Numerical Simulations of Myanmar Cyclone Nargis and the Associated Storm Surge Part I: Forecast Experiment with a Nonhydrostatic Model and Simulation of Storm Surge

Home , 1991 Bangladesh cyclone, Cyclone Sidr

Journal of the Meteorological Society of Japan, Vol. 88, No. 3, pp. 521--545, 2010. 521 DOI:10.2151/jmsj.2010-315

Tohru KURODA, Kazuo SAITO, Masaru KUNII

Meteorological Research Institute, Tsukuba, Japan

and

Nadao KOHNO

Japan Meteorological Agency, Tokyo, Japan

(Manuscript received 27 May 2009, in ﬁnal form 25 February 2010)

Abstract

Numerical simulations of the 2008 Myanmar cyclone Nargis and the associated storm surge were conducted using the Japan Meteorological Agency (JMA) Nonhydrostatic Model (NHM) and the Princeton Ocean Model (POM). Although the JMA operational global analysis (GA) and the global spectral model (GSM) forecast underestimated Nargis’ intensity, downscale experiments by NHM with a horizontal resolution of 10 km using GA and GSM forecast data reproduced the development of Nargis more properly. Sensitivity experiments to study the e¤ects of ice phase, sea surface temperature (SST), and horizontal resolutions to Nargis’ rapid development were conducted. In a warm rain experiment, Nargis developed earlier and the eye radius became larger. It was shown that a high SST anomaly preexistent in the Bay of Bengal led to the rapid intensiﬁcation of the cyclone, and that SST at least warmer than 29C was necessary for the development seen in the experiment. In a simulation with a horizontal resolution of 5 km, the cyclone exhibited more distinct development and attained a center pressure of 968 hPa. Numerical experiments on the storm surge were performed with POM whose horizontal resolution is 3.5 km. An experiment with POM using GSM forecast data could not reproduce the storm surge, while a simulation using NHM forecast data predicted a rise in the sea surface level by over 3 m. A southerly sub-surface current driven by strong surface winds of the cyclone caused a storm surge in the river mouths in southern Myanmar facing the Andaman Sea. Our results demonstrate that the storm surge produced by Nargis was predictable two days before landfall by a downscale forecast with a mesoscale model using accessible operational numerical weather prediction (NWP) data and application of an ocean model.

1. Introduction tion is important for preventing and mitigating meteorological disasters. In the areas around the Bay Severe meteorological phenomena such as tropi- of Bengal, historically, there have been several cal cyclones (TCs) sometimes cause catastrophic cases in which storm surges induced by TCs gave damage to human society; therefore, their predic- rise to severe floods (Obashi 1994). In cases such as the 1970 Bohla cyclone (Frank and Husain 1971) Corresponding author: Tohru Kuroda, Meteorological and the 1991 Bangladesh cyclone (Katsura et al. Research Institute, 1-1, Nagamine Tsukuba, Ibaraki 1992; Bern et al. 1993), cyclones generated in the 305-0052, Japan. E-mail: [email protected] central area of the bay moved northward and 6 2010, Meteorological Society of Japan made landfall in Bangladesh, and the associated 522 Journal of the Meteorological Society of Japan Vol. 88, No. 3 storm surges destroyed the lowlands of that coun- experiments, in which the JMA global spectral try. In 2007, cyclone Sidr struck the same area and model (GSM) forecast data are used as the lateral caused considerable destruction (MFDM Bangla- boundary conditions are conducted. Also, the re- desh 2008; Hasegawa et al. 2008). sults are compared with a reproduction experiment, In contrast with the above cases, cyclone Nargis in which global analyses (GA) are used as the lat- that was generated at the end of April 2008, moved eral boundary conditions instead of the GSM fore- eastward. On May 2, it made landfall in southern cast in order to observe the impact of the accuracy Myanmar during its strongest period and caused a of the lateral boundary value. destructive storm surge over the Irrawaddy Delta 2) To investigate the impact of the sea surface and other low-lying areas that claimed more than temperature (SST) and the physical process on the one hundred thousand lives (Webster 2008). For di- rapid development of Nargis: saster prediction in the areas mentioned above, SST is an important factor controlling the devel- forecasts of TCs and the associated storm surges opment of tropical cyclones. McPhaden et al. (2009) based on numerical weather prediction (NWP) are pointed out that there was a preexisting warm particularly important. anomaly of SST in the Bay of Bengal in late April Since 2007, a research project called ‘‘Interna- 2008 and inferred that this had contributed to the tional Research for Prevention and Mitigation of rapid intensification of Nargis. Lin et al. (2009) Meteorological Disasters in Southeast Asia’’ has determined that there were warm anomalies not been conducted by the Kyoto University, the Mete- only in the SST but also in the temperatures of orological Research Institute (MRI), and other in- the subsurface layer, and showed that this situation stitutes in Southeast Asian countries (Yoden et al. reduced the cyclone-induced ocean cooling by 2008; Koh and Teo 2009). The goals of this project using numerical experiments with a one-dimensional are to demonstrate the applicability of downscale ocean mixed layer model. However, neither Mc- NWP in Southeast Asia and to propose a decision Phaden et al. (2009) nor Lin et al. (2009) have con- support tool for preventing and mitigating meteo- ducted numerical simulation of Nargis using a full- rological disasters. From this point of view, we se- scale atmospheric model. In this paper, we examine lected the devastating disasters caused by Nargis as the impact of SST on Nargis’ development through one of the most important targets that we should sensitivity experiments using di¤erent SST datasets. study in this project. We assumed a minimum lead The impact of ice phase on Nargis’ development is time of two days before the landfall in order to ef- also examined, and the results are compared with fectively mitigate Nargis’ storm surge damage and the study by Sawada and Iwasaki (2007), in which set the initial time of our simulation as 12 UTC on simplified conditions including a horizontally uni- April 30, 2008, the time when Nargis started its form background were used. The impact of hori- eastward movement and one day before its rapid zontal resolution is also examined. These sensitivity development. Considering the project’s purpose experiments are not necessarily comprehensive, but and the real-time accessibility to data required for they give us information that help us to understand downscale NWP, the Japan Meteorological Agency the magnitude of the influence of the model un- (JMA) nonhydrostatic model (NHM) and the certainty with respect to the influence of the initial JMA’s operational global data are used as the fore- and boundary conditions on Nargis’ rapid develop- cast model and for obtaining the initial and/or ment. boundary conditions, respectively; the model and 3) To examine the predictability of storm surges the data are available and accessible to registered using downscale NWP and an ocean model: users in Southeast Asia. As for storm surges on the Bay of Bengal, In this paper, we conduct numerical simulation Flather (1994) studied storm surges associated with of Nargis and the associated storm surge for the the 1970 Bohla cyclone (Frank and Husain 1971) following purposes and scientific interests: and the 1991 Bangladesh cyclone using a numerical 1) To examine the predictability of Nargis two ocean model. However, this was a two-dimensional days before its landfall by downscale NWP using open sea model, with surface winds and pressures NHM and data available to Southeast Asian re- derived from a semi-analytical cyclone model using searchers: the best track data supplied by the US Navy Joint Considering the practical availability of the ex- Typhoon Warning Center (JTWC). periment in the case of real-time operation, forecast Recently, Kim et al. (2006) conducted numerical June 2010 T. KURODA et al. 523 simulation of the storm surge of Hurricane Ka- proached southern Myanmar. The minimum center trina, which damaged the city of New Orleans pressure estimated by JTWC was 937 hPa, while in the United States of America in 2005. The RSMC estimated its intensity as 962 hPa. The rain- simulation results obtained using a sophisticated fall rate observed by the Tropical Rainfall Measur- atmosphere-wave-ocean coupled model were in ing Mission’s Microwave Imager (TRMM/TMI) at agreement with the actual observations. However, 0137 UTC on May 2 is indicated in Fig. 2, which for achieving a practical disaster prevention, sim- depicts the typical structure of a developed cyclone pler surge predictions using a downscale mesoscale with a compact central dense overcast (CDO) and NWP and a one-way nested ocean model may be distinct spiral rainbands. After landfall at around more desirable, with application to the Bay of 09–12 UTC on May 2, the cyclone moved inland Bengal as an urgent subject (Dube et al. 2009). to the northeast, passing over southern Myanmar In this study, we conduct a numerical simulation and rapidly decayed. of the storm surge of Nargis, applying the Prince- 2.2 Storm surge of Nargis ton Ocean Model (POM) to the Bay of Bengal. The destructive damage in southern Myanmar The advantages of the NHM-simulated surface during the passage of Nargis was primarily caused winds and pressures over the GSM forecast will be by the storm surge, though the estimated maximum shown. wind speed exceeded 40 m/s. Since the river deltas This paper is organized as follows: Section 2 re- in southern Myanmar are low-lying, the storm views the characteristic features of Nargis and its surge reached inland several tens of kilometers associated storm surge. Section 3 presents JMA’s from the coastal areas facing the Andaman Sea global analysis and the performance of the JMA causing extensive floods. The shaded areas in Fig. GSM forecast. These data are used for obtaining 3 indicate the resultant water/wet regions or the the initial and boundary conditions of NHM. In vegetation loss. A field survey around the Yangon Section 4, we describe the numerical simulations of River was conducted by Shibayama et al. (2008). Nargis that were conducted using NHM. The sensi- The Yangon River has a wide mouth of about tivity of Nargis’ development to the SST, the ice 8 km, with the river becoming narrower on the up- phase, and the horizontal resolution are examined. stream side. Water level deviation due to the storm In Section 5, we describe the numerical simulations surge was estimated to be more than 3 m at a point of the storm surge that were carried out using fore- about 30 km upriver from the river mouth. Even casts from GSM and NHM. The advantages of around Yangon city, 40 km upstream from the the downscale high resolution simulation over the river mouth, a water level deviation of more than global model forecast are demonstrated. The sum- 1 m was reported. A numerical simulation for a mary and concluding remarks are given in Sec- particular point in the river (the Yangon point, in- tion 6. dicated by ‘‘Y’’ in Fig. 3) is presented later in this 2. Cyclone Nargis and storm surge study. In the Ayeyarwaddy district, including the region around the Irrawaddy River mouth (the Ir- 2.1 Characteristic features of cyclone Nargis rawaddy point, indicated by ‘‘I’’ in Fig. 3), a higher Cyclone Nargis, known as the ‘‘Myanmar Cy- storm surge might have occurred, but no detailed clone’’ was first generated as a tropical depression reports have been made yet. in the center of the Bay of Bengal and was detected as a tropical storm on April 27, 2008. After April 3. Global analysis and operational forecast of 29, it moved eastward as it developed and made JMA landfall in southwestern Myanmar at around 09– In Section 4, we will describe the downscale ex- 12 UTC on May 2 (Fig. 1a). periments conducted using NHM in which JMA’s Figure 1b presents the time sequence of the cen- operational global analysis and forecast data were ter pressure of Nargis as estimated by the Regional used as the initial and boundary conditions, respec- Specialized Meteorological Center (RSMC), New tively. Prior to discussing the experiments, we de- Delhi and JTWC. The cyclone was analyzed as a scribe the above data and observe how Nargis was tropical storm of around 970–980 hPa until 06 expressed in the NWP operation of JMA. UTC on May 1, and it developed rapidly after that. Nargis reached its maximum intensity of cate- 3.1 JMA global analysis gory 4 around 06–12 UTC on May 2, as it ap- The JMA global analysis is a 6-hourly analysis 524 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 1. a) Best track of cyclone Nargis. b) Time sequence of sea-level center pressure estimated by RSMC, New Delhi and JTWC. produced by the 8th grade Numerical Analysis and Kadowaki 2005) with 6 h assimilation windows Prediction System (NAPS) of JMA. An incremen- which assimilate conventional observation data (ra- tal four-dimensional variational analysis (4DVAR; dio sonde, surface, ship, buoy, and aircraft) and June 2010 T. KURODA et al. 525

