Simulation of tornado over Brahmanbaria on 22 March 2013 using Doppler weather radar and WRF model

Transcription

1 Geomatics, Natural Hazards and Risk ISSN: (Print) (Online) Journal homepage: Simulation of tornado over Brahmanbaria on 22 March 2013 using Doppler weather radar and WRF model Mohan K. Das, Someshwar Das, Md. Abdul Mannan Chowdhury & Samarendra Karmakar To cite this article: Mohan K. Das, Someshwar Das, Md. Abdul Mannan Chowdhury & Samarendra Karmakar (2016) Simulation of tornado over Brahmanbaria on 22 March 2013 using Doppler weather radar and WRF model, Geomatics, Natural Hazards and Risk, 7:5, , DOI: / To link to this article: Informa UK Limited, trading as Taylor & Francis Group Published online: 27 Nov Submit your article to this journal Article views: 841 View related articles View Crossmark data Citing articles: 1 View citing articles Full Terms & Conditions of access and use can be found at Download by: [ ] Date: 29 December 2017, At: 22:32

2 GEOMATICS, NATURAL HAZARDS AND RISK, 2016 VOL. 7, NO. 5, ARTICLE Simulation of tornado over Brahmanbaria on 22 March 2013 using Doppler weather radar and WRF model Mohan K. Das a,b, Someshwar Das c, Md. Abdul Mannan Chowdhury b and Samarendra Karmakar d a SAARC Meteorological Research Centre (SMRC), Dhaka 1207, Bangladesh; b Department of Physics, Jahangirnagar University, Savar, Dhaka 1342, Bangladesh; c India Meteorological Department, New Delhi , India; d Dhaka 1212, Bangladesh Centre for Advanced Studies, Gulshan, Dhaka 1212, Bangladesh ABSTRACT A tornado occurred at Brahmanbaria in Bangladesh in the afternoon of 22 March The tornado event has been studied based on tropical rainfall measuring mission (TRMM) data, radar observations and model simulations. The maximum reflectivity and the vertical extent of the system have been recorded to be about 54.7 dbz and 15 km, respectively, by the Doppler weather radar (DWR) at Agartala, India. The event has been simulated by using the WRF model at 3- and 1-km horizontal resolution nested domains based on six hourly final (FNL) re-analysis data and boundary conditions of National Centers for Environmental Prediction (NCEP). Results show that, while there are differences of 40 minutes before the observed time of the storm, the distance between observed and simulated locations of the storms is 0.5. The maximum amount of vorticity transferred by directional shear in the storm updraft (helicity) due to convective motion simulated by the model is found to be 1774 m 2 s 2, and the highest value of bulk Richardson number shear that defines the region in which low-level mesocyclogenesis is more likely has been m 2 s 2, which is generally supposed to produce rotating storms according to the prescribed range. The highest vertical velocity simulated by the model is about 28 to 58 m s Introduction ARTICLE HISTORY Received 28 September 2014 Accepted 25 October 2015 KEYWORDS Tornado; DWR; WRF; vertical velocity Tornado is a rare weather phenomenon involving a violently rotating column of air which is in contact with a cumulonimbus cloud or, in rare cases, a cumulus cloud base and the surface of the Earth. Although this dangerous phenomenon occurs mostly in the United States, it can cause death and destruction almost anywhere on the globe, except Antarctica. Spawned from powerful thunderstorms, tornadoes can cause fatalities and devastate a neighbourhood in seconds. Tornadoes come in many sizes, but are typically in the form of a visible condensation funnel, whose narrow end touches the ground and is often encircled by a cloud of debris. Bangladesh is not free from tornadoes. Finch and Dewan (2003), Das et al. (2014) and Das, Chowdhury, et al. (2015) have shown that many parts of Bangladesh, particularly Brahmanbaria, Tangail, Mymensingh and Dhaka, are vulnerable to tornadoes during pre-monsoon season (March May). A tornado lashed Brahmanbaria District ( N Nand E E) in Bangladesh at coordinated universal time (UTC) ( local standard time CONTACT Mohan K. Das mohan28feb@yahoo.com 2015 Informa UK Limited, trading as Taylor & Francis Group

