Journal of the Meteorological Society of Japan, Vol. 88, No. 3, pp. 547--570, 2010.

547

DOI:10.2151/jmsj.2010-316

Numerical Simulation of Myanmar Cyclone Nargis and the Associated Storm Surge Part II: Ensemble Prediction

Kazuo SAITO, Tohru KURODA, Masaru KUNII Meteorological Research Institute, Tsukuba, Japan

and Nadao KOHNO Japan Meteorological Agency, Tokyo, Japan (Manuscript received 26 May 2009, in final form 24 February 2010)

Abstract A mesoscale ensemble prediction system (EPS) employing the Japan Meteorological Agency’s (JMA’s) highresolution global analysis and forecast for initial and boundary conditions of the control run and perturbations from JMA’s one-week global EPS for initial and boundary perturbations is developed and applied to numerical simulations of cyclone Nargis. Using the JMA nonhydrostatic model (NHM) with a horizontal resolution of 10 km, the system reproduces Nargis’ development and the associated storm surge in southwestern Myanmar with plausible ensemble spreads. In the ensemble prediction with initial boundary perturbations, predicted positions of cyclone centers are distributed in an elliptic area whose major axis is oriented east-northeast, suggesting that track forecast errors tend to increase in the moving direction of Nargis. The location of the minimum surface pressure of the ensemble mean is closer to the best track than the control run, and root mean square errors (RMSEs) of the ensemble mean against analyses are smaller than those of the control run in all forecast variables. However, ensemble spreads tend to decrease in the latter half of the forecast period, and the cyclone center does not disperse enough compared with the track forecast error without the lateral boundary perturbation. When lateral boundary perturbations are implemented in addition to the initial perturbations, dispersion of the cyclone center and spread of the center pressure increase by about 50% at forecast time (FT) ¼ 42. The location of the minimum surface pressure in the ensemble mean shifts westward, reducing the track error. RMSEs of ensemble means become smaller than the ensemble prediction without lateral boundary perturbations. Ensemble forecasts of storm surge were conducted using the Princeton Ocean Model (POM). When surface wind and sea level pressure from JMA’s global EPS were input, the maximum surge was no more than 0.6 m even in the highest ensemble member. The POM simulation driven by the mesoscale ensemble prediction with NHM predicted a storm surge near 4 m in southwestern Myanmar, where the timings of the peak surge were dispersed widely from FT ¼ 33 to FT ¼ 56. When the ensemble mean was input to POM, the maximum surge was 1.5 m, despite the better accuracy of the ensemble mean in terms of RMSE. This result shows that the scenario is more important than the ensemble mean when applying the mesoscale ensemble prediction to disaster prevention.

Corresponding author: Kazuo Saito, Meteorological Research Institute, 1-1, Nagamine Tsukuba, 305-0052, Japan. E-mail: [email protected] 6 2010, Meteorological Society of Japan

1. Introduction Meteorological phenomena that occur in Southeast Asia sometimes cause severe damage. In particular, tropical cyclones (TCs) and associated

548

Journal of the Meteorological Society of Japan

storm surges occasionally cause devastating disasters in the coastal regions facing the Bay of Bengal. The 1970 Bhola cyclone struck East Bangladesh and India’s West Bengal and claimed 500,000 lives, primarily as a result of the storm surge that flooded over the Ganges Delta (Frank and Husain 1971). The 1991 Bangladesh cyclone struck southeastern Bangladesh and forced a storm surge inland, killing at least 138,000 people (Obashi 1994). More recently, in 2007, Cyclone Sidr slammed the densely populated coastal areas of Bangladesh and was blamed for more than 3,000 deaths (Shibayama et al. 2008a). On 2 May 2008, cyclone Nargis made landfall in southwestern part of Myanmar and caused the worst natural disaster in the country, claiming more than 100,000 lives mainly due to a storm surge (Webster 2008). Flather (1994) studied storm surges on the Bay of Bengal by the 1970 Bohla cyclone and by the 1991 Bangladesh cyclone using a numerical ocean model. However, the model was a two-dimensional open sea model and surface winds and pressures were derived from a semi-analytical cyclone model using best track data supplied by the US Navy Joint Typhoon Warning Center (JTWC). The numerical weather prediction (NWP)-based storm surge prediction had not been applied to the Bay of Bengal (Dube et al. 2009). To prevent and mitigate the meteorological disasters in Southeast Asia, a three-year international research project supported by the Ministry of Education, Culture, Sports, Science and Technology in Japan (MEXT) Special Coordination Funds for Promoting Science and Technology was conducted from FY2007 in partnership with Kyoto University, the Meteorological Research Institute (MRI), and institutes in southeast Asian countries such as the Institut Teknologi Bandung (ITB) of Indonesia. One of the main goals of this project, International Research for Prevention and Mitigation of Meteorological Disasters in Southeast Asia1 CIQ is the utilization of probability information obtained by ensemble NWP and development of a decision support tool for disaster prevention. In conjunction with the above project, the authors recently conducted numerical simulations of Nargis and the associated storm surge using the Japan Meteorological Agency (JMA) nonhydrostatic model (NHM) and the Princeton Ocean model

1 http://www-mete.kugi.kyoto-u.ac.jp/project/MEXT/

Vol. 88, No. 3

(POM) (Kuroda et al. 2010; hereafter referred as Part 1). A storm surge of about 3.2 m was simulated in southern Myanmar, but the cyclone center lagged about 120 km in position and the predicted intensity of the cyclone was weaker than the analyses by the Regional Specialized Meteorological Center (RSMC) New Delhi and the Joint Typhoon Warning Center (JTWC). The magnitude of a storm surge highly depends on the track and intensity of TC, and NWP has unavoidable forecast errors due to the uncertainties of initial and boundary conditions and the model’s dynamics and physics. Considering the destruction caused by the storm surge, warnings should be issued and measures taken to prepare for the worst case scenarios. Ensemble prediction is one way to consider the errors in NWP, but currently most operational ensemble predictions systems (EPS) are based on global models whose horizontal resolutions are around 100 km. As described later, global EPSs are unable to predict storm surges if their results are used as input data for a dynamical storm surge model. Recently, Flowerdew et al. (2007) tested the performance of the mesoscale EPS of the United Kingdom Met O‰ce Global and Regional Ensemble Prediction System (MOGREPS; Boeler et al. 2008) on the storm surge prediction. They ran a barotropic storm surge model and estimated the risk of damaging events given the forecast uncertainties sampled by the ensemble. However, they used a North Atlantic and European domain with a 24 km grid length and did not target severe storm surges produced by tropical cyclones. Mesoscale ensembles have not been applied to predict storm surges by tropical cyclones over the Bay of Bengal and are seldom applied even in other regions. In this study (Part 2 of Kuroda et al. (2010)), we conduct a mesoscale ensemble forecast of cyclone Nargis using the JMA nonhydrostatic mesoscale model with a horizontal resolution of 10 km and simulate the associated storm surge. The purposes of this paper are as follows. 1) To develop a simple mesoscale EPS using a mesoscale atmospheric model and data available for Southeast Asian scientists to consider the forecast errors in NWP. 2) To compare results of the mesoscale EPS system with the deterministic downscale NWP conducted by Kuroda et al. (2010). The influence of the lateral boundary perturbation in the TC ensemble forecast is also examined. 3) To compare the results of the mesoscale EPS