Fig. 2. Rainfall rate (mm/h) observed by TRMM/TMI at 0137 UTC on May 2, 2008. After JAXA/EORC Tropical Cyclone Database (http://sharaku.eorc.jaxa.jp/TYP_DB/index_e.shtml). satellite data [NASA/NOAA TIROS Operational analysis procedures are the same but the resolutions Vertical Sounder (TOVS), QuikSCAT, the Mod- are di¤erent. In this study, we refer to the high res- erate Resolution Imaging Spectroradiometer olution global analysis as ‘‘GA’’, and the coarse (MODIS), the Multi-functional Transport Satellite mesh pressure plane data as ‘‘GA-p’’. (MTSAT) Cloud Motion Vector, etc.] is employed. Figure 4a shows the tracking of Nargis on the The resolutions are T159L60 (about 80 km) in the basis of the JMA global analyses from 12 UTC on inner model of the 4DVAR and TL959L60 (about April 30, to 06 UTC on May 2, 2008. In this figure, 20 km) in the outer model. For a typhoon in the the track of GA (thin solid line) is depicted after in- northwestern Pacific, typhoon bogus data are as- terpolation on Mercator grids with a resolution of similated, while no bogus data are used for others 10 km. The dotted line is the track of GA-p with including those in the Bay of Bengal. Further de- the original grids (1.25), while the broken line tails of the JMA global analysis are given by Narui shows a track that was interpolated on the grids (2007). with a horizontal resolution of 0.25 using the Bes- The global analysis data with two di¤erent reso- sel interpolation method. It is seen that the starting lutions are archived at MRI. The first is a high- points in both GA and the interpolated GA-p at 12 resolution analysis as the initial condition of the UTC on April 30 deviate by about 100 km east- operational global NWP at JMA with an original wardly from the best track (thick solid line). At 06 resolution of 20 km (0.1875 Gaussian grids) and UTC on May 2 (east end points), the eastward po- 60 model (h-) planes. The second, a coarse mesh sitional lags are smaller than those at 12 UTC on analysis with 1.25 (latitude-longitude grids) and April 30, but northward positional lags are seen in 11 pressure planes, has been more widely used by both GA and GA-p at most analysis times after 12 researchers due to its relative ease of handling. The UTC on May 1. 526 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 3. Storm surge-a¤ected areas in southern Myanmar observed by Terra SAR-X micro wave radiometer. The shaded areas indicate water/wet regions or vegetation losses and the two rectangles show footprints of Terra SAR-X on May 8, 2008. The Irrawaddy and Yangon points are indicated by the circled ‘‘I’’ and ‘‘Y’’, respectively. Source: the Information Technology for Humanitarian Assistance, Cooperation and Action (ITHACA; www.ithacaweb.org) in cooperation with the United Nations World Food Programme (WFP) and the German Aerospace Center (DLR).

