ORIGINAL PAPER

Simulations of Cyclone Sidr in the Bay of Bengalwith a high-resolution model: sensitivity to large-scaleboundary forcing

Anil Kumar • James Done • Jimy Dudhia •

Dev Niyogi

Received: 13 July 2009 / Accepted: 19 August 2011

� Springer-Verlag 2011

Abstract The predictability of Cyclone Sidr in the Bay of

Bengal was explored in terms of track and intensity using

the Advanced Research Hurricane Weather Research

Forecast (AHW) model. This constitutes the first applica-

tion of the AHW over an area that lies outside the region of

the North Atlantic for which this model was developed and

tested. Several experiments were conducted to understand

the possible contributing factors that affected Sidr’s

intensity and track simulation by varying the initial start

time and domain size. Results show that Sidr’s track was

strongly controlled by the synoptic flow at the 500-hPa

level, seen especially due to the strong mid-latitude wes-

terly over north-central India. A 96-h forecast produced

westerly winds over north-central India at the 500-hPa

level that were notably weaker; this likely caused the

modeled cyclone track to drift from the observed actual

track. Reducing the model domain size reduced model

error in the synoptic-scale winds at 500 hPa and produced

an improved cyclone track. Specifically, the cyclone track

appeared to be sensitive to the upstream synoptic flow, and

was, therefore, sensitive to the location of the western

boundary of the domain. However, cyclone intensity

remained largely unaffected by this synoptic wind error at

the 500-hPa level. Comparison of the high resolution,

moving nested domain with a single coarser resolution

domain showed little difference in tracks, but resulted in

significantly different intensities. Experiments on the

domain size with regard to the total precipitation simulated

by the model showed that precipitation patterns and 10-m

surface winds were also different. This was mainly due to

the mid-latitude westerly flow across the west side of the

model domain. The analysis also suggested that the total

precipitation pattern and track was unchanged when the

domain was extended toward the east, north, and south.

Furthermore, this highlights our conclusion that Sidr was

influenced from the west side of the domain. The dis-

placement error was significantly reduced after the domain

size from the western model boundary was decreased.

Study results demonstrate the capability and need of a

high-resolution mesoscale modeling framework for simu-

lating the complex interactions that contribute to the for-

mation of tropical cyclones over the Bay of Bengal region.

1 Introduction

Accurate cyclone track and intensity predictions remain a

challenging task for atmospheric scientists and the research

community. A large number of cyclones form in the Bay of

Bengal (hereafter BOB) region and make landfall along the

coastal regions of India, Bangladesh, and Myanmar. These

cyclones have been responsible for billions of dollars in

property damage, loss of agriculture crops, and thousands

of human lives (e.g., Paul 2010). Between October and

December, cyclonically favorable, large-scale atmospheric

conditions are typical over BOB.

This study concerns the simulation of a recent, notable

BOB storm—Cyclone Sidr using the Advanced Research

Responsible editor: C. Simmer.

A. Kumar � J. Done � J. Dudhia

National Center for Atmospheric Research, Boulder, CO, USA

A. Kumar � D. Niyogi

Purdue University, West Lafayette, IN, USA

Present Address:A. Kumar (&)

Hydrological Science Branch, NASA/GSFC, Code-614.3,

Greenbelt, MD 20771, USA

e-mail: anil.kumar@nasa.gov

123

Meteorol Atmos Phys

DOI 10.1007/s00703-011-0161-9

Hurricane Weather Research Forecast (AHW) model. This

would be the first test of the AHW model (Davis et al.

2008; Xiao et al. 2009) outside the Atlantic basin or the

region for which it was developed and evaluated. We chose

Sidr mainly because of the (a) very strong wind and Saffir–

Simpson equivalent category 5 intensity associated with

this cyclone, (b) most of the operational models used for

forecasting purposes failed to capture track as well as

intensity, and (c) to help clarify the influence of strong

upper-level mid-latitude westerlies over north India on the

simulated Sidr cyclone under different domain dimensions.

Cyclone Sidr has been used as a test case by a number of

modeling groups. Pattanayak and Mohanty (2008) and then

Bhaskar Rao and Srinavas (2010) reported on the perfor-

mance of MM5 and the Weather Research Forecast (WRF)

modeling system on track and intensity changes. They

showed that there is no significant improvement in the

model after a 36-h model forecast, because the model

boundary and initial conditions provided by the coarser

resolution NCEP forcing data dominated the results. Sim-

ilarly, Badarinath et al. (2009) used the Sidr case with the

MM5 model to assess aerosol loading. Kotal et al. (2008)

tested a statistical–dynamical approach to understand the

errors in the Sidr track and found a northwest directional

bias. More recently, Akter and Tsuboki (2010) simulated

the supercells in the Sidr rainbands with a cloud resolving

model to understand the synoptic latent heat and storm axis

interactions.

In the following section, Sidr’s track and intensity

changes are discussed. This is followed by the AHW model

description in Sect. 3. Model performance and track anal-

ysis are presented in Sect. 4. The impact of domain size on

model track is presented in Sect. 5. Details of the cyclone

structural features of the core are given in Sect. 6. Study

conclusions are summarized in Sect. 7.