June 2010

K. SAITO et al.

system using the global EPS and show the advantage of the downscale EPS using a high-resolution mesoscale model. 4) To conduct ensemble simulation of storm surges with a dynamical ocean model and show the feasibility of NWP to mitigate meteorological disasters in Southeast Asia considering forecast reliability and errors. Through the above simulations, we examine the possibility of issuing forecasts of storm surges two days before the landfall of the cyclone. The ensemble dataset produced in this study will be utilized as the input data for a prototype of a decision support system for disaster prevention in the aforementioned MEXT international research project. This paper is organized as follows. Section 2 briefly describes cyclone Nargis and the associated storm surge quoting from Part 1. Section 3 presents a mesoscale EPS using NHM as the forecast model with the rationale of using EPS for disaster prevention. Results of the JMA operational global EPS and mesoscale EPS are compared. The influence of lateral boundary perturbations on mesoscale EPS is examined, and the EPS performance is verified by checking statistical scores. Section 4 presents the results of ensemble prediction of storm surges with POM using the global and mesoscale EPS results as input data. Summary and concluding remarks will be given in Section 5. 2. Cyclone Nargis and its storm surge Cyclone Nargis, known as the ‘‘Myanmar Cyclone’’ developed on April 27 in the central area of the Bay of Bengal. It tracked slowly north-northwestward initially and moved eastnortheastward while developing after April 29 (Fig. 1). On May 2, Nargis attained its peak intensity of a weak Category 4 cyclone. The cyclone made landfall in southern Myanmar around 09–12 UTC May 2 and passed near the major city of Yangon. Southern part of Myanmar was severely damaged by a storm surge caused by Nargis. As depicted in the Terra SAR-X microwave image (Fig. 2), the storm surge flooded wide areas in the Aeyawaddy and Yangon Divisions of Myanmar. The most severe damage was at river mouths facing the Andaman Sea, as in the Irrawaddy and Yangon deltas. More detailed descriptions of Nargis’ characteristics, including the TRMM/TMI satellite imagery are given in Part 1.

549

Fig. 1. Track of Nargis from 00 UTC April 29, 2008 to 12 UTC May 3, 2008. Positions at 00 UTC are indicated by triangles. Broken rectangle shows the domain of Princeton Ocean Model.

3. Ensemble prediction 3.1 Rationale of using ensemble prediction for preventing meteorological disasters Accuracies of the operational NWP in world forecast centers have been considerably improved in recent years by advances in numerical modeling and data assimilation techniques and increases in computer powers. However, there are still many di‰culties in the predicting mesoscale severe phenomena with specifications of the intensity, location, and timing. Performance of the quantitative precipitation forecast is notoriously bad even in the latest mesoscale NWP if we focus on the statistical scores of intense rains. The TC track forecast is one of the most improved items in the recent NWP, while statistically there are still about 150 to 200 km errors in the 48-h track, even in the forecasts produced by the most advanced centers as ECMWF (Komori et al. 2007). Since NWP analysis techniques are theoretically based on the maximum likelihood estimation, the initial condition of numerical models inevitably includes certain analysis errors due to errors in the background field and observations. Local damage caused by TCs depends highly on the track of the cyclone. Winds in the right semi-

550

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 2. Storm surge a¤ected areas in southern Myanmar observed using Terra SAR-X microwave radiometer (shaded). Shaded areas indicate water/wet regions or vegetation losses and two rectangles show the footprints of Terra SAR-X on May 8, 2008. The Irrawaddy and Yangon points are indicated by circled ‘‘I’’ and Y CIQ 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).

circular area of a tropical cyclone are generally stronger than those in the left semicircular area. Thus, if a tropical cyclone moving northward landfalls in a coastal region in Southeast Asian countries facing the Pacific or Indian oceans, southerly strong wind in the right semicircular area is more likely to cause orographic heavy rains and/or storm surges. In Japan, Typhoon Vera (T5915) in 1959 caused a storm surge in the Bay of Isewan (central Japan) that claimed more than 5,000 lives when the typhoon made landfall to the west of the Bay of Isewan (Japan Meteorological Agency 1961). Considering the serious damage caused by storm surges, errors in the TC track forecast should be considered for taking proper measures and decision

making. To incorporate the error of the TC track forecast, JMA’s operational warning system employs a storm surge model with the following five scenarios (Higaki et al. 2009): (1) Center of the forecast error circle, (2) Rapid course, (3) Right course, (4) Slow course, and (5) Left course (Fig. 3). Surface winds in each scenario are determined by bogus data that employs the Fujita (1952) empirical formula and relationship of gradient wind2. However, the Fujita’s formula assumes a simple symmetric structure of the typhoon, and no oro2

In September 2007, a sixth scenario based on the JMA operational mesoscale model (MSM) forecast was added (Higaki et al. 2009).

June 2010

K. SAITO et al.

Fig. 3. Concept of ‘‘five scenarios’’ in Japan Meteorological Agency’s (JMA’s) operational storm surge waning system. Radius of forecast error circle is defined by the probability circle within which a tropical cyclone is forecasted to exist with a probability of 70% after Higaki et al., (2009).

graphic e¤ect is considered in computing surface winds. For more reliable risk management, a probabilistic forecast of storm surge based on the ensemble prediction is desired. 3.2 JMA one-week global ensemble prediction JMA has been operating a one-week EPS (hereafter, WEP) since 2001. The current WEP version is operated with 51 members produced by JMA’s TL319L60 global spectral model (corresponding horizontal resolution is about 60 km), whose initial perturbations are given by a global singular vector method of T63L40 (Sakai 2009)3. This study employs 6 hourly surface forecast data and 12 hourly pressure plane (11 levels) forecast grid point values (GPVs) with a horizontal grid distance of 1.25 degrees. These data are archived at MRI and can be

3 JMA began operating another typhoon EPS in 2008. This EPS employs a targeted global singular vector method with an 11 member TL319L60 GSM (Yamaguchi et al. 2009).

