Use of Advanced Techniques to Model the Dispersion of Chlorine in Complex Terrain

Use of Advanced Techniques to Model the Dispersion of Chlorine in Complex Terrain

0957±5820/01/$10.00+0.00 q Institution of Chemical Engineers Trans IChemE, Vol 79, Part B, March 2001 USE OF ADVANCED TECHNIQUES TO MODEL THE DISPERS...

1MB Sizes 0 Downloads 34 Views

0957±5820/01/$10.00+0.00 q Institution of Chemical Engineers Trans IChemE, Vol 79, Part B, March 2001

USE OF ADVANCED TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN M. A. MCBRIDE1 , A. B. REEVES1 , M. D. VANDERHEYDEN2 , C. J. LEA3 and X. X. ZHOU3 2

1 Environmental Resources Management, Manchester, UK Rowan Williams Davies & Irwin Inc., Guelph, Ontario, Canada 3 Health and Safety Laboratory, Buxton, UK

T

he EU Seveso II Directive requires operators of major hazard facilities to prepare safety reports for sites storing quantities of dangerous substances in excess of speci®ed levels. The safety report should include an assessment of the risk associated with the facility, which will include an evaluation of the effects of releases of dangerous substances to the environment. The models commonly used for assessing the dispersion of dense gases in the atmosphere are based on the `ideal’ of ¯at, unobstructed terrain. For `real’ situations such models may be unduly pessimistic in their predictions and, in certain circumstances, may even be optimistic. This paper describes work undertaken by Environmental Resources Management (ERM), in association with sub-consultants Rowan Williams Davies & Irwin Inc. (RWDI) and the Health and Safety Laboratory (HSL) to model the dispersion of chlorine in complex terrain. The work involved physical modelling of releases in a Boundary Layer Wind Tunnel (BLWT) and the use of Computational Fluid Dynamics (CFD). The paper focuses on the key ®ndings of the study, which provide a dramatic insight into how terrain and buildings can fundamentally alter the dispersion behaviour of dense gases. The results show how ¯at terrain models may overestimate the chlorine hazard range by as much as a factor of 5, whilst the predicted direction of travel of the cloud may err by up to 908 . This has implications not only for the assessment of risks associated with major hazard facilities, but also for land-use planning in the vicinity of the site and emergency preparedness. Keywords: dispersion; chlorine; wind tunnel; Computational Fluid Dynamics; major hazards; complex terrain.

BACKGROUND

small in relation to the cloud height; in other words that signi®cant obstacles, which may act to channel or divert a dense gas cloud, cannot simply be treated as additional surface roughness1 . Apart from their application in QRA studies, dispersion models are also used to inform decisions on land-use planning and emergency planning at major hazard sites (e.g., to set consultation distances or locate off-site emergency control centres, police road check-points, etc.). This may involve undertaking dispersion calculations down to much lower concentration levels (e.g., a few ppm) than of interest in the QRA. For ¯at terrain models, this often leads to unsatisfactory results (e.g., distances of many kilometres, or even tens of kilometres), when it is clear that topography will exert a signi®cant in¯uence on dispersion behaviour. There is a need therefore for more realistic modelling which will provide improved information about the extent of the off-site hazard.

Quantitative Risk Assessment (QRA) studies of major hazard facilities commonly employ `¯at terrain’ dispersion models in their evaluation of the dispersion of hazardous dense gas releases. This is because these models are relatively easy to use and generate dispersion distances which are generally perceived as being conservative, but not unduly so. Many major hazard sites, however, differ markedly from the ideal of ¯at, unobstructed terrain. They may be characterized by the presence of large on-site buildings (e.g., administration buildings, workshops or warehouses) or areas of congested, multi-level process plant. The surrounding environment may also contain signi®cant natural or man-made features such as hills, valleys, embankments, cuttings, tall vegetation and multi-storey buildings. The question therefore arises as to what extent such features in¯uence the dispersion of the gas and the circumstances in which the use of conventional ¯at terrain models could either signi®cantly overestimate, or even underestimate, the extent of the hazard zone. In ¯at terrain models, the presence of buildings or tall vegetation is often taken into account by simply increasing the surface roughness parameter. However it is an inherent limitation of such models that the surface roughness height must be

INTRODUCTION This paper describes work undertaken by Environmental Resources Management (ERM), in association with Rowan Williams Davies & Irwin Inc. (RWDI) and the Health and Safety Laboratory (HSL). The work concerns the modelling of the dispersion of chlorine in complex terrain with 89

MCBRIDE et al.

90

buildings. This work was undertaken as part of a QRA study of eight major hazard installations using chlorine in 1 tonne or 50 kg containers. The study involved a fundamental reappraisal of the methodology for undertaking such assessments and included a critical review of emergency planning practice, safety management and land-use planning at each of the eight sites. A key requirement of the study was to model the in¯uence of topography and buildings (on-site and off-site) on chlorine dispersion in the atmosphere. Accordingly, ERM developed an approach based around the use of `state of the art’ Boundary Layer Wind Tunnel (BLWT) testing and Computational Fluid Dynamics (CFD) techniques, supplemented by conventional ¯at terrain dispersion modelling. This paper describes the key ®ndings of the chlorine dispersion modelling work, revealing the fundamental in¯uence buildings and terrain can have on dispersion behaviour. It also identi®es the circumstances in which conventional ¯at terrain modelling could be signi®cantly in error. The implications of this work extend beyond chlorine to other types of dense gas release. OVERALL APPROACH Overview of Hazard The work undertaken concerns eight major hazard installations using chlorine in 1 tonne or 50 kg containers (hereinafter referred to as Sites 1±8). The sites are characterized by the diversity of the topography which surrounds them, ranging from relatively ¯at terrain to complex topography featuring valleys and ridges. Large onsite buildings are a feature of all of the sites and many are also located near to major off-site buildings, including multi-storey residential accommodation. Two types of release were of concern: instantaneous releases of chlorine due to catastrophic failure of a container; and continuous releases due to leakage from ®ttings on the container, e.g., the container valve, fusible plugs, etc. Some examples of