2 Sidr description

Cyclone Sidr was the fourth named storm of the 2007

northern Indian Ocean cyclone season. Sidr formed in the

central BOB region and quickly strengthened to reach

1-min sustained winds of 225.3 km h-1 (150 mph),

according to the Joint Typhoon Warning Centre (JTWC).

This report qualified Sidr as a category 5 equivalent trop-

ical cyclone on the Saffir–Simpson Scale from 06 UTC 15

November 2007. The storm eventually made landfall in

Bangladesh on 15 November 2007. According to the media

reports, the storm caused large-scale evacuations of about

650,000 people and resulted in more than 2,400 fatalities.

Most of the deaths were attributed to falling trees that

flattened many coastal structures. Cyclone Sidr was

described as the most severe storm (in terms of fatalities

and damage) to strike Bangladesh since 1991. JTWC

issued a forecast on 9 November 2007 identifying a trop-

ical disturbance with weak, low-level circulation near the

Nicobar Islands. Initially, a moderate upper-level wind

shear with strong diffluence aloft aided in the developing

convection zone. The vertical shear decreased greatly as

the circulation became better defined. As a result, a tropical

cyclone formation alert was issued on 11 November, at a

time when the circulation was located a short distance

south of the Andaman Islands. JTWC warnings were based

upon both Windsat microwave images that showed a low-

level circulation center and upper-level analyses that

showed enhanced convection due to a strong diffluent flow

over the disturbance. Around the same time, the India

Meteorological Department (IMD) designated the system

as a depression and issued a warning stating that a

‘‘depression has formed over the southeast Bay of Bengal

and adjoining Andaman Sea and lay centered at 1430 hours

IST (India Standard Time) of 11 November 2007 near

10.0�N and 92.0�E about 200 km south–southwest of Port

Blair and the system is likely to intensify further and move

in a west north mid-latitude westerly direction.’’ Figure 1a

shows the tracks that were issued every 6 h by the JTWC,

the US National Hurricane Center (NHC), and the Central

Pacific Hurricane Center (CPHC). The JTWC upgraded

Sidr to a tropical cyclone after Dvorak estimates indicated

winds of 65 km h-1 (40 mph) on 11 November. Moreover,

as the day progressed, the storm intensified into a deep

depression as it moved slowly northwestward. The track is

shown in Fig. 1a with the global sea surface temperature

(RTG_SST) analysis at 00/11 November 2007, developed

through the National Centers for Environmental Prediction/

Marine Modeling and Analysis Branch (NCEP/MMAB).

The IMD observed track is plotted in Fig. 1b with the

intensity and track discussions. Figure 1a and b shows

slightly different tracks: at 0600 UTC 15 November, IMD

estimated 132.5 mph surface winds, whereas JTWC shows

135 mph. There is no surface wind speed data available

from IMD after 1800 UTC 15 November, which causes us

to rely solely on JTWC track records for that information.

The cyclone intensified to reach peak winds of 132.5 mph

at 0600 UTC 15 November based on IMD observations

and agrees with the JTWC estimates of 135 mph peak wind

speed for the same time. Sidr officially made landfall at

1600 UTC 15 November as per IMD track (IMD 2008).

3 Model description

The Advanced Hurricane WRF (AHW) is a derivative of

the Advanced Research WRF model. The model is capable

of resolving multiscale cyclone features from about 1 km

to synoptic-scale feedbacks. The technical details are

A. Kumar et al.

123

available at the WRF repository (http://www.mmm.

ucar.edu/wrf/users/docs/arw_v3.pdf). For our simulations,

the outermost domain was fixed (Fig. 2) with 12-km grid

spacing (423 9 324 grid points), two nested movable

domains at 4 km (201 9 201 grid points), and a 1.33-km

grid spacing (240 9 240 grid points) that covered an area

of 320 km 9 320 km and was configured with a two-way

nesting option. The choice of inner domain grid spacing

follows the findings of Chen et al. (2007) that for the WRF

model, proper treatment of the inner core requires a grid

spacing of less than 2 km. All domains had 35 vertical

layers with a terrain that followed sigma coordinates with

the model top at 0.5 hPa. The nest positions were updated

every 15 min of the simulation and the track was updated

with the center of the cyclone. The model used the WSM3

microphysics scheme (Hong et al. 2004), while the Rapid

Radiative Transfer Model (RRTM, Mlawer et al. 1997) and

the Dudhia scheme (Dudhia 1989) were used for the

longwave and shortwave radiation calculations, respec-

tively. The thermal diffusion scheme was used to represent

surface physics with the Yonsei University (YSU) plane-

tary boundary layer scheme (Noh et al. 2003). The initial

and boundary conditions for the large-scale atmospheric

fields were derived from the 1 91 degree NCEP global

final analysis (FNL) using the WPS (WRF Pre-processing

System) software package. The model run started at 00Z 11

November 2007 and ended at 00Z 17 November 2007. Sea

surface temperatures were derived from the high-resolution

real-time global sea surface temperature (RTG_SST) at

1/12-degree resolution analyses from NCEP/MMAB.

