Sunteți pe pagina 1din 19

Nat Hazards (2012) 63:13931411

DOI 10.1007/s11069-011-9839-z
ORIGINAL PAPER

Tropical cyclone prediction over Bay of Bengal:


a comparison of the performance of NCEP operational
HWRF, NCAR ARW, and MM5 models
D. V. Bhaskar Rao Vijay Tallapragada

Received: 7 September 2010 / Accepted: 28 April 2011 / Published online: 27 May 2011
US Government 2011

Abstract Much progress has been made in the area of tropical cyclone prediction using
high-resolution mesoscale models based on community models developed at National
Centers for Environmental Predication (NCEP) and National Center for Atmospheric
Research (NCAR). While most of these model research and development activities are
focused on predicting hurricanes in the Atlantic and Eastern Pacific domains, there has
been much interest in using these models for tropical cyclone prediction in the North
Indian Ocean region, particularly for Bay of Bengal storms that are known historically
causing severe damage to life and property. In this study, the advanced operational hurricane modeling system developed at NCEP, known as the Hurricane Weather Research
and Forecast (HWRF) model, is used to simulate two recent Bay of Bengal tropical
cyclonesNargis of November 2007 and Sidr of April 2008. The advanced NCEP
operational vortex initialization procedure is adapted for simulating these Bay of Bengal
tropical cyclones. Two additional regional models, the NCAR Advanced Research WRF
and NCAR/Penn State University Mesoscale Model version 5 (MM5) are also used in
simulating these storms. Results from these experiments highlight the superior performance of HWRF model over other models in predicting the Bay of Bengal cyclones. These
results also suggest the need for a sophisticated vortex initialization procedure in conjunction with a model designed exclusively for tropical cyclone prediction for operational
considerations.
Keywords

Tropical cyclones  Numerical models

D. V. Bhaskar Rao
TLGVRC, Jackson State University, Jackson, MS 39217, USA
V. Tallapragada (&)
NCEP/EMC, 5200 Auth Road, Camp Springs, MD 20746, USA
e-mail: vijay.tallapragada@noaa.gov

123

1394

Nat Hazards (2012) 63:13931411

1 Introduction
Although only about 7% of the global tropical cyclones occur in the North Indian Ocean
(Bay of Bengal and Arabian Sea), they are most deadly. The shallow waters of the Bay of
Bengal, the low flat coastal terrain, and the funneling shape of the coast line can lead to
devastating losses of life and property. About 56 cyclones form in this basin each year,
with a primary maximum in November and a secondary maximum in AprilMay.
Current conventional numerical weather prediction (NWP) models are capable of
making skillful forecasts in middle and high latitudes for approximately 1 week; however,
predictability is limited to 23 days in the tropics. The situation is a combined result of
several factors such as lack of adequate data, problems with the analysis of primary
variables like wind and moisture and inadequate parameterization of physical processes.
The analysis and forecasting of tropical cyclones may also suffer from inadequate model
resolution. Much progress has been made in recent years in addressing real-time tropical
cyclone prediction, especially for the Atlantic and Eastern Pacific basins (Bender et al.
2007; Surgi et al. 2008).
Concerted effort of community model development at the National Center for Atmospheric Research (NCAR) in association with the Environmental Modeling Center (EMC)
of National Centers for Environmental Prediction (NCEP) resulted in making significant
progress in terms of meso-scale modeling for various weather applications. The weather
research and forecasting (WRF) model hosted by NCAR has two distinct dynamical cores
non-hydrostatic mesoscale model (NMM) developed at NCEP for operational forecasts
for the North American region; and an advanced research version of WRF (ARW) widely
used by research community across the world for various applications. NCEP/EMC has
adopted WRFNMM for hurricane prediction (HWRF) for operational use starting with the
2007 hurricane season, and ever since has been improved substantially for increased
forecast skill for tropical cyclone tracks and intensity. The operational HWRF modeling
system along with its scientific and technical documentation is provided by the developmental testbed center (DTC) (Gopalakrishnan et al. 2010).
A few studies on the comparative performance of the mesoscale models for severe
weather events were available. Sousounis et al. (2004) reported superior performance of
WRF model as compared to MM5, RUC, and ETA models in the prediction of a heavy
precipitation event. Cheng and Steenburgh (2005) made a comparison study between WRF
and ETA on the surface sensible weather forecast over Western United States and found that
the WRF model has better forecast skill. Patra et al. (2000) reported better performance of
MM5 compared to RAMS model in simulating tropical cyclones over the Bay of Bengal.
The present study aims at the application of the NCEP operational hurricane model, the
hurricane weather research and forecast (HWRF) system for two very severe cyclone
storms over Bay of Bengal. Results from experiments with two other meso-scale models,
the NCAR ARW and NCAR MM5, and their relative performance in the prediction of
intensity as well as track are presented here in comparison to HWRF forecasts. The
performances of these models were evaluated through comparison with India Meteorological Department (IMD) and Joint Typhoon Warning Center (JTWC) observations. Both
IMD and JTWC primarily use Dvorak technique to estimate tropical cyclone intensity. The
Dvorak technique is based on the analysis of cloud patterns in visible and infrared imagery
from geostationary and polar-orbiting satellites. Tropical cyclone intensity estimates are
made using two temperatures derived from the IR imagery. The first is the warmest pixel in
the eye, and second is the warmest pixel on the coldest circle between 24 and 111 km from
the cyclone center. Using these values, a Raw T-number can be created by using the locally