actual incidents involving 1 tonne chlorine containers are given in Table 1. Approach to Chlorine Dispersion Modelling Wind tunnel testing (undertaken by RWDI) and CFD modelling (by HSL) were used to model the effects of buildings and terrain at the eight sites. Wind tunnel testing was found to be the most cost-effective method and was therefore applied to all eight sites in order to investigate the dispersion behaviour of a range of chlorine releases at different wind speeds. Typically, 30 tests were undertaken for each site, i.e., 30 combinations of release rate, wind direction and wind speed. However, as the wind tunnel testing was limited to the consideration of near-neutral atmospheric stability, CFD was used to investigate the in¯uence of stability on typical chlorine releases at two sites. A total of 12 CFD simulations were undertaken. Conventional ¯at terrain modelling was undertaken to provide the `source term’ for both the wind tunnel and CFD simulations. This is because of the important near-®eld momentum and thermodynamic (i.e., aerosol) effects associated with releases of chlorine from pressurized, lique®ed containment, which cannot easily be simulated with either the wind tunnel or CFD. The models chosen for this work were EJECT2 and DRIFT3,4 . For consistency between the wind tunnel and CFD simulations it was determined that the source term be de®ned by the point at which the aerosol was 99% vaporized. The following sections describe each aspect of the dispersion modelling in further detail. WIND TUNNEL TESTING Background Wind tunnel simulations of heavy gas releases have been undertaken for several years, but recently interest in this

Table 1. Examples of incidents involving 1 tonne chlorine containers (Source: MHIDAS). Date

Location

25/9/78

Vancouver, Canada

16/6/81

Atlanta, Georgia, US

9/2/87

Morristown, Tennessee, US

5/6/89

New Delhi, India

13/6/89 7/4/95

Description A chlorine leak from a 1 te drum on a truck. The truck was carrying 12 drums, and 6 drums broke loose when the truck braked at an intersection and fell onto the street. 3 hours later, a truck with caustic soda arrived and the chlorine was neutralized by piping it into this solution. A chlorine leak from a cylinder on a truck. The truck was at a loading dock in the garage of a hotel. A 68 kg cylinder fell from the truck and began to leak. 33 people were injured, including 6 ®remen, 4 police and 23 civilians. Rupture of 1 te container of chlorine gas at water treatment plant caused 1000 residents to be evacuated from homes ahead of an advancing chlorine gas cloud. Two company employees and 2 ®re®ghters injured. A chlorine leak from a 1 te drum on a truck. A fusible plug was found to be failed. One hour later, a policeman drove the truck to an isolated spot and the leak was plugged.

Release quantity, kg

Number of fatalities

Number of injuries

600

0

77

68

0

33

1000

0

4

±

0

200

Khabarovsk, Soviet Union

A chlorine drum burst at a water pumping station, sending a cloud of chlorine over a riverside area.

800

±

70

China

Steel chlorine canister exploded at the Chengdu Chemical Plant.

±

3

19

Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN ®eld has increased. Examples of recent studies involving both heavy gases and complex terrain include Davies and Hall5 , Heidorn et al.6 , Ayrault et al.7 , KoÈnig-Langlo and Schatzmann8 and Schatzmann et al.9 . Most of these studies concerned instantaneous releases; however, continuous releases have also been investigated.

Wind Tunnel Scaling Scaling from full to model scale involves holding signi®cant physical parameters constant. In the case of a wind tunnel simulation of heavy gas dispersion, the simulation is considered to be representative of full scale if similarity is ensured for the following: · Reynolds number of the ¯ow, Re LUra /m; · Richardson number, Ri g Dr/ra L/U 2 or densimetric Froude number, Fr U/ Dr/ra gL ; · Froude number, Fr U/ gL ; and · initial density ratio, rg /ra . All of these criteria cannot be satis®ed in practice and a review of the literature shows that, for dense gas releases, the scaling emphasis is either on the Richardson number or the Froude number and density ratio. For the present study Richardson number scaling was used, which simpli®es to the relationship: Lm Drm /Um2

Lf Drf /Uf2

Wind Tunnel Simulation RWDI has two BLWTs. Figure 1 shows a schematic of the tunnel used for this study. Wind tunnel simulations are constrained by the tunnel’s physical characteristics, especially in simulating the boundary layer. These limitations include: · wind speeds (must be greater than 1 ms 1 at gradient height); · atmospheric stability (limited to neutral stability); and · height of topography (must not cause more than 20% blockage of the tunnel). For this study a model scale of 1 : 500 was chosen, permitting representation of chlorine dispersion over a full-scale distance of 1.2 km for a model diameter of 5 m. The wind tunnel simulations were performed in two phases: the ®rst phase was to visualize the heavy gas ¯ow using smoke and the second phase was to use tracer gases which were emitted from the modelled sources to measure actual gas concentrations. The scope of the wind tunnel simulations included modelling the wind®eld upstream of the release location as well as the downwind dispersion of the chlorine cloud. This was an important consideration as the complexity of the terrain meant that wind conditions (i.e., speed and direction) at the release location could be signi®cantly different from the nominal weather conditions recorded at the meteorological station local to the installation.

1

Within this relationship L, r and U can be adjusted to allow for the wind tunnel limitations. However it is clear that there is a limitation to modelling dense gas releases over large dispersion distances if the wind tunnel speed and model scale density ratio are to remain within reasonable limits. Referring to equation (1) above, if Lf /Lm is large, then Uf /Um 2 Drm /Drf must also be large, which may create practical dif®culties if the model scale wind speed Um is required to be too low or density ratio Drm /Drf too high.