3.1 Surface flux parameterization

Hurricane intensity is sensitive to the parameterization of

momentum and enthalpy fluxes between the surface and

the atmosphere (Rosenthal 1971; Emanuel 1995). In the

storm core, maximum wind speed depends on the square

root of the ratio of the drag and enthalpy exchange coef-

ficients, ðCk=CDÞ1=2following Emanuel (1986). The sur-

face drag parameterization in the AHW model is based on

Donelan et al. (2004) which defines the relation between

roughness length (Z0) and frictional velocity (u*) as,

Z0 ¼ 10 expð�9=u1=3� Þ;

Fig. 1 a JWTC estimated track and spatial pattern of sea surface

temperature (from 1/12-degree real-time Global SST analysis) at

0000 UTC 11 November 2007 in the Bay of Bengal, and b IMD

recorded observed track (no data is available after 1506 UTC in IMD

source)

Fig. 2 Model, nested domains (at 12-, 4-, and 1.33-km resolution and

terrain height in m)

Simulations of Cyclone Sidr in the Bay of Bengal

123

where Z0 is bound by a limiting range between

0.125 9 10-6 and 2.85 9 10-3 m, respectively. Further-

more, the Ck formulation was modified with the so-called

ramped Ck approach by introducing a ramping effect in the

enthalpy roughness length as described in Dudhia et al.

(2008). This ramped Ck up with wind speeds of hurricane

strength.

3.2 Coupling with a 1D ocean model

Ocean temperature feedback was applied to every grid

point in the AHW model through a 1D ocean model based

on Pollard et al. (1973). The ocean model was initialized

for this case with a 30-m ocean mixed-layer depth (MLD).

Rao et al. (1989) studied the mean monthly MLD in the

Arabian and BOB regions and found it to be between 30

and 40 m in November. The NCEP Global Ocean Data

Assimilation System (GODAS) showed a 25- to 35-m

MLD for the same month and was considered appropriate.

The model does not consider lateral heat transfer

between individual ocean columns, so heat only propagated

vertically. This model accounted for the Coriolis effect, but

there was no advection or pressure gradient. A MLD of

30 m produced a maximum cooling of about 3.1 K when

considering a deep-layer lapse rate of 0.05 km-1 (Davis

et al. 2008). Frictional velocity estimations were through

surface layer physics and net radiation. Surface fluxes

accounted for thermal forcing as secondary forcing only

with ocean thermal mixing being the primary forcing. The

atmospheric model called the ocean 1D column model at

every time step and also updated the SSTs.

4 Results

4.1 Model track analysis initialized at 0000 UTC 11

November 2007

Figure 3 shows the model-predicted track from the simula-

tion initialized at 0000 UTC 11 November. The model-

simulated results presented in this section are from a moving

nest at 1.33-km domain resolution. The model track deflec-

ted to the left of the observed track, and resulted in landfall on

the Orissa coast, which was far from the actual landfall

location. However, the model was able to capture Sidr’s

intensity reasonably well. This model simulation at

0000 UTC 15 November was indicative of a category 5

cyclone with maximum sustained winds of 141 knots. The

model-estimated maximum sustained winds refer to 10-m

winds and minimum surface pressure of 931.6 hPa posi-

tioned at 14.80�N and 86.43�E. The model-simulated tem-

poral evolution of intensity and minimum surface pressure is

shown in Fig. 4. The track started to diverge from the actual

track at 0000 UTC 13 November toward the northwest and

continued simulating the incorrect track with later prediction

times. The model also simulated a slower moving cyclone by

up to 3� latitude at 0000 UTC 15 November when compared

to the observed location (17.8�N, 89.2�E).

4.2 Model track analysis initialized at 0000 UTC 12

November 2007

In an attempt to improve the predicted track and lag time,

the model was initiated at 0000 UTC 12 November. The

modeled track, shown in Fig. 3, again diverges from the

actual track toward the northwest direction with only a

small improvement on the simulation with the earlier

model initialization time. The predicted intensity reached

127.3 knots with a center pressure as low as 933.11 hPa at

16.21�N and 86.92�E for 0000 UTC 15 November corre-

sponding to a category 4 cyclone, but the displacement

error at this time was 1.17� (128 km) from the actual

cyclone position. With the later model initialization times,

the model improved the track by 0.5� toward the east and

also improved the position and timing of the cyclone. Still

the predicted intensity was slightly weaker (by 14 knots) in

comparison to the model simulation initialized at

0000 UTC 12 November 2007. The model overpredicted

the maximum sustained winds (by 12 knots) in comparison

with observed data. In summary, changing model initiali-

zation time not only made a small difference to track and

timing, but also had an impact on intensity (Fig. 4). To

gain further insight into the impact of model initialization

Fig. 3 Simulated 1.33-km resolution-based track and intensity from

different model initialization time and observed track (white line)

A. Kumar et al.

123

time on cyclone track and intensity, we conducted further

tests described in the next section.

4.3 Model track analysis initialized at 1200 UTC 12

November 2007

Figure 3 shows that the predicted track for a simulation

initialized at 1200 UTC 12 November still diverged from

the actual track toward the northwest. The model simula-

tion at 0000 UTC 15 November showed maximum sus-

tainable winds of about 101 knots, surface pressure at

958.9 hPa, with its location at 14.59�N and 87.81�E for the

first 24 h. The model follows the observed track, but

thereafter diverges toward the northwest. Model tracks

were approximately the same as those seen in the simula-

tion initialized at 0000 UTC 12 November, yet the maxi-

mum sustained winds dropped from 127 to 101 knots,

while surface pressure increased from 933.11 to 958.9 hPa

(Fig. 4). One possibility for the reduction in maximum

winds may be a cooled sea surface. It was thought that

initializing the model later in time may improve track and

intensity due to more realistic lateral and boundary

conditions. This experiment is discussed in the following

section.