123

Nat Hazards (2012) 63:13931411

1395

developed table that expands upon the table published in Dvorak (1984). Each T-number
has an intensity, in terms of maximum 1-min sustained winds, associated with it and can be
converted to an intensity.
At present, IMD is receiving and processing meteorological data from two Indian
satellites namely Kalpana-1 and INSAT-3A. Kalpana-1, and INSAT-3A both have three
channel very high resolution radiometer (VHRR) for imaging the Earth in visible
(0.550.75 lm), infra-red (10.512.5 lm) and water vapor (5.77.1 lm) channels having
resolution of 2 9 2 km in visible and 8 9 8 km in water vapor (WV) and infra red (IR)
channels. At present about 48 satellite images are taken daily from Kalpana-1 which is the
main operational satellite and 9 images are taken from INSAT-3A. Satellite application
unit of IMD issues hourly and half-hourly bulletins of tropical cyclones (http://www.imd.
gov.in/section/nhac/dynamic/rsmc.pdf).
One significant difference between the IMD and JTWC agencies is that different wind
speed averaging periods are used in reporting the tropical cyclones maximum sustained
wind (MSW). IMD reports a 3-min MSW (maximum sustained winds) using Dvorak wind
table without any adjustment. JTWC reports 1-min MSW data at 6 h interval derived from
Meteosat-7 (at present and previously Meteosat-5) for the North Indian Ocean. Meteosat-7
provides 1 km resolution digital VIS imagery and 3 km resolution digital IR imagery
(NATIONAL HURRICANE OPERATIONS PLAN, FCM-P12-2009; http://www.ofcm.
gov/nhop/09/pdf/06-chap6-09.pdf). The difference in MSW estimation may arise due to (1)
the estimation of 1-min MSW from different satellites with different resolutions although
using the same Dvorak technique and (2) 1- and 3-min averaging by JTWC and IMD,
respectively, which means that JTWC will report higher maximum sustained surface wind
speeds than non-U.S. Tropical Cyclone Warning Centers for the same cyclone (Knapp and
Kruk 2010).
Section 2 provides brief description of the tropical cyclone cases selected for this study
followed by description models in Sect. 3. The advanced vortex initialization procedure is
described in Sect. 4. Results from several experiments in terms of track and intensity
prediction are provided in Sect. 5. Section 6 provides summary, conclusions, and future
work.

2 Recent Bay of Bengal tropical cyclones


2.1 Tropical cyclone Sidr (November 2007)
Sidr was graded as a very severe cyclonic storm, nominated as super cyclone at landfall
time, and had the life cycle during 1116 November 2007, with an attained intensity in
terms of the minimum central sea-level pressure (CSP) of 944 hPa and the maximum wind
speed (MW) of 115 knots. It was identified as a low pressure system at 03 UTC of 11
November 2007 over southeast BOB (Bay of Bengal) and neighborhood. It developed into
a depression at 09 UTC of the same day, moved slightly northwestwards, and intensified
into a deep depression at 19 UTC of 11 November. It intensified into a cyclonic storm,
Sidr and lay centered at 03 UTC of 12 November about 220 km near 10.5N, 91E. It
moved in the northwesterly direction, further intensified into severe cyclonic storm,
remained stationary and then intensified into a very severe cyclonic storm at 18 UTC of 12
November. The system moved slightly northwestwards thereafter till 00 UTC of 13
November, followed by a nearly northward direction up to 12 UTC of 15 November and
then recurved and moved in a northnortheasterly direction. Finally, it crossed west

123

1396

Nat Hazards (2012) 63:13931411

Bangladesh coast around 17 UTC of 15 November. After landfall, the system weakened
into cyclonic storm at 21 UTC of 15 November and further weakened into depression at 03
UTC and remained as depression until 06 UTC on 16 November.
Cyclone Sidr was the most powerful cyclone to impact Bangladesh since 1991, with the
death toll at approximately 3,406 and with extensive damage to crop and infrastructure
across 30 districts. Most damage and deaths were caused by the 4 m plus storm surge. A
comprehensive analysis undertaken by a team of Bangladesh Government and international
experts, using state of the art assessment methodologies, estimated the total damage and
loss caused by the cyclone to be US$1.6 billion.
2.2 Tropical Cyclone Nargis (April 2008)
Nargis was graded as a very severe cyclonic storm, had the life cycle during 27 April03
May 2008, with an attained intensity in terms of the minimum central sea-level pressure
(CSP) of 962 hPa and the maximum wind speed (MW) of 90 knots. Nargis started as a
depression over the southeast Bay of Bengal in the morning of 27 April, 2008 and was later
intensified into a cyclonic storm at 0000 UTC of 28 and into a very severe cyclonic storm
at 0300 UTC of 29. Initially, the depression moved westwards, intensified into a deep
depression and lay centered at 1200 UTC of 27 near 12N and 86.5E. Due to the strong
steering flow from the southeast, the system started to move in a northwesterly direction
and intensified into a cyclonic storm at 0000 UTC of 28 near 13N, 85.5E. Nargis intensified
into a severe cyclonic storm at 0900 UTC of 28 and into a very severe cyclonic storm at
0300 UTC of 29. The storm started to move in an eastnortheasterly direction till 1200
UTC of 1 May, followed by movement in easterly direction while intensifying further, and
crossed southwest coast of Myanmar between 1200 to 1400 UTC of 2 May near 16N. After
crossing the coast, the system gradually weakened and it lay as a severe cyclonic storm,
centered at 0300 UTC of 3 May over Myanmar near 17N 96E, close to Yangon. The
system weakened into a low pressure area over northeast Myanmar and adjoining Thailand
in the evening of 3rd May, 2008. The maximum sustained wind speed of 90 knots prevailed around system center during 0600 to 1,800 UTC of 2 May. The minimum CSP of
the system was 962 hPa recorded at 0600 UTC of 2 May with a pressure drop of about
40 hPa.
Nargis was the most devastating cyclonic storm over the north Indian Ocean in recent
years in terms of loss of life and property. The storm surge of about 35 m over the
Ayeyarwady delta region of Myanmar has been reported in the media. Maximum estimated
wind speed of 190 kmph has been reported over delta region during landfall. The affected
population was estimated at 11 millions with 84,000 human deaths, 54,000 missing
humans and 745,764 houses damaged, 155,248 loss of livestock, 72,798 acres area covered
by sea water and all amounting to a total loss of about US $ 4 billion.