Heavy Gas Release Simulation Propane was used as the tracer gas for this study. The propane was introduced at each source through a sintered bronze plate at an initial concentration of 100%. Two types of chlorine release were simulated: instantaneous and continuous releases. The design criteria for the source of the release were as follows: · · · ·

shape similar to full scale; release ¯ux similar to full scale; minimum interference from release; and no leaks from the source.

Figure 1. Schematic of RWDI’s boundary layer wind tunnel No. 2.

Trans IChemE, Vol 79, Part B, March 2001

91

MCBRIDE et al.

92

The release characteristics (volume, density, etc.) were provided by the ¯at terrain modelling of the near ®eld dispersion phase. For the instantaneous releases, an HP Arbitrary Waveform Generator (AWG) was programmed with waveforms which produced a release volume simulating a 1 tonne instantaneous release of chlorine. Following KoÈnig-Langlo and Schatzmann8 , each instantaneous release was repeated ®ve times to provide a statistically suf®cient representation of peak concentration and concentration variability with time. Model Design The wind tunnel scale models were 5 m in diameter or 2.5 km (full scale). The models were placed on a turntable in the wind tunnel, which allowed them to be rotated to simulate different wind directions. The models included all topographic features in the study area, as well as all signi®cant on-site and off-site buildings (Figure 2). Receptor Design Two multi-plexed Flame Ionization Detectors (FIDs) were used to record propane concentrations at each of 30 receptor locations at a height of 1.5 m above grade (full scale). For this study, a frequency response of about 12 Hz was required and the sample tube dimensions for the FIDs were selected accordingly. A digitizing unit converted the FID signals to counts, which were then converted to ppm by using constants determined from the FID calibration for each test. Simulation Outputs The main outputs from each wind tunnel simulation were: · a plot of lethal dose (LD) contours, i.e. LD03 (3% probability of fatality), LD50 and LD90 contours; and

· isopleths corresponding to chlorine concentrations of 10, 100 and 1000 mg m 3 ; Lethal dose was calculated according to the following equation: … Lethal dose C 2.3 dt 2

where C gas concentration and the period of integration was set at 10 minutes (for continuous releases) and 10 minutes or the cloud passage time (whichever was the shorter) for instantaneous releases. The probit equation used for this study was the TNO `Green Book’ probit10: Probit

14.3

ln C 2.3 t

3

giving LD03 LD50 LD90

3.68 ´ 107 mg m 2.41 ´ 108 mg m 8.67 ´ 108 mg m

3 2.3

min; min; and 2.3 min.

3 2.3 3

COMPUTATIONAL FLUID DYNAMICS Overview The CFD code employed for this study was AEA Technology’s CFX-4 code, which is representative of the `state of the art’ of commercial CFD packages. It is a structured grid, multi-block, boundary-®tted code based on the ®nite volume method. The code incorporates multiple discretization schemes and solution algorithms, allowing the choice of appropriate schemes and algorithms according to the requirements of the study. CFX-4 has a track record of application to safety-related ¯ows, including dense gas dispersion11 . Betts and Haroutunian12 and Sayma and Betts13 have also demonstrated the ability of CFD (FEMSET code) to reproduce the results of the Burro 8 ®eld trial (Koopman et al.14 )Ðcomprising dense gas dispersion over mildly sloping terrain in stable atmospheric conditionsÐto within the uncertainty of the experimental data.

Figure 2. Example wind tunnel scale model (Site 7).

Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN

upper boundary is that the overall cross-sectional area available to ¯ow through the domain is constant, ensuring that, at altitude, the wind speed remains constant with that speci®ed at the upstream domain boundary. A further advantage is that the grid cell distribution in the vertical direction can be ®xed at the domain boundary and then propagated throughout the domain, ultimately easing representation of on- and off-site buildings. The computational grid comprised 30 connected subblocks with a total of over 300,000 grid cells at each of the two sites. This is in excess of previously published work15 . The multi-block grid system of CFX-4 was used to control the disposition of grid cells over the domain, allowing local grid re®nement over the site and regions immediately downstream. Further from the site, the grid cells were expanded to allow coverage of a wide area of topography. In the vertical direction, grid cells were clustered close to the ground, to aid resolution of building wake ¯ows and in recognition of the fact that the density of the chlorine will, at least at high concentrations, ensure that much of the important ¯ow and dispersion dynamics would occur close to the ground.

Table 2. Extent of CFD computational domain. Site Site 1 Site 3

Upwind extent, m

Downwind extent, m

Width, m

Height above local ground level, m

230 550

1545 1200

1500 1400

500 400

93

The main role of CFD in this study was to examine the in¯uence of atmospheric stability on chlorine dispersion in complex terrain with buildings. CFD modelling was undertaken for D and F stability conditions at two sites for two types of chlorine release: a 1 tonne instantaneous release and a continuous release of 1.4 kg s 1 (the latter corresponding to a liquid release from guillotine failure of the container draw-off valve). Additionally, a simulation of a 1 tonne instantaneous release was undertaken in B conditions at one of the sites. Simulations of the two types of chlorine release were also undertaken for ¯at terrain to provide a comparison with the DRIFT ¯at terrain modelling (see below). The following paragraphs provide the details of the application of CFD in this study. Computational Grid and Domain

Boundary Conditions

The computational domain de®nes the region of space in which ¯ow and dispersion will be simulated. The dimensions of the domain for each of the two sites studied are shown in Table 2. It can be seen that the domain encompasses a signi®cant upwind fetch, to allow the speci®ed atmospheric conditions to respond to the local terrain. Also the lateral and vertical extent of the domain have been set so as to avoid any interference on chlorine dispersion in the vicinity of the release location. A signi®cant advantage of employing a terrain-following