4.4 Model track analysis initialized at 0000 UTC 13

November 2007

Figure 3 shows good track yet poor intensity for a simu-

lation initialized at 0000 UTC 13 November. For the large

model domain used in the study region, it is well known

that the large-scale processes in the model diverge from

those in the boundary conditions (Denis et al. 2003).

Therefore, it is plausible that for simulations with an earlier

initialization time the model has time to develop large-

scale errors that result in larger cyclone track errors. Ini-

tializing the model closer in time to landfall limits the error

growth at large scales, which may be the reason for the

improved model cyclone track. However, the simulation

with improved track also predicted poor intensity.

Thus, the results show a large variation in minimum sea

level pressure, intensity, and track. The results also suggest

that the cyclone track is controlled by large-scale features

such as synoptic winds while intensity strongly depends on

both local and large-scale conditions.

5 Impact of domain size on cyclone track

To further assess the large-scale/local-scale interactions,

we ran simulations initialized at 0000 UTC 11 November

out to 144 h for four different domain sizes, each at 12-km

grid spacing. The domain sizes are shown in Fig. 5. This

Fig. 4 Time series for central minimum sea level pressure (hPa) and

maximum velocity (knots) of cyclone for simulations (at 1.33-km

resolution) beginning at different initialized times

Fig. 5 Four domains (at 12-km resolution) used in the domain size

sensitivity study

Simulations of Cyclone Sidr in the Bay of Bengal

123

section discusses the model results corresponding to the

12-km resolution domain. The largest domain, domain 1,

had 424 9 325 grid points (longitude: 65–115�E, latitude:

1�S–35�N). The second largest domain, domain 2, had

364 9 285 grid points (longitude: 70–110�E, latitude:

0–30�N). The third largest domain, domain 3, had

264 9 215 grid points (longitude: 75–105�E, latitude:

5–28�N), and, the smallest domain (domain 4), shown in

Fig. 5, had 164 9 185 grid points (longitude: 80–98�E,

latitude: 6–25�N). All grids were measured in west–east

and north–south directions. Figure 6 shows the model track

produced by the different domain-sized simulations. As the

domain size decreased, the track improved. The simulation

on the smallest domain, domain 4, simulated a reasonable

track. Although there are only small differences in model

Fig. 6 Predicted track using different domain sizes (at 12-km

resolution) with model start time at 0000 UTC 11 November. DS1is the largest domain, DS2 is the second largest, DS3 is the third

largest, and DS4 is the smallest domain size as shown in Fig. 5

Fig. 7 NCEP FNL analysis (interpolated to 12-km resolution from

1� 9 1�) wind speed and vector (m s-1) at the 500-hPa level at

0000 UTC 15 November 2007

Fig. 8 Wind speed and vector (m s-1) at the 500-hPa level at

0000 UTC 15 November in domain 1 (at 12-km resolution), a model

analysis, and b difference field NCEP minus model

A. Kumar et al.

123

track between the third and fourth domain sizes, the track

using the smallest domain (fourth domain size) is in good

agreement with the actual track during and after landfall.

The track is slightly better for domain 3 than for domain 4

(i.e., the smallest domain track) until landfall. After land-

fall, the cyclone’s low pressure center moved into the

BOB. Since this opposes the observed track, domain 3

cannot be considered a good track. The variability of

cyclone intensity with model domain size is discussed in

the next section.

We used NCEP/NCAR FNL data to verify large-scale

flow in the model. First we focus on the mid-latitude

westerly flow over northern India of winds at the 500-hPa

level in order to find differences between the various

domain-sized simulations with regard to their impact on

cyclone track. Figure 7 shows a FNL analysis wind vector

and speed at 0000 UTC 15 November. Figure 8a shows the

corresponding model-simulated wind vector and wind

speed for domain 1 while the difference field is shown in

Fig. 8b. The analysis time of 0000 UTC 15 November was

chosen because at this time there were significant differ-

ences among the tracks from various domain sizes when

model forecast time is 96 h. Figure 7 shows strong north-

westerly (NW) winds over central India in the NCEP data

while the model-simulated winds at the 96-h forecast time

show a generally weak flow. The field shows wind speed

differences of 8–10 m s-1. Hence, for the largest domain,

the model errors are largest for the 500-hPa NW flow over

central India during the 96-h forecast. One possible reason

is that the distances to the boundary conditions, which are

approximately 1,200 km from central India in the west as

well as in the east direction, allow freedom within the

interior of the domain for the model to diverge at a large

scale from the driving analysis. A similar analysis is

Fig. 9 a Model-simulated (at 12-km resolution) wind speed and wind

vector at the 500-hPa level in domain 2, b difference NCEP FNL

analysis (shown in Fig. 6) minus model winds in domain 2 Fig. 10 Same as Fig. 9a, b except in domain 3

