Full text

advertisement
Characteristics of Atmospheric Disasters and Rainfall in Pre- and Mature
Summer Monsoon Seasons over Northeast India and Bangladesh
Toru Terao1*, Fumie Murata2, Md. Nazrul Islam3 and Taiichi Hayashi4
1 Kagawa University, Japan
2 Kochi University, Japan
3 SAARC Meteorological Research Centre, Bangladesh
4 Kyoto University, Japan
Abstract
Bangladesh is one of the most atmospheric disaster prone countries in the world. Tropical
cyclones, tornadoes, river floods, and various kinds of atmospheric disasters attack this country
almost every year. We summarized the characteristics of disasters that are different from
season to season. Our raingauge network clearly detected the difference of rainfall intensity
between pre- and mature summer monsoon seasons in Bangladesh. Using these raingauges, we
constructed a gridded hourly rainfall amount dataset for the use of the hydrological modeling of
flash floods that take place in Sylhet area, the northeasetern most part of the country. It is
confirmed that the gridded dataset is useful for the eastern part of Sylhet division, where dense
raingauge network has been established. We also discuss about the validity of Dhaka radar
data which is potentially highly important for the rainfall estimation.
1. Introduction
The area of Bangladesh and Meghalaya state in India just north of Bangladesh is well known
as one of the heaviest rainfall areas in the world. This heavy rainfall heats atmosphere over this
area, playing a major role in the atmospheric heating that is responsible for the entire Asian
monsoon circulation. On the one hand, this heavy rainfall supplies precious water resources,
and on the other hand, that causes severe atmospheric disasters (Karmakar and Alam, 2005)1.
The tremendous catastrophe brought by the cyclone Sidr is still a fresh memory of ours.
We are making a research project on the mechanisms of rainfall in this region. Firstly, we
conducted rawin sonde observation to observe the upper layer atmosphere. And recently, we
are developing an automatic raingauge network over this area for the observational study of
rainfall characteristics.
This paper outlines our research activities and major results over Bangladesh area. In section 2,
we summarize the characteristics of seasonal march of atmospheric circulation and disasters
over this area. Section 3 will be devoted to the description of the development of the
atmospheric observation network and the knowledge comes from them. In section 4, the focus
is on an assessment of the impact of the usage of our raingauge network data for the prediction
of flash floods in Sylhet area. Two case studies on the severe local storms developed over
Bangladesh during pre-monsoon season is shown in section 5. A recent attempt for calibration
of radar data using our raingauge data is shortly described in section 6. The last section is
summary.
2. The seasonal march of the characteristics of disasters in Bangladesh
The aspects of the characteristics of atmospheric disasters in Bangladesh largely vary as the
seasonal march. In other words, Bangladesh is a disaster prone country in any season.
Tropical cyclones are one of the most popular causes of disaster in Bangladesh (Table 1).
They form over the Indian Ocean, and some of them Table 1: Damages caused by
go up north through the Bay of Bengal, and attack
tropical cyclones over Bangladesh.
Bangladesh. However, as is shown in Table 1, most The number of death in 2007 is
of cyclones striking Bangladesh develop in April- based on the latest announcement
May and October-November. Not like typhoons over by Bangladesh government.
western North Pacific and hurricanes over eastern
D eath
Year M onth
Pacific and Atlantic Ocean, they form not in summer
1970
11
500000
but seasons before and after summer. The reason is
1985
5
10000
probably because of the strong vertical wind shear
1991
4 140000
prevailing over Indian Ocean in summer season. In
2007
11
4166 ?
summer, as the establishment of the Asian summer
monsoon circulation, lower tropospheric westerly and upper tropospheric easterly evolve to
form a strong vertical shear. This suppresses the development of tropical cyclones with axissymmetric vertical structures (Gray, 1967)2.
However, the atmospheric disaster in Bangladesh is caused not only by tropical cyclones, but
also by various different types of atmospheric disturbances. The summer season from May to
September can be divided into two different stages from the view point of the feature of the
atmospheric disasters.
Figure 1 shows the average seasonal march of monthly rainfall, monthly averaged daily
sunshine duration, and monthly averaged daily maximum temperature at Dhaka, the capital of
Bangladesh (The average is calculated as 1961-1999 mean). Roughly speaking, it is a typical
monsoon climate with summer rainy season and winter dry season.
Looking more closely, we find that the rainfall attains its peak value in May, although the
sunshine duration is still rather high in May in comparison with June. That is, although we have
much rainfall in May
comparable to that in mature
monsoon season, cloud is not
so much, and it is dry and hot.