At the ground, no-slip conditions with a uniform roughness length of 0.2 m were used for both sites. This is representative of trees with a few buildings16. For the ¯at terrain simulations, a uniform roughness length of 0.01 m was used, consistent with the DRIFT modelling. For D2 weather, heat transfer from the ground to the cold chlorine cloud was modelled using a standard wall function approach17. For F2 weather, a ®xed heat ¯ux to the ground was speci®ed. At the upstream boundary, vertical pro®les of velocity,

Table 3. CFD boundary conditions. Stability D

Parameter Velocity Turbulence

F

Velocity Turbulence Temperature

B

Velocity Turbulence

Temperature

Pro®le u K e u Lmo e f K T T u Wm X K e fK fe fm T

WT Y

Values

u /k ln z/z0

u u Cm

2

u 3 / Cm u / kz u /k ln z/z0

Lmo 107 m (Hunt et al.20 ) 0.1598 ms 1 u

5 z/Lmo

3

u3 Cp r0 T0 / kgH u /kz f 1 2.5 z/Lmo 0.6 1.5 eLmix /Cm0.75 2/3 1/k ln z/z0 5 z/Lmo T0 /T H/ rCp u u /k ln z/z0 Wm z/Lmo 2 ln 1 X /2 ln 1 X 2 /2 0.25 1 2 15 z/Lmo u3 /kCm0.5 fK u /kz fe fe /fm 0.5 fm z/Lmo 1 16 z/Lmo 0.25 T0 /T j/k ln z/z0 WT z/Lmo

2 ln 1 Y 2 /2 1 9 z/Lmo 0.25

Trans IChemE, Vol 79, Part B, March 2001

0.1013 for z0 0.01 m 0.1789 for z0 0.2 m 0.09, k 0.35

Lmix Lmix

2 tan

1

X

kz for z < Cm d/k Cm d for z > Cm d/k

p/2

j 0.85 10.218 m Lmo for H 150 W m (Hunt et al.20 ) 0.246 ms 1 u

2

MCBRIDE et al.

94

turbulence and temperature (F2 only) were speci®ed as detailed in Table 3. At the downstream boundary, a `mass ¯ow’ boundary condition was speci®ed, which ®xes the total mass out¯ow rate to be the same as that entering the domain from all sources. Additionally it imposes a zerogradient condition on ¯ow variables at the boundary. At the remaining vertical sides of the domain, symmetry conditions were speci®ed, which constrains the bulk ¯ow to be parallel to the lateral boundaries at these locations. (For the ¯at terrain simulations a symmetry condition was also speci®ed on a vertical plane coincident with the wind direction and passing through the centre of the release.) At the upper surface of the domain, a uniform velocity was speci®ed consistent with the appropriate atmospheric ¯ow pro®le. The continuous chlorine releases were modelled as steady state, whereas the instantaneous releases required a time-dependent simulation. The initial conditions for this time-dependent simulation were generated by running the model for the ambient wind®eld alone until a converged steady-state solution was obtained for the ¯ow through the domain. Chlorine Source Terms As for the wind tunnel testing, the chlorine source terms for the CFD simulations were provided by the DRIFT ¯at terrain modelling. For the 1.4 kg s 1 continuous release case, the code EJECT was also used to simulate the initial momentum-dominated two-phase ¯ashing jet ¯ow. Physical Sub-models The physical sub-models used in this study comprised: · a buoyancy-modi®ed k-e turbulence model; · solution of a scalar transport equation for chlorine; · solution of an enthalpy transport equation to account for heat transfer; and · solution of a weakly compressible form of the transport equations by use of an equation of state. Numerical Sub-models To minimize the possibility of numerical instability arising, the computational grids for this study were constructed as far as possible to be near-orthogonal over most regions. Higher-order discretization schemes were used, wherever possible, to minimize the deleterious effects of numerical diffusion arising from the use of ®rst-order schemes. Convergence of the iterative solution algorithm was judged by a number of means: · examination of the overall mass balance for bulk ¯ow through the domain; · examination of the mass balance for chlorine released into the domain; and · monitoring of solution values at key locations. Typical converged mass imbalances for neutral and stable atmospheric conditions were of the order of 0.1%.

Simulation Outputs Outputs from the CFD simulations were as per the wind tunnel testing, together with concentration-time histories and concentration-height pro®les at key locations throughout the domain. The chlorine concentrations predicted by the CFD model were ensemble mean values. FLAT TERRAIN DISPERSION MODELLING Conventional ¯at terrain dispersion modelling was undertaken using the codes EJECT and DRIFT. EJECT is a two-phase jet dispersion model using the same thermodynamic model as TRAUMA18 , but with the two-phase jet dynamics extended to cover bending due to the wind ®eld and/or gravity. DRIFT is a `second generation’ dispersion model applicable to heavy gases, passive gases and aerosol clouds. For continuous releases, EJECT and DRIFT were used in combination to simulate the near ®eld dispersion of chlorine, i.e. EJECT for the initial two-phase jet dispersion and DRIFT for the subsequent dilution and vaporization of the chlorine vapour/aerosol cloud. For instantaneous releases (e.g., catastrophic failure of a chlorine container), DRIFT only was required. For both types of simulation it was assumed that the liquid fraction of the chlorine release was fully-entrained as an aerosol, i.e. no rain-out. From the EJECT and DRIFT modelling it was found that, for 1.4 kg s 1 continuous releases, the chlorine aerosol was 99% vaporized at a point 20 m downwind of the release location. For instantaneous releases of 1 tonne of chlorine the corresponding transition point was at 60 m downwind. These distances therefore represent the ¯at terrain `off-set’ from which the wind tunnel and CFD could then be used to model the effects of terrain and buildings on the dispersion behaviour. To provide a comparison of the wind tunnel/CFD results against the equivalent ¯at terrain predictions, the DRIFT runs were extended to simulate the entire downwind phase of the dispersion. Another commonly-used ¯at terrain code HGSYSTEM19 was also run to provide lethal dose and concentration contours for comparison against the wind tunnel and CFD predictions. RESULTS OF THE DISPERSION MODELLING Wind Tunnel Testing The wind tunnel testing revealed the signi®cant in¯uence complex terrain and buildings can have on the dispersion of dense gases in the atmosphere. Such features affect not only the downwind hazard range, but also the width of the cloud and its direction of travel. The effects observed included: · channelling of dense gas clouds along valley ¯oors or cuttings; · diversion of gas clouds by tall buildings or hills; · preferential ¯ow of dense gas clouds down slopes (including upwind ¯ow); · entrapment of dense gas clouds within the wake zones formed in the lee of hills; · lateral spreading of dense gases at the foot of upslopes; · deviations of the direction of travel of dense gas clouds Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN Table 4. Selection of wind tunnel test results (in all cases wind speed at upstream boundary Maximum extent of cloud, m