3 1578 M.K. DAS ET AL. (LST); LST D UTC C 6 h) according to eye witness and media reports on 22 March The daily newspaper reported a powerful tornado pounded outskirts of Brahmanbaria in Bangladesh killing 38 people, injuring 388 apart from totally destroyed 2635 houses and partially destroyed 752 houses and damaging crops. Comilla-Sylhet highway was disrupted for 5 h following uprooting trees on the road. As per eyewitness, rain and hailstorm started at around 05:00 PM (1100 UTC) on 22 March 2013 (Friday), but immediately a tornado hit the area damaging crops and houses and killing people of the villages of Ramrail, Machihata and Sultanpur Upazila. A large number of uprooted trees fell on the highway resulting traffic jam of vehicle movement. The tornado also blew away a passenger bus that fell into a ditch. It also broke away a backside wall of the district jail (200-feet boundary wall and 420- feet security wall). The twister ravaged more than 15 villages of Brahmanbaria. There was strong proof of a visible vortex wind or funnel-shaped cloud in the event. A supercell has a vortex the air is spinning around it as it moves up. When that happens, it takes the moisture above the freezing level (FL) and keeps it there for long time. Rain particles keep developing bigger and bigger into hail. Also, because the draft is so strong in supercells, they tend to stay for a long time up in the air, and when they reach the right weight, they drop down. Some people chased the tornado and captured the video with their mobile phone which is now available at YouTube. This is the first tornado video and picture over Bangladesh. Generally, the devastation a storm causes demands more attention be paid to rescuing and caring for victims than to documenting exactly what happened meteorologically. Frequent occurrences of such phenomena in the same region warranted investigation. It is particularly important because tornadoes occur generally during the premonsoon season (March May) in this region. Finch and Dewan (2003) ( have made a detailed documentation of the tornadoes of Bangladesh. Several factors lead to the active thunderstorm season across Bengal (Bangladesh and its adjoining regions in India). North and Central India heats up and dries out in late March or early April. A deep, dry mixed layer develops. Low-level flow from the Bay of Bengal increases markedly during this time. Westerly mid-level flow around the Tibetan Plateau advects the Indian mixed layer over the Bengal moist tongue. This leads to the elevated mixed layer. It may be noted that parts of the East Indian Plateau are elevated ( ft) compared to Bangladesh which is near sea level. The mid-level flow is fairly strong in April with knots (»15 25 m s 1 ) speed at 700 hpa and knots (»18 25 m s 1 ) at 500 hpa. The high-level jet is usually over or just north of Bengal in April. The southern branch of the polar jet often retreats north of the Tibetan Plateau by May, leaving light, mid- to high-level flow across the Bengal region. By June, the high-level flow is light. Nocturnal storms over the Khasi Hills near Cherrapunji leave outflow boundaries over northern Bangladesh. These nocturnal storms are probably caused by the low-level jet (LLJ) impinging on the Khasi Hills of Meghalaya, India. All these factors result in a tornado maximum in early to mid-april, even in May. In short, vertical wind shear and instability are maximized and the jet is in a favourable position during this time. Litta et al. (2010, 2012), successfully simulated severe thunderstorms, which produced tornadoes close to Ludhiana Airport (Punjab), the north-west region of India, on 15 August 2007 and over Odisha on 31 March 2009 using WRF NMM model. Akter and Ishikawa (2014) studied synoptic features and environmental conditions of Brahmanbaria tornado by using the Japanese 55-year Reanalysis (JRA-55) data (50-km horizontal resolution) and multi-functional transport satellite images. The results were encouraging in terms of time and the place of occurrence of the event. In the previous studies, the data resolution was poor, so that the regional preference of storm genesis is not known properly. The use of good resolution, reliable and homogeneous data may produce better prediction of severe storm in this area. Chevuturi et al. (2014) showed in a study that the moisture incursion from the Arabian Sea and the Bay of Bengal along with the release of convective available potential energy (CAPE) in the lower levels leads to development of the lower instability over National Capital Region, India, in winter. On the other hand, the western disturbances in the subtropical westerly jet cause baroclinic instability in the mid/upper troposphere. Both the sources of instability are able to attain condition favourable for rare hail formation. An attempt is made to simulate the Brahmanbaria tornado using the Weather Research and Forecasting (WRF) model at nested domains of 3- and 1-km resolution. The outputs are obtained

4 GEOMATICS, NATURAL HAZARDS AND RISK 1579 for every minute. Several convective parameters, such as CAPE, lifted index (LI), Showalter index (SWI), total totals index (TTI), etc., are studied based on observations. Studies showed that none of these parameters can be taken as a stand-alone predictor for the occurrence of thunderstorms with 100% accuracy. It indicates that Numerical Weather Prediction (NWP) models are very important for obtaining guidelines for forecasting local severe storms. A brief climatology of tornadoes over Bangladesh is presented in the Section 2. Section 3 presents field investigation of the tornado event. Section 4 presents the large-scale synoptic situations prevailing during the period. The observations of Cox s Bazar, Agartala Doppler weather radar (DWR) and tropical rainfall measuring mission (TRMM) are discussed in Section 5. A brief description of the WRF model configuration used in this study is given in Section 6. The results of simulation are also discussed in this section. Finally, the summary of the results and concluding remarks are presented in Section Climatology of tornadoes in Bangladesh By definition, a tornado is a violently rotating column of air, in contact with the ground, either pendant from a cumuliform cloud or underneath a cumuliform cloud, and often (but not always) visible as a funnel cloud. In practice, for a vortex to be classified as a tornado, it must be in contact with both the ground and cloud bases. Tornado refers to the vortex of wind, not the condensation cloud. Most tornadoes have wind speeds between 40 mph (64 km h 1 ) and 110 mph (177 km h 1 ), are approximately 250 feet (75 m) across and travel a few miles (several kilometres) before dissipating. Some attain wind speeds of more than 300 mph (480 km h 1 ), stretch more than a mile (1.6 km) across and stay on the ground for dozens of miles (more than 100 km). Most of the tornadoes (about 80%) occur in Bangladesh during the pre-monsoon season (March May) with a majority in early to mid-april. Many researchers have documented the tornadoes in Bangladesh since 1888 (Petersen & Mehta 1981; Ono 1997; Goldar et al. 2001; Finch & Dewan 2003; Das, Chowdhury, et al. 2015; Das et al. 2015; Das, Sarkar, et al. 2015). According to the world map of tornadoes by Fujita (1973), the relative intensities of tornadoes in Bangladesh are ranked as F-4; no other places are ranked as stronger than F-3 except the United States. The highest concentration of the most violent tornados in Bangladesh occurs over the region N, E, and almost all of them occur in the afternoon or evening with a peak around 1030 UTC (Finch & Dewan 2003). The preferred movements of violent tornados in Bangladesh are towards the south-east or south south -east with speed ranging from km h 1. The paths are not always along a straight line, sometimes they move in a zigzag fashion having path length of about 25 km and width of about 1.5 km (Ono & Schmidlin 1996). It is believed that the world s deadliest tornado occurred in Bangladesh on 29 April 1989 at Saturia in Manikganj, with an estimated maximum wind speed of km h 1 (BMD 1989) that killed approximately 1300 people, injured 12,000 people and made 80,000 people homeless (Grazulis 1991). Although majority of tornados occur during the pre-monsoon season, there are some reports of fall and winter tornados in Bangladesh (Finch & Dewan 2003; Das et al. 2015). 3. Field investigations Considering the severity of the event, a field survey was conducted at Brahmanbaria, Bangladesh, to verify the severity of the event, and assess the damages caused by the tornadic storm. Figure 1 shows the tornado track and area of influence of the tornado at Brahmanbaria in Bangladesh (Siddiqui & Hossain 2013). The areas affected by the tornadic event (table 1) is surrounded by water mass (the Meghna and the Titas rivers), which might have supplied enough moisture in the boundary layer. Several villages in Brahmanbaria district affected by the tornado were visited and the villagers were interviewed. The damages were found similar to the report by different media. All of them confirmed the occurrence of an unusual severe storm in their area. According to the villagers, a strange