It is in June that the cloudy
wet season begins.
The
former is called the premonsoon season, and the
latter the mature monsoon
season.
During the mature monsoon
season, well known and
frequently reported extensive
and
long-lasting
flood
prevails over this country.
This type of floods is called
the “river flood,” which will
be discussed in Section 3 later.
On the other hand, during
the pre-monsoon season,
Fig. 1: Climatological seasonal march at Dhaka
disasters are caused by the
Bangladesh. The monthly rainfall in mm (a), the daily
severe local storms formed as
sunshine duration in hour (b), and the daily maximum
the meso-scale convective
temperature in C . Values are averaged for 39 years from
complex with 20 to 100 km
1961 to 1999.
horizontal scale. Those meso-scale disturbances are characterized by strong ascending motion
due to the unstable stratification, and gusty wind associated with the formation of relatively cool
air through the evaporation of precipitation.
Such disturbances are dreaded as “Kalbaishaki” by local people. They sometimes accompany
tornadoes. On 14 April 2004, tornadoes attacked the village Netrokona, killing 70. Bangladesh
is one of the most tornado prone areas in the world (Yamane and Hayashi, 2006)3. Even if they
do not accompany tornadoes, strong gusty winds often cause severe atmospheric disasters. On
23 May 2005, strong gust wind made two ferry boats capsized on the Meghna river near Dhaka,
killing more than 200 people. Although it seems to be behind the toll caused by tropical
cyclones and monsoon river floods, severe damage caused by severe local storms in premonsoon season is also reported almost every year. Quick action is expected. Section 2 will be
devoted to the description of severe local storms.
Such difference of types of disasters reflects that of global circulation pattern. In May, as the
subtropical westerly jet flows just south of the Tibetan Plateau, Bangladesh is under the
influence of extratropical climate systems. However, in middle of June, the subtropical
westerly shifts to the north of Tibetan Plateau. Asian summer monsoon circulation covers
entire south Asia including Bangladesh. Upper tropospheric easterly replaces the subtropical
westerly over south Asia. The lower tropospheric monsoon westerly originated from south
Indian Ocean flows through Madagascar, off the Somali Coast, Arabian Sea, over the Indian
Subcontinent and the Bay of Bengal, that transport huge amount of water vapor resulting in the
abundant monsoon rain.
Bangladesh is attacked by several different types of disasters every year as the seasonal march.
Such differences are associated with the changing global circulation patterns; the meridional
migration of the subtropical jet, the reversal of the temperature contrast between huge landmass
of Eurasian Continent and the Indian and Pacific Oceans, along with the annual solar cycle.
2. The observation network in Bangladesh and India
In 1999, we started the observational
study to investigate the mechanisms of
atmospheric disasters over Bangladesh.
Firstly, we conducted the 4-times
daily
upper
layer
sounding
observation using rawin sonde at
Dhaka (Terao et al., 2005)4. We
described the pattern of diurnal
rainfall variations over this country.
We reconfirmed that, in northeastern
part of this country where heaviest
rain falls, the midnight-early morning
rainfall peak is prominent. Further,
we reported the existence of nocturnal
jet that is a candidate of the causes of
the midnight-early morning rainfall
peak in northeastern part of country.
Later, we developed the raingauge
networks (Fig. 2). One is the country
wide scale raingauge network,
Fig. 2: Map of Bangladesh. Shade indicates
elevations. Country wide scale raingauge network
is shown by open boxes.
consists of 6 raingauges at Dhaka,
Chittagong, Dinajpur, Rajshahi, Sylhet and
Mymensingh, established mainly in
summer 2004 (open boxes in Fig. 2).
Another is the local scale dense raingauge
network distributed over Sylhet and
southern slope of Meghalaya Plateau
installed in the pre-monsoon season in
2006. This network comprises more than
15 gauges including 5 gauges in India.
Three automatic weather stations (AWSs)
are installed at Dhaka in 2004, and
Jaintiapur Bangladesh and Cherrapunjee
Meghalaya in 20075,6.
3. Analysis of rainfall intensity.
It is qualitatively well known that the
rainfall intensity is quite different between
pre- and mature summer monsoon seasons
over Bangladesh. However, it is not well
described quantitatively, since there is no
attempt to analyze rainfall characteristics
missing
Fig. 3: Monthly rainfalls in mm at 6 raingauges
from July 2004 to July 2005.