3 Tropical cyclone prediction models


3.1 Hurricane Weather Research and Forecast (HWRF) model
The HWRF modeling system was developed based on the NMM (non-hydrostatic mesoscale model) core of the WRF model with an objective to improve the skills in predicting
tropical cyclone tracks and intensity. This model was designed to take into account the
strengths of the WRF software system, the use of the well-tested NMM dynamic core, and

123

Nat Hazards (2012) 63:13931411

1397

the physics packages of the highly successful GFDL hurricane forecast system. The model
has a two-way interactive movable nested grid that follows the forecasted path of tropical
cyclone. The configuration of HWRF system consists of outermost (parent) domain and the
movable nested grid with resolutions of 27 and 9 km, respectively, and 42 vertical levels.
The operational HWRF model at NCEP is coupled in the Atlantic basin to a high-resolution Princeton University 3-D ocean model (POM). Atmospheric component of this
model is being adopted for simulating Bay of Bengal cyclones in this study.
HWRF uses a non-hydrostatic system of equations on a rotated latitude/longitude
Arakawa-E grid. HWRF employs the simplified Arakawa Schubert (SAS) for the cumulus
convection scheme (Arakawa and Schubert 1974; Grell 1993), Ferrier (2005) for microphysics parameterization, GFS non-local planetary boundary layer (Troen and Mahrt 1986;
Hong and Pan 1996) scheme with surface layer physics of Moon et al. (2007) and GFDL
radiation scheme. The model uses initial and boundary conditions from GFS model, which
are subjected to relocation and bogusing of the vortex structure (Liu et al. 2008) using
3DVAR data assimilation using a high-resolution grid point statistical interpolation (GSI)
described in Sect. 4. More HWRF model information is available from Gopalakrishnan
et al. (2006), Tallapragada et al. (2008) and the HWRF website at (http://www.emc.ncep.
noaa.gov/HWRF/index.htm).
3.2 Advanced Research WRF (ARW) model
ARW mesoscale atmospheric modeling system has been developed and sourced from
National Center for Atmospheric Research (NCAR), as the next generation model after
MM5, incorporating the advances in atmospheric simulation system suitable for a broad
range of applications. This model system has versatility to choose the domain region of
interest; horizontal resolution; interactive nested domains and with various options to
choose parameterization schemes for convection, planetary boundary layer (PBL), explicit
moisture; radiation and soil processes. ARW is designed to be a flexible, state-of-the-art
atmospheric simulation system that is portable and efficient on available parallel computing platforms and a detailed description was provided by Skamarock et al. (2008).
ARW model system was used in this study for its accurate numerics, higher order mass
conservation characteristics, and advanced physics. The model consists of fully compressible non-hydrostatic equations and the prognostic variables include the threedimensional wind, perturbation quantities of pressure, potential temperature, geo-potential,
surface pressure, turbulent kinetic energy, and scalars (water vapor mixing ratio, cloud
water, etc.). The model equations are formulated using mass-based terrain-following
coordinate system, and solved in Arakawa-C grid using RungeKutta third-order time
integration techniques. The model has several options for spatial discretization, diffusion,
nesting, and lateral boundary conditions. The ARW solver is the key component of the
modeling system, which is composed of several initialization programs for idealized, and
real-data simulations, and the numerical integration program. ARW supports horizontal
nesting that allows resolution to be focused over a region of interest by introducing an
additional grid (or grids) into the simulation with the choice of one-way and two-way
nesting procedures.
3.3 Mesoscale Model Version 5 (MM5) model
NCAR MM5, a non-hydrostatic primitive equation model developed by Pennsylvania State
University (PSU)/National Center for Atmospheric Research (NCAR) (Dudhia 1993; Grell