5 1580 M.K. DAS ET AL. Figure 1. Track and area of influence of Brahmanbaria tornado (adapted from Siddiqui and Hossain 2013). noise was heard for several minutes during which their homes were blown away and trees were uprooted. They also saw something like fire red in the sky (which could be intense lightning). The storm passed from east to south-east direction as seen from the movement of deep clouds from the Cox s Bazar DWR and lasted for about 10 minutes or more. The tornado began at Ramrail Village of Ramrail Union and ended at Ahmedabad Village of North Akhaura Union. The tornado travelled a distance of more than 10 km. It affected an area of 17 km 2 (Siddiqui & Hossain 2013). The neighbouring villages also had much impact and many damages. The villagers also said that the storm moved over the water bodies nearby their houses and lifted water as it passed. This means that there was a water spout found due to the passing of the tornado over the water bodies. Fishes of the pond were spread out over the land area. Due to the water spout, the water level of the pond went down more than one foot. As it moved over the pond and marshy land, a lot of clay was spread over the path of tornado. Most of them also said that they saw rotating funnel-shaped cloud like the Table 1. Tornado, squall, gust, hail and 24-h rainfall occurred over many places of Bangladesh before and after the tornado day. Date Station Occurrence time (UTC) Wind speed (m s 1 ) Direction 24-h rainfall (mm) 21 March 2013 Bogra N 3 Sylhet Hail 15 Rangpur 4 Mymensingh Trace 22 March 2013 Brahmanbaria F2 ornado No observatory Dhaka S 17 Comilla Hail/past hail SW 27 Chittagong Main SW 3 Meteorological Office Sylhet Trace Sitakunda 13 Feni 23 Faridpur March 2013 Dhaka 1

6 GEOMATICS, NATURAL HAZARDS AND RISK 1581 trunk of an elephant that damaged everything which lay in its way. The survey was also carried out in many villages of the affected district. The villagers described similar experiences as earlier. 4. Large-scale synoptic features Past studies of tornadoes over the Indian and Bangladesh region (Hussain & Karmakar 1998; Prasad 2006) have indicated that a horizontal vortex sheet is created by the presence of horizontal and vertical wind shears in association with low-/middle-level wind core. The vortex is fuelled by the advection of dry air by westerly current into a region of warm and moist air. The vortex is tilted by the wind shear. The cyclonic vorticity thus created is helpful for the formation of severe storms such as tornadoes. On 22 March 2013, cloud clusters were present over eastern Bangladesh since morning. Convection developed further and a tornado was reported in the Brahmanbaria area in Bangladesh around 1055 UTC (table 1). The large-scale synoptic conditions at 500 and 850 hpa have been analysed during the days of tornadic events (diagrams are not shown for brevity). It has been seen that, during most of the tornadic events, a strong trough exists at 500 hpa over the middle of Bangladesh. At lower level 850 hpa, the flow is southerly or south-westerly feeding moisture from the Bay of Bengal over Bangladesh. The low-level moisture incursion in the zone of convergence coupled with the upper level trough has made conditions conducive for the development of severe thunderstorms over the region. Owing to lack of dense network of observatories in the area, the exact measurement of maximum wind speed, pressure drop, pressure tendency, the observed vertical profiles of temperature, moisture, etc. are not available in the region. Some of these features are diagnosed through the model simulations. The synoptic conditions have led to the formation of convective cloud clusters over the region (Akter & Ishikawa 2014). On 22 March 2013, cloud clusters were present over east Bangladesh since morning (figure 2). Before the occurrence of the tornado, there was no cloud top temperature (CTT). As the time of the day passes, a deep convection was found and CTT became 40 to 50 C when the tornado occurred over Brahmanbaria. The convection developed further and the tornado was reported in the Brahmanbaria area in Bangladesh at around 1100 UTC (table 1). Hourly analysis of CTT obtained from the Kalpana-1 satellite (figure 2) reveals that, on 22 March 2013, large cloud clusters were found to exist over eastern Bangladesh and the north-east region in India. An in situ cloud formed over the eastern part of Bangladesh around 1000 UTC, which quickly developed into deep clouds (CTT» 40 to 60 C) by 0700 to 1200 UTC. 5. Characteristics derived from radar and TRMM Three echoes were noticed to the east of Bangladesh at 1011 UTC, which intensified into an E SE oriented echo, and the middle echo was of 50 km length at 1011 UTC. The analysis of DWR radar picture at 1011 UTC at Agartala is shown in figure 3(a). This figure shows that the vertical length of the system is about 15 km. Bangladesh Meteorological Department has DWR at Cox s Bazar. Precipitation rates (mm h 1 ) are derived from the DWR at Cox s Bazar during 1100 UTC, on 22 March The precipitation rates observed by Cox s Bazar DWR on 22 March 2013 are shown in figure 3(b). The analysis of the radar observations indicates that three convective cells have developed at around Brahmanbaria and Agartala areas at about 1100 UTC. The maximum precipitation rates observed by radar are about mm h 1. The observed accumulated precipitation from TRMM during UTC (figure 5(a)) shows about 31.5 mm. However, higher precipitation rates observed by radar may not imply that the storm was at its peak intensity at that time. It has been often reported that the strongest wind and the vortex are generally followed by the occurrences of hail and rainfall. The gust front and downdrafts arrive before the heaviest rainfall.