551

accessed by the research community as the data of the meteorology research consortium between JMA and the Meteorological Society of Japan (http:// www.mri-jma.go.jp/Project/cons/index.html); the original model plane data are not archived even at JMA. Figure 4a depicts the tracks of Nargis predicted by WEP until forecast time (FT) ¼ 42 whose initial time is 12 UTC April 30, 2008. Compared with the best track, the positions of the center of Nargis at the initial time in WEP are located about 120 km east-northeast. After the start-up, cyclone centers tend to move northward in the first six hours and then move eastward. The movement of the cyclone predicted in WEP is consistent with the JMA’s global analysis (see Fig. 3a in Part 1). Figure 4b presents the time sequence of center pressures of Nargis predicted by WEP. At the initial time (FT ¼ 0; 12 UTC April 30) center pressures of the cyclone were analyzed as weak depressions of about 1001 hPa. The control run indicated a weak development after FT ¼ 18, but its central pressure was only about 998 hPa even in the Nargis’s strongest period (FT ¼ 24–48). Some ensemble members exhibited more distinct development, but the lowest central pressure is 995 hPa, which is much weaker than the estimated intensity of Nargis (962 hPa in the RSMC New Delhi’s best track (dotted line) and 937 hPa in the JTWC’s best track; see Fig. 1b in Part 1). However, the cyclone did not develop in some ensemble members, and we could not specify their center positions after FT ¼ 36. Figure 7a presents positions predicted by WEP at FT ¼ 42 relative to the best track. Only detected members are plotted in this figure. The center position of the control run (^) is located 144 km eastnortheast of the best track. Since the initial position in the global analysis was located about 120 km east-northeast of the best track, a part of this positional lag is attributable to the error of analysis in the initial condition. Positions simulated by all members are distributed in a rectangle area oriented east to west and having long sides of about 4
in longitude and short sides of about 2
in latitude. Since position errors of TC track forecasts at FT ¼ 48 by major forecast centers are about 200 km over the western North-Pacific domain, as reported in the World Meteorological Organization (WMO) Working Group of Numerical Experimentation (WGNE) TC track forecast intercomparisons (Komori et al. 2007), the dispersion of the cyclone tracks in WEP is roughly comparable to the

552

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 4. a) Predicted tracks of Nargis until valid time 06 UTC May 2, 2008 by Japan Meteorological Agency’s one-week global ensemble prediction systems (EPS). Initial time is 12 UTC April 30, 2008 (FT ¼ 42). First 21 members with positive 10 members (p01–p10; broken line), negative 10 members (m01–m10; dotted line), and the control run (solid line) are shown. Positions at 00 UTC May 1 (FT ¼ 12) and 00 UTC May 2 (FT ¼ 36) are indicated with triangles, while the tracks of some members are not depicted after FT ¼ 36 because their central positions could not be determined. Best track is indicated by a thick solid line. b) Time evolution of the central pressures of Nargis predicted by global EPS. Central pressures are depicted until their central positions were retrievable. Control run is depicted by a thick line. The best track of the Regional Specialized Meteorological Center New Delhi is plotted by a dotted line.

track errors in the current NWP tropical cyclone forecast. The average position of the detected 17 detected members is shown by a cross (), and has a positional lag error of 126 km, slightly smaller than that of the control run. Overall, the forecast of Nargis in WEP basically followed JMA’s global analyses in terms of the track, and the magnitude of positional spread

seems acceptable compared with the statistical TC track forecast error. However, predicted intensities were clearly inadequate compared with the best track data. 3.3 Mesoscale ensemble prediction using NHM A mesoscale EPS is developed to consider errors in the forecast of the cyclone Nargis. The JMA

June 2010

K. SAITO et al.

NHM (Saito et al. 2006a; 2007) with a horizontal resolution of 10 km is employed as the forecast model. It covers the Bay of Bengal and its surrounding areas with 341  341 grid points (Fig. 6 in Part 1). Hybrid-vertical coordinates with 40 stretched levels are used, and the lowest level is located at 20 m AGL. A modified Kain-Fritsch convective parameterization and a three-ice bulk cloud microphysics scheme are used as the moist processes; these specifications are identical to the forecast experiment in Part 1. One of the simplest ways to conduct a highresolution ensemble numerical experiment may be downscaling WEP using a high-resolution mesoscale model. However, we did not take this option because the expression of Nargis in WEP is totally inappropriate for intensity both at the initial time and in the forecast. In this study, we regard ‘‘GAGSM’’ run in Part 1 as the control run of the NHM ensemble prediction. JMA’s high-resolution global model plane analysis (Narui 2007; 0.1875  0.1875
, 60 levels) at 12 UTC April 30, 2008 is used as the initial condition and the 6 hourly GSM forecast GPV (0.5  0.5
, 17 levels) supplied from the Japan Meteorological Business Support Center (JMBSC) is used as the boundary condition of the control run. Figure 5 illustrates preparation procedures for initial (upper figure) and boundary (lower figure) conditions in the mesoscale ensemble prediction. To provide the initial conditions of ensemble runs, perturbations from WEP are extracted by subtracting the control run forecast from the first 10 positive ensemble members in WEP. Since the highest level of the archived pressure plane forecast GPV of WEP is 100 hPa and is lower than the model top of the 40 level NHM (22.1 km@40 hPa), forecast GPVs of WEP are first interpolated to the 32 level hybrid NHM (NHM L32) model planes (model top is located at 13.8 km@160 hPa) and perturbations are extracted by subtracting the interpolated field of the control run from perturbed runs. The perturbations are then normalized and added to the initial condition of the control run of the 40 level hybrid NHM (NHM L40). To normalize the perturbation, normalization coe‰cients are determined so that root mean square values of the perturbations at each level do not exceed prescribed upper limits of the standard error of analysis (0.7 hPa for MSL pressure, 1.8 m/s for horizontal winds (U and V), 0.7 K for potential temperature, and 15% for relative humidity). These values are