123

1398

Nat Hazards (2012) 63:13931411

et al. 1995) was used in the present study. This is a three-dimensional, limited-area model
with a terrain-following vertical coordinate. The model is based upon a set of equations for
a fully compressible atmosphere in a rotating frame of reference. This model has versatility
to choose the domain region of interest; horizontal resolution; interacting nested domains;
non-hydrostatic dynamics, which allows the model to be used at a few-kilometer scale;
multitasking capability on shared and distributed-memory machines; a four-dimensional
data assimilation capability, and with various options to choose parameterization schemes
for convection, planetary boundary layer, explicit moisture; radiation and soil processes.
The non-hydrostatic version employs the terrain-following sigma (r) coordinate, and the
pressures at the r levels are determined from a reference state that is estimated using the
hydrostatic equation from a given sea-level pressure and temperature with a standard lapse
rate. The model equations are formulated using terrain-following r-coordinates, solved in
Arakawa-B grid with Leapfrog time integration scheme with time-splitting technique. The
model has several options for spatial discretization, diffusion, nesting, and lateral boundary
conditions. MM5 supports horizontal nesting that allows resolution to be focused over a
region of interest by introducing an additional grid (or grids) into the simulation with the
choice of one-way and two-way nesting procedures.

4 Vortex initialization procedure for Bay of Bengal tropical cyclones


4.1 HWRF model
HWRF model uses a vortex initialization and relocation algorithm based on observed
tropical cyclone position and intensity parameters. These parameters are provided operationally in real-time by National Hurricane Center (NHC) for the Atlantic and Eastern
Pacific basins, and by the Joint Typhoon Warning Center (JTWC) for all other oceanic
basins. Regional Specialized Meteorological Center (RSMC), along with India Meteorological Department (IMD) in New Delhi, is responsible to provide observed track and
intensity for tropical cyclones originating in the North Indian Ocean region.
The grid point statistical interpolation (GSI) 3D-VAR assimilation was implemented for
the vortex initialization as a potential improvement for bogusing the vortex. The relocation
procedure allows for vortex relocation and intensity adjustment to better match the actual
tropical cyclone observations. The usage of real-time data in the HWRF hurricane core,
together with the higher resolution of the model, allows for more accurate hurricane
predictions of intensity and structure. A composite synthetic vortex is inserted in the storm
location (based on observed estimates) preceded by the data assimilation.
The initialization of tropical cyclones in the HWRF model consists of four major steps:
(1) interpolate the global analysis fields from the global forecast system (GFS) onto the
operational HWRF model grids; (2) remove the GFS vortex from the global analysis; (3)
add the axi-symmetric synthetic vortex (constructed based on a series of HWRF model
forecasts); and (4) add through data assimilation any available observational data in the
vicinity of tropical cyclone area. Steps (3) and (4) provide major advancements over the
conventional GFDL tropical cyclone initialization procedure (Kurihara et al. 1995).
The operational vortex initialization procedure used at NCEP is modified for application
over the Bay of Bengal basin. The changes include reconstruction of axi-symmetric synthetic vortex for Bay of Bengal basin, use of JTWC/IMD provided parameters for observed
location and maximum wind speed and modification to GSI procedure that utilizes
available satellite radiance datasets over the Indian Ocean region.

123

Nat Hazards (2012) 63:13931411

1399

4.2 ARW model and MM5


Both ARW and MM5 models use somewhat similar methods for vortex initialization. To
insert an initial vortex closer to the observed storm intensity, the erroneously large vortex
in the background must be first removed. The first step of the removal process is to identify
the vortex corresponding to the storm of interest in the first-guess field. This is accomplished by searching for the maximum vorticity on the analysis pressure level nearest to the
surface within a prescribed radial distance (*400 km) from the best track location of the
tropical cyclone. The point of maximum vorticity then serves as the center of the vortex to
be removed. The approach we adopted is to modify the vorticity, geostrophic vorticity, and
divergence, then solve for the change in the non-divergent stream function, geopotential,
and velocity potential and compute a modified velocity field. The bogus storm to be added
to the background field is axi-symmetric in the current version of the scheme and hence,
will not affect the storm motion.
The vorticity is set to zero outside a radius of 400 km. The Dirichlet boundary conditions stream function is set to zero and used the successive over relaxation (SOR) method
to solve the equations. Once the non-divergent wind is calculated, it is subtracted from the
first-guess U and V wind fields. Similarly, the divergent wind is subtracted from the firstguess wind fields followed by removal of the geopotential height anomaly from the firstguess field. The temperature anomaly field and sea-level pressure perturbations are then
removed from the first-guess fields, leaving with only the background flow where the firstguess storm was originally located. The input data to the bogusing scheme consists mainly
of storm location and estimated maximum winds. The specification of a three-dimensional
vortex structure is arbitrary. The bogus storm profile chosen here is based on the following
assumptions: (1) Axi-symmetry; (2) Vorticity specified within 400 km of the bogus storm
center; (3) Radius of maximum wind (RMW) fixed (90 km on 45 km grid); (4) Mass and
wind fields in non-linear balance; (5) Nearly saturated (w.r.t. water or ice) core; no eye (on
45 km grid); and (6) Maximum winds of bogus storm are a predetermined fraction of
maximum winds observed.
For ARW model, maximum winds of the bogus storm are a user-specified fraction of
maximum winds observed. This is an input parameter in the ARW name list input. The
wind profile of the vortex is given by a Rankine vortex. The relative humidity field within
the radius of maximum winds is defined as nearly saturated. Outside twice the radius of
maximum winds, the relative humidity is not affected. Between the two radii, a linearly
weighted relative humidity is constructed.

