysics (Eta-GDb), the simu- to resolve the convective process within clouds. In lated intensity is slightly weaker than the other two addition, the vertical motions on the 6-km grid may not sets of runs without cumulus parameterization for the be large enough to lift an air parcel to reach saturation 6-km domain (NC2011-CTRL and Eta-GD). Never- in a relatively short time during the initial stage of theless, the simulated track for Eta-GDb is slightly model simulations. With cumulus parameterization, better compared to the other two sets of runs without subgrid-scale precipitation is allowed to occur on the cumulus parameterization for the 6-km domain (Fig. 12c). grid points of the 6-km domain. We perform model For all three sets of model runs, it is apparent that in runs with the same combination of cumulus parame- addition to better initial TC structure and intensity terization and precipitation scheme (Eta-GD) but with (Chen et al. 2014), our initialization scheme will lead to the Grell–Devenyi cumulus parameterization scheme better TC simulations as compared with the CTRL for both the 18- and 6-km (Eta-GDb) in the last phase runs. of model configuration. The composite error statistics from all three sets of For the model runs using Eta microphysics (Eta- runs (Fig. 13) show improvements for both the track and GD), the storm intensity is slightly weaker than that in intensity simulations over the CTRL runs, especially the WSM6 (NC2011-CTRL) (Figs. 12b,c). However, the intensity simulations within the first 24–48 h before track simulations are slightly better (Fig. 12a). When making the landfall. In addition to uncertainties in the

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4353

We also test the performance of the new initialization scheme using the WRF Model for four typhoons that made landfall over southern China or Vietnam in 2006. Sensitivity tests show that the prescribed SLP has very little impact on the spinup process of the typhoon vor- tex. For an intense supertyphoon with Pmin as low as 916 hPa, the scheme can spin up the TC vortex to close to the best track data without prescribing SLP. These results attest to the importance of the storm environment on the TC intensity. Because the prescribed SLP is not needed, there is no need to specify the radius of the maximum wind in the original scheme developed by NC2011. Thus, the size and storm structure are determined mainly by the given large-scale conditions and the model employed. Our results show that a prescribed warm core for the first 10 cycles or so significantly reduces the number of cycles needed for the TC vortex to spin up to the best track intensity. The error statistics for these four landfalling typhoons in 2006, using the modified NC2011 scheme (NC2011- CTRL), are compiled and compared with those from the CTRL runs. A total of thirty-six 48-h simulations are performed for both sets of runs. Over the open ocean, typhoon intensity is usually estimated by satellite observations with some uncertainties. Despite these uncertainties, the modified NC2011 scheme shows re- markable skill in improving tropical cyclone simulations, especially intensity, in this data-sparse region without using in situ data, prescribed SLP, or the radius of max- imum wind to initialize the model. The key element for success of the NC2011 scheme is simply that the initial TC vortex is well adapted to the model employed with a TC intensity close to the best track data at the very beginning of the model integration. For all the cases considered, including those simulated by Nguyen and Chen (2011) [Morakot (2009), Jangmi (2008), and Kalmaegi (2008)], FIG. 12. Mean absolute errors for all four typhoons at different the NC2011 scheme’s simulations of intensity and track 21 forecast hours for (a) track (km), (b) maximum wind speed (m s ), are superior to the CTRL runs in which the initial vortex and (c) minimum SLP (hPa) for the CTRL (green), NC2011-CTRL (red), Eta-GD (blue), and Eta-GDb (black) runs. in the model needs to adjust to environmental conditions during the first 24–36 h of model simulations. We also show that with uncertainties in the precipitation physics precipitation processes used in the models, there are used in the models and uncertainties in the estimation of uncertainties in the estimation of storm intensity over storm intensity in the open ocean where conventional the open ocean where convection data are sparse. It data are sparse, a small ensemble using high-resolution appears that a small ensemble of high-resolution models models initialized by the modified NC2011 scheme will initialized by the modified NC2011 scheme will improve improve both the track and intensity simulations over the both the track and intensity simulations over the ocean ocean before TC making landfall. before TC landfall. Acknowledgments. This work was supported jointly by the Pacific Disaster Center, Kihei, Hawaii, and the 5. Summary and discussion National Oceanic and Atmospheric Administration In this study, we made improvements to the typhoon ini- (NOAA) Award NA11NOS0120039. We would also tialization scheme developed by Nguyen and Chen (2011). like to thank the University of Hawaii/Maui High