Simulations of Cyclone Sidr in the Bay of Bengal

123

carried out for the second largest domain, domain 2. Fig-

ure 9a shows the model field while the difference field is

presented in Fig. 9b. The modeled 96-h forecast field

shows weak synoptic winds that are shifted slightly toward

the north. The difference field shows 12–15 m s-1 devia-

tions in wind speed over east-central India. Figure 10a

shows model-simulated winds for domain 3 with

improvements seen in a confined region of strong NW flow

in this domain. The difference field, Fig. 10b, shows values

of 6–10 m s-1 wind speed difference as well as differences

in wind vector fields. The model boundary forcing on the

west side of the domain is located at 75�E longitude which

is approximately 500 km away from the strong mid-lati-

tude westerlies that is confined over central India. The

closer proximity of the lateral boundary conditions,

updated every 6 h, provided stronger control on the large

scales in the model, which helped to improve the 500-hPa

level westerlies as well as the cyclone track. Figure 11a

shows model-simulated winds on the smallest domain,

domain 4, which has similar patterns and magnitudes at

large scales as the reanalysis has. Differences from the

driving analysis reach 5 m s-1 as shown in Fig. 11b.

Domain 4 is small enough to capture the synoptic pattern

over the BOB region and parts of Central India, Bangla-

desh and Myanmar, and also sufficiently resembles the

observed cyclone track. We caution that our results should

be considered true only for such cases where strong syn-

optic flow influences a cyclone and may not be true for all

cases over BOB.

Fig. 11 Same as Fig. 10a, b except in domain 4

Fig. 12 a Model-simulated (at 12-km resolution) wind speed profile

over Raipur station at 0000 UTC 15 November, and b temporal wind

speed comparison at 500-hPa level over Raipur station

A. Kumar et al.

123

The vertical wind speed profiles at 0000 UTC 15

November over Raipur station (21�N, 82�E) is shown in

Fig. 12a. We chose Raipur station mainly because it is

situated under the region where we see significant changes

in wind speed at the 500-hPa level in the four different

domain-sized experiments. We anticipate that the winds

over broader central India including the Raipur region may

be affecting the model’s lateral boundary from the west

side which can cause a track deflection despite the wind

direction remaining the same in all four domain-size sim-

ulations. The observed wind speeds were obtained from the

Wyoming atmospheric sounding data archive and were

compared with modeled wind speed profiles of the differ-

ent domain-size simulations. As expected, the wind profiles

using the smaller domain sizes (domains 3 and 4) are

closest to the observed wind speed profile, especially

around the 500-hPa level. The temporal evolution of wind

speed at 500 hPa in the observation and model-simulated

winds is shown in Fig. 12b, for the location (21�N, 82�E)

where NW winds are strong. Winds in the smallest domain

more closely follow the observed winds at the 500-hPa

level. With this analysis, we concluded that the interior

flow on the smaller domains is more strongly constrained,

which results in an improved track prediction.

The improvement on the smallest domain is explored

further by extending domain 4 toward the east, north, and

south directions by 5�. The model was run for this domain

with nested inner domains at 4 and 1.33 km. One of the

objectives of this experiment was to determine the impor-

tance of the location of the western domain lateral

boundary condition that controls the simulated track and

compare those findings to the importance of domain size.

The simulation for the extended domain did not show any

significant difference in track (not shown). Therefore, this

further highlights that model track is controlled by the west

domain boundary flow conditions rather than domain size

or other lateral boundaries. The peak intensity of

120 knots, however, was not maintained for long in the

simulation.

Simulated total precipitation patterns from the different

domain-size experiments can be helpful in understanding

the impact of domain size on the model’s output. Figure 13

shows the total precipitation that occurred between

0000 UTC 11 November and 0000 UTC 16 November

Fig. 13 Model-simulated (at

12-km resolution) total

precipitation (mm) observed

between 0000 UTC 11

November and 0000 UTC 16

November and simulated wind

barb (m s-1) at 0000 UTC 16

November is plotted for

a domain 1 b domain 2,

c domain 3, and d domain 4. For

domain information see Fig. 5

Simulations of Cyclone Sidr in the Bay of Bengal

123

2007 (120-h rain) along with the 10-m surface wind barb.

The heaviest precipitation was around the eyewall of the

storm and followed the track. As seen in Fig. 13, the pre-

cipitation patterns are significantly different in each of the

four domain-size simulations and heavy precipitation fol-

lows the model track. Figure 13d shows that projected

rainfall was close to the high-resolution global precipita-

tion map (not shown) based off the satellite TRMM-PR

estimates. We also investigated precipitation patterns from

an extended domain experiment which expanded domain

4’s area toward the east, south, and north (Fig. 5). Fig-

ure 14a shows accumulated precipitation from 11 to 16

November 2007 in domain 4 while Fig. 14b shows the

accumulated precipitation for the same period in the

extended domain. Overall precipitation patterns were the

same and followed the heavy rain along the cyclone track.