553

the same as those used in the mesoscale ensemble experiments by Saito et al. (2006b) and Seko et al. (2009). In the highest eight levels of the NHM L40, the perturbation at the 32 nd level of the NHM L32 is extrapolated. At the kz th level, the perturbation amplitude is multiplied by f1  cosðp  ðkz  40Þ= 8Þg=2 so that the perturbation amplitude becomes zero at the model top of the NHM L40. Adding 10 negative members, a total of 20 initial perturbations is prepared, and the saturation adjustment is applied to the initial conditions of all perturbed members. In subsection 3.6, the influence of lateral boundary perturbations is examined. Similar procedures are conducted to prepare lateral boundary conditions for ensemble members. Since the WEP forecast GPV on pressure planes is archived only in 12 hourly, perturbations for the NHM L32 are extrapolated to L40 with space and interpolated to 6 hourly with time, and added to (or subtracted from) the 6 hourly NHM L40 lateral boundary conditions produced from the GSM forecast GPV by JMBSC (Fig. 5; lower). The saturation adjustment is applied to all perturbed lateral boundary conditions. Sea surface and ground temperatures are important lower boundary conditions in atmospheric models. However, for simplicity, sea surface and ground temperatures for the control run by the operational global land and sea surface temperature (SST) analyses of JMA at 12 UTC April 20 (see Fig. 12 of Part 1) are given for initial conditions of all ensemble members. Although the four layer soil temperatures are predicted over land in each model, SST is fixed, not predicted, in the simulation. 3.4 Lagged-ensemble with NHM Prior to presenting the results of the mesoscale ensemble prediction with initial perturbations described in the previous subsection, we first refer to the downscale lagged-ensemble with the 10 km NHM. Figure 6a plots the predicted tracks of Nargis until the valid time of 06 UTC May 2, 2008 predicted by the 10 km NHM with five initial conditions using 6 hourly JMA global analyses from 00 UTC April 30 to 00 UTC May 1. In this experiment, lateral boundary conditions were given by GSM forecasts whose initial times are identical to those of NHM; the result from 12 UTC April 30 corresponds to ‘‘GAGSM’’ in Part 1 and the control run in this paper. Among the five simulations, the Nargis’ position

554

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 5. Schematic chart of the preparation procedures for initial (upper figure) and boundary (lower figure) conditions in the mesoscale ensemble prediction.

June 2010

K. SAITO et al.

Fig. 6. a) Predicted tracks of Nargis until valid time 06 UTC May 2, 2008 by the 10 km nonhydrostatic model (NHM) using 6 hourly Japan Meteorological Agency’s global analyses from 00 UTC April 30 to 00 UTC May 1 as the initial condition. Track by initial time 12 UTC April 30 (control run) is indicated by a thin solid line. The tracks by initial times of 00 UTC and 06 UTC April 30 are shown by dotted lines, while the tracks by initial times of 18 UTC April 30 and 00 UTC May 1 are depicted by broken lines. Corresponding best track from 12 UTC April 30 is shown by a thick solid line. b) Time evolution of central pressures of Nargis predicted by NHM with various initial times.

555

556

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 7. Relative positions of Nargis at valid time 06 UTC May 2, 2008 against the best track data. Horizontal axis is the longitudinal position error in degree while vertical axis is the latitudinal error. a) Japan Meteorological Agency’s one-week ensemble prediction systems (WEP). Control run is depicted by a solid diamond. Positive members are indicated by triangles while negative members by rectangles. Average position is marked by a cross mark. Unspecified members (p01, p02, m05, and m09) are not plotted. b) Lagged ensemble from the initial times of 18 UTC April 30 to 00 UTC May 1. Control run (initial time at 12 UTC April 30) is depicted by a diamond. Initial times at 00 UTC and 06 UTC April 30 are indicated by triangles while initial times at 18 UTC April 30 and 00 UTC May 1 by rectangles. Average position is marked by a cross mark. c) Same as in a) but mesoscale ensemble with initial perturbations. The position of ensemble mean is indicated by an asterisk. d) Same as in a) but mesoscale ensemble with initial and lateral boundary perturbations.

at FT ¼ 48 predicted by NHM whose initial time is 06 UTC April 30 (a dotted line) is very close to the best track (a thick solid line). The NHM run with initial times after 12 UTC April 30 tended to predict the landfall of Nargis earlier than the best track. A similar tendency was seen in the GSM

forecast (Fig. 5 in Part 1). As a rule of thumb, position errors in the five forecasts above are large in the moving direction of Nargis (WSW to ENE direction), and their relative track errors are located along this direction (Fig. 7b). The average position with the lagged ensemble of five initial times is

June 2010

K. SAITO et al.

closer to the best track than the control run, and the position error is about 60 km. Figure 6b depicts the evolution of central pressures of Nargis predicted by NHM with various initial times. As indicated in this figure, NHMs with earlier initial times intensify Nargis more than those with later initial times; central pressure with an initial time of 00 UTC April 30 reached 968 hPa at 12 UTC May 2 (FT ¼ 60). This tendency di¤ers from the GSM forecast. In the case of GSM, as shown in Fig. 4 in Part 1, the center pressure of Nargis was predicted to be lower with later initial times. The reason for this discrepancy is not clear, but it may be due to the di¤erence of horizontal resolutions and physical processes of the two models. Figure 6c illustrates the evolution of central pressures of Nargis predicted by NHM as a function of the center position in longitude. The central pressure of Nargis in NHM became the lowest when its center reached 94
E. NHM could thus reproduce the development of Nargis up to a point if the forecast period is su‰cient. In GSM, the tropical cyclone develops more slowly and is weaker than that of NHM due to its coarser resolution (Fig. 4 in Part 1), and the initial intensity in the analysis field is relatively more important than that in NHM. 3.5 Ensemble prediction with initial perturbations In this subsection, we present the results of an ensemble prediction using NHM with initial perturbations. The initial time is 12 UTC April 30, 2008, as in Part 1, and the ‘‘GAGSM’’ run in Part 1 is regarded as the control run. Hereafter, we refer to this ensemble prediction as ‘‘WepNone’’ because the initial perturbations are given by WEP but no perturbations are applied to the lateral boundary condition. Figure 8a plots the tracks of Nargis until the valid time of 06 UTC May 2 (FT ¼ 42) predicted by the 10 km NHM ensemble with initial perturbations. Most ensemble members predicted center positions of the cyclone east of the best track, and the landfall times are earlier than those observed. One of the reasons for this error is the positional lag of the cyclone center at the initial time (FT ¼ 0). Predicted positions of the cyclone center at FT ¼ 42 are distributed in an elliptical area elongated in the moving direction (WSW to ENE; Fig. 7c), and all ensemble members except one made landfall on the west coast of southern Myanmar. In a rough estimate, the major axis of the ellipse is about 300 km,