2 ms 1 ).

Wind direction (at upwind boundary)

LD03*

Site 1 Site surrounded by hills to 1 tonne instantaneous N, W and S. Tall buildings to E of site. Site 3 Site located on ridge 1 tonne instantaneous running N-S. Surrounding countryside ¯at.

WSW

670 (1.12)

700

420

Cloud travels towards NNE

N

490 (0.82)

1200

480

Site 4 Site cut into side of broad 1 tonne instantaneous slope. Tall buildings to E of site. Site 5 Large hills to immediate S 1 tonne instantaneous and W of site. Surrounding terrain ¯at.

W

570 (0.95)

860

645

S

225 (0.38)

550

340

Cloud splits either side of ridge towards SSW and SSE. Cloud also travels upwind towards NW. Cloud splits either side of high-rise estate towards NE and ESE ±

N SE

490 (0.82) 270 (0.45)

1040 310

520 450

± Cloud travels towards SW

Site

Description of topography and signi®cant buildings

Release case

Site 7 Large hills to SE of site. 500 kg instantaneous Tall buildings to the immediate S and W of site.

100 mg m

3

Maximum width to LD03, m

95

Direction of travel of cloud

* ®gure in brackets is fraction of equivalent LD03 distance predicted by DRIFT ¯at terrain modelling (i.e., 600 m, from Table 8).

due to features upwind of the release location (affecting local wind patterns); and · enhanced dispersion of dense gases released on ridges or hill tops as a result of local acceleration of the wind. Table 4 presents some of the key results from the wind tunnel testing where topography and/or buildings were observed to have a signi®cant in¯uence on the dispersion behaviour. Figures 3±8 show the lethal dose contours predicted for these cases directly compared against the equivalent results from the ¯at terrain modelling (HGSYSTEM results). It is emphasized that data presented in Table 4 and Figures 3±8 represent only a small fraction of the total data generated in the wind tunnel testing for this study. The cases selected are illustrative of extreme conditions where the wind tunnel predictions of the chlorine hazard zone

differed markedly from those of ¯at terrain dispersion models. Figure 3 shows an example of the diversion of a gas cloud by high-rise buildings near to a major hazard installation and subsequent channelling of the cloud along the valley ¯oor. In this case it was important to account for the diversion/channelling effect as the valley ¯oor contains some signi®cant outdoor populations (playgrounds and a swimming pool), which are relatively more exposed to the effects of chlorine than people within buildings in the highrise estate. Figure 4 shows an example of a dense gas cloud responding strongly to local topographic in¯uences. The gas cloud splits either side of the ridge on which the release occurs. It also travels a signi®cant distance (300 m) back down the slope, directly against the wind. In this case, accounting for the effects of terrain would not have had a

Figure 3. Example wind tunnel test results for Site 1. Release case: 1 tonne (instantaneous). Wind direction: WSW. Weather class: D2.

Trans IChemE, Vol 79, Part B, March 2001

96

MCBRIDE et al.

Figure 4. Example wind tunnel test results for Site 3. Release case: 1 tonne (instantaneous). Wind direction: N. Weather class: D2.

signi®cant effect on the risk assessment, as the area surrounding the site is sparsely populated. However it would have implications for emergency planning in that signi®cant concentrations of chlorine would be expected on the major roads to the north and east of the site (not predicted by the ¯at terrain modelling). Figure 5 shows the dramatic effect clusters of high-rise buildings can have on the dispersion of dense gases. In this case the gas cloud diverts preferentially either side of the high-rise blocks, adopting a U-shape, compared to the classical elliptical shape for dispersion of gas over ¯at terrain. As for Site 1, the diversion of the gas cloud has implications for the risk assessment in that the cloud is predicted to travel preferentially along a busy highway. The concentration of gas on the road (in excess of 1000 ppm) could be suf®cient to bring vehicles to a halt, potentially exposing large numbers of people to the toxic effects of the gas. Figure 6 and 7 show the contrasting effects of dense gas dispersion for different wind directions at the same site (Site 5). The site is bounded to the south and west by large hills.