This then confirms that the east side of the domain

boundary was not controlling the Sidr track. The dis-

placement error of the simulated cyclone eye location was

Fig. 14 Model-simulated (at 12-km resolution) total precipitation

(mm) observed between 0000 UTC 11 November and 0000 UTC 16

November and simulated wind barb (m s-1) at 0000 UTC 16

November is plotted for a experiment with domain 4, and b model

experiment with extended domain toward east, north and south with

reference to domain 4

Table 1 Displacement error, this error is calculated using center of

observed cyclone at 24 h interval

Synoptic time (UTC) Displacement error (km)

YYYYMMDDHHMM DOM-1 DOM-2 DOM-3 DOM-4

200711160000 725 592 488 210

200711150000 260 225 188 190

200711140000 165 156 120 125

200711130000 112 72 55 58

200711120000 115 95 40 60

200711110000 70 65 62 62

Table 2 Date, location, and wind speed (mph) based on observed

JWTC (Fig. 1a)

Synoptic Time (UTC) Latitude Longitude Wind speed (mph)

(YYYYMMDDHHMM)

200711160000 25.0 91.9 105

200711151200 22.8 90.3 130

200711151200 20.9 89.5 130

200711150600 19.3 89.3 135

200711150000 17.8 89.2 130

200711141800 16.6 89.3 130

200711141200 15.7 89.3 130

200711140600 15.0 89.4 120

200711140000 14.3 89.6 115

200711131800 13.7 89.5 115

200711131200 13.0 89.6 115

200711130600 12.5 89.8 115

200711130000 12.1 89.8 115

200711121800 11.6 90.0 105

200711121200 11.0 90.3 75

200711120600 10.8 90.4 55

200711120000 10.4 90.8 50

200711111800 10.4 91.4 45

200711111200 10.2 91.9 35

200711110600 10.0 92.3 35

A. Kumar et al.

123

also calculated from domain-size experiments and is shown

in Table 1. Incidentally, the cyclone eye displacement error

was less in domain 3 and domain 4 than in domains 1 and

2. Domain 4’s displacement error was lowest during the

0000 UTC 16 November 2007 simulation. On the whole, a

210-km displacement error was found among all three

domain sizes.

6 Sidr eyewall structure and intensity investigation

To examine Sidr more closely in terms of intensity and

minimum sea level pressure (MSLP), we compared the

intensity and central pressure obtained from the JWTC site

with the domain 4 simulation (Table 2). Figure 15 shows

close agreement for MSLP and maximum winds. After

00 UTC 14 November (a 72-h model forecast), the model

predicted higher MSLP (940 hPa) than the satellite-derived

value of 920 hPa pressure. Also, the model-estimated

maximum winds at 120 knots, whereas the satellite-derived

estimates were closer to 140 knots. This confirms that the

model can predict intensity and MSLP reasonably well

while maintaining a good track throughout the 144-h

forecast duration.

We also conducted an experiment with a single domain

and compared it to a nested domain simulation for domain

4. The simulated tracks and minimum central pressures

presented in Fig. 16a and b showed that the single domain

simulation displayed an improved track but had poor

intensity. This indicates that the simulated track was less

dependent on the domain size and the simulated intensity

of the cyclone is more dependent on model grid resolution.

Model winds at the 700-hPa level were compared with

satellite-based wind analysis (at 700 hPa). Winds estimates

include satellite data product reference datasets from the

Advanced Microwave Sounding Unit (AMSU), Cloud-

drift/IR/WV winds, IR-proxy winds and Scatterometer

winds, QuikSCAT, and Advanced-Scatterometer (A-

SCAT). A variational approach described in Knaff and

DeMaria (2006) in conjunction with these five data sources

were used to create a mid-level (near 700 hPa) wind. Two

dimensional (2D) flight winds are estimated from IR

imagery (Mueller et al. 2006). These 2D winds were

obtained following AMSU derived wind fields and are used

in solving the non-linear balance equations as described in

Bessho et al. (2006). Figure 17a and b shows a comparison

between satellite-derived winds and model-predicted winds

at 4-km grid resolution at 0000 UTC 15 November 2007

(96-h model forecast). The model-predicted wind direction

at the 700-hPa level (Fig. 17c) was within reasonable

agreement of the satellite-derived wind direction. It was

also noted that during the 96-h model forecast, at 0000

UTC 15 November, the displacement error was 190 km

and the model-simulated wind near the periphery was

50 knots in the domain 4 simulation conducted with a 1.33-

km inner nested domain. This is close to the satellite-

Fig. 15 Model and observed MSLP and TC intensity comparison.

Observed MSLP and TC intensity is denoted as filled triangle andcircle, where as model estimate shows in open triangle and circle.

Model results are from the simulation conducted using domain size 4

at 1.33-km resolution

Fig. 16 a Single domain (at 12-km resolution) and three nested

domains (at 1.33-km resolution) simulated tracks versus observed,

and b time series of minimum central pressure from single and nested

model results and compared with observation

Simulations of Cyclone Sidr in the Bay of Bengal

123

derived winds of 60 knots (Fig. 18a, b). Due to a lack of

good quality data in and around the inner cyclone core,

further verification is limited.

Satellite-derived winds suggested the eye was circular

and symmetric in nature, yet the model predicted the storm

eye was neither symmetric nor circular. To visualize the

model-predicted storm eye, potential vorticity parameters

were used at different model forecast times. At 0000 UTC

14 November, the storm’s eye shape was both circular and

symmetric (Fig. 19). At later times, for instance, at 14

November 1200 UTC and 15 November 0000 UTC, the

eye took on a more triangular shape. Furthermore, at

1200 UTC 15 November, the eye was oval shaped. A

similar triangular-shaped eye was documented in a mod-

eling study of Hurricane Katrina (Corbosiero et al. 2008).