557

while the minor axis is about 100 km. This means the average dispersion of cyclone centers in perturbed runs from that in the control run is mainly along the moving direction, which suggests that the landfall timing in the forecast of Nargis was relatively di‰cult to determine. The average of relative displacements of the cyclone centers from the control in ensemble members is about 90 km. Figure 8b indicates the time evolution of center pressures. Despite the small initial perturbation in pressure, predicted cyclone center pressures range from 972 to 982 hPa at FT ¼ 42. One member intensified Nargis up to 966 hPa at FT ¼ 38 (02 UTC May 2). This intensity is weaker than the JTWC’s best track data (941 hPa at 00 UTC and 937 hPa at 06 UTC May 2), but comparable to the best track data of RSMC New Delhi (Fig. 4b; 972 hPa at 00 and 03 UTC and 962 hPa at 06 UTC). Figures 9a (9b) indicates sea level pressure (threehour accumulated precipitation) at 06 UTC May 2 predicted by member p01 (m01). At this time (FT ¼ 42), Nargis is predicted to be at the Bay of Bengal near the west coast of southern Myanmar in p01 (Fig. 9a), while it has already made landfall and reached southern Myanmar near Yangon in m01 (Fig. 9b). Since in the control run Nargis is predicted to be near the west coast of southern Myanmar (Fig. 6a), these positions in the positivenegative pair of ensemble members are located symmetrically as compared to the control run. The center pressure of the cyclone is predicted to be lower in m01 (974 hPa) but weaker in p01 (978 hPa) at this forecast time. Figure 9c shows the ensemble mean sea level pressure. The minimum pressure is about 990 hPa on the west coast of southern Myanmar. Contours are elongated in the moving direction of Nargis (WSW-ENE), which reflects that the locations of Nargis’ center are dispersed in the ellipse depicted in Fig. 7c. Since the Nargis’ center in the control run was simulated to be inland of Myanmar (Fig. 6a), the position of the cyclone center in the ensemble mean is closer to the best track (indicated by a star). As shown in Fig. 7c, the average position is close to the control run, but the position of minimum pressure (*) is shifted further toward the best track. Figure 9d plots the ensemble spread of sea level pressure at FT ¼ 42. The largest ensemble spread is seen in southern Myanmar, and the spread has a dipole pattern where another maximum spread is

Journal of the Meteorological Society of Japan

558

Fig. 8.

Vol. 88, No. 3

Same as in Fig. 4 but by the 10 km nonhydrostatic model ensemble prediction (WepNone).

located o¤ the west coast of southern Myanmar. In Fig. 7c, there is a group of slowly moving cyclones west of the control run. The spread maxima over sea may correspond to this group. Figure 10a presents the time evolution of ensemble spreads of the predicted field (U, V, T, T-TD, and Z) at the 700 hPa level in the model domain. Ensemble spreads of all variables increase gradually until FT ¼ 42, while the spreads of Z (height field) and V decrease after FT ¼ 42 and FT ¼ 48, respectively. Spreads of T (temperature) and TTD (dew point depression) are almost constant during most of the forecast period. Magnitudes of ensemble spreads at FT ¼ 24 are about 1.2 m/s for U, 1.4 m/s for V, and 5.5 m for Z, while at FT ¼ 48 they are 1.4 m/s for U, 1.8 m/s for V, and 6 m for Z. The magnitudes of ensemble spreads are about

0.3 K for T and 2 K for TTD at FT ¼ 24 and at FT ¼ 48. Root mean square errors (RMSEs) of the control run and ensemble mean against analyses at FT ¼ 24 and FT ¼ 48 (initial conditions at 12 UTC May 1 and 12 UTC May 2) are presented by dark shaded bars and white bars in Fig. 13. RMSEs in ensemble means are smaller than those of the control run, which shows the properness of the ensemble forecast. However, the magnitudes of RMSE are larger than the ensemble spreads (e.g., the RSME of Z is about 11 m at FT ¼ 48 in Fig. 13b but the spread is 6 m as described above), which means that the ensemble spreads are underestimated. Figures 10b depicts the spreads of surface fields. Spreads of T and TTD exhibit distinct diurnal

June 2010

K. SAITO et al.

559

Fig. 9. a) Sea level pressure and 3 hour accumulated precipitation at 06 UTC May 2, 2008 (FT ¼ 42) by member p01 in the NHM ensemble prediction. b) Same as in a) but by member m01. c) Sea level pressure at FT ¼ 42 by ensemble mean. Position of the Nargis best track is shown by a star symbol. d) Ensemble spread of sea level pressure.

cycles but decrease with time after FT ¼ 18. Spreads of PSEA (surface pressure) and RR3H (three-hour accumulated precipitation) also reach maxima at around FT ¼ 18 and gradually decrease after that. 3.6

Ensemble prediction using initial and lateral boundary perturbations The previous subsection considered only initial perturbations, and no lateral boundary perturbations were given in the ensemble forecast. Since the influence of the lateral boundary condition propa-

gates inside the model domain with time (Saito et al. 2008), lateral boundary perturbations become important in regional ensemble prediction if the forecast period is long. Figure 11 illustrates tracks and evolution of center pressures of Nargis predicted by the NHM ensemble using initial and lateral boundary perturbations (hereafter, referred to as ‘‘WepWep’’). The center positions at FT ¼ 42 are distributed in an elliptical area with a major axis in the moving direction (Fig. 7d) but are obviously dispersed over a wider area than that of ‘‘WepNone’’ (Fig. 7c). The

560

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 10. a) Time series of ensemble spreads of U, V, T, T-TD, and Z at 700 hPa level by the nonhydrostatic model (NHM) ensemble prediction (WepNone). Numerals on the left axis are digits for U (m/s), V (m/s), T (K), and T-TD (K), while the right axis is for Z (m). b) Same as in a) but for surface field (U, V, T-TD, PSEA (hPa, left axis), and RR3H (%, right axis)). c) Same as in a) but the NHM ensemble prediction with initial and lateral boundary perturbations (WepWep). d) Same as in c) but for surface field.

major axis of the ellipse is about 400 km; the minor axis is about 200 km. Evolution of cyclone pressures also exhibits a larger spread, about 15 hPa in intensity, and the timing of minimum pressures ranges from FT ¼ 36 to FT ¼ 60. Another point we should note is that most cyclone tracks predicted by WepNone (Fig. 8) have northerly biases, but some members in WepWep depicted in Fig. 11 have no northerly biases. Figure 12a presents sea level pressure and RR3H at FT ¼ 42 predicted by members p01, and Fig. 12b presents that for member m01. In p01, Nargis is predicted to be in the Bay of Bengal o¤ the west coast of southern Myanmar (Fig. 12a), but in m01, the center is located over the Andaman Sea close to the southeast coast of southern Myanmar (Fig. 12b). These positions in the positive-negative pair

of ensemble members are located symmetrically with respect to the control run, but the distance is greater than the prediction in the previous subsection. Center pressures of the predicted cyclones are 976 hPa for p01 and 972 hPa for m01, and are similar to the ensemble prediction without lateral boundary perturbations. The minimum ensemble mean sea level pressure (Fig. 12c) is 992 hPa and located o¤ the west coast of southern Myanmar. This is closer to the best track (star; error distance is about 80 km) than that predicted by the ensemble without lateral boundary perturbations (Fig. 9c). Contours are elongated in the moving direction of Nargis, reflecting the dispersion of cyclone centers in the ensemble prediction. Figure 12d depicts the ensemble spread of mean