For winds from the south (Figure 6) the chlorine release occurs in the wake zone of the nearby hills. This signi®cantly retards the downwind dispersion of the cloud, with correspondingly reduced consequences for the population lying to the north and east of the site. For winds from the north (release location near northern site boundaryÐ Figure 7) the dispersion of the gas is curtailed by the signi®cant upslope. Accordingly, the gas cloud spreads laterally, resulting in a shorter, wider cloud than would be predicted by ¯at terrain modelling. It is noted however that harmful concentrations of the gas (above 100 mg m 3 or 34 ppm) could still occur at signi®cant elevations (in excess of 200 m). Similar to Figure 6, Figure 8 shows the effects of a chlorine release into the lee of a nearby hill. Instead of moving in the downwind direction, the gas cloud moves at 908 to the mean wind ¯ow towards the south-west. Flow reversal situations, such as depicted in Figures 6 and 8, are clearly of interest from an emergency planning perspective, revealing the uncertainty in predicting the direction and extent of the off-site effects of hazardous dense gas releases

Figure 5. Example wind tunnel test results for Site 4. Release case: 1 tonne (instantaneous). Wind direction: W. Weather class: D2.

Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN

Figure 6. Example wind tunnel test results for Site 5. Release case: 1 tonne (instantaneous). Wind direction: S. Weather class: D2.

Figure 7. Example wind tunnel test results for Site 5. Release case: 1 tonne (instantaneous). Wind direction: N. Weather class: D2.

Figure 8. Example wind tunnel test results for Site 7. Release case: 500 kg (instantaneous). Wind direction: SE. Weather class: D2.

Trans IChemE, Vol 79, Part B, March 2001

97

MCBRIDE et al.

98

Table 5. CFD modelling results for ¯at terrain (D stability conditions, wind speed 2 ms 1 ).

Table 7. CFD modelling results for Site 3 (in all cases wind speed wind direction SW).

Downwind distance (m) Release case 1.4 kg s

1

Maximum extent of lethal dose contours, m

LD90

LD50

LD03

246

365

650

Release case

continuous

1.4 kg s

1

continuous

1 tonne instantaneous

in complex terrain. Advice given to the emergency services on whether it is safe to approach an installation (and if so, from which direction) could be in error if based solely on ¯at terrain dispersion results. Table 4 indicates that, overall, the wind tunnel chlorine hazard range may be as low as 40% of the equivalent hazard range predicted by the DRIFT ¯at terrain modelling. Computational Fluid Dynamics The results of the CFD modelling are summarized in Tables 5±7. The lethal dose contours are shown in Figures 9±17. Table 5 shows the results of the CFD modelling for ¯at terrain (1.4 kg s 1 continuous release) which may be compared with the equivalent results from the DRIFT modelling (Table 8). It can be seen there is close agreement between the two sets of data, which lends con®dence to the CFD predictions, since DRIFT has been validated against ¯at terrain ®eld and wind tunnel data. Table 6 and Figures 9±12 compare the results of the CFD modelling for different atmospheric stability conditions at Site 1. It can be seen that there is comparatively little difference in the extent of the lethal dose contours between D and F stability conditions. This indicates that, for chlorine concentrations over the range of interest, the in¯uence of atmospheric stability may be relatively slight when buildings and complex terrain are taken into account. It is also apparent from these results that the predicted extent of the lethal dose contours is signi®cantly reduced compared to the DRIFT and HGSYSTEM ¯at terrain results (Table 8), i.e., may be as low as 20% of the DRIFT predictions (1.4 kg s 1 continuous release in F2 conditions). It is possible to speculate that this stems from the increased mixing generated by ¯ow over the site buildings and, possibly, the increased ground roughness. It is noted that, whilst CFD predicts a reduced downwind hazard range compared to the ¯at terrain models, the vertical hazard range is correspondingly greater, e.g., the predicted cloud height for a 1.4 kg s 1 Table 6. CFD modelling results for Site 1 (in all cases wind speed wind direction WSW).

2 ms 1 ,

Stability

LD90

LD50

LD03*

D F D F B

135 130 200 180 75

165 180 215 255 95

265 (0.48) 330 (0.29) 255 (0.43) 355 (0.44) 105 (0.22)

* ®gure in brackets is fraction of equivalent LD03 distance predicted by DRIFT ¯at terrain modelling

continuous release in D2 weather conditions is typically 60 m). This is an important parameter in assessing risk to occupants of multi-storey buildings. Table 7 and Figures 13±17 compare the results of the CFD modelling for different atmospheric stability conditions at Site 3. For the 1.4 kg s 1 continuous release, Table 7 shows comparatively little difference in the downwind extent of the LD03 contour (D2Ð265 m, F2Ð330 m) and certainly not as great a difference as would be predicted by DRIFT (a factor of 2) or HGSYSTEM (a factor of 1.6). However it is apparent from Figures 13 and 14 that the predicted direction of travel of the cloud is somewhat different. It is possible that, under F2 conditions, the chlorine is initially responding more directly to changes in local elevation. This is because the consequences of stable atmospheric conditions can be expected to be a suppression of vertical mixing of air into the plume and a reduction in wind shear, resulting in a ¯ow more strongly in¯uenced by gravity.

2 ms 1 ,

Maximum extent of lethal dose contours, m Release case 1.4 kg s

1

continuous

1 tonne instantaneous

Stability

LD90

LD50

LD03*

D F D F

149 145 170 220

188 155 200 255

260 (0.47) 225 (0.20) 255 (0.43) 275 (0.34)

* ®gure in brackets is fraction of equivalent LD03 distance predicted by DRIFT ¯at terrain modelling

Figure 9. CFD modelling results for Site 1. Release case: 1.4 kg s (continuous). Wind direction: WSW. Weather class: D2.

1

Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN

99

1

Figure 12. CFD modelling results for Site 1. Release case: 1 tonne (instantaneous). Wind direction: WSW. Weather class: F2.