To view cyclone structure and associated bands, side-by-

side comparisons of satellite IR imagery and model cloud

top temperature (not shown) were made and were found to

be in good agreement in terms of band and overall cyclonic

structure.

7 Conclusion

In this study, we applied the AHW modeling system for an

intense tropical cyclone in the Bay of Bengal region. The

study investigated the impact lateral and boundary forcing

of four different domain sizes has on cyclone track and

intensity and found that the reduction in domain size both

minimized the substantial model error growth in synoptic

winds and improved the cyclone track and storm intensity

during a complete 144-h model forecast. Our analysis

showed that the model simulated cyclone Sidr’s track was

significantly influenced by a large-scale 500-hPa level mid-

latitude westerlies relative to flow on south and east of the

domain. For the large domain, the model underestimated

Fig. 17 Wind speed at

0000 UTC 15 November for

a satellite-estimated winds,

b model winds in a movable

nest domain at 4-km resolution,

and c model wind vectors

A. Kumar et al.

123

wind speeds by 8–10 m s-1 at the 500-hPa level over the

central part of India resulting in a poor cyclone track

projection. The underpredicted large-scale flow could be

corrected by reducing the domain size. This highlights the

importance of the mid-tropospheric flow for the tropical

cyclone simulation. The reasons for the larger domains

failing to capture the feature accurately will need to be

addressed.

The main findings from this study are as follows.

Experimentation with model initialization time using the

large domain size showed that later initialization times

not only improved the model-predicted track, but also

produced poor cyclone intensity. To understand the rel-

ative importance of the location of the western boundary

versus the domain size when predicting cyclone track,

we extended the domain toward the east, north, and

south directions and kept the western boundary the same.

Results indicated that the simulated track and intensity

are reasonable and hence, confirmed that the western

boundary played a significant role in controlling the

track. Our analysis also highlighted the impact of model

domain resolution on track and intensity. Furthermore, it

was found that with coarser resolution, the model pre-

dicted a good track but failed in terms of intensity. The

domain size also affected the total simulated precipita-

tion patterns, but the precipitation amount was not much

different in different domain-size simulated experiments.

Extending the domain toward the east, north, and south

did not affect the simulated precipitation patterns, which

implied that was only influenced by the westward large-

scale boundary forcing. The displacement error in Sidr’s

storm eye was significantly affected by changing the

domain size used in modeling experiments, which

implied that the displacement error decreased after

reducing the domain size from west to east. Interestingly,

the difference in displacement error between second

smallest domain (domain 3) and smallest domain

(domain 4) is small and many times domain 3 track is

better by few kilometers but after making landfall,

domain 3 simulated track is moved back in the ocean

and get off the track completely on last day of simula-

tion period. Hence, we made conclusion that smallest

domain simulated track is better and follow actual track

even after making landfall. For the smallest domain,

where the model predicted both track and intensity in

good agreement with observation, the model-predicted

eyewall and structure was captured well. However, the

triangular shape of the storm eye was not consistent with

the more circular eye inferred from satellite-derived

winds and imagery. The model’s predicted storm loca-

tion was generally within 150–200 km of the actual

storm location. It is likely that the impact of domain size

and boundary flow significantly affected the cyclone’s

motion at times when there was strong synoptic flow in

this region, as seen here in the case of cyclone Sidr.

However, this may be less important for cyclones that

occur in weaker synoptic flows.

Our experimentation with domain size and analysis

boundary conditions highlighted the importance of the

location of the western ‘inflow’ boundary. In forecast

models, improvements may be possible in simulating the

BOB cyclones by reducing the errors in the 500-hPa wind,

and the role of sounder data assimilation, better initial

Fig. 18 Wind speed and direction along Sidr eye at 0000 UTC 15

November for a satellite-estimated winds, and b model-simulated

winds in a movable nest domain at 1.33-km resolution

Simulations of Cyclone Sidr in the Bay of Bengal

123

conditions, and improved model physics needs to be

investigated.

Acknowledgments The authors would like to thank Qingnong

Xiao from MMM Division at National Center for Atmospheric

Research (NCAR) for the internal review on an initial draft. We

also thank NCAR supercomputing resources for providing com-

puting GAUS. NCAR is sponsored by the National Science

Foundation. The study also benefited from the NSF CAREER grant

(ATM-0847472).

References

Akter N, Tsuboki K (2010) Characteristics of supercells in the

rainband of numerically simulated Cyclone Sidr. SOLA

6A:25–28

Badarinath KVS, Kharol SK, Sharma A, Ramaswamy V, Kaskaoutis

DG, Kambezidis HD (2009) Investigations of an intense aerosol

loading during 2007 cyclone SIDR—a study using satellite data

and ground measurements over Indian region. Atmos Environ

43:3708–3716

Bessho K, DeMaria M, Knaff JA (2006) Tropical cyclone wind

retrievals from the Advanced Microwave Sounder Unit

(AMSU): application to surface wind analysis. J App Meteorol

45:399–415

Bhaskar Rao DV, Srinivas D (2010) Real-time prediction of SIDR

Cyclone over Bay of Bengal using high-resolution mesoscale

models. Indian Ocean Trop Cyclones Clim Chang 3:159–167.