June 2010

K. SAITO et al.

561

Fig. 11. Same as in Fig. 8 but nonhydrostatic model ensemble prediction with initial and lateral boundary perturbations (WepWep).

sea level pressure. The maximum spread is not located in southern Myanmar but o¤ the west coast. Another local maximum is seen on the southeast coast of Myanmar. As a whole, the area of large spread extends over a greater area than that in Fig. 9d. It is interesting that the large spread area mainly extends to the western and southern sides compared with Fig. 9d, that is, the direction toward the best track position. If we examine each ensemble member forecast in detail, predicted positions of the cyclone center in members p02, m05, m09, and p10 were better than those in the control run,

but the intensities were weaker than those in the control run (figure not shown). Time series of ensemble spreads are plotted in Figs. 10c and 10d. Ensemble spreads at 700 hPa (Fig. 10c) increase gradually throughout the forecast period. Their magnitudes are 1.6 m/s for U, 1.8 m/s for V, and 6 m for Z at FT ¼ 24; at FT ¼ 48 these are 2.3 m/s for U and V, and 7 m for Z. Spread magnitudes become about 50% larger than those for the case without lateral boundary perturbations (Fig. 10a) at FT ¼ 72. RMSEs of ensemble means at FT ¼ 24 and

562

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 12. Same as in Fig. 9 but nonhydrostatic model ensemble prediction with initial and lateral boundary perturbations (WepWep).

FT ¼ 48 against the analysis are represented by lightly shaded bars in Fig. 13. RMSEs of ensemble means are smaller than those of the control run and also smaller than those for the case without lateral boundary perturbations (WepNone). Magnitudes of ensemble spreads are still smaller than RMSE but reach about 70% of RMSEs. These results suggests that the ensemble forecast is further improved by including lateral boundary perturbations and that the magnitudes of ensemble spreads in WepWep are plausible compared with the forecast errors (and statistical TC track errors of operational global forecasts in the Northwestern Pacific region). Surface fields spreads (Fig. 10d) are over 50%

larger at FT ¼ 72 than those for the case without lateral boundary perturbations (Fig. 10b), although they decrease slightly with time after FT ¼ 18. One reason for the underestimated ensemble spreads at the surface could be neglecting the lower boundary perturbation; the same SST was input to all ensemble members, and the SST is not predicted but fixed throughout the 72-hour simulation period. 4. Storm surge simulation 4.1 Design of experiments Storm surge simulations are performed using surface winds and pressure from ensemble predictions. The POM (Blumberg and Mellor 1987) with a hor-

June 2010

K. SAITO et al.

563

Fig. 13. a) Root mean square errors of U, V, T, T-TD, and Z at 700 hPa level at FT ¼ 24 against the analysis of 12 UTC May 1, 2008. From left to right, control run (dark shaded bar), ensemble mean without the lateral boundary perturbation (WepNone; white bar), and ensemble mean with the lateral boundary perturbation (WepWep; light shaded bar). b) Same as in a) but at FT ¼ 48 against the analysis of 12 UTC May 2, 2008.

izontal resolution of 3.5 km is used as in Part 1. Vertically 12 level sigma coordinates are employed. In this study, 12 UTC April 30, 2008 was taken as the initial time, and the ocean model was initiated from a static state. As the input data, 10 m horizontal winds and sea level pressures are given in every 10 minutes from GSM and NHM ensemble forecasts. The computation domain of POM covers the Bay of Bengal (84–99
E, 10–23
N), which is indicated by a broken rectangle in Fig. 1. Further details on POM are given in Part 1. 4.2

Ensemble prediction of storm surge using global EPS First, POM was driven by surface winds and pressure predicted by the global ensemble predic-

tion of Nargis (WEP). Figure 14 plots the time sequence of wind speeds, wind directions, and water levels predicted by WEP at the Irrawaddy point (16.10
N, 95.07
E; for the location, see Fig. 2) and the Yangon point (16.57
N, 96.27
E; Fig. 2). Maximum surface wind speed from the GSM control run at the Irrawaddy point is about 5 m/s at 12 UTC May 2 (FT ¼ 48). Maximum wind speed in all ensemble members is about 7.5 m/s (Fig. 14a). In Fig. 14b, wind directions of most ensemble members became clockwise, but became counterclockwise in two members. As depicted in Fig. 2, Nargis passed north of Irrawaddy point. However, the above changes of wind direction mean that in two of 21 ensemble members in WEP, the cyclone passed south of the Irrawaddy point. A small storm

564

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 14. a) Time sequence of wind speeds by the GSM ensemble prediction at the Irrawaddy point. Control run is depicted by a thick line. b) Same as in a) but for wind directions. c) Same as in a) but water levels simulated by Princeton Ocean Model. d)–f) Same as in a)–c) but at the Yangon point.

June 2010

K. SAITO et al.

surge was simulated by GSM winds. As indicated in Fig. 14c, the maximum water level is about 0.3 m in the control run and 0.6 m in all ensemble members. At the Yangon point, the maximum surface wind speed reached 7 m/s in the control run and 8.5 m/s at 06 UTC May 2 (FT ¼ 42) in all ensemble members (Fig. 14d). Wind directions (Fig. 15e) suggest that the cyclone simulated in four members passed south of the Yangon point. Despite relatively stronger surface wind speeds, the maximum water level at the Yangon point (0.4 m) is lower than that at the Irrawaddy point (Fig. 14f). 4.3

Ensemble prediction of storm surge using mesoscale ensemble prediction Next, we present the results of the POM simulation using the mesoscale ensemble prediction. The ensemble prediction with initial and lateral boundary perturbations described in subsection 3.6 (WepWep) provides surface winds and pressure that drive POM. Figure 15 plots the time sequence of wind speeds, wind directions, and water levels at the Irrawaddy and the Yangon points. At the Irrawaddy point, the maximum surface wind (25 m/s) was attained by the control run (Fig. 15a). Several members predicted strong winds exceeding 20 m/s, but the timings of the strongest wind are dispersed within 30 hours from 20 UTC May 1 (FT ¼ 32) to 02 UTC May 3 (FT ¼ 62). Wind directions (Fig. 15b) in most members were southerly until 20 UTC May 1 (FT ¼ 32) and changed to westerly after 17 UTC May 2 (FT ¼ 53), suggesting that the simulated cyclones passed north of the Irrawaddy point (except in two members). The water level at the Irrawaddy point reached 3.2 m at 07 UTC (FT ¼ 43) in the control run. As indicated by streamlines in Figs. 16a and 16b, southwesterly strong wind in the right semicircle of Nargis blew toward the mouth of the Irrawaddy River, which opens in the southsouthwest direction. Two members predict high water levels near 4 m at FT ¼ 33 and FT ¼ 37 (solid triangles in Fig. 15c), while two other members simulated 3.1 m storm surges at FT ¼ 45 and FT ¼ 56 (shown by open triangles). It should be noted that the latter simulation was by member m05, where the simulated cyclone passed south of the Irrawaddy point (Figs. 16c and 16d). This storm surge is caused by southerly winds in front of the cyclone, and it means that the storm surge may occur even in the left semicircle of the cyclone