Figures 15 and 16 show that for the 1 tonne instantaneous release the difference in cloud trajectory between D and F stability conditions is more marked. In F conditions the cloud slumps down the hill to the north (at

right angles to the mean wind ¯ow) resulting in a larger hazardous `footprint’ in which some 350 m of the adjacent road are impacted. For the simulation of B stability conditions, great

Figure 11. CFD modelling results for Site 1. Release case: 1 tonne (instantaneous). Wind direction: WSW. Weather class: D2.

Figure 13. CFD modelling results for Site 3. Release case: 1.4 kg s (continuous). Wind direction: SW. Weather class: D2.

Figure 10. CFD modelling results for Site 1. Release case: 1.4 kg s (continuous). Wind direction: WSW. Weather class: F2.

Trans IChemE, Vol 79, Part B, March 2001

1

100

MCBRIDE et al.

1

Figure 16. CFD modelling results for Site 3. Release case: 1 tonne (instantaneous). Wind direction: SW. Weather class: F2.

dif®culty was found in generating a steady-state wind®eld and a number of convergence strategies were tried before solution residuals fell within acceptable bounds. Convergence was achieved, but not to the same degree as the

previous cases. Interpretation of these results must therefore be approached with more caution than those of the preceding case. Nevertheless Figure 17 indicates that mixing of the chlorine in the atmosphere occurs much more rapidly under

Figure 15. CFD modelling results for Site 3. Release case: 1 tonne (instantaneous). Wind direction: SW. Weather class: D2.

Figure 17. CFD modelling results for Site 3. Release case: 1 tonne (instantaneous). Wind direction: SW. Weather class: B2.

Figure 14. CFD modelling results for Site 3. Release case: 1.4 kg s (continuous). Wind direction: SW. Weather class: F2.

Trans IChemE, Vol 79, Part B, March 2001

USE OF TECHNIQUES TO MODEL THE DISPERSION OF CHLORINE IN COMPLEX TERRAIN Table 8. Flat terrain dispersion modelling results (in all cases wind speed

101

2 ms 1 ). Downwind distance, m

Model

Release case

DRIFT*

1.4 kg s

1

continuous

1 tonne instantaneous HGSYSTEM²

1.4 kg s

1

continuous

1 tonne instantaneous

* Surface roughness

0.01 m, ² surface roughness

Stability

LD90

LD50

LD03

D F B D F B D F B D F B

268 460 184 325 400 293 200 300 100 450 300 375

362 650 234 425 550 350 250 400 125 550 550 450

550 1120 323 600 800 470 375 600 170 775 800 600

0.1 m

B stability conditions than under either D or F conditions with a consequent reduction in the downwind distance to a given toxic load. Qualitatively, this is consistent with the behaviour predicted by DRIFT and HGSYSTEM (Table 8).

approach was adopted of using the CFD results for the smaller continuous releases and the wind tunnel results for the larger instantaneous releases. Both sets of results offered signi®cant improvement over the ¯at terrain estimates conventionally used in QRA studies.

COMPARISON OF WIND TUNNEL AND CFD In this study the wind tunnel and CFD were used in complimentary roles to address different aspects of the dispersion of chlorine in complex terrain with buildings. However, at two sites, both techniques were used to study the same release scenarios. The results are compared in Table 9. These results show that, for the small continuous release case, the CFD predicts the longer hazard range (by a factor of 2 or more), whilst for the large instantaneous release, the wind tunnel predicts the longer dispersion distance (by a factor of 2.5). It should be noted, however, that there are important differences in the modelling approach between the CFD and the wind tunnel. These concern: · representation of buildings (simpler in the CFD model than in the wind tunnel); · representation of the incoming wind pro®le; · the extent of the upwind fetch (greater in the wind tunnel than the CFD model); · modelling of heat transfer (included in the CFD model, but not in the wind tunnel); and · the precise representation of the source term. Due to these differences it is not possible to draw reliable conclusions as to the comparative performance of the wind tunnel and CFD in this study. For the QRA, a conservative Table 9. Comparison of wind tunnel and CFD results for Sites 1 and 3 (neutral atmospheric stability, wind speed 2 ms 1 ). Maximum extent of LD03 contour, m Site Site 1 Site 3

Release case

Wind tunnel

CFD

1.4 kg s 1 continuous 1 tonne instantaneous 1.4 kg s 1 continuous 1 tonne instantaneous

<125* 670 <125* 680

260 255 265 255

* LD03 contour did not extend off-site

Trans IChemE, Vol 79, Part B, March 2001

CONCLUSIONS The work reported in this paper indicates that complex terrain and tall buildings can exert a signi®cant in¯uence on the dispersion of dense gases in the atmosphere. These effects include: · diversion of dense gas clouds by hills or tall buildings; · channelling of clouds along valleys or cuttings; · entrapment of releases within the wake zone on the lee side of hills; · preferential ¯ow of dense gases down slopes (including upwind ¯ow); and · lateral spreading of dense gas clouds at the foot of upslopes. Whilst many of these effects would result in a reduction in hazard range, it has been shown that, in certain circumstances, off-site risks could increase. This is the case where the terrain-modi®ed trajectory of the cloud could impact populations which might otherwise have been assumed not to be affected (based on ¯at terrain model predictions). The CFD component of this study indicates that the in¯uence of atmospheric stability (i.e. neutral vs stable conditions) on the chlorine hazard range may be relatively slight in the presence of complex terrain and buildings. However it is shown that the trajectory of the cloud could be signi®cantly modi®ed in stable conditions, i.e. that the cloud responds more strongly to local topographic features. The ®ndings of this study are relevant not only to the assessment of risks for major hazard installations, but also to land-use planning and emergency preparedness. In both these cases an improved understanding of topographical in¯uences on dense gas dispersion may assist in determining such parameters as consultation distances, evacuation distances and locations for off-site emergency control centres, police road check-points, etc.

MCBRIDE et al.