Fig. 4: Contribution ratio of each rainfall intensity category for pre- and mature monsoon
seasons at (a)Mymensingh, (b)Sylhet, (c)Dhaka, and (d)Chittagong.
using automatic rain gauges that measures rainfall variability with high temporal resolution.
Our raingauge network enables us to analyze the variability of rainfall intensity with 10
minutes resolution. The rain measurements with finest temporal resolution used for the past
studies in this area were 3 hourly operational rainfall observations by BMD. Figure 3 shows the
monthly precipitation for four observatories, Mymensingh, Sylhet, Dhaka and Chittagong. In
this analysis, rainfall in January-May and that in July-December are defined as pre- and mature
monsoon rainfall, since rainfall in June is considered to be a mixture of pre- and mature
monsoon rainfall. In Fig. 4, we show the difference in the rainfall intensity between pre- and
mature summer monsoon seasons. It is a series of histograms showing the contribution ratios of
rainfall amount within different categories of rainfall intensity.
In the pre-monsoon season, except for Chittagong, the contribution of rain whose intensity is
greater than 5 mm per 10 minutes is more than half of total precipitation. The difference of
rainfall intensity in this area is shown qualitatively at the first time by using our raingauge
network. As is shown in Fig. 3, pre-monsoon rainfall amount is basically larger in northern part
of area. Sometimes, monthly precipitation in May even exceed those during mature monsoon
season
4. The impact assessment of hourly rainfall data upon flash flood prediction.
Recently, we started another project to assess the impact of introduction of hourly rainfall data
upon the prediction of flash flood using hydrological model. Especially in areas near the steep
slopes such as Sylhet area in the northeastern part of the country, flood type other than the river
flood, flash flood occurs frequently (Hofer and Messerli., 2006)7. For the diagnosis and
prediction of this type of flood, good hydrological model using rainfall data with temporally
high resolution is needed. The Institute for Water Modelling (IWM), an NGO for research
purpose in Dhaka, has a hydrological model, which is used for daily description and prediction
of flood in this area. They run this model with the daily rainfall and river water discharge near
boundaries measurements by Bangladesh Water Development Board (BWDB). For river floods,
the temporal resolution of observation is enough, since the time scale of river flood is rather
long. However, for flash floods, it may not be sufficient. Since the observation by BWDB is
based on the man power, it is difficult to increase the temporal resolution of observation. The
density of observatory of BMD is rather low in Sylhet area. Therefore, the rainfall
measurement by our research project is potentially very important for the improvement of
description and prediction of flash floods. Following this consideration, we started the project
to evaluate the model performance using our dense raingauge network in Sylhet and Meghalaya
area.
In Fig. 5, the temporal data coverage of raingauges and automatic weather stations of our
project in Bangladesh and
north-eastern India is shown
schematically. In Sylhet and
Meghalaya area, many raingauges are installed in spring
2006. So, we decided to begin a
pilot diagnostic calculation for
this season. The river network
in this area flows into the
Meghna river near Bhairab
Bazar (Fig. 6). The network is
Fig. 5: Schematic diagram of the temporal data coverage
so complicated, that it is
of our raingauges installed in this area.
NE-36
NE-37
Bhairab Bazar
Fig. 6: River systems in Sylhet area used for the calculations of IWM model. NE-36 and
NE-37 are the names of catchments.
difficult to choose sub-catchments. We calculated for model domain, whole catchments shown
in Fig. 6. The locations of raingauges utilized for calculation are shown in Fig. 7. Clearly, they
are concentrated in eastern half of the model domain. There is only one raingauge in the
western half.
To make data appropriate for the hydrological model, we constructed the gridded dataset by
an interpolation method. The grid interval is set to 0.05 degrees in both longitudinal and
latitudinal directions. The interpolation procedure follows Shepard (1968)8. It is based on
1 / r 2 weighting functions and developed with several intuitive manipulations to overcome
some shortcomings of simple weighting function method.
Fig. 7: Locations of raingauges used for the model
calculation.
Firstly, we calculated only using
the daily data, which are
calculated as the accumulation of
hourly values. The purpose of
this calculation is to validate the
rainfall data of our raingauge
network by comparison with
results using BWDB operational
daily rainfall data. Results are
shown in Fig. 8.
For the catchments NE-36 (Fig.