565

if the cyclone center passes close to a point. At the Yangon point, the maximum wind speeds were about 20 m/s (Fig. 15d), but the simulated storm surges were relatively lower than those of the Irrawaddy point, where the maximum water levels were 1.5 m in the control run and about 2.5 m in all ensemble members (Fig. 15f). These water levels are somewhat lower than the estimated water level (about 3 m) in the Yangon River (Shibayama et al. 2008b; for details, see Part 1) for the control run but almost comparable in the maximum member. In Fig. 15d, wind speeds in some members have sharp minima on May 2, corresponding to passage of the cyclone’s ‘‘eye’’ PIQ Wind directions (Fig. 15e) changed at later timings than for the Irrawaddy point. In some members, water levels in Fig. 15f became negative due to the northerly wind. Figure 17 presents a time sequence of wind speeds, wind directions, and water levels simulated by POM using ensemble mean values of WepWep. As indicated in Fig. 13, the accuracy of the ensemble mean is superior to that of the control run in terms of RMSE. However, if the ensemble mean surface wind and pressure were given as input data for POM, the maximum water levels were about 1.5 m at the Irrawaddy point and less than 1 m at the Yangon point. This underestimation of storm surge is due to the relatively weaker surface winds (12 m/s at the Irrawaddy point and 10 m/s at the Yangon point) in the ensemble mean. This result suggests that the local structures of surface winds, which are o¤set in the ensemble mean, are important for reflecting realistic scenarios. From the time sequences plotted in Fig. 15c, we can compute the maximum, minimum, and center magnitudes of water levels. If we assume equal weight for all ensemble members, we can compute 25% and 75% probability values from the number of members that exceed the corresponding thresholds (Fig. 18). We can see that at the Irrawaddy point, a storm surge of about 1.8 m is expected with a probability exceeding 50%, and 2.2 m with a probability of about 25%. In the worst cases, water levels may reach about 4 m. We can notice that the possibility of the peak water level becomes maximum around FT ¼ 42, but the highest water level may occur early or late such as FT ¼ 33 or FT ¼ 60. This kind of figure is obtained only using the high resolution ensemble prediction and gives important information on forecast errors and reliability for e¤ective risk management.

566

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 15. Same as in Fig. 14 but for the nonhydrostatic model ensemble prediction with initial and lateral boundary perturbations (WepWep). Ranges of vertical axes in wind speed and water level are di¤erent from Fig. 14.

June 2010

K. SAITO et al.

567

Fig. 16. Stream lines of surface wind by nonhydrostatic model given to Princeton Ocean Model. a) Control run, at FT ¼ 30. b) Control run at FT ¼ 42. c) Member m05 at FT ¼ 42. d) Members m05 at FT ¼ 56.

5. Summary and concluding remarks Mesoscale ensemble forecast experiments of cyclone Nargis and the associated storm surge were conducted. JMA’s operational global analysis and forecast of high resolution GSM were used for initial and boundary conditions for the control run, while JMA’s global one-week EPS was used to make initial and boundary perturbations. Although

Nargis’ intensity in the JMA’s global one-week EPS is too weak, an ensemble forecast using NHM with a horizontal resolution of 10 km reproduces Nargis’s development up to a point. The predicted positions of cyclone centers were distributed in an elliptical area. This means that in Nargis’ case, forecast errors tend to increase in the moving direction, and this result is consistent with the tendency of the lagged ensemble.

568

Journal of the Meteorological Society of Japan

Vol. 88, No. 3

Fig. 17. Same as in Fig. 15 but by ensemble mean.

When only initial perturbations are given in the mesoscale ensemble prediction, the dispersion of cyclone positions at FT ¼ 42 was around 90 km from the control run, which is smaller than statisti-

cal track errors at FT ¼ 48 of typhoon over Northwestern Pacific predicted by the JMA global forecast and smaller than the forecasted track error. The minimum surface pressure in the ensemble

June 2010

K. SAITO et al.

Fig. 18. Time sequence of the maximum, minimum and center magnitudes of tide levels at the Irrawaddy point. Widths between 25% and 75% probability values are depicted with solid rectangles, whose upper and lower sides correspond to 25% and 75%, respectively.

mean is located closer to the best track than the control run. Although ensemble spreads decrease in the latter half of the forecast period, RMSEs of ensemble mean are smaller than those of the control run. When lateral boundary perturbations are implemented, the dispersion of the cyclone center increases, and the spread of center pressures also increases by 50%. This magnitude of track dispersion is comparable with the forecasted track error. The minimum surface pressure in the ensemble mean is shifted westward, reducing the track error from the best track. Ensemble spreads also increase by 50% and the RMSEs of ensemble mean against analyses become smaller than those predicted by the ensemble without the lateral boundary perturbations. Ensemble forecasts of storm surge were conducted using the POM. When surface wind and sea level pressure from the global EPS were given as the input data, the maximum storm surge was no more than 0.6 m even in the highest member. This result means that the current global ensemble forecast is totally inadequate for the purposes of quantitative disaster prevention. The POM simulation driven by the mesoscale ensemble prediction using the NHM reproduced storm surges near 4 m at the Irrawaddy point, where the timings of the highest water level dispersed widely from FT ¼ 33 to