102

At present large wind tunnel or CFD modelling studies of the type described in this paper are largely con®ned to the research domain. However there is an increasing interest in the application of physical modelling and computational techniques and a growing body of data on the effects of terrain and buildings on dense gas dispersion. Therefore operators of major hazard facilities should at least be aware of the important physical effects, particularly where these could act to increase risks. Finally it should be noted that there are still signi®cant uncertainties associated with the application of both wind tunnel testing and CFD modelling in this context and a need for further full-scale comparative studies. Nonetheless, the combination of these techniques offers a signi®cant insight into how hazardous dense gases behave in `real’ environments.

5.

6.

7. 8. 9.

10.

NOMENCLATURE Cp f g H K L Lmo m T T0 U u u z z0

speci®c heat sub-script denoting full scale acceleration due to gravity heat ¯ux turbulent kinetic energy length Monin-Obhukov length sub-script denoting model scale temperature temperature of ground velocity wind speed friction velocity height above ground level surface roughness height

d e ra rg r0 m j

boundary layer height rate of dissipation of kinetic energy density of air gas density density of air viscosity turbulent Prandtl number

11. 12. 13. 14. 15. 16. 17. 18. 19.

REFERENCES 1. Center for Chemical Process Safety, 1996, Use of Vapor Cloud Dispersion Models, 2nd Edition (American Institute of Chemical Engineers). 2. Tickle, G. A., Jones, S. J., Martin, D., Ramsdale, S. A. and Webber, D. M., 1997, Development and Validation of Integral Models of TwoPhase Jets (AEA Technology/1389). 3. Webber, D. M., Jones, S. J., Tickle, G. A. and Wren, T., 1992, United Kingdom Atomic Energy Authority Safety and Reliability Directorate, SRD/HSE R587, A Model of a Dispersing Gas Cloud and the Computer Implementation DRIFT-II Steady Continuous Releases. 4. Webber, D. M., Jones, S. J., Tickle, G. A. and Wren, T., 1992, United Kingdom Atomic Energy Authority Safety and Reliability Directorate,

20.

SRD/HSE R587, A Model of a Dispersing Gas Cloud and the Computer Implementation DRIFT-II Instantaneous Releases. Davies, J. K. W. and Hall, D. J., 1996, An analysis of some replicated wind tunnel experiments on instantaneously released heavy gas clouds dispersing over solid and crenellated fences, Journal of Hazardous Materials, 49: 311±328. Heidorn, K.C., Murphy, M.C., Irwin, P.A., Sahota, H., Misra, P.K. and Bloxam, R., 1992, Effects of obstacles on the spread of a heavy gasÐ wind tunnel simulations, Journal of Hazardous Materials, 30: 151±194. Ayrault, M., Balint, J-L. and Morel, R., 1991, An experimental study on the evolution and dispersion of a cloud of gas heavier than air, Journal of Hazardous Materials, 26: 1±26. KoÈnig-Langlo, G. and Schatzmann, M., 1991, Wind tunnel modelling of heavy gas dispersion, Atmospheric Environment, 25A(7): 1189±1198. Schatzmann, M., Marotzke, K., and Donat, J., 1991, Research on continuous and instantaneous heavy gas clouds, Contribution of subproject EV 4T-0021-D to the ®nal report of the joint CEC-project (Meteorological Institute, University of Hamburg). Committee for the Prevention of Disasters, 1992, Methods for the Determination of Possible Damage to People and Objects Resulting from Releases of Hazardous Materials, Rep. CPR 16E (Voorburg) (the `Green Book’). Witlox, H. W. M., 1990, Mathematical Modelling of Threedimensional Dense-gas Dispersion Problems, Report No: TRCP.3268R (Shell Research Ltd, Thornton Research Centre). Betts, P. L. and Haroutunian, V., 1988, Finite element calculations of transient dense gas dispersion, in Puttock, J. S., Stably Strati®ed Flow and Dense Gas Dispersion (Clarendon Press, Oxford), 349±384. Sayma, A. I. and Betts, P. L., 1997, A ®nite element model for the simulation of dense gas dispersion in the atmosphere, International Journal for Numerical Methods in Fluids, 24: 291±317. Koopman, R. P. and Baker, J., 1982, Burro Series Data Report: LLNL/ NWC 1980 LNG Spill Tests, Technical Report UCID-19075 (Lawrence Livermore National Laboratory). Hall, R. C., 1997, Evaluation of Modelling Uncertainty: CFD Modelling of Near-®eld Atmospheric Dispersion, Report on contract: EV5V-CT94-0531. Panofsky, H. A. and Dutton, J. A., 1984, Atmospheric Turbulence (Wiley). Launder, B. E. and Spalding, D. B., 1974, The numerical computation of turbulent ¯ows, Computer Methods in Applied Mechanics and Engineering, 3: 269±289. Wheatley, C. J., 1987, A User’s Guide to TRAUMA, a Computer Code for Assessing the Consequences of Accidental Two-Phase Releases of NH3 into Moist Air, HSE/SRD/079. Post, L., 1994, HGSystem 3.0 Technical Reference Manual, Report No: TNER.94.059 (Shell Research Ltd, Thornton Research Centre, UK). Hunt, J. C. R., et al., 1991, in Dopp, H. V. and Steyn, D. G. (eds), Developments in Modelling Air Pollution for Regulatory Purposes. Air Pollution Modelling and its application VIII (Plenum Press), 17±59.

ADDRESS Correspondence concerning this paper should be addressed to Mr M. A. McBride, Health and Safety Laboratory, Harpur Hill, Buxton, Derbyshire SK17 9JN, UK. The manuscript was received 7 June 2000 and accepted for publication after revision 13 February 2001.

Trans IChemE, Vol 79, Part B, March 2001