7 1582 M.K. DAS ET AL. Figure 2. Cloud top temperatures derived from the Kalpana-1 satellite at (a) 0900, (b) 1000, (c) 1100 and (d) 1200 UTC, 22 March Model-simulated characteristics 6.1. WRF ARW model main features The WRF model (version 3.5.1) has been used for simulation of the tornadic storm in this study. The WRF model is a new-generation mesoscale NWP system designed to serve both operational forecasting and atmospheric research needs (NCAR 2009). It features multiple dynamical cores, a three-dimensional variational (3DVAR) data assimilation system, and a software architecture allowing for computational parallelism and system extensibility. WRF is suitable for a broad spectrum of applications across scales ranging from metres to thousands of kilometres. Applications of WRF include research and operational NWP, data assimilation and parameterized-physics research, downscaling climate simulations, driving air quality models, atmosphere ocean coupling and idealized simulations (i.e., boundary layer eddies, convection, baroclinic waves). There are two dynamics solvers in the WRF system: the Advanced Research WRF (ARW) solver (originally referred to as the Eulerian mass or em ) developed primarily at NCAR, and the Non-hydrostatic Mesoscale Model (NMM) solver developed at National Centers for Environmental Prediction (NCEP). The ARW system consists of the ARW dynamics solver with other components of the WRF system needed to produce a simulation. For the purpose of simulating the tornado, the model has been run on doublenested domains at 3- and 1-km resolutions (figure 4) with 31 vertical levels using initial and boundary conditions data obtained from NCEP FNL (Final) Operational Global Analysis, which is at about 1 1 horizontal resolution. Das, Sarkar, et al. (2015) conducted several sensitivity experiments with different combinations of physical parameterization schemes of the model and found that the best skill scores were obtained by the combinations of no-cumulus, Milbrandt and YSU schemes for cumulus convection, cloud microphysics and planetary boundary layer,

8 GEOMATICS, NATURAL HAZARDS AND RISK 1583 Figure 3. (a) Reflectivity derived from the DWR Agartala at 1011 UTC and (b) precipitation rates (mm h 1 ) derived from the DWR at Cox s Bazar at 1100 UTC on 22 March respectively, for the simulation of nor westers over the Indian and Bangladesh region. This combination of physical processes provided least RMSE values for rainfall, wind speed at surface and time of occurrences of storms in the model simulations. This combination of physics has been used in the present study. The model has been integrated for 24 h for the tornado event; the starting time of integration is 0000 UTC, on 22 March Main features of the model employed for this study are summarized in table 2.

9 1584 M.K. DAS ET AL. Figure 4. Locations of tornado-affected district Brahmanbaria (BRB) in Bangladesh with topography shaded in metres. The two domains (outer Domain-1 and inner Domain-2) nested within each other illustrate the WRF model domains used for simulations Model-simulated diagnostics In this section, many simulated characteristics of the tornadic event (Das et al. 2015) are investigated such as the CAPE, precipitation, surface wind speed, flow patterns, skew-t diagram, storm-relative environment helicity (SREH), bulk Richardson number shear (BRNSHR), dew-point depression, potential vorticity (PV) and vertical velocity. All results discussed here correspond to the inner Figure 5. Accumulated rainfall at UTC on 22 March 2013 (a) retrieved from TRMM and (b) model simulated.

10 GEOMATICS, NATURAL HAZARDS AND RISK 1585 Table 2. Summary of the WRF model. Model features Configurations Horizontal resolution Nested 3 and 1 km Vertical levels 31 Topography USGS Dynamics Time integration Semi-implicit Time steps 15 s Vertical differencing Arakawa s Energy Conserving Scheme Time filtering Robert s method Horizontal diffusion Second-order over-quasi-pressure, surface, scale selective Physics Convection No CU PBL YSU Cloud microphysics Milbrandt and Yau (2005) Surface layer Monin-Obukhov Radiation RRTM (LW), Mlawer et al. (1997) SW (Dudhia 1989) Gravity wave drag No Land surface processes Unified NOAH Land Surface Model domain at 1-km resolution. Every minute variation of these parameters has been studied over Brahmanbaria in Bangladesh. The model has simulated the event at 1022 UTC which is 40 minutes ahead of the occurrence of the event. This result is based on the initial condition of 0000 UTC, 22 March Simulated precipitation and CAPE Simulated precipitation by the model during UTC over the region (figure 5(b)) shows isolated pockets of rainfall (8 16 mm) which are less than the TRMM observations (figure 5(a)) but the pattern is almost similar. The CAPE is the positive buoyancy of an air parcel. It is the amount of energy a parcel of air would have if lifted a certain distance vertically through the atmosphere. It is an indicator of atmospheric instability. It is defined as Z Zn CAPE D g ðt vp T veþ dz (1) Z f T ve where, Z f and Z n are the levels of free convection and neutral buoyancy, respectively. T vp and T ve are the virtual temperatures of the air parcel and environment respectively. The threshold values of CAPE for different stability regimes are given as follows: CAPE < 1000 CAPE > 1000 < 2500 CAPE > 2500 : Instability is weak : Moderate instability : Strong instability The simulated CAPE by the model is not shown for brevity. However, the analysis showed that CAPE has started building up in the middle of the analysis area since morning 0400 UTC and is found to be maximum of 3650 J kg 1 at around 1100 UTC. It is found to decrease afterwards Surface wind, SLP, RH and temperature analyses The wind speed at 10 m above the ground as simulated by the model is shown in figure 6. The maximum surface wind speed simulated by the model is about 31 m s 1 near the place of occurrence of