569

FT ¼ 56. When wind and pressure predicted by the ensemble mean were used as the input data for POM, the maximum water level was 1.5 m at the Irrawaddy point, despite the better accuracy of ensemble mean in terms of the RMSE. Our results indicated that countermeasures relying only on a single deterministic forecast are inadequate and often dangerous. They also suggest that countermeasures should not be taken for the ensemble mean field but for each possible scenario. Quantitative information on forecast errors and reliability based on the high resolution ensemble prediction are very important for e¤ective risk management and will become necessary in the future disaster mitigation systems. Our results are applicable to decision making tools as the input data in the research project for the prevention and mitigation of meteorological disasters in Southeast Asia, where a prototype version using a web-based tool (Gfdnavi; http://www.gfd-dennou.org/arch/davis/ gfdnavi/index.en.htm) is under development at Kyoto University. Although the mesoscale ensemble forecast presented in this study reproduced the storm surge in southern Myanmar (Irrawaddy and Yangon points), simulated water level was still lower than the estimated level. Furthermore, the simulated intensity of the cyclone was somewhat weaker than that of the best track analysis by JTWC. To reduce this underestimation of the cyclone intensity, e¤orts should be made in the area of data assimilation. Assimilation studies for Nargis using the meso 4DVAR have been conducted by Kunii et al. (2010) and Shoji et al. (2010). Data assimilation and ensemble prediction of Nargis using LETKF are also underway by the authors’ group. Acknowledgments A part of this work was supported by the MEXT Special Coordination Funds for Promoting Science and Technology ‘‘International Research for Prevention and Mitigation of Meteorological Disasters in Southeast Asia’’ represented by Professor Shigeo Yoden of Kyoto University. The authors are grateful to Munehiko Yamaguchi, Masayuki Kyouda, and Ryota Sakai of JMA for their help in using JMA’s global EPS data. We also thank Mitsuru Ueno and Yoshinori Shoji of MRI and Masakazu Higaki of JMA for their comments and information. Thanks are extended to two anonymous reviewers, whose comments considerably improved the quality of this paper.

Journal of the Meteorological Society of Japan

570

References Blumberg, A. F., and G. L. Mellor, 1987: A description of a three-dimensional coastal ocean circulation model. Three-Dimensional Coastal ocean Models, American Geophysical Union., 208 pp. Bowler, N. E., A. Arribas, K. R. Mylne, K. B. Robertson, and S. E. Beare, 2008: The MOGREPS short-range ensemble prediction system. Quart. J. Roy. Meteor. Soc., 134, 703–722. Dude, S. K., Indu Jain, A. D. Rao, and T. S. Murty, 2009: Storm surge modelling for the Bay of Bengal and Arabian Sea. Natural Hazards, 51, 3–27. Flather, R. A., 1994: A storm surge prediction model for the northern Bay of Bengal with application to the cyclone disaster in April 1991. J. Phys. Oceanogr., 24, 172–190. Frank, N. L., and S. A. Husain, 1971: The Deadliest Tropical Cyclone in History. Bull. Amer. Meteor. Soc., 52, 438–445. Fujita, T., 1952: Pressure distribution within typhoon. Geo. Phys. Mag., 23, 437–451. Higaki, M., H. Hayashibara, and F. Nozaki, 2009: Outline of the Storm Surge Prediction Model at the Japan Meteorological Agency. RSMC Tokyo— Typhoon Center Technical Review, No.11, 25–38. Flowerdew, J., K. Horsburgh, and K. Mylne, 2007: Ensemble forecasting of storm surges. Marine Geodesy, 32, 91–99. Japan Meteorological Agency, 1961: General investigation report of Isewan Tyhoon. JMA General Investigation Report, 7, 899 pp, (in Japanese). Komori, T., M. Yamaguchi, R. Sakai, and Y. Takeuchi, 2007: WGNE Intercomparison of tropical cyclone forecasts with operational global models: Quindecennial Report. Science Highlights, WCRP. (available online at http://wcrp.wmo.int/documents/ WGNE_TC_Intercomparison_Quindecennial _Quicklook15Anniversary.pdf) Kunii, M., Y. Shoji, M. Ueno, and K. Saito, 2010: Mesoscale Data Assimilation of Myanmar Cyclone Nargis. J. Meteor. Soc. Japan, 88, 455–474. Kuroda, T., K. Saito, M. Kunii, and N. Kohno, 2010: Numerical simulations of Myanmar cyclone Nargis and the associated storm surge Part I: Forecast experiment with a nonhydrostatic model and simulation of storm surge. J. Meteor. Soc. Japan, 88, 521–545. Narui, R., 2007: Global Analysis. Outline of the operational numerical weather prediction at the Japan

Vol. 88, No. 3

Meteorological Agency. 18–28. (Available online at http://www.jma.go.jp/jma/jma-eng/jma-center/ nwp/outline-nwp/pdf/pdf3/outline3_5.pdf ) Obashi, G. O. P., 1994: WMO’s Role in the International Decade for Natural Disaster Reduction. Bull. Amer. Meteor. Soc., 75, 1655–1661. Saito, K., T. Fujita, Y. Yamada, J. Ishida, Y. Kumagai, K. Aranami, S. Ohmori, R. Nagasawa, S. Kumagai, C. Muroi, T. Kato, H. Eito, and Y. Yamazaki, 2006a: The operational JMA Nonhydrostatic Mesoscale Model. Mon. Wea. Rev., 134, 1266–1298. Saito, K., M. Kyouda, and M. Yamaguchi, 2006b: Mesoscale Ensemble Prediction Experiment of a Heavy Rain Event with the JMA Mesoscale Model. CAS/JSC WGNE Research Activities in Atmospheric and Oceanic Modelling, 36, 5.49–5.50. Saito, K., J. Ishida, K. Aranami, T. Hara, T. Segawa, M. Narita, and Y. Honda, 2007: Nonhydrostatic atmospheric models and operational development at JMA. J. Meteor. Soc. Japan., 85B, 271–304. Saito, K., M. Kunii, H. Seko, M. Yamaguchi, and K. Aranami, 2008: Implementation of lateral boundary perturbations into mesoscale EPS. CAS/JSC WGNE Research Activities in Atmospheric and Oceanic Modelling, 37, 3.11–3.12. Sakai, R., 2009: The one-week ensemble prediction system at JMA. Annual report of Numerical Prediction Division, 55, 117–125, (in Japanese). Seko, H., K. Saito, M. Kunii, and M. Kyouda, 2009: Mesoscale Ensemble Experiments on Potential Parameters for Tornado Formation. SOLA, 5, 57–60. Shibayama, T., and Investigation Team of Japan Society of Civil Engineering, 2008a: Investigation report on the storm surge disaster by Cyclone SIDR in 2007. Japan Society of Civil Engineering, 81 pp. Shibayama, T., and Investigation Team of Japan Society of Civil Engineering, 2008b: Prompt report on the storm surge disaster by the Myanmar Cyclone. JSCE Magazine, 93, No.7, 41–43, (in Japanese). Shoji, Y., M. Kunii, and K. Saito, 2010: Mesoscale data assimilation of Myanmar Cyclone Nargis Part II: Assimilation of GPS derived precipitable water vapor. J. Meteor. Soc. Japan, (submitted). Webster, P. J., 2008: Myanmar’s deadly da¤odil. Nature Geoscience, 1, 488–490. Yamaguchi, M., R. Sakai, M. Kyoda, T. Komori, and T. Kadowaki, 2009: Typhoon Ensemble Prediction System developed at the Japan Meteorological Agency. Mon. Wea. Rev., 137, 2592–2604.