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4354 MONTHLY WEATHER REVIEW VOLUME 142

FIG. 13. Mean absolute errors from the WRF-GD, Eta-GD, and Eta-GDb ensemble mean (red) and CTRL run (green) for all four typhoons at different forecast hours for (a) track (km), 2 (b) maximum wind speed (m s 1), and (c) minimum SLP (hPa).

Performance Computing Center (UH/MHPCC) for REFERENCES providing part of the computing resources for this re- search;Profs.F.F.Jin,P-S.Chu,D.E.Stevens,K.F. Bao, J.-W., S. G. Gopalakrishnan, S. A. Michelson, F. D. Marks, and M. T. Montgomery, 2012: Impact of physics representa- Cheung, and two anonymous reviewers for their com- tions in the HWRFX on simulated hurricane structure and ments; C.-Y. Chen for his assistance; and May Izumi for pressure–wind relationships. Mon. Wea. Rev., 140, 3278–3299, editing the text. doi:10.1175/MWR-D-11-00332.1.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC NOVEMBER 2014 N G U Y E N A N D C H E N 4355

Betts, A. K., and M. J. Miller, 1993: The Betts–Miller scheme. Amer. Meteor. Soc., 9C.6. [Available online at https://ams. The Representation of Cumulus Convection in Numerical confex.com/ams/pdfpapers/168732.pdf.] Models, Meteor. Monogr., No. 46, Amer. Meteor. Soc., 107– Kurihara, Y., M. A. Bender, and R. J. Ross, 1993: An initialization 121. scheme of hurricane models by vortex specification. Mon. Wea. Chen, C.-Y., Y.-L. Chen, and H. V. Nguyen, 2014: The spin-up Rev., 121, 2030–2045, doi:10.1175/1520-0493(1993)121,2030: process of a cyclone vortex in a tropical cyclone initialization AISOHM.2.0.CO;2. scheme and its impact on the initial TC structure. SOLA, 10, Kwon, I.-H., and H.-B. Cheong, 2010: Tropical cyclone initial- 93–97, doi:10.2151/sola.2014-019. ization with spherical high-order filter and idealized three- Davidson, N. E., and H. C. Weber, 2000: The BMRC high-resolution dimensional bogus vortex. Mon. Wea. Rev., 138, 1344–1367, tropical cyclone prediction system: TC-LAPS. Mon. Wea. doi:10.1175/2009MWR2943.1. Rev., 128, 1245–1265, doi:10.1175/1520-0493(2000)128,1245: Lin, Y.-L., R. D. Farley, and H. D. Orville, 1983: Bulk parameteri- TBHRTC.2.0.CO;2. zation of the snow field in a cloud model. J. Climate Appl. Davis, C., and S. Low-Nam, 2001: The NCAR-AFWA tropical cy- Meteor., 22, 1065–1092, doi:10.1175/1520-0450(1983)022,1065: clone bogussing scheme. Air Force Weather Agency (AFWA) BPOTSF.2.0.CO;2. Rep., 12 pp. [Available online at http://sgi200.ust.hk/mm5/ Liu, Y., D.