11 1586 M.K. DAS ET AL. Figure 6. Vector wind at 850 hpa and surface wind speed (m s 1 ) forecasts valid on the day of the tornado at 1022 UTC, 22 March (To view this figure in colour, see the online version of the journal.) the tornado (figure 6) and is associated with a zone of mass convergence. The gust is from the east. The prevailing wind direction is westerly south-westerly. Figure 6 shows the confluence of southerly wind with westerly wind which is important for the formation of convective system. Figure 7(a) (d) depicts the time series of observed and model-simulated sea-level pressure (SLP), wind speed, relative humidity (RH) and temperature at the surface on 22 March 2013 at Agartala (23.53 N, E), which is a nearby observatory of the tornado occurrence area. The observed and simulated values show fairly good correspondences. The model-simulated wind speeds are generally underestimated by about 1 2 ms 1. The maximum similarity between observed and simulated RH and temperature is seen at 0900 UTC when the convection is generally at peak. Good agreement trends between observed and simulated values (overestimation/underestimation) are seen at the surface Skew-T diagrams analysis Skew-T diagrams based on observed upper air soundings at Dhaka (23.77 N, E), Agartala (23.88 N, E) and from model simulation are shown in figure 8 for 22 March Indian Meteorological Observatory at Agartala is the nearest observatory to the place of occurrence of the tornado. However, the observation is taken at 0000 UTC. Therefore, the model-simulated skew-t diagrams on 22 March 2013 have been compared with those at Dhaka and Agartala, but the diagrams are not shown here for brevity. The convective characteristics of the air parcel such as the LI, TTI, SWI, CAPE, convective inhibition energy (CINE), wind hodograph and wind profiles have

12 GEOMATICS, NATURAL HAZARDS AND RISK 1587 Figure 7. Time series of observed and model-simulated (a) SLP, (b) wind speed, (c) relative humidity and (d) temperature at surface on 22 March 2013 at Agartala (23.53 N, E). been examined from the Skew-T diagram. The Skew-T diagrams of the event showed instability in the atmosphere. The CAPE values are greater than 1200 J kg 1 and the CINE values are negative. The LIs are usually less than 3. When the LI values are between 2 and 6, the atmosphere is unstable with a possibility of severe thunderstorms. The TTI values are found to range between 41 Figure 8. Skew-T diagrams of maximum convection area on 22 March 2013 simulated by model.

13 1588 M.K. DAS ET AL. Table 3. Observed and simulated diagnostics. Indices Observed Model KI CAPE 1690 J kg J kg 1 LI TTI FL 645 hpa 640 hpa LCL 897 hpa 880 hpa and 46. The TTI values above 40 are indicative of the possibility of severe thunderstorms. Values greater than 47 are indicative of severe thunderstorms with tornado intensity. Model-simulated freezing point is near 850-hPa level which is a notable indication of tornado event as compared to study of hail event by Chevuturi and Dimri (2015). The model has produced better simulation of these parameters on the day of occurrence (table 3). From the model-simulated diagram (figure 8), it is found that the FL is at 640 hpa and lifted condensation level (LCL) is at 880 hpa, whereas the observed FL and LCL are at 645 and 897 hpa, respectively (table 3). This value indicates that, on the date of occurrence of the tornado, the FL and the LCL have become low, which is significant for severe weather formation Storm-relative environment helicity (SREH) The helicity is a measure of the amount of rotation found in a storm s updraft air. If there is significant rotation in a storm s updraft air, the storm will more than likely become a supercell and possibly spawn one or more tornadoes. Helicity is a parameter that defines the amount of streamwise vorticity (i.e., directional shear) a steady storm updraft will ingest as a result of a given storm motion. In meteorology (Thompson et al. 2007), helicity corresponds to the transfer of vorticity from the environment to an air parcel in convective motion. Here the definition of helicity is simplified to only use the horizontal component of wind and vorticity: Z H D Z! V h z! h dz D! V h r V h dz (2) where Z is the altitude, V h is the horizontal velocity and z h is the horizontal vorticity. According to this formula, if the horizontal wind does not change direction with altitude, H will be zero as the product of V h and r! Vh is perpendicular to one another making their scalar product nil. H is then positive if the wind turns (clockwise) with altitude, and negative, if it backs (counter-clockwise). Helicity has energy units per units of mass (m 2 s 2 ) and thus is interpreted as a measure of energy transfer by the wind shear with altitude, including directional. This notion is used to predict the possibility of tornadic development in a thundercloud. In this event, the vertical integration is limited below cloud tops (generally 3 km or 10,000 feet) and the horizontal wind is calculated to wind relative to the storm in subtracting its motion: Z SREH D ð! V CÞ r V h dz (3) where C is the cloud motion to the ground. Critical values of storm-relative helicity (SRH) for tornadic development, as obtained in North America, are (Thompson et al. 2007) as follows: SREH D supercells possible with weak tornadoes according to Fujita (1973) scale, SREH D very favourable to supercell development and strong tornadoes, SREH > 450 violent tornadoes, When calculated only below 1 km (4000 feet), the cut-off value is 100.