5 Results and discussion


Numerical experiments were performed to predict the intensification and movement of two
cyclones Sidr and Nargis with three different meso-scale models MM5, ARW, and
HWRF. Since the initialization of these models is different, a description of the structure of
the cyclonic systems at the initial time of model integration was briefly presented in this
section. The model predicted tracks were compared with both the IMD and JTWC track
positions and the vector track errors were presented. Model predicted minimum central
surface pressure (CSP) and the maximum wind (MW) at 925 hPa level were analyzed and
discussed. These estimates are instantaneous values obtained from the finest model grid at
respective forecast output times.

123

1400

Nat Hazards (2012) 63:13931411

5.1 Sidr
The characteristics of the Sidr cyclone were presented in Sect. 2.1. The spatial distribution of mean sea-level pressure and wind speed of 925 hPa level, corresponding to
2007111300 (initial time of model integration) for MM5, ARW, and models are analyzed
in this section The initial conditions for these models were deduced with the initialization
of bogus vortex as per the methods described in Sect. 4. With MM5 model, (Fig. 1a) the
spatial distribution of surface pressure shows concentric isobars with a minimum of
991 hPa and with the location at 89.6E, 12.1N and the horizontal pressure gradients are
stronger in the northwestsoutheast direction as compared to southeastnorthwest. Correspondingly the distribution of wind speed at 925 hPa level (Fig. 1b) shows a maximum
of 35 m/s with the radius of maximum (RMW) wind at 100 km. The vertical variation of
vorticity (Fig. 1c) at RMW shows maximum cyclonic vorticity of 36 9 1e-5/s below
600 hPa level, decreasing rapidly to reach 6 9 1e-5/s at 200 hPa level. The vertical
variation of wind speed (Fig. 1d) shows maximum at 850 hPa level and the wind speed
decreasing rapidly above 600 hPa level. With ARW model, spatial distribution of surface
pressure (Fig. 2a) shows concentric isobars with a minimum of 964 hPa and with the
location at 12N and 89.5E. Correspondingly, Fig. 2b shows wind distribution at 925 hPa
level with a maximum of 45 m/s with RMW as 150 km. The vertical variation of vorticity
at RMW (Fig. 2c) shows maximum cyclonic vorticity of 36 9 1e-5/s below 600 hPa level,
decreasing rapidly to reach 0.0/s at 100 hPa level. The vertical variation of wind speed
(Fig. 2d) shows maximum at 920 hPa level and the wind speed decreases rapidly above
600 hPa level.
The distributions of mean sea-level pressure at initial time for HWRF shown in Fig. 3a
indicate that the cyclone vortex has a minimum CSP of 960 hPa with the center located at
90.1E and 11.1N. The sea-level pressure distribution shows more axi-symmetric concentric
isobars with strong pressure gradients compared to initial conditions for other models.
Correspondingly, the maximum wind distribution shown in Fig. 3b indicates a maximum
value of 44 m/s near the storm center, and is much closer to the observed intensity of Sidr.
The EW cross-section along the storm center is presented in Fig. 3c, which shows vertically stacked vortex with a very narrow eye-region and maximum winds extending to just
about 50 km from the storm center. The simulated storm has a maximum wind of 135 kts
(about 70 m/s) at about 800 hPa. This is a direct result of advanced HWRF initialization
procedure, where the model initial fields are adjusted to the observed parameters and the
3-d fields of winds, temperature and moisture are rebalanced over the assimilation domain.
The warm core (dashed lines) in the initial vortex is extended all the way to about 200 hPa.
The above description of initial conditions show that ARW and HWRF models have the
initial vortex of nearly same strength in terms of minimum CSP and MW, whereas MM5
has a weaker vortex. As such, the features of initial vortex are more nearer to the observations in the ARW and HWRF models.
The model predicted track positions at 6 h interval during the period of model integration are shown in Fig. 4a along with IMD and JTWC estimations. The vector track
errors were computed for the three models predictions with respect to both IMD and JTWC
and are presented in Fig. 4b. For the track errors with respect to IMD, MM5 prediction
shows errors gradually increasing to reach about 147, 154, and 356 km; ARW model has
the track errors of 137, 407, and 464 km; and HWRF had 109, 145, and 167 km at 24, 48,
and 72 h predictions, respectively. The track prediction with MM5 was noted to be slower
and ARW to be faster as compared to the observed track whereas HWRF nearly follows

123

Nat Hazards (2012) 63:13931411

1401

Fig. 1 Initial conditions for cyclone Sidr from MM5 model valid at 00Z Nov. 13, 2007. a Distribution of
sea-level pressure (hPa); b Distribution of winds at 925 hPa (m/s); c Vertical variation of relative vorticity
(s-1) at radius of maximum winds and d Vertical variation of wind speed at radius of maximum winds

the observed track. This indicates better prediction of speed of the cyclone system with
HWRF as compared to MM5 and ARW models.
The time variation of minimum CSP and MW with the three models is presented in
Fig. 5a, b. It was noted that the estimates from IMD and JTWC differ significantly with
JTWC estimates higher by 2030%. The maximum attainment in terms of CSP (MW) was
944 hPa (59 m/s) by IMD and 918 hPa (72 m/s) by JTWC. A comparison of the prediction
from the three models shows that HWRF produces the best prediction both in terms of CSP
and MW. MM5 fails to predict the intensification as the initial vortex was much weaker
than the observations.