The sea level pressure of the cyclone center repre- 3.2 Operational global forecast of JMA sented in GA and GA-p is seen in Fig. 4b. Center Forecasts made by the operational NWP at JMA pressures in GA are lower than those in GA-p at using a global model (GSM) on Nargis will now be almost all analysis times. This shows that GA can reviewed. GSM is a global spectral model of JMA represent the cyclone more properly than GA-p. with the world’s highest resolutions of TL959L60 Although both analyses capture the evolution of as an operational global model. Details of the Nargis to some extent, the represented center pres- model are given in Kitagawa et al. (2007). sure is quantitatively insu‰cient compared with the We used GRIB2 formatted GSM data distrib- central pressure obtained from the best track (Fig. uted by the Japan Meteorological Business Support 1b). The insu‰ciency of Nargis’ expression in the Center (JMBSC). The data comprise Grid Point JMA global analysis mentioned above is probably Values (GPVs) obtained every 6 h in the GRIB2 due to a lack of the TC bogus data and the short- format with a horizontal resolution of 0.5 and 17 age of assimilated observation data in the Bay of levels of pressure planes. The real-time data are dis- Bengal. seminated mainly to commercial users but archived June 2010 T. KURODA et al. 527

Fig. 4. a) Tracks of Nargis by the JMA global analyses from 12 UTC on April 30 to 06 UTC on May 2, 2008. The thin solid and dotted lines indicate the tracks by GA and GA-p, respectively, and the thick solid line indicates the best track. The broken line was depicted by the interpolation of GA-p. The triangles indicate the positions at 00 UTC on 1 and May 2 eastwardly on the track. b) Time sequence of center pressure of the cyclone. The solid and broken lines represent GA and GA-p, respectively. data are available to the research community. De- times from 00 UTC on April 30 to 00 UTC on pressions corresponding to Nargis in GSM fore- May 1 are plotted in Fig. 6a, and the eastward casts with four (12-hourly) initial times (00 UTC motion of the tracks is depicted in Fig. 6b. These and 12 UTC on April 30 and May 1) are depicted tracks were obtained from interpolated data with a in Fig. 5. Forecasts with later initial times tended horizontal resolution of 0.1 using Bessel interpola- to predict lower center pressures. This tendency tion. The forecasted center positions at 06 UTC on was probably due to the representation of the cy- May 2 deviate northwardly from the best track, clone at the initial conditions; the later the initial and the landfall times were earlier than those of times, the lower the center pressures that appear in the best track, except in the forecast with an initial the global analysis. The maximum decrease of time of 00 UTC on April 30. center pressures in the GSM forecasts was about As mentioned previously, GSM, one of the world 10 hPa at best. highest resolution operational global models, is be- Tracks of Nargis by GSM forecasts with initial lieved to have the ability to simulate some meso-b 528 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 5. Time evolution of sea-level cyclone center pressures by GSM forecasts. Broken line, thick solid line, thin solid line, and dotted line indicate the forecasts with initial times of 00 UTC on April 30, 12 UTC on April 30, 00 UTC on May 1 and 12 UTC on May 1, respectively. scale phenomena, but the forecasted minimum cen- 10 km as described by Saito et al. (2006; 2007). ter pressure of Nargis at an initial time of 12 UTC The Mellor-Yamada-Nakanishi-Niino’s level 3 on April 30 (two days before landfall) was around (MYNN-3) turbulent closure model developed by 992 hPa. It was di‰cult to foresee the cyclone’s cat- Nakanishi and Niino (2004) was implemented as astrophic disaster from this forecast. the first operational NWP model in the world (Hara 2008). Surface momentum, heat, and mois- 4. Numerical simulation with NHM ture fluxes over the sea were computed using Bel- 4.1 Numerical model and design of experiments jaars and Holtslag’s (1991) scheme. Here, we con- The JMA nonhydrostatic model (NHM) was sidered the wind to be at 10-m height as the used for performing the numerical simulations in surface wind, which is diagnosed from the lowest this study. The model was originally developed as level wind by the similarity law using a bulk mo- a community mesoscale model for research and mentum coe‰cient (see Eq. (4.5.89) in Japan Mete- weather forecasting by a collaboration between orological Agency 2007). MRI and the Numerical Prediction Division of In this study, we selected 12 UTC on April 30, JMA (Saito et al. 2001) and has been used for oper- 2008 as the initial time of our numerical simula- ation at JMA since September 2004 (Saito et al. tions with NHM. This initial time was about 48 h 2006). The horizontal resolution of operational before the landfall of Nargis and was chosen con- mesoscale forecasts has been enhanced from 10 km sidering the two day lead time required to issue cy- to 5 km since March 2006 (Saito et al. 2007). A clone or storm surge warnings. We paid attention three-ice bulk cloud microphysics scheme based on to the valid time of 06 UTC on May 2 (FT ¼ 42), Murakami (1990) that predicts cloud water, rain, just before landfall. In the latter half of this 42-h cloud ice, snow, and graupel, and a Kain-Fritsch period, Nargis rapidly developed from Category-1 convective parameterization scheme (Kain and to Category-4 as seen in Fig. 1b. Fritsch 1993) were included as the moist processes, NHM with a horizontal resolution of 10 km was whereas several points were modified for an opera- used in most experiments. Its domain was a square tional NWP with horizontal resolutions of 5 to of 3400 km size (1S–30N, 73E–107E) that cov- June 2010 T. KURODA et al. 529

Fig. 6. a) Cyclone tracks until 06 UTC on May 2 by GSM forecasts with various initial times. The points specify the locations of the cyclone center every 6 h and the triangles depict the 00 UTC positions during the forecast period. The thin solid line indicates the forecasted track whose initial time is 12 UTC on April 30. The initial times represented by the two dotted lines are 00 UTC and 06 UTC on April 30 (earlier than 12 UTC on April 30), and broken lines are 18 UTC on April 30 and 00 UTC on May 1 (later than 12 UTC on April 30). The thick solid line is the best track from 00 UTC on April 30. b) Eastward motion of cyclone tracks appeared in a). ers the Bay of Bengal and the surrounding area, States Geographical Survey (USGS) Global Land including Myanmar (Fig. 7). Forty-level terrain- Cover Characterization (GLCC) with 1-km hori- following hybrid coordinates were employed verti- zontal resolution. These high resolution data were cally, with vertical grid distances stretching from averaged and smoothed in the NHM grids. JMA 40 m near the surface to 1180 m at the model top global analysis and forecast data were used as ini- and with the lowest level located 20 m above tial and/or boundary conditions. No bogus data ground level. were used. A JMA global SST analysis with 0.25 The topography was obtained from the Global resolution and a JMA global land analysis with 30 Arc Second Elevation Data Set (GTOPO30). 0.1875 resolution were used with interpolation to Also, the land use data was based on the United the grid of the NHM experiment. SST and land 530 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 7. Domain used for NHM forecasts with 10-km horizontal resolution. The dotted square is the domain of nested experiments (NEST1 and NEST2 in Table 1), and broken rectangle indicates the domain of POM for storm surge simulations.