14 GEOMATICS, NATURAL HAZARDS AND RISK 1589 Helicity in itself is not the only component of severe thunderstorms and those values are to be taken with caution. Therefore, the energy helicity index (EHI) has been created. It is the result of SRH multiplied by the CAPE and then divided by a threshold CAPE. This incorporates not only the helicity but the energy of the air parcel and thus tries to eliminate weak potential for thunderstorms even in strong SRH regions. The critical values of EHI are as follows: EHI D 1 indicates possible tornadoes, EHI D 1 2 indicates moderate to strong tornadoes, EHI > 2 indicates strong tornadoes Figure 9(a) presents the distribution of SREH for the event. The values are integrated for 0 3km layer. The SREH values reached the maximum around 1006 UTC over the western part of Brahmanbaria, where the storm is reported. High CAPE values are also simulated at around the same place. As mentioned earlier, when the SREH value is greater than 150 m 2 s 2, supercells are possible with weak tornadoes, and when the values are greater than 450 m 2 s 2, violent tornadoes are possible. Figure 10(a) presents every minute time evolution of the maximum SREH obtained from the model simulations over Bangladesh for the event. The diagrams indicate that the SREH values have generally 2 3 maxima within an hour; one at 1006 UTC, the second at 1012 UTC and the third one at 1025 UTC. Comparison between the observed time of occurrence of the storm and the peak values indicates that the storm has occurred when SREH is persistently high for at least minutes. At most of the time, the SREH values have been higher than the threshold value (150 m 2 s 2 ) for the formation of weak tornadoes. The model-simulated value of SREH for the tornado under study is 1774 m 2 s 2. This value is highly favourable for the formation of a severe tornado. Figure 9. (a) Storm-relative environment helicity (SREH) (m 2 s 2 ) simulated by the model for 1022 UTC, 22 March Values above 50 are shaded. (b) Bulk Richardson number shear (m 2 s 2 ). Values above 25 are shaded.

15 1590 M.K. DAS ET AL. Figure 9. (Continued) Bulk Richardson number shear (BRNSHR) The BRN is used to quantify the relationship between buoyant energy and vertical wind shear (Moncrieff & Green 1972), and is defined as BRN D CAPE 0:5ðu 2 C v 2 Þ (4) Figure 10. Every minute time series of (a) storm-relative environment helicity (SREH) and (b) bulk Richardson number shear (BRNSHR) on 22 March 2013.

16 GEOMATICS, NATURAL HAZARDS AND RISK 1591 where u and v are the wind components of the difference between the density-weighted mean winds over the lowest 6000 m and the lowest 500 m above ground level. As discussed by Droegemeier et al. (1993), the BRN is only a gross estimate of the effects of vertical wind shear on convective storms, since it does not measure the turning of the wind profile with height. However, Weisman and Klemp (1984) showed using cloud-scale model simulations that the BRN can distinguish between supercell and multicell storms, with modelled supercells likely when 10 BRN 50 and multicells storms are likely when BRN > 35. It is important to note that there is no well-defined threshold value for BRN, since there is an overlap in these values used to specify storm type. BRNSHR is defined by the denominator of equation (4) and has been found to be highly correlated with the maximum vertical vorticity of modelled thunderstorms by Droegemeier et al. (1993), despite the fact that it does not account for the turning of the wind vector with height, or the magnitude of the low-level storm-relative winds (Lazarus & Droegemeier 1990). Brooks, Doswell, and Wilhelmson (1994) and Brooks, Doswell, and Cooper (1994) hypothesized that the mid-level stormrelative winds are important to the development of low-level rotation in thunderstorms. Since their conceptual model indicated that the strength and lifetime of low-level mesocyclones is a function of the balance between low-level baroclinic generation of vorticity and outflow development, they examined the redistribution of rain in modelled supercells. Their results indicated that, for very weak mid-level storm-relative winds, the low-level mesocyclones are short-lived, occur early in the storm life cycle and low-level outflow dominates the storm. Storms forming in this type of environment are more likely to evolve into squall lines owing to the strong organizing influence of the outflow. For very strong storm-relative winds, low-level mesocyclones develop very slowly, or do not develop at all, and the outflow is weak, since the rain is being blown away from the storm by the strong mid-level winds. In the middle of these two extremes, the results of Brooks, Doswell, and Wilhelmson (1994) and Brooks, Doswell, and Cooper (1994) show that low-level mesocyclones tend to be long-lived, owing to the balance between the mesocyclone circulation and the storm-relative winds. These results are related to the values of BRNSHR, since an examination of the supercell thunderstorm proximity sounding data-set from Brooks, Doswell, and Cooper (1994) indicated that the BRNSHR can be used as a proxy for the storm-relative wind. The use of BRNSHR instead of the storm-relative wind is a valuable simplification, since BRNSHR is both independent of storm motion and vertically integrated, making BRNSHR values better behaved than values of storm-relative mid-level winds calculated from mesoscale model output where the storm motion must be estimated. In addition, using the proximity data-set of Brooks, Doswell, and Cooper (1994) and subjectively determining the best-fit line to discriminate between tornadic and non-tornadic thunderstorms using only the values of SREH and BRNSHR. Stensurd et al. (1997) found that, as the value of BRNSHR increases, the value of SREH also must increase to support mesocyclogenesis. No observed tornadic storms occur with BRNSHR values less than 20 m 2 s 2. Thus, in more highly sheared environments it is expected that the value of SREH must be significantly higher than the guidance value of 100 m 2 s 2 in order to increase the likelihood of developing tornadic supercell thunderstorms. Small values of BRNSHR correspond to low values of mid-level storm-relative winds and storms that are outflow dominated with a tendency to produce damaging winds (Stensurd et al. 1997). The results of Weisman (1993), who examined bow echoes using a cloud-scale model, have shown that bow echoes are more prevalent for lower values of BRNSHR, while supercells are more prevalent for larger values of BRNSHR, assuming that there is sufficient shear to generate long-lived rotating storms. For the largest values of BRNSHR used, the results of Weisman (1993) indicate that no organized convective activity has occurred in the numerical simulations. Thus, in general agreement with the conceptual model, his results show that it is in the middle range of BRNSHR values that supercell thunderstorms develop. The results of Stensurd et al. (1997) suggest that values of BRNSHR between 40 and 100 m 2 s 2 indicate a greater likelihood of tornadic supercell thunderstorms, if the SREH values are large enough to produce rotating storms. The value of 40 m 2 s 2 used for the modelled BRNSHR threshold is larger than that suggested by the proximity sounding data, likely owing to the difficulties in