-L. Zhang, and M. K. Yau, 1997: A multiscale numerical mm5v3/tc-report.pdf.] study of Hurricane Andrew (1992). Part I: Explicit simulation Dudhia, J., 1989: Numerical study of convection observed during and verification. Mon. Wea. Rev., 125, 3073–3093, doi:10.1175/ the winter experiment using a mesoscale two- 1520-0493(1997)125,3073:AMNSOH.2.0.CO;2. dimensional model. J. Atmos. Sci., 46, 3077–3107, doi:10.1175/ Mlawer, E. J., S. J. Taubman, P. D. Brown, M. J. Iacono, and 1520-0469(1989)046,3077:NSOCOD.2.0.CO;2. S. A. Clough, 1997: Radiative transfer for inhomogeneous Dunnavan, G. M., and J. W. Diercks, 1980: An analysis of Super atmosphere: RRTM, a validated correlated-k model for the Typhoon Tip (October 1979). Mon. Wea. Rev., 108, 1915–1923, long-wave. J. Geophys. Res., 102, 16 663–16 682, doi:10.1029/ doi:10.1175/1520-0493(1980)108,1915:AAOSTT.2.0.CO;2. 97JD00237. Fujita, T., 1952: Pressure distribution within typhoon. Geophys. Monin, A. S., and A. M. Obukhov, 1954: Basic laws of turbu- Mag., 23, 437–451. lent mixing in the surface layer of the atmosphere (in Grell, G. A., and D. Devenyi, 2002: A generalized approach to Russian). Contrib. Geophys. Inst. Acad. Sci. USSR, 151, parameterizing convection combining ensemble and data as- 163–187. similation techniques. Geophys. Res. Lett., 29, doi:10.1029/ Nguyen, H. V., and Y.-L. Chen, 2011: High-resolution initialization 2002GL015311. and simulations of Typhoon Morakot (2009). Mon. Wea. Rev., Hawkins, H. F., and D. T. Rubsam, 1968: Hurricane Hilda, 1964. Mon. 139, 1463–1491, doi:10.1175/2011MWR3505.1. Wea. Rev., 96, 617–636, doi:10.1175/1520-0493(1968)096,0617: Noh, Y., W. G. Cheon, S.-Y. Hong, and S. Raasch, 2003: Im- HH.2.0.CO;2. provement of the K-profile model for the planetary boundary Holland, G. J., 1980: An analytic model of the wind and pressure layer based on large eddy simulation data. Bound.-Layer profiles in hurricanes. Mon. Wea. Rev., 108, 1212–1218, doi:10.1175/ Meteor., 107, 401–427, doi:10.1023/A:1022146015946. 1520-0493(1980)108,1212:AAMOTW.2.0.CO;2. Pu, Z. X., and S. A. Braun, 2001: Evaluation of bogus vortex Hong, S.-Y., and J.-O. J. Lim, 2006: The WRF single-moment 6-class techniques with four dimensional variational data as- microphysics scheme (WSM6). J. Korean Meteor. Soc., 42, 129–151. similation. Mon. Wea. Rev., 129, 2023–2039, doi:10.1175/ ——, J. Dudhia, and S.-H. Chen, 2004: A revised approach to ice 1520-0493(2001)129,2023:EOBVTW.2.0.CO;2. microphysical processes for the bulk parameterization of Rogers, E., T. Black, B. Ferrier, Y. Lin, D. Parrish, and G. DiMego, clouds and precipitation. Mon. Wea. Rev., 132, 103–120, cited 2001: Changes to the NCEP Meso Eta Analysis and doi:10.1175/1520-0493(2004)132,0103:ARATIM.2.0.CO;2. Forecast System: Increase in resolution, new cloud micro- Janjic, Z. I., 1994: The step-mountain eta coordinate model: Further physics, modified precipitation assimilation, modified 3DVAR development of the convection, viscous sublayer, and turbulence analysis. NWS Tech. Procedures Bull. 488, NOAA/NWS. closure schemes. Mon. Wea. Rev., 122, 927–945, doi:10.1175/ [Available online at http://www.emc.ncep.noaa.gov/mmb/mmbpll/ 1520-0493(1994)122,0927:TSMECM.2.0.CO;2. eta12tpb.] ——, 2000: Comments on ‘‘Development and evaluation of a convection Skamarock, W. C., J. B. Klemp, J. Dudhia, D. O. Gill, D. M. scheme for use in climate models.’’ J. Atmos. Sci., 57, 3686–3686, Barker, W. Wang, and J. G. Powers, 2005: A description of the doi:10.1175/1520-0469(2000)057,3686:CODAEO.2.0.CO;2. advanced research WRF version 2. NCAR Tech. Note Kain, J. S., 2004: The Kain–Fritsch convective parameteriza- NCAR/TN-4681STR, 88 pp. [Available online at http://www. tion: An update. J. Appl. Meteor., 43, 170–181, doi:10.1175/ mmm.ucar.edu/wrf/users/docs/arw_v2.pdf.] 1520-0450(2004)043,0170:TKCPAU.2.0.CO;2. Velden, C. S., B. M. Goodman, and R. T. Merrill, 1991: Western ——, and J. M. Fritsch, 1993: Convective parameterization for North Pacific tropical cyclone intensity estimation from mesoscale models: The Kain- Fritsch scheme. The Represen- NOAA polar-orbiting satellite microwave data. Mon. Wea. tation of Cumulus Convection in Numerical Models, Meteor. Rev., 119, 159–168, doi:10.1175/1520-0493(1991)119,0159: Monogr., No. 24, Amer. Meteor. Soc., 165–170. WNPTCI.2.0.CO;2. Kidder,S.Q.,M.D.Goldberg,R.M.Zehr,M.DeMaria,J.F.W. Wu, C.-C., and Coauthors, 2005: Dropwindsonde observations for Purdom, C. S. Velden, N. C. Grody, and S. J. Kusselson, 2000: typhoon surveillance near the Taiwan region (DOTSTAR): Satellite analysis of tropical cyclones using the Advanced Micro- An overview. Bull. Amer. Meteor. Soc., 86, 787–790, wave Sounding Unit (AMSU). Bull. Amer. Meteor. Soc., 81, 1241– doi:10.1175/BAMS-86-6-787. 1260, doi:10.1175/1520-0477(2000)081,1241:SAOTCU.2.3.CO;2. ——, K.-H. Chou, Y. Wang, and Y.-H. Kuo, 2006: Tropical cyclone Kucas, M. E., 2010: Challenges of forecasting tropical cyclone in- initialization and prediction based on four-dimensional vari- tensity change at the Joint Typhoon Warning Center. 29th ational data assimilation. J. Atmos. Sci., 63, 2383–2395, Conf. on Hurricanes and Tropical Meteorology, Tucson, AZ, doi:10.1175/JAS3743.1.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC 4356 MONTHLY WEATHER REVIEW VOLUME 142

——, Y.-H. Huang, and G.-Y. Lien, 2012: Concentric eyewall Zhao, Y., B. Wang, and Y. Wang, 2007: Initialization and simula- formation in Typhoon Sinlaku (2008). Part I: Assimilation of tion of a landfalling typhoon using a variational bogus mapped T-PARC data based on the ensemble Kalman filter (EnKF). data assimilation (BMDA). Meteor. Atmos. Phys., 98, 269– Mon. Wea. Rev., 140, 506–527, doi:10.1175/MWR-D-11-00057.1. 282, doi:10.1007/s00703-007-0265-4. Zhang, S., T. Li, X. Ge, M. Peng, and N. Pan, 2012: A 3DVAR- Zou, X., and Q. Xiao, 2000: Studies on the initialization and sim- based dynamical initialization scheme for tropical cyclone ulation of a mature hurricane using a variational bogus data predictions. Wea. Forecasting, 27, 473–483, doi:10.1175/ assimilation scheme. J. Atmos. Sci., 57, 836–860, doi:10.1175/ WAF-D-10-05066.1. 1520-0469(2000)057,0836:SOTIAS.2.0.CO;2.

Unauthenticated | Downloaded 10/03/21 03:33 AM UTC