123

1402

Nat Hazards (2012) 63:13931411

Fig. 2 Same as Fig. 1 except for ARW model

Though ARW model predicted intensification with minimum CSP as 956 hPa (59 m/s),
the time of attainment of maximum intensity was earlier by 1218 h. The prediction with
HWRF nearly coincides with the JTWC estimations, although with 1020% underestimation. These results indicate that HWRF model produces the best prediction both in terms
of track as well as intensification for tropical cyclone Sidr, when compared to observed
estimates from both IMD as well as JTWC. The comparison of track errors with JTWC
estimates shows the errors to be 149, 189, 415 km with MM5; 154, 376, 428 km with
ARW and 85, 112, 189 km with HWRF at 24, 48, and 72 h, respectively. These errors

123

Nat Hazards (2012) 63:13931411

1403

Fig. 3 Initial conditions for cyclone Sidr from HWRF model valid at 00Z Nov. 13, 2007. a Distribution of
sea-level pressure (hPa), b Distribution of 10 m winds (m/s) and c Vertical cross-section of winds (kts,
shaded) and temperature (oC, dashed lines) along the EW direction centered at 12.1N

clearly show the prediction with HWRF model to be significantly superior to ARW and
MM5 model as the track errors were below 150 km up to 66 h of prediction whereas MM5
and ARW models have the errors rapidly increasing after 24 h.

123

1404

Nat Hazards (2012) 63:13931411

Fig. 4 a Comparison of track forecasts for cyclone Sidr from HWRF, ARW and MM5 models. Initial
forecast time is valid at 00Z, Nov. 13, 2007. Estimated observed track positions from JTWC and IMD are
also shown. b Vector track errors in km, computed with respect to IMD estimates

5.2 Nargis
The characteristics of the Nargis cyclone were presented in Sect. 2.2. Cyclone Nargis
moved northeast and with an attained intensity of about 960 hPa. This cyclone system
maintained nearly the same intensity from 06 UTC of 29 April 0818 UTC of 1 May 08
and then intensified with a pressure drop of 8 hPa and 10 hPa within the next 26 h period
finally attaining the mature stage at 06 UTC of 2 May 08. All three models were integrated
starting with initial forecast time valid at 12 Z of April 29, 2008. With the MM5 model, the
spatial distribution of sea-level pressure shown in Fig. 6a depicts 4 concentric isobars with
minimum CSP of 995 hPa located at 85.5E and 13N. The pressure gradients were stronger
toward south as compared to northern side. The distribution of wind speed at 925 hPa
level, shown in Fig. 6b, indicates a maximum of 25 m/s with the contours oriented
northeastsouthwest with a RMW at 110 km. The vertical variation of vorticity at RMW
(Fig. 6c) shows a maximum of 15 9 1e-5/s below 600 hPa level and rapidly decreases
above to reach 3 9 1e-5/s at 200 hPa level. Correspondingly, the vertical variation of
wind speed (Fig. 6d) shows the maximum at 925 hPa level of 26 m/s, decreasing to 17 m/s

123

Nat Hazards (2012) 63:13931411

1405

Fig. 5 Time variation of model predicted (a) central surface pressure (hPa) and (b) maximum winds (m/s)
for cyclone Sidr. Initial forecast time is valid at 00Z, Nov. 13, 2007. Observed estimates from IMD (for CSP
and MW) and JTWC (for MW) are also shown

at 700 hPa level, increases from 700 to 550 hPa and then decreases above 550 hPa level to
reach about 9 m/s at 200 hPa level. With ARW model, the initial sea-level pressure distribution (Fig. 7a) shows 3 concentric isobars with a minimum CSP of 995 hPa located at
85.5E and 13.5N. The isobars show stronger pressure gradient toward southeast. The
spatial distribution of wind speed (Fig. 7b) shows a maximum of 19 m/s with the contours
oriented northeastsouthwest and with a RMW at 120 km. The vertical variation of vorticity at RMW (Fig. 7c) shows a maximum of 8 9 1-5/s below 600 hPa level, decreasing
above to reach 3 9 1e-5/s at 250 hPa level. The vertical variation of wind speed (Fig. 7d)
shows the wind speed maximum 21 m/s at 950 hPa level, decreasing to 12 m/s at 650 hPa
level; an increase from 650 hPa up to 550 hPa level and then decreasing up to 300 hPa
level, increasing slightly at 200 hPa level and decreases above 200 hPa level reaching
above 6 m/s at 100 hPa level.
Sea-level pressure (Fig. 8a) and 10 m wind distribution (Fig. 8b) from HWRF model
shows asymmetric structure of the initial vortex with maximum winds concentrated in a
region within a radius of about 50 km from the storm center. Application of vortex initialization procedure for this case enhanced the wind field from GFS initial fields of about
20 m/s to the observed value of 44 m/s. HWRF model has the initial vortex with a minimum CSP of 958 hPa with the center located at 13.6N and 85.2E. These values are
consistent with the JTWC estimates of initial location and intensity of tropical cyclone
Nargis. Initial conditions for MM5 and ARW were closer to IMD estimates, and hence are
weaker than that of HWRF.