17 1592 M.K. DAS ET AL. simulating low-level winds. Although BRNSHR and SREH are measures of the vertical wind profile, they can vary in the opposite directions. It is not unusual for BRNSHR to decrease as SREH increases, as can occur with the development of an LLJ. Figure 9(b) illustrates the spatial distribution of BRNSHR for the event. High values of BRNSHR (457.3 m 2 s 2 ) are simulated on the day around 1019 UTC, indicating sheared environment. Such values of BRNSHR may help the formation of tornado under study. Since both CAPE and SREH values are also higher in this region, the necessary convective energy and rotation of wind field required to produce the storms are sufficient, leading to the occurrence of the tornado. Figure 10(b) presents time evolution of the maximum BRNSHR obtained from the model simulations over south-west of Brahmanbaria for the event. The diagram indicates that the BRNSHR has generally two maxima; one at 1019 UTC and the second at 1025 UTC. At most of the time, the BRNSHR values are higher than the threshold value (40 m 2 s 2 ) for the formation of supercell thunderstorms. The simulated maximum values of both SREH and BRNSHR are higher than the threshold values in the case of the tornado on 22 March In thunderstorm forecasting, CAPE is used to define the region in which convection is possible, SREH is used to define the region in which thunderstorms are likely to be supercells and BRNSHR is used to define the region in which low-level mesocyclogenesis is more likely. These results highlight the potential value of analysing various severe weather parameters in forecasting tornadic thunderstorms. By combining the storm characteristics suggested by these parameters, it is possible to use mesoscale model output to infer the dominant mode of severe convection. There are many other parameters that should be used in forecasting severe weather threat (Johns & Doswell 1992; Thompson 1998). However, the study focuses here on these three parameters CAPE, SREH and BRNSHR for simplicity. It should be recognized that there remains great uncertainty in, and debate about, the best parameters to use for forecasting tornadoes. Figure 11 depicts the combined graphic of the three fields of CAPE (>1200 J kg 1 in blue contours), SREH (>300 m 2 s 2 in green contours) and BRNSHR (>150 m 2 s 2 in red contours) as simulated by the model for 1022 UTC on 22 March It highlights the area in which all the three fields are prominent, especially near the place of occurrence of the tornado. The values are in the ranges that are favourable for low-level mesocyclones in the event of tornado. Rainfall values have been shaded in the diagram to indicate areas of convection. Results show that the most favourable region for the development of tornado is near Brahmanbaria Dew-point depression The presence of a dry line is often considered as a signal to the convective initiation and tornadogenesis. The dry line is defined by the leading edge of the dew-point temperature gradient 15 C (100 km) -1 (Roebber et al. 2002; Stensurd & Weiss 2002). The dry line acts as a boundary between dry and moist air mass, and may be conducive for the development of convection by forcing a deep layer of dry air above a moist boundary layer. Sometimes a double dry-line structure is also found in the regions of tornado genesis (Roebber et al. 2002). For simplicity, in this study, the dew-point depression (Tdd) has been analysed and is shown in figure 12. The dry and moist areas are demarcated by shading the Tdd values greater than 15 C and less than 10 C, respectively. Figure 12 shows that the dew-point depression values were as much as 5 20 C in the vicinity of the place of occurrence of the tornado. Analysis of the simulated values revealed that there was swapping of dry (moist) air by moist (dry) air many times prior to and after the occurrences of the storm. However, the dry lines generally remained fairly well defined Potential vorticity The PV is the absolute circulation of an air parcel that is enclosed between two isentropic surfaces. In adiabatic frictionless flow, PV is defined as a product of absolute vorticity and static stability on a

18 GEOMATICS, NATURAL HAZARDS AND RISK 1593 Figure 11. CAPE (>1200 J kg 1 in blue contours), SREH (>300 m 2 s 2 in green contours) and BRNSHR (>150 m 2 s 2 in red contours) as simulated by the model for 1022 UTC, 22 March Rainfall is shaded. (To view this figure in colour, see the online version of the journal.) constant potential temperature surface, i.e., PV D gð& Q C f Þ du dp & C f D (5) where g is the acceleration due to gravity, & Q is the relative isoentropic vorticity, z is the relative vorticity, f is the Coriolis parameter, u is the potential temperature, p is the pressure and D is the depth of layer. If PV is displayed on a surface of constant potential temperature, then it is called isentropic potential vorticity. Of course, PV could also be displayed on another surface, for example, a pressure surface. From the above relation, it may be noted that PV is simply the product of absolute vorticity on an isentropic surface and static stability. So PV consists, in contrast to vorticity on isobaric surfaces, of two factors, a dynamical element and a thermodynamical element. The PV has the SI unit m 2 s 1 Kkg 1. It has become accepted to define m 2 s 1 Kkg 1 as one potential vorticity unit (1 PVU). PV remains conserved in adiabatic, frictionless, non-compressible and homogeneous conditions. PV can only be changed by diabatic heating (such as latent heat released from condensation) or frictional processes. When the air converges to maintain potential vorticity, the air speed increases resulting in a stretched ring vortex. Divergence causes the vortex to spread, slowing down the rate of spin. PV is a useful concept for understanding the generation of vorticity in cyclogenesis and flow over mountains.