temperature at the initial time were used and were and boundary conditions is named ‘‘GAGA’’, and assumed to be constant throughout the forecast pe- the experiment using GA-p is ‘‘GAPGAP’’. riod until FT ¼ 72. The NHM experiments con- The cyclone tracks up until FT ¼ 42 (valid time ducted in this study are listed in Table 1. 06 UTC on May 2) predicted by the two experiments appear in Fig. 8a. The positional deviation 4.2 Numerical simulation using global analysis of GAPGAP (broken line) was remarkable in the data (reproduction experiment) northeast direction. The GAGA track (thin solid Prior to the forecast experiment with NHM us- line) was close to the best track (thick solid line), ing the JMA global analysis as the initial condition but the positions at each forecast time deviated and the JMA global forecast as the lateral bound- east-northeastward from the best track. As men- ary condition, we conducted numerical experiments tioned in previous sections, global analyses also using the global analysis data (GA and GA-p) as have east-northeastward positional lags (Fig. 4a), the initial and lateral boundary conditions to ob- and the GSM forecast for the same initial time had serve their quality and ability to reproduce Nargis’ an even larger positional lag (Fig. 6). The position track and intensity. The coarse mesh pressure plane at FT ¼ 42 was (16.4N, 95.0E) for GAGA and global analysis data (GA-p) are easy to handle and (16.0N, 95.5E) for GSM, and the positional lags have been widely used in the research community of GAGA from the best track was 150 km while in Japan, while we developed a new preprocessing that of GSM was 193 km. GAGA thus represents tool to use high resolution Gaussian grid model a better track than GSM. In GAPGAP, the cyclone plane data (GA) for numerical experiments with exhibited unnatural northward movement after NHM over Southeast Asia. As listed in Table 1, the initial start-up and the positional deviation at the reproduction experiment using GA as the initial FT ¼ 42 was very large in the northeast direction. June 2010 T. KURODA et al. 531

Table 1. List of numerical experiments. Kain–Fritsch convection parameterization is used except for NEST2. Horizontal Initial Boundary Minimum center Name resolution Initial time condition condition SST pressure GAGA 10 km 12 UTC 30 Apr GA GA JMA 976 GAPGAP 10 km 12 UTC 30 Apr GA-p GA-p JMA 976 GAGSM (CNTL) 10 km 12 UTC 30 Apr GA GSM JMA 974 WR 10 km 2 UTC 30 Apr GA GSM JMA 971 GAGSM_SST30 10 km 12 UTC 30 Apr GA GSM JMA 974 (Max. 30C) GAGSM_SST29 10 km 12 UTC 30 Apr GA GSM JMA 985 (Max. 29C) GAGSM_SSTN 10 km 12 UTC 30 Apr GA GSM NCEP 971 NEST1 5 km 00 UTC 01 May GAGSM GAGSM JMA 968 NEST2 3 km 00 UTC 01 May GAGSM GAGSM JMA 982

Its center position at that time was (17.6N, lated rainfall rates were 10 to 20 mm/h and were 95.3E), and the positional lag from the best track thus slightly lesser than those in the TRMM image, was 252 km. the characteristics of Nargis are well reproduced in The forecasted center pressures are given in Fig. the simulation as a whole. 8b. The minimum values for GAGA and GAP- GAP were almost the same (976 hPa), but the in- 4.3 Forecast experiment tensiﬁcation of the cyclone started earlier in GAGA In this section, we used the GSM forecast data than in GAPGAP. Both forecasts with NHM re- from JMBSC, described in Section 3.2, as the lat- produced the development of Nargis more accu- eral boundary condition. Considering the result rately than GA and GA-p themselves. in the previous section, we used a high resolution Next, we will compare the GAGA forecast with global analysis (GA) at 12 UTC on April 30 as the satellite observations. The TRMM/TMI-observed initial condition. The domain, resolution, and other 1-h rainfall rate at 0137 UTC on May 2, 2008 is experiment conditions were the same as those in the presented in Fig. 2. At this time, the cyclone center reproduction experiment. We refer to this forecast was located o¤ the coast of southwestern Myanmar experiment as GAGSM. (15.8N, 93.2E). The rainfall intensity was 10 to The cyclone tracks predicted by NHM 20 mm/h in the major part of the spiral bands with (GAGSM; thin solid line) and GSM (broken line) maximum values of 20 to 25 mm/h. The diameter are plotted in Fig. 10a. The predicted position of the CDO was less than 200 km, and a wide spi- (16.6N, 94.6E) at FT ¼ 42 in GAGSM again de- ral rainband adjoined the western part of the CDO. viates northwardly from the best track (thick solid Another distinct spiral band was seen south of the line), but its positional lag of 124 km was lower CDO. Figure 9a gives the rainfall rate predicted by than that in GSM (193 km). Positional lags in the GAGA at FT ¼ 31, when the position of the simu- initial and boundary conditions mentioned in Sec- lated cyclone center (16.0N, 93.1E) was close to tion 3 probably caused the positional lag in the the cyclone center seen in the TRMM image. The NHM forecast. predicted CDO was also compact, with a size com- The change in the center pressures over time parable to Fig. 2. There were two major spiral predicted by NHM (GAGSM) and GSM is illus- bands, one to the west and the other to the south trated in Fig. 10b. The minimum sea-level pressure of the cyclone center, while weak orographically- of GSM was 993 hPa and that of GAGSM was induced rains had already started in the coastal 974 hPa, which implied that NHM could predict areas of southern Myanmar. Although the simu- Nargis’ rapid development more accurately than 532 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 8. a) Tracks of Nargis by GAGA (thin solid line) and GAPGAP (broken line) and the best track (thick solid line). The points represent locations of the cyclone center every 6 h. Every track starts from 12 UTC on April 30 and is depicted until 06 UTC on May 2. The triangles indicate the positions at 00 UTC on May 1 and May 2 eastwardly on the track. b) Time sequence of cyclone center pressures obtained from GAGA (solid line) and GAPGAP (broken line). Initial time is 12 UTC on April 30, 2008.

GSM. The higher resolution of NHM contributed phoon stronger than GSM with the A–S scheme. to the better representation of the cyclone center. The maximum surface wind speed predicted by Another possible cause is the di¤erence in the phys- GSM was less than 20 m/s, while that predicted by ical processes, especially the convective parameter- GAGSM was more than 30 m/s (Fig. 10c). ization schemes. The Kain–Fristch (K–F) scheme The rainfall intensity at FT ¼ 33 is indicated employed in NHM tends to heat the atmosphere at with the cyclone center (16.4N, 93.2E) in Fig. 9b. lower levels than does the Arakawa–Schubert (A– The features are similar to those in GAGA (Fig. S) scheme in GSM. Recently, E. Shindo (private 9a), where Nargis’ observed characteristics (com- communication) has reported that NHM tends to pact CDO with spiral rainbands to the west and develop a typhoon more rapidly than GSM in its south) were simulated. Precipitation areas with early stage with the same horizontal resolution and rates of 5 to 20 mm/h in spiral bands were larger that GSM with the K–F scheme develops a ty- than in GAGA. June 2010 T. KURODA et al. 533

Fig. 9. a) Rainfall rate (mm/h) simulated by GAGA at FT ¼ 31. b) Same as in a) but GAGSM (CNTL) at FT ¼ 33. c) NEST1 at FT ¼ 21.