8a), the results are far different
from each other, showing that the
calculation of hydrological model
(a)
(b)
Fig. 8: Calaulated river runoff from different catchments by IWM model using daily
rainfall data obtained from BWDB (solid line) and our raingauge network (thin line with
triangle). (a)For the catchments NE-36. (b)For the catchments NE-37. See Fig. 6 for the
locations of catchments.
is not realistic if our raingauge network is used as the data source of daily rainfall. On the other
hand, the result for the catchments NE-37 (Fig. 8b) shows good agreements with each other.
The bad result in Fig. 8a reflects the fact that the former catchments are in the western part of
the model domain, where our raingauges are scantily distributed. From the result shown in Fig.
8b, it is concluded that the model calculations using our raingauge are at least as valid as those
from BWDB rainfall data. We are now trying to estimate the impact of the usage of hourly
rainfall data upon the flash flood predictions over Sylhet area.
5. Observation of severe local storms.
The severe local storm that
occurred on 23 May 2005 is
observed by the AWS installed at
Dhaka (Fig. 9). Intense rain with 3.5
mm/min, i.e. greater than 210
mm/hour,
is
observed.
Corresponding with this rainfall,
strong northwesterly gust blows and
temperature falls very rapidly. The
pressure seems to show small rise,
showing the structure of meso-high
pressure
system.
These
characteristics are well coincide with
those of the rain bands well
described by many past studies
(Houze, 1993)9. The radar image
(Fig. 10a) also indicates the line
shape clearly.
Fig. 9: Time series of the meteorological elements
On the other hand, for the case on
observed at Dhaka AWS during the passage of severe
14 April 2004, the AWS could not
local storm on 23 May 2005. Horizontal axis shows
observe the disturbance that caused
hour in local time on 23 May 2005. (a)Rainfall in mm
tornadoes. The radar image (Fig.
/min, (b)temperature in C , (c)wind speed in m/s, and
10b) captured the structure of meso(d)pressure in hPa.
scale disturbances developed in
northern and northeastern part of Bangladesh. These disturbances with about 20 km scales
moved southeastward from Assam area beyond the Meghalaya Plateau, causing severe storms
in Bangladesh. The shape is, interestingly, rather different from the system shown in Fig. 10a.
(a)
(b)
Fig. 10: Dhaka radar images at times designated on the top of each plate. Rainfall
intensity is shown by the legend at the bottom left corner of plate (a).
6. Evaluation of radar data using raingauge data network.
Dhaka radar of BMD is a S-band radar with the 250 km effective radius, providing gridded
and categorized rainfall intensity data in 600 km times 600 km rectangular area with 2.5 km
times 2.5km pixel intervals. Radar data is of course extremely useful for analysis of rainfall
distribution. However, there are some problems to be solved in the data obtained by the Dhaka
radar.
The first problem is the accuracy of the rainfall intensity. The rainfall radar estimates the
rainfall intensity from the refractivity of radar radiation. We should determine the relation
between the refractivity and the rainfall intensity empirically. This relationship is different from
place to place and season to season regarding the difference of characteristics of rainfall and
many other factors. Unfortunately, the Dhaka radar have never rendered to this sort of
evaluation. In spite of some difficulties, Islam et al. (2005)10 have tried to estimate the accuracy
of radar data using the operational 3-hourly BMD observations as the ground truth. They found
overall underestimation of radar rainfall.
The second problem to note is the small resolution of rainfall intensity. The radar system
automatically produces the converted rainfall intensity data. However, the intensity is classified
into only 6 categories, 1-4, 4-16, 16-32, 32-64, 64-128, and over 129. The radar refractivity
data are not archived at all. Thus, there is a difficulty in determining the rain intensity from the
categorized data.
The third one is essential for the consideration in flash flood forecast. The radar is operated
for seven 1 hour durations; 5-6, 8-9, 11-12, 14-15, 17-18, 20-21 and 23-24 in local time, in a
day. The rainfall pattern is so variable in this area, that it is impossible to interpolate the data
without radar observation. Therefore, at this moment, the radar data cannot be used for the
input data of hydrological model, in spite of their big potential importance.
To further utilize the radar data, the evaluation of radar data and the improvement of
observation pattern are keenly needed.
Here, to evaluate the rainfall data estimated by radar, we conducted a comparison between our
Fig. 11: Comparison between hourly rainfall observed by raingauges (solid lines) and radar
(dashed lines) at four stations (a)Mymensingh, (b)Sylhet, (c)Rajshahi, and (d)Chittagong.