123

1406

Nat Hazards (2012) 63:13931411

Fig. 6 Initial conditions for cyclone Nargis from MM5 model valid at 12Z Apr. 29, 2008. a Distribution of
sea-level pressure (hPa); b Distribution of winds at 925 hPa (m/s); c Vertical variation of relative vorticity
(s-1) at radius of maximum winds and d Vertical variation of wind speed at radius of maximum winds

The model predicted track positions with 6 h interval for three models along with IMD
and JTWC are shown in Fig. 9a and the track errors were presented in Fig. 4b. Both JTWC
and IMD estimates have nearly the same track positions. Of the 3 models, HWRF produces
best track nearly coinciding with the observations up to 30 h and then with increasing
errors due to higher speed of the model storm. Both MM5 and ARW models have larger
errors as their predictions are much to the north of the observations. However, the northeast
ward movement of the storm was predicted by all the models. MM5 model predicts the
speed of the cyclonic storm further than the other two models and ARW model has higher
speed of motion than HWRF. The vector track errors, as compared with IMD. Show that
the track errors to be 58, 277 and 668 km for MM5; 187, 207, and 346 km for ARW model
and 90, 235, and 273 for HWRF models at 24, 48, and 72 h, respectively. The vector track
errors with respect to JTWC estimates are 149, 189, and 415 km for MM5; 154, 376, and
428 km for ARW and 85, 112, and 189 km for HWRF models at 24, 48, and 72 h,

123

Nat Hazards (2012) 63:13931411

1407

Fig. 7 Same as Fig. 6 except for ARW model

respectively. These errors show that HWRF produced better track prediction than MM5
and ARW models and that ARW is slightly better than MM5 model.
The time variation of CSP and MW are shown in Fig. 10 (alb). It is to be noted that the
estimates from IMD and JTWC differ significantly with JTWC higher than those of IMD.
A comparison of prediction from three models show that ARW predicts the highest
intensity and HWRF predicts intensity nearer to the JTWC estimations. ARW also produced a very good prediction of intensity whereas MM5 model had weaker intensity
compared to ARW and HWRF models. However, both HWRF and ARW models predict
the time of attainment of maximum intensity at 48 h and MM5 at 60 h as compared to the
observed intensity maximum at 72 h. The faster intensification could be a reason for the
faster model cyclone as compared to observations. HWRF predicted a minimum CSP of
942 hPa and maximum wind of 57 m/s as compared to 952 hPa and 76 m/s with ARW;
962 hPa and 51 m/s with MM5 model. These results suggest that HWRF model provided
better skill in predicting track and intensity for tropical cyclone Nargis.

123

1408

Nat Hazards (2012) 63:13931411

Fig. 8 Initial conditions for cyclone Nargis from HWRF model valid at 12Z Apr. 29, 2008. a Distribution
of sea-level pressure (hPa) and b Distribution of 10 m winds (m/s)

Fig. 9 a Comparison of track forecasts for cyclone Nargis from HWRF, ARW and MM5 models. Initial
forecast time i valid at 12Z, Apr. 29, 2008. Estimated observed track positions from JTWC and IMD are also
shown. b Vector track errors in km, computed with respect to IMD estimates

123

Nat Hazards (2012) 63:13931411

1409

6 Summary and future work


Research in the area of meso-scale modeling of tropical cyclones gained momentum with
the increased computational abilities, advanced initialization methods, and community
modeling frame work that supports wide range of parameterization schemes specially
designed to represent physical processes applicable to the tropical systems. The advanced
HWRF modeling system has been developed at NCEP/EMC to provide operational
guidance to the National Hurricane Center (NHC) for the tropical cyclones originated
Atlantic and Eastern Pacific basins. This study is an attempt to utilize the HWRF modeling
system along with the advanced operational vortex initialization procedure in simulating
the tropical cyclones of the Bay of Bengal region. Two other popular meso-scale models,
the NCAR ARW and NCAR/PSU MM5 models were also used in this study to demonstrate the superior skill of HWRF modeling system. Much of the skill of HWRF model can
be attributed to the specialized vortex initialization procedure that includes vortex relocation, use of a synthetic vortex generated exclusively for the Bay of Bengal cyclones, and
use of GSI that facilitates assimilating any available satellite data in the tropical cyclone
core region. The advanced two-way interactive movable nest is another unique feature of
the HWRF model that is designed to accurately track and predict intensity of tropical
storms. HWRF model is known to have some inherent issues like erroneous wind-pressure

Fig. 10 Time variation of model predicted (a) Central surface pressure (hPa) and (b) Maximum winds
(m/s) for cyclone Nargis. Initial forecast time is valid at 12Z, Apr. 29, 2007. Observed estimates from IMD
(for CSP and MW) and JTWC (for MW) are also shown

123

1410

Nat Hazards (2012) 63:13931411

relationship (HWRF tends to have lower sea-level pressure for a prescribed wind speed
compared to observations). Results from the two experiments in simulating tropical
cyclones Nargis and Sidr revealed better intensity prediction skill from HWRF model in
terms of wind speeds compared to prediction of sea-level pressure. Among the other two
models, ARW demonstrated comparable track prediction skill, but the model intensity
prediction suffered from both low and high biases. Discrepancy in the initial maximum
sustained winds arising from the use of observed estimates from different sources (JTWC
and IMD) was another reason why initial conditions for ARW and MM5 differ from those
of HWRF.
This study is a first-of-its kind where the NCEP operational hurricane modeling system
was tuned for simulating tropical cyclones originated in the Bay of Bengal region. Much
work still needs to be done in improving the specification of synthetic vortex, use of
analysis cycling, use of additional observations in the model domain, and further tuning of
model parameterizations schemes to take into account the characteristics of tropical
cyclones in the North Indian Ocean.