4.4 Sensitivity of Nargis’ development to ice phase tions including a horizontally uniform background. Experiment GAGSM in the previous subsection Since their simulation was performed with 2-km employed a three-ice cloud microphysics scheme resolution, convective parameterization was not with K–F convective parameterization. Sawada employed. In this subsection, we describe a warm and Iwasaki (2007) studied the impact of the ice rain experiment (WR) that we conducted and ex- phase on the typhoon development. They con- amine the sensitivity of the ice phase to Nargis’ de- ducted numerical experiments using NHM with velopment, and thus, confirm the results of the and without ice phase in cloud microphysics and study conducted by Sawada and Iwasaki (2007) in examined the results. In their study, the ice-phase a practical situation. process delays the intensification of the cyclone, Figure 11a plots the maximum azimuthally aver- though the maximum intensity is not very di¤erent. aged tangential surface wind velocity around the The radius of the eye becomes smaller with an ice cyclone center (averaged VT). In the WR (broken phase due to the weaker tangential wind. However, line), the averaged VT was larger than that in the their experiments were done with simplified condi- CNTL (¼ GAGSM) experiment with the ice phase 534 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 10. a) Predicted tracks of Nargis until FT ¼ 42 (06 UTC on May 2) by NHM (GAGSM, thin solid line) and GSM (broken line), and the best track (thick solid line). The triangles indicate the positions at 00 UTC on May 1 or May 2. b) Time evolution of sea-level center pressure of Nargis by NHM (GAGSM) and GSM forecasts. Initial time is 12 UTC on April 30, 2008. c) Maximum predicted surface wind speed by NHM (GAGSM) and GSM. June 2010 T. KURODA et al. 535

Fig. 11. a) Time sequence of the maximum averaged VT with the control run (CNTL) and the warm rain experiment (WR). b) Same as in a) but eye radius. Initial time is 12 UTC on Apr. 30, 2008. during most of the simulation period. This means in WR tended to be larger than that in CNTL that Nargis develops earlier without the ice phase during most of the period after the mature stage in the development stage. The peak value of the (FT ¼ 36). Again this tendency is consistent with maximum averaged VT in the warm-rain experi- that reported by Sawada and Iwasaki (2007). ment was comparable to that in CNTL. These re- However, in the developing stage, up until sults are consistent with those reported by Sawada FT ¼ 36, an opposite tendency is observed, i.e., and Iwasaki (2007). the eye radius in WR was smaller than that in Figure 11b indicates the radius that gives the CNTL. The earlier organization of WR represented maximum averaged VT. We refer to this radius as as a rapid intensiﬁcation in Fig. 11a may bring the ‘‘eye radius’’ for simplicity. The eye radius about this tendency. 536 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 12. a) Surface temperature by the JMA land and SST analyses at 12 UTC on April 30, 2008. b) Same as in a) but NCEP SST analysis. c) Di¤erence between the JMA SST and NCEP SST. d) Same as in a) but SST in GAGSM_SST29.

4.5 Impact of SST In order to observe the impact of SST, we In this subsection, we examine the impact of SST conducted another forecast experiment (GAGSM on Nargis’ development through sensitivity experi- _SSTN) by replacing the JMA SST with the ments that were conducted over a wide range of National Centers for Environmental Prediction SSTs. Figure 12a indicates the surface temperature (NCEP) SST (Fig. 12b; the horizontal resolution is at 12 UTC on April 30, 2008 as employed in the 1). The di¤erence between the JMA and NCEP control forecast experiment (GAGSM), in which a SST is evident in Fig. 12c, where higher SST areas JMA SST analysis with a horizontal resolution of are seen in the eastern and southern part of the Bay 0.1875 is used. High SST areas warmer than 30C of Bengal in the NCEP SST. are seen in the central and western parts of the Bay Thick (thin) solid lines in Fig. 13 indicate the of Bengal. McPhaden et al. (2009) reported this time evolution of cyclone center pressure pre- pre-existing warm anomaly in the SST in the Bay dicted by GAGSM (GAGSM_SSTN). The center of Bengal and suggested that it might have contrib- pressure of Nargis predicted by GAGSM_SSTN uted to the rapid intensiﬁcation of Nargis. reached 970 hPa. The warmer SST in the NCEP June 2010 T. KURODA et al. 537

Fig. 13. Time evolution of cyclone center pressures. The thick solid line shows the GAGSM (CNTL) case, and the thin solid line represents the case when NCEP SST is used (GAGSM_SSTN). The broken and dotted lines indicate the results of SST suppressed to 30C and 29C, respectively (GAGSM_SST30 and GAGSM_SST29). Initial time is 12 UTC on April 30, 2008. analysis resulted in greater intensification in conducted with a horizontal resolution of 5 km, ex- GAGSM_SSTN. periments with higher resolutions are worth con- To confirm the influence of the warm anomaly in ducting to observe the performance of NHM. Fore- SST in the Bay of Bengal, we prepared two artifi- cast experiments with horizontal resolutions of cial SSTs in which the JMA SST was suppressed 5 km (NEST1) and 3 km (NEST2) were conducted to 30Cor29C, and conducted additional sensitiv- using the GAGSM (CNTL) forecast at 00 UTC on ity experiments (GAGSM_SST30 and GAGSM May 1 as the initial condition and using 3-hourly _SST29). The sea level pressure of the cyclone cen- GAGSM forecasts as the boundary conditions. ter in these experiments is indicated in Fig. 13 by The domain is a square of about 2000 km size indi- the broken and dotted lines. GAGSM_SST30 ex- cated by a dotted line in Fig. 7 (5.4N–24.2N, hibits slower development and earlier decay, but 80.0E–100.0E). The physical processes in NEST1 the minimum cyclone center pressure (975 hPa) were the same as those in CNTL. NEST2 does not was similar to that of GAGSM. In GAGSM use convective parameterization; only cloud micro- _SST29, the development of Nargis was drastically physics was used for moist processes. suppressed, and the minimum center pressure was The track plotted by NEST1 was similar to that only 985 hPa. of GAGSM, as can be seen in Fig. 14a. The rainfall The above results indicate that the high SST rate (mm/h) by NEST1 at FT ¼ 21 (21 UTC on anomaly preexistent in the Bay of Bengal led to May 1) is illustrated with the cyclone center the rapid intensification of the cyclone and that an (16.4N, 93.2E) in Fig. 9c. Areas with rates greater SST over 29C at least was indispensable in this than 20 mm/h were seen not only in CDO but also forecast experiment. in spiral bands. The track plotted by NEST2 was similar to that of NEST1 until FT ¼ 18, but the 4.6 Experiments with higher resolutions eastward deviation from the best track became In previous subsections, we performed experi- larger after that, and the landfall was too early. ments with a horizontal resolution of 10 km. Since Time evolutions of the cyclone center pressures in the current operational mesoscale NWP at JMA is these experiments are presented in Fig. 14b. In 538 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 14. a) Tracks of the cyclone until 06 UTC on May 2 forecasted by GAGSM (CNTL, thin solid line), the 5-km nesting run (NEST1, broken line) and the 3-km run (NEST2, dotted line). The GAGSM track and the best track (thick solid line) start from 12 UTC on April 30, while tracks of NEST1 and NEST2 start from 00 UTC on May 1. The triangles indicate the positions at 00 UTC on May 1 or May 2. b) Time evolution of cyclone center pressures by GAGSM (solid line), NEST1, (broken line) and NEST2, (dotted line). Initial time of GAGSM is 12 UTC on Apr. 30, 2008, while initial time for NEST1 and NEST2 is 00 UTC on May 1.