doi:10.1007/978-90-481-3109-9_20

Chen SS, Price JF, Zhao W, Donelan MA, Walsh EJ (2007) The

CBLAST-Hurricane Program and the next-generation fully

coupled atmosphere–wave–ocean models for hurricane research

and prediction. Bull Am Meteorol Soc 88:311–317

Corbosiero KL, Molinari J, Vollaro D, Wang W, Done JM (2008) The

distribution of helicity and intense convection in tropical

cyclones. Paper P2F.13, 28th AMS Conference on hurricane

and tropical meteorology, Orlando, 28 April–2 May 2008

Davis C, Wang W, Chen SS, Chen Y, Corbosiero K, DeMaria M,

Dudhia J, Holland G, Klemp J, Michalakes J, Reeves H, Rotunno

R, Snyder C, Xiao Q (2008) Prediction of landfalling hurricanes

with the Advanced Hurricane WRF model. Mon Weather Rev

136:1990–2005

Denis B, Laprise R, Caya D (2003) Sensitivity of a regional climate

model to the resolution of the lateral boundary conditions. Clim

Dyn 20:107–126

Fig. 19 Model storm eye shapes at different times seen in the field of potential vorticity at the 700-hPa level (1.33-km resolution). (1

PVU = 10-6 m2 s-1 K kg-1)

A. Kumar et al.

123

Donelan MA, Haus BK, Reul N, Plant WJ, Stiassnie M, Graber HC,

Brown OB, Saltzman ES (2004) On the limiting aerodynamic

roughness of the ocean in very strong winds. Geophys Res Letts.doi:10.1029/2004GL019460

Dudhia J (1989) Numerical study of convection observed during the

winter monsoon experiment using a mesoscale two-dimensional

model. J Atmos Sci 46:3077–3107

Dudhia J, Done J, Wang W, Chen Y, Xiao Q, Davis C, Holland G,

Rotunno R, Torn R (2008) Prediction of Atlantic tropical

cyclones with the Advanced Hurricane WRF (AHW) model.

28th AMS Conference on hurricanes and tropical meteorology,

Paper 18A2, Orlando, 28 April–2 May 2008. http://ams.confex.

com/ams/28Hurricanes/techprogram/paper_138004.htm

Emanuel KA (1986) An air–sea interaction theory for tropical

cyclones. Part I: steady-state maintenance. J Atmos Sci

43:585–604

Emanuel KA (1995) Sensitivity of tropical cyclones to surface

exchange coefficients and a revised steady-state model incorpo-

rating eye dynamics. J Atmos Sci 52:3969–3976

Hong SY, Dudhia J, Chen SH (2004) A revised approach to ice

microphysical processes for the bulk parameterization of clouds

and precipitation. Mon Weather Rev 132:103–120

IMD (2008) Track of storm and depressions over the Indian Seas

during 1891–2007, Cyclone e-Atlas of IMD. IMD, New Delhi

Knaff JA, DeMaria M (2006) A multi-platform satellite tropical

cyclone wind analysis system. http://rammb.cira.colostate.

edu/products/tc_realtime/images/MPSW.pdf

Kotal SD, Roy Bhowmik SK, Kundu PK (2008) Application of

statistical–dynamical scheme for real time forecasting of the Bay

of Bengal very severe cyclonic storm ‘‘Sidr’’ of November 2007.

Geofizika 25:139–158

Mlawer EJ, Taubman SJ, Brown PD, Iacono MJ, Clough SA (1997)

Radiative transfer for inhomogeneous atmosphere: RRTM, a

validated correlated-k model for the long-wave. J Geophys Res

102:16663–16682

Mueller KJ, DeMaria M, Knaff JA, Vonder Haar TH (2006) Objective

estimation of tropical cyclone wind structure from infrared

satellite data. Weather Forecast 21:990–1005

Noh Y, Cheon WG, Hong SY, Raasch S (2003) Improvement of the

k-profile model for the planetary boundary layer based on large

eddy simulation data. Bound Layer Meteorol 107:401–427

Pattanayak S, Mohanty UC (2008) A comparative study on perfor-

mance of MM5 and WRF models in simulation of tropical

cyclones over Indian seas. Curr Sci 95:7

Paul B (2010) Human injuries caused by Bangladesh’s Cyclone Sidr:

an empirical study. Nat Hazards 54:483–495

Pollard RT, Rhines PB, Thompson RORY (1973) The deepening of

the wind mixed layer. Geophys Fluid Dyn 4:381–404

Rao R, Molinari RL, Festa JF (1989) Evolution of the climatological

near-surface thermal structure of the tropical Indian Ocean.

J Geophys Res 94:1081–10815

Rosenthal SL (1971) The response of a tropical cyclone model to

variations in boundary layer parameters, initial conditions, lateral

boundary conditions and domain size. Mon Weather Rev

99:767–777

Xiao Q, Zhang X, Davis C, Tuttle J, Holland G et al (2009)

Experiments of hurricane initialization with airborne Doppler

radar data for the Advanced-research Hurricane WRF (AHW)

model. Mon Weather Rev. doi:10.1175/2009MWR2828.1

Simulations of Cyclone Sidr in the Bay of Bengal

123