References
Arakawa A, Schubert WH (1974) Interaction of a cumulus cloud ensemble with the large-scale environment.
Part I. J Atmos Sci 31:674701
Bender MA, Ginis I, Tuleya R, Thomas B, Marchok T (2007) The operational GFDL coupled hurricaneocean prediction system and a summary of its performance. Mon Weather Rev 135:39653989
Cheng WYY, Steenburgh WJ (2005) Evaluation of surface sensible weather forecasts by the WRF and the
Eta models over the Western United States. Weather Forecast 20:812821
Dudhia J (1993) A nonhydrostatic version of the Penn State/NCAR mesoscale model: validation tests and
simulations of an Atlantic cyclone and cold front. Mon Weather Rev 121:14931513
Dvorak V (1984) Tropical cyclone intensity analysis using satellite data. NOAA Technical Report NESDIS
11, 47 pp. Available from NOAA/NESDIS, 5200 Auth Rd. Washington, DC, 20233
Ferrier BS (2005) An efficient mixed-phase cloud and precipitation scheme for use in Operational NWP
Models. EOS Trans. AGU, 86, Jt. Assem. Suppl., Abstract A42A-02
Gopalakrishnan SG, Surgi N, Tuleya R, Janjic Z (2006) NCEPs two-way-interactive-moving-nest NMMWRF modeling system for hurricane forecasting. Preprints, 27th conference on Hurricanes and
Tropical Meteorology, Monterey, CA, Amer. Meteor. Soc., Ar. 7A.3
Gopalakrishnan SG, Liu Q, Marchok T, Sheinin D, Surgi N, Tuleya R, Yablonsky R, Zhang X (2010)
Hurricane weather research and forecasting (HWRF) Model scientific documentation. In: Bernardet L
(ed) Developmental Testbed Center, 80 pp. Available online at http://www.dtcenter.org/HurrWRF/
users/docs/scientific_documents/HWRF_final_2-2_cm.pdf
Grell GA (1993) Prognostic evaluation of assumptions used by cumulus parameterizations. Mon Weather
Rev 121:764787
Grell GA, Dudhia J, Stauffer DR (1995) A description of the fifth-generation Penn State/NCAR mesoscale
model (MM5). NCAR Tech. Note NCAR/TN-398 ?STR, p 122
Hong S-Y, Pan H-L (1996) Nonlocal boundary layer vertical diffusion in a medium-range forecast model.
Mon Weather Rev 124:23222339
Knapp KR, Kruk MC (2010) Quantifying interagency differences in tropical cyclone best-track wind speed
estimates. Mon Weather Rev 138:14591473
Kurihara Y, Bender MA, Tuleya RE, Ross RJ (1995) Improvements in the GFDL hurricane prediction
system. Mon Weather Rev 123:27912801
Liu Q, Surgi N, Lord S, Tuleya RE, Tallapragada V, Kwon Y (2008) Initalizations in advanced operational
hurricane WRF (HWRF) model. Extended abstract, 28th conference on Hurricanes and Tropical
Meteorology, Orlando, Florida, Amer. Meteo, Soc, Paper 4A.2
Moon IL-JU, Ginis I, Hara T, Thomas B (2007) A physicalbased parameterization of airsea momentum
flux at high wind speeds and its impact on hurricane intensity predictions. Mon Weather Rev
135:28692878

123

Nat Hazards (2012) 63:13931411

1411

Patra KP, Santhanam MS, Potty KVJ, Tewari M, Rao PLS (2000) Simulation of tropical cyclones using
regional weather prediction models. Curr Sci 79(1):7078
Sousounis PJ, Hutchinson TA, Marshall SF (2004) A comparison of MM5, WRF, RUC, ETA performance
for great plains heavy precipitation events during the spring of 2003. In: Preprints 20th conference on
weather analysis and forecasting, Seattle, Amer. Meteor. Soc., J24.6
Surgi N, Tuleya R, Liu Q, Tallapragada V, Kwon Y (2008) Advancement of the HWRF for the Next
Generation Prediction at NCEPs Environmental Modeling Center. In: Proceedings of the 62nd
interdepartmental hurricane conf, Charleston, SC, Office of the Federal Coordinator for Meteorology,
32 pp. Available online at http://www.ofcm.gov/ihc08/linking_file_ihc08.htm
Tallapragada V, Surgi N, Liu Q, Kwon Y, Tuleya R, OConnor W (2008) Performance of the advanced
operational HWRF modeling system during pre-implementation testing and in real-time 2007 hurricane season. Extended abstract, 28th conference on Hurricanes and Tropical Meteorology, Orlando,
Florida, Amer. Meteo, Soc, Paper 4A.5
Troen I, Mahrt L (1986) A simple model of the atmospheric boundary layer; sensitivity to surface evaporation. Boundary Layer Meteorol 37:129148

123

S-ar putea să vă placă și