NEST1, the simulated cyclone exhibited more dis- convective parameterization as mentioned by Noda tinct development than that in GAGSM, attaining and Niino (2003), and Lean and Clark (2003). 968 hPa around FT ¼ 36 even with JMA’s SST 5. Numerical simulation of storm surge analysis. This intensity was still weaker than that obtained in the JTWC analysis in Fig. 1a but was 5.1 Princeton Ocean Model and design of comparable to the estimate made by the RSMC experiment (972 hPa at 00 and 03 UTC and 962 hPa at 06 Storm surge simulations were performed using UTC). The reason for the positional lag and insuf- the Princeton Ocean Model (POM; Blumberg ficient development in NEST2 has not been investi- and Mellor 1987). POM is a free surface, three- gated fully. It seems that the fast motion of Nargis dimensional, community, general circulation ocean in NEST2 impeded the cyclone’s full organization model developed at Princeton University. Oceanic by its early landfall. We should note that the 3-km currents and water levels were calculated with horizontal resolution is too coarse to remove the sigma (terrain-following) coordinates using the sur- June 2010 T. KURODA et al. 539 face pressure and winds as the input data as de- interpolated independently. The computation do- scribed later. The bathymetry and topography data main of POM covered the Bay of Bengal, indicated were obtained from the National Geophysical Data by a broken rectangle in Fig. 7 (10N–23N, 84E– Center (NGDC) ETOPO2 databases of seafloor 99E). The horizontal resolution was 3.5 km, and and land elevations on a 2-min latitude-longitude 12 layers were used vertically. The layers were de- grid. The coastal boundary was assumed to be a fined by depth normalized by the seabed depth rigid wall where the land height in ETOTO2 is pos- (as 0, 0.016, 0.031, 0.062, 0.125, 0.250, itive and inundation was not considered. The mini- 0.375, 0.500, 0.625, 0.750, 0.875, 1.000). mum depth of the ocean was assumed to be 3 m in Though the thickness varies in the Bay of Bengal, order to prevent dry up; this assumption may sup- the di¤erences in POM level thickness had little ef- press the surge overestimation near the wall that fect on the result of the simulation because strong can be caused by the wind setup e¤ect since this ef- current is usually generated in shallow-water re- fect was inversely proportional to the water-depth. gions rather than in the deep sea during storm surge The maximum ocean depth was assumed to be phenomena. 1000 m, since the water depth a¤ects the wind 5.2 Storm surge simulation with the GSM forecast setup only in shallow-water regions and never alters First, we prepared a storm surge simulation us- the inverse barometer e¤ect in storm surge phe- ing the GSM forecast. Here, we pick up two nu- nomena. Faster horizontal oceanic motion was e‰- merical ocean grid points of POM, the Irrawaddy ciently simulated using vertically averaged current point (16.10N, 95.07E) and the Yangon point every time step (2-dimensional; external mode), (16.57N, 96.27E), indicated in Fig. 3. The Yan- with slower vertical motion calculated every 30 gon point corresponds to the location where the time steps (3-dimensional; internal mode). These as- Yokohama National University conducted a field sumptions enabled us to represent a storm surge in- survey along the Yangon River (Shibayama et al. cluding these major e¤ects with reasonable calcula- 2008). The Irrawaddy point is the location where tion cost. the maximum water-level deviation occurs in the The open-sea boundary was assumed to follow a NHM forecast, as discussed below. static balance with the atmospheric surface pres- Figures 15a–c plot the GSM-predicted surface sure, and deviations from the statically balanced wind and POM-predicted water level at the Irra- level caused inflow or outflow current and gravita- waddy point during the 72-h simulation. As seen in tional waves. Note that JMA has been operating Fig. 15a, the surface wind predicted by GSM was an original two-dimensional storm surge prediction weak (about 6 m/s). The wind direction (Fig. 15b) model (Higaki et al. 2009) that di¤ers from POM. changed from south to west before the wind speed In this study, 12 UTC on April 30, 2008 was tak- became maximum at around 06 UTC on May 2. en as the initial time, and the ocean model was ini- This clockwise change in the wind direction indi- tiated from a static state. The astronomical tide cated that the simulated cyclone passed north of was not taken into account, and thus, only the devi- the Irrawaddy point. The simulated maximum wa- ation of water level was computed with respect to ter level was about 0.7 m (Fig. 15c). At the Yangon the ocean’s vertical motion. Although the e¤ect of point, the wind direction (Fig. 15e) changed be- waves was not considered in order to save calcula- tween 06 UTC and 12 UTC with a maximum wind tion cost, the assumptions mentioned above can speed of about 6 m/s and water level of 0.5 m. su‰ciently represent major processes of a storm These simulated water levels were quantitatively surge. The momentum flux balance across the air- too small to foresee the storm surge disaster caused sea interface based on the law of the wall (see Eq. by Nargis. B2 of Appendix B in Mellor 2004) is achieved by setting the wind stress coe‰cient to 2:6 103 N/ 5.3 Storm surge simulation with the NHM m2. Input winds and pressures given by NWP in forecast NHM forecasts were then used as external driving Next, we used the NHM forecast (GAGSM) as forces. Sea level pressures were determined for the input data to the storm surge simulation by POM. input data every 10 min from GSM or NHM fore- The results for the two points specified in the previ- casts through temporally and spatially linear inter- ous section are depicted in Fig. 16. As seen in Figs. polation to the grids of POM. For a 10-m horizon- 16a and 16d, the surface wind speeds obtained were tal wind, the wind speed and direction are linearly much higher than those obtained in the GSM fore- 540 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Fig. 15. a) Time sequence of the surface wind speed by the GSM forecast at the Irrawaddy point (16.10N, 95.07E). b) Same as in a) but for wind directions. c) Same as in a) but water levels simulated by POM. d)–f) Same as in a)–c) but at the Yangon point (16.57N, 96.27E). cast, reaching 25 m/s at the Irrawaddy point and in Figs. 16c and 16f. At the Irrawaddy point, the 20 m/s at the Yangon point. The trend of the wind water level became highest when the southerly direction (Figs. 16b and 16e) was similar to that in wind was strongest. At 07 UTC on May 2 (FT ¼ the GSM case. Simulated water levels are depicted 43), the displacement of the sea surface level June 2010 T. KURODA et al. 541

Fig. 16. Same as in Fig. 15 but with the NHM forecast (GAGSM).