19 1594 M.K. DAS ET AL. Figure 12. Dew-point depression ( C, shaded) and surface pressure (hpa, contours) as simulated by the model for 1022 UTC, 22 March The percentage of supercell storms that produce low-level mesocyclones is not known. Moreover, the fraction of low-level mesocyclones that subsequently produce tornadoes is not known. Studies of the conceptual model of mesocyclones by Brooks, Doswell, and Wilhelmson (1994) and Brooks, Doswell, and Cooper (1994) have indicated that the strength and lifetime of low-level mesocyclones are a function of the balance between low-level baroclinic generation of vorticity and outflow development. An important question is whether non-tornadic supercell storms are simply ones that fail to produce low-level mesocyclones or are unable to produce tornadoes once a low-level mesocyclone develops. Preliminary observations from the Verification of the Origins of Rotation in Tornadoes Experiment (VORTEX) examined some of these issues (Rasmussen et al. 1994). Several storms intercepted during the experiment generated moderate to strong low-level mesocyclones yet failed to produce a tornado (Wakimoto & Atkins 1996; Trapp 1997; Wakimoto & Cai 2000). Trapp (1997) examined three tornadic and three non-tornadic supercells using airborne Doppler radar data. At the time of tornado formation or failure, his results have suggested that tornadic mesocyclones have smaller core radii, stronger low-level vertical vorticity and are associated with stronger vortex stretching. Given that many severe convective outbreaks are associated with mobile upper level troughs, it is natural to ask whether the characteristics of severe convection are sensitive to the presence and intensity of such mobile troughs. Several researchers have used PV concept to infer the balanced dynamics of predominately synoptic-scale weather features. Gold and Nielsen-Gammon (2008) showed that local increases in PV produced proportional increases of CAPE. They also studied the response of various shear

20 GEOMATICS, NATURAL HAZARDS AND RISK 1595 Figure 13. Vertical cross section of potential vorticity simulated by the model for 1045 UTC, 22 March Shaded region indicates positive values. The contours are drawn at the intervals of five. The cross sections are drawn from N, Eto N, E representing the areas of maximum convection. parameters and vertical motion to upper level PV modifications during an outbreak of 56 tornadoes in six states from Texas to Illinois on 13 March Their results showed that the 0 6 km vertical shear and the SREH are useful for distinguishing between environments supporting tornadic and non-severe thunderstorms. The SREH depends on both the vertical shear and the estimated storm motion vector. Amplifying (reducing) the magnitude of the upper level PV anomalies reduces (increases) the SREH down shear of the modification. A positive upper level PV perturbation, such as the one produced when an upper level trough is amplified, will always produce a cyclonic circulation of the perturbation shear vector such that the shear is generally opposed (enhanced) on the trough s cyclonic (anticyclonic) shear side. Davenport (2009) used the non-linear balance PV inversion developed by Davis and Emanuel (1991) to atmospheric features and motions on the order of the mesoscale and storm scale. He examined the low-level thunderstorm dynamics from a PV perspective for an idealized supercell using the WRF model, and showed that the PV diagnostic can be applied to thunderstorm dynamics. The vertical cross section of the PV simulated by the model is presented in figure 13 for the tornado under study. The cross sections have been drawn across the areas of maximum convection. The simulation of the event shows mostly positive values having maximum of about PVU located between 225 and 175 hpa (figure 13). It is not known what the critical value of PV for the tornadic mesocyclones is. However, thunderstorms of tornadic intensity have occurred in the event presented here. More research is required to differentiate between the cases of supercell thunderstorms, supercell thunderstorms with tornadic intensity and supercell thunderstorms that eventually develop into tornadoes.

21 1596 M.K. DAS ET AL. Figure 14. Vertical velocity diagnosed by the WRF model for 1045 UTC, 22 March The cross sections are drawn from latitude N representing the areas of maximum convection Updraft and downdraft (vertical velocity) The vertical cross section of the vertical velocity (m s 1 ) simulated by the model is shown in figure 14. The cross sections in x z plane are taken along the lines of maximum rainfall. The diagram shows cores of upward motion reaching up to the hpa. At 300 hpa, generally there is jet stream. Figure 14 shows the maximum vertical velocity of about 58 m s 1. Hence, the wind speed of 58 m s 1 may be a part of the jet stream (Roy 1953). The magnitude of the maximum updrafts is about 58 m s 1, which appears to be higher than the values obtained by Das et al. (2014), Das, Chowdhury, et al. (2015) and Das, Sarkar, et al. (2015) for nor westers over Bangladesh. The updraft cores are generally surrounded by mesoscale downdrafts with values ranging from 28 to 58 m s 1. These values are comparable to those obtained by Wakimoto et al. 2004, Wakimoto and Atkins (1996) and Wakimoto and Cai (2000). The cores of updrafts are the regions where the maximum cloud hydrometeor contents are located (figure 14). In a study by Wakimoto et al. (2004), they found that the first echo appeared in the rising thermal s cap, where the cloud s oldest surviving cloud and ice particles had time to both accumulate and become small graupel particles. In the next few minutes, the echo grew brighter and started wrapping around the updraft, undoubtedly assisted in doing so by the flanking descent. The maximum updraft strength of 60 m s 1 was attained at t D 845 s. 7. Summary and concluding remarks Bangladesh is prone to severe thunderstorms of tornadic intensity due to its geophysical location. The tornado was reported at Brahmanbaria in Bangladesh on 22 March The tornado lashed Brahmanbaria district around 1700 LST (1100 UTC). In this study, detailed model analyses of the event are presented. The tornadic thunderstorm has been found to form due to low-level moisture incursion by southerly flow from the Bay of Bengal coupled with upper level westerly jet stream causing intense instability and shear in the wind fields. Owing to the lack of dense network of