Comparisons are made for every possible hour when radar observation is conducted no less
than 10 times during that hour. Horizontal axis is the number of cases. Vertical axis is
observed rainfall intensity in mm/hour.
raingauge network and radar data of the corresponding pixels, as a preliminary first attempt.
We conducted the comparison of hourly rainfall amount for four raingauges, Mymensingh,
Sylhet, Rajshahi and Chittagong, only for August 2004. The ground truth, the hourly rainfall in
mm measured by the raingauge is simply calculated by the tip count multiplied by 0.5.
Corresponding radar estimated hourly rainfall is calculated as follows. Firstly, we neglected
errorneous data. The rainfall intensity at a pixel for rainfall intensity categories 1 to 6 are
assumed as 2.5, 10.5, 24.5, 48.5, 96.5 and 129 mm/h, respectively, following the method of
Islam et al. (2005). Considering the movement of the rainfall systems, we calculated the
averaged rainfall intensity for 3 times 3 pixels with the corresponding pixel the central. All
radar snapshots observed within that 1 hour duration are gathered and simply averaged to make
the hourly rainfall value at the location. Only hours when there are no less than 10 times
observations are assumed to be valid.
Results are shown in Fig. 11. There are about 100 hours for each raingauge when the
corresponding hourly radar data are available. The difference in number of hours is due to the
difference of the starting time of the observation. It is seen that the radar tends to detect rainfall
at the times when the raingauge records rainfall more than 0.5 mm. On the other hand, the
amount of radar rainfall is highly underestimated. The total rainfall measured by raingauge for
Mymensingh, Sylhet, Rajshahi and Chittagong are 34.5, 58.0, 14.5 and 23.5 mm, respectively.
However, those by radar are 9.5, 29.9, 9.2, and 3.1, respectively, which are one-third or half of
ground truth. At Chittagong, as was also pointed out by Islam et al. (2005), severe
underestimation occurs, partly because of the longest distance from the radar site.
Since we have collected much more cases after 2005, we will be extend this result to find any
good way to calibrate the radar data using raingauges.
7. Summary
Bangladesh is one of the most atmospheric disaster prone countries in the world. Tropical
cyclones, tornadoes, river floods, and various kinds of atmospheric disasters attack this country
almost every year. We summarized the characteristics of disasters that are different from
season to season. Our raingauge network clearly detected the difference of rainfall intensity
between pre- and mature summer monsoon seasons in Bangladesh. Using these raingauges, we
constructed a gridded hourly rainfall amount dataset for the use of the hydrological modeling of
flash floods that take place in Sylhet area, the northeasetern most part of the country. It is
confirmed that the gridded dataset is useful for the eastern part of Sylhet division, where dense
raingauge network has been established. We also discuss about the validity of Dhaka radar
data which is potentially highly important for the rainfall estimation. It is found that the Dhaka
radar highly underestimates the rainfall for all over the country. Some correction is needed to
obtain quantitatively reliable results.
Acknowledgement
We highly appreciate Dr. S. Karmakar the director of Bangladesh Meteorological Department
for continuous support for our research. For drawing figures, GFD DENNOU Library is
utilized.
References
1. S. Karmakar, and M. M. Alam, Mausam, 56, 671-680 (2005).
2. W. M. Gray, Mon. Wea. Rev., 95, 55-73, (1967).
3. Y. Yamane, and T. Hayashi, Geophys. Res. Lett., 33, L17806, doi:10.1029/ 2006GL026823,
(2006).
4. T. Terao, T. Hayashi, M. N. Islam, and T. Oka, Geophys, Res. Lett., 33, doi: 10.1029/
2006GL026156, (2006).
5. T. Terao, M. N. Islam, F. Murata, and T. Hayashi, Natural Hazards., on the web, doi:
10.1007/s1 1069-007-9128-z, (2007).
6. F. Murata, T. Terao, T. Hayashi, H. Asada, and J. Matsumoto, Natural Hazards., on the web,
doi: 10.1007/s1 1069-007-9125-2, (2007).
7. T. Hofer and B. Messerli, Floods in Bangladesh: History, dynamics and rethinking the role
of the Himalayas, United Nations Univ. Press, 468 pp., (2006).
8. D. Shepard, Proceedings in 1968 ACM National Conference, 517-524, (1968).
9. R. A. Houze Jr., Cloud Dynamics, Academic Press, 573pp., (1993).
10. M. N. Islam, T. Terao, H. Uyeda, T. Hayashi, and K. Kikuchi, J. Meteor. Soc. Japan., 83,
21-39, (2005).
Download