reached 3.2 m (Fig. 16c), which was the largest val- a height of several meters as mentioned above, ue both spatially and temporally in this simulation while the water level at the river mouth, closer to and was roughly of the same magnitude as the dis- the sea, was about 1 m. placement due to the storm surge at the Yangon At the Yangon point, a similar tendency was River reported by Shibayama et al. (2008). The wa- seen in the sense that the maximum water level ter level was quite high in the upriver area, reaching was observed at 10 UTC (Fig. 16f) just before the 542 Journal of the Meteorological Society of Japan Vol. 88, No. 3 wind speed reached the maximum value at 13 UTC (Fig. 16d) but the water level value (1.5 m) was somewhat lower than that at the Irrawaddy point (Fig. 16f), and the water level in the estuary was 1 m at that time. Figure 17 illustrates the displacement of the sea surface level simulated by POM at 00 UTC on May 2, 2008 (FT ¼ 36). At this time, the center of the simulated Nargis was located o¤ the west coast of southern Myanmar. The rise in sea level due to low pressure near the cyclone center (the inverse barometer e¤ect) is seen as a circular contour (Fig. 17a). In the enlarged view (Fig. 17b), we can see that a southerly ocean current generated by strong surface winds caused by the cyclone ﬂows into river mouths. The accumulated water brought about a rise in water level (the wind setup e¤ect) in the coastal region of southern Myanmar facing the An- daman Sea. Since the sea-level rise due to pressure depression was less than 0.5 m, the major part of the storm surge was caused by the ocean current generated by strong wind. The ocean current speed reaches the maximum value at the shallowest level while the speed decreases with depth. Our simulation suggests that the storm surge generated by Nargis was predictable two days before landfall using a downscale forecast with a high resolution regional model and the appropriate application of an ocean model. Though some errors still exist, the prediction was shown to be possible over- all using the resources introduced in this study. 6. Summary and concluding remarks The 2008 Myanmar cyclone Nargis was numeri- cally simulated with NHM using JMA global data. First, the quality of the JMA analysis data and the operational global forecast from GSM were exam- Fig. 17. a) Displacement of the sea surface ined. The JMA analyses captured the evolution level simulated by POM at 00 UTC on of Nargis to some extent, but the intensities were May 2, 2008 (FT ¼ 36). b) Enlarged view quantitatively insu‰cient and northeastward posi- (FT ¼ 36) depicts the beginning of the sea tional lags were seen in most analysis times before level rise (gray scale) with showing the landfall. The operational forecast of JMA using sea-level pressure (thick contour indicates GSM exhibits the same tendencies—weak inten- 1000 hPa and contour interval is 1 hPa) sity and northeastward positional lags. The GSM- and vertically averaged current (arrows, forecasted center pressure of Nargis with an initial m/s) which ﬂows into the river mouths in southern Myanmar. The triangle and circle time of two days before landfall was only 992 hPa indicate the Irrawaddy and Yangon points, and thus too weak to foresee the cyclone’s disas- respectively. trous impact. Downscale experiments by NHM with a horizontal resolution of 10 km were performed using pressure-plane global-analysis data (GA-p) were JMA global analyses to establish the initial and lat- used, the forecasted track had a large positional eral boundary conditions. When the low-resolution lag. With high-resolution global-analysis data June 2010 T. KURODA et al. 543

(GA), however, the track forecast was considerably sidering forecast errors and reliability. In Part 2 ameliorated. NHM reproduced the development of (Saito et al. 2010), we will perform ensemble pre- Nargis more properly than GSM and more accu- dictions of Nargis and the associated storm surge rately than even the global analyses themselves. to consider forecast errors due to uncertainties in A forecast experiment for Nargis with NHM was the initial and boundary conditions. Nargis’ inten- conducted using the GA and GSM forecast as the sity in the JMA global analysis was very weak, initial and boundary conditions. Despite the small which may be due to the lack of the TC bogus northward bias in the track forecast in NHM, data and the shortage of assimilated observation quantitatively better forecasts than GSM were sim- data in the Bay of Bengal. A data assimilation ulated, and a maximum surface wind speed of more study to improve the accuracy of the initial condi- than 30 m/s was obtained. The NHM-predicted cy- tions will be conducted by Kunii et al. (2010). clone exhibited characteristics similar to those of the TRMM/TMI satellite observation. Acknowledgements Experiments were conducted to study the sensi- This work was supported by the Ministry of tivity of Nargis’ rapid development to ice phase, Education, Culture, Sports, Science and Technol- SST, and horizontal resolution. In a warm rain ex- ogy in Japan (MEXT) and its Special Coordination periment, Nargis developed earlier and the eye ra- Funds for Promoting Science and Technology ‘‘In- dius became larger; these results were consistent ternational Research for Prevention and Mitigation with those of Sawada and Iwasaki’s (2007) ideal of Meteorological Disasters in Southeast Asia’’, experiments. It was demonstrated that a high SST represented by Professor Shigeo Yoden of Kyoto anomaly preexistent in the Bay of Bengal led to University. The authors are grateful to Mitsuru the rapid intensiﬁcation of the cyclone, and that an Ueno, Shunsuke Hoshino, Eiki Shindo, and Yoshi- SST over 29C at least was required in Nargis’ case. nori Shoji of MRI for their valuable comments and In a simulation with a horizontal resolution of information. Thanks are extended to two anony- 5 km, the cyclone showed more distinct develop- mous reviewers whose comments signiﬁcantly im- ment and attained a center pressure of 968 hPa, proved the quality of this paper. but a 3-km simulation without convective parameterization resulted in a larger positional lag. Numerical experiments on the storm surge were References performed with POM. Though the experiment us- Beljaars, A. C. M., and A. A. M. Hotslag, 1991: Flux pa- ing the GSM forecast could not represent the storm rameterization over land surfaces for atmospheric surge, the simulation using the NHM forecast pre- models. J. Appl. Meteor. 30, 327–341. dicted a storm surge of more than 3 m. A southerly Bern, C., J. Sniezek, G. M. Mathbor, M. S. Siddiqi, C. ocean current driven by the strong surface winds of Ronsmans, A. M. Chowdhury, A. E. Choudhury, the cyclone caused the disastrous storm surge at the K. Islam, M. Bennish, E. Noji, and R. I. Glass, 1993: Risk factors for mortality in the Bangladesh river mouths in southern Myanmar facing the An- cyclone 1991. Bulletin of WHO, 71, 73–78. daman Sea. Blumberg, A. F., and Mellor, G. L. 1987: A description Although our results demonstrated the predict- of a three-dimensional coastal ocean circulation ability of Nargis’ storm surge given a lead time of model. Three-Dimensional Coastal Ocean Models, two days, there were several quantitative discrepan- edited by N. Heaps, American Geophysical Union, cies between the forecast and the real situations 208 pp. involving the cyclone intensity, track, and timing. Dude, S. K., Indu Jain, A. D. Rao, T. S. Murty, 2009: For example, the storm surge at the Yangon River Storm surge modelling for the Bay of Bengal and was about 4 m (Shibayama et al. 2008b), while the Arabian Sea. Nat. Hazard, 51, 3–27. maximum level at the Yangon point in our simula- Flather, R. A., 1994: A storm surge prediction model for tion was 1.5 m. Errors in the initial and boundary the northern Bay of Bengal with application to the cyclone disaster in April 1991. J. Phys. Oceanogr., conditions and SST, as well as insu‰ciencies of the 24, 172–190. model resolutions and physics, cause forecast er- Frank, Neil L., and S. A. Husain, 1971: The Deadliest rors. Thus, if the northward bias of the TC track Tropical Cyclone in History. Bull. Amer. Meteor. predicted by NHM were reduced, a higher water Soc., 52, 438–445. level might have been simulated at the Yangon Hasegawa, K., and Investigation Team of Japan Society point. Risk management should be undertaken con- of Civil Engineering, 2008: Prompt report on the 544 Journal of the Meteorological Society of Japan Vol. 88, No. 3

Bangladesh Cyclone disaster, JSCE Magazine, 93, teractions during cyclone Nargis. Eos Trans. AGU, No. 3, 46–51, (in Japanese). 90, 54–55. Hara, T., 2008: Turbulent process. Suuchiyohoka Hou- Mellor, G. L., 2004: Users guide for a three dimen- koku Bessatsu, 54, 117–145, (in Japanese). sional, primitive equation, numerical ocean model. Higaki, M., H. Hayashibara, and F. Nozaki, 2009: Out- (Available online at http://www.aos.princeton.edu/ line of the Storm Surge Prediction Model at the WWWPUBLIC/htdocs.pom/PubOnLine/POL Japan Meteorological Agency. RSMC Tokyo— .htm) Typhoon Center Technical Review, No. 11, 25–38. Ministry of Food and Disaster Management, Bangladesh Japan Meteorological Agency, 2007: Meso-Scale Model Secretariat, Dhaka, Bangladesh, 2008: Super Cy- (JMA-MSM0603), Outline of the operational; Nu- clone Sidr 2007, Impacts and Strategies for Inter- merical Weather Predction at the Japan Meteoro- ventions. (Available online at http://www.cdmp logical Agency (available online at http://www .org.bd/reports/Draft-Sidr-Report.pdf) .jma.go.jp/jma/jma-eng/jma-center/nwp/outline- Murakami, M., 1990: Numerical modeling of dynamical nwp/pdf/pdf4/outline4_5.pdf). and microphysical evolution of an isolated convec- Kadowaki, T., 2005: A 4-dimensional variational assimi- tive cloud—the 19 July 1981 CCOP cloud—. J. lation system for the JMA Global Spectrum Model. Meteor. Soc. Japan, 68, 107–128. CAS/JSC WGNE Research Activities in Atmo- Nakanishi, M., and H. Niino, 2004: An improved spheric and Oceanic Modelling, 34, 1–17. Mellor-Yamada level 3 model with condensation Kain, J., and J. Fritsch, 1993: Convective parameteriza- physics: Its design and verification. Bound.-Layer tion for mesoscale models. Meteor. Monogr., 24, Meteor., 112, 1–31. 165–170. Narui, R., 2007: Global Analysis. Outline of the opera- Katsura, J., and Cyclone Disaster Research Group, 1992: tional numerical weather prediction at the Japan Storm surge and strong wind disaster due to 1991 Meteorological Agency. 18–28. (Available online cyclone in Bangladesh. Annuals of Disas. Prev. at http://www.jma.go.jp/jma/jma-eng/jma-center/ Res. Inst., Kyoto Univ., 35A, 119–159, (in Japa- nwp/outline-nwp/pdf/pdf3/outline3_5.pdf) nese). Noda, A., and H. Niino, 2003: Critical grid size for sim- Kim, K. O., H. S. Lee, M. Haggag, and T. Yamashita, ulating convective storms: A case study of the Del 2006: Storm surge field simulation on Hurricane City supercell storm, Geophys. Res. Lett., 30(16), Katrina using an atmosphere-wave-ocean coupled 1844, doi:10.1029/2003GL017498. model. Annual J. Coastal Eng., JSCE, 53, 416– Obashi, G. O. P., 1994: WMO’s Role in the International 420, (in Japanese). Decade for Natural Disaster Reduction. Bull. Kitagawa, H., K. Tamiya, M. Nakagawa, T. Komori, K. Amer. Meteor. Soc., 75, 1655–1661. Yamada, M. Hirai, K. Iwamura, and T. Sakashita, Saito, K., T. Kato, H. Eito, and C. Muroi, 2001: 2007: Global Spectral Model (JMA-GSM0603). Documentation of the Meteorological Research Outline of the operational numerical weather pre- Institute/Numerical Prediction Division unified diction at the Japan Meteorological Agency. 41– nonhydrostatic model. Tech. Rep. MRI, 42, 133 66. (Available online at http://www.jma.go.jp/jma/ pp. jma-eng/jma-center/nwp/outline-nwp/pdf/pdf4/ Saito, K., T. Fujita, Y. Yamada, J. Ishida, Y. Kumagai, outline4_2.pdf) K. Aranami, S. Ohmori, R. Nagasawa, S. Kuma- Koh, Tieh-Yong, and Chee-Kiat Teo, 2009: Toward a gai, C. Muroi, T. Kato, H. Eito, and Y. Yamazaki, Mesoscale Observation Network in Southeast Asia. 2006: The Operational JMA Nonhydrostatic Mes- BAMS, 90, 481–488. oscale Model. Mon. Wea. Rev., 134, 1266–1298. Kunii, M., Y. Shoji, M. Ueno, and K. Saito, 2010: Meso- Saito, K., J. Ishida, K. Aranami, T. Hara, T. Segawa, M. scale Data Assimilation of Myanmar Cyclone Narita, and Y. Honda, 2007: Nonhydrostatic at- Nargis Part I. J. Meteor. Soc. Japan, 88, 455–474. mospheric models and operational development at Lean, H. W., and P. A. Clark, 2003: The e¤ects of chang- JMA. J. Meteor. Soc. Japan, 85B, 271–304. ing resolution on mesocale modelling of line con- Saito, K., T. Kuroda, M. Kunii, and N. Kohno, 2010: vection and slantwise circulations in FASTEX Numerical Simulation of Myanmar Cyclone Nar- IOP16. Quart. J. Roy. Meteor. Soc., 125, 2255– gis and the associated storm surge Part 2: Ensem- 2278. ble prediction J. Meteor. Soc. Japan, 88, 547–570. Lin, I., C. Chen, I. Pun, W. Liu, and C. Wu, 2009: Warm Sawada, M., and Iwasaki T., 2007: Impacts of Ice Phase ocean anomaly, air sea fluxes, and the rapid inten- Processes on Tropical Cyclone Development. J. sification of tropical cyclone Nargis (2008). Geo- Meteor. Soc. Japan, 85, 479–494. phys. Res. Lett., 36, L03817. Shibayama, T., and Investigation Team of Japan Society McPhaden, M. J., G. R. Foltz, T. Lee, V. S. N. Murty, of Civil Engineering, 2008a: Investigation report M. Ravichandran, G. A. Vecchi, J. Vialard, J. D. on the storm surge disaster by Cyclone SIDR in Wiggert, and L. Yu, 2009a: Ocean-atmosphere in- 2007. Japan Society of Civil Engineering, 81 pp. June 2010 T. KURODA et al. 545

Shibayama, T., and Investigation Team of Japan Society Yoden, S., K. Saito, T. Takemi, and S. Nishizawa, 2008: of Civil Engineering, 2008b: Prompt report on the New Phase of International collaborative study to storm surge disaster by the Myanmar Cyclone. contribute to mitigation of meteorological disasters JSCE Magazine, 93, No. 7, 41–43, (in Japanese). in SouthEast Asia, Tenki, 55, 705–708, (in Japa- Webster, P. J., 2008: Myanmar’s deadly da¤odil. Nature nese). Geoscience, 1, 488–490.