You are on page 1of 15

Atnmsphmic Ewironment Vol. 23, No.4,pp. 731-745, 1989. Printed inGnatBritain.

ooo16981/89 s3.at+o.o0 pqmm pnas plc

A REVIEW OF RECENT FIELD TESTS AND MATHEMATICAL MODELLING OF ATMOSPHERIC DISPERSION OF LARGE SPILLS OF DENSER-THAN-AIR GASES
RONALD P. KOOPMAN, DONALD L. ERMAK

and

STEVENS T. CHAN

Lawrence Livermore National Laboratory, Box 808, Livermore, CA 94550, U.S.A. (First received 15 February 1988 and injnulform
2 Nouember 1988)

Abstract-Large-scale spills of hazardous materials often produce gas clouds which are denser than air. The dominant physical processes which occur during dense-gas dispersion are very different from those recognized for trace gas releases in the atmosphere. Most important among these processes are stable stratification and gravity Sow. Dense-gas flows displace the ambient atmospheric flow and modify ambient turbulent mixing. Thermodynamic and chemical reactions can also contribute to dense-gas effects. Some materials flash to aerosol and vapor when released and the aerosol can remain airborne, evaporating as it moves downwind, causing the cloud to remain cold and dense for long distances downwind. Dense-gas dispersion models, which include phase change and terrain effects have been developed and are capable of simulating many possible accidental releases. A number of large-scale field tests with hazardous materials such as liquefied natural gas (LNG), ammonia (NH,), hydrofluoric acid (HF) and nitrogen tetroxide (N204) have been performed and used to evaluate models. The tests have shown that gas concentrations up to ten times higher than those predicted by trace gas models can occur due to aerosols and other dense-gas effects. A methodology for model evaluation has been developed which is based on the important physical characteristics of dense-gas releases. Key word index: Atmospheric dispersion, dense gas, accidental release, dispersion modelling, hazardous materials, gravity flow, aerosol, flashing, complex terrain.

INTRODUCIION

could be developed. This approach involves work in three complementary areas: modeling based on physical laws in the form of conservation equations and submodels of the important physical processes, (2) field experiments to validate models and discover important phenomena, (3) scaled simulations in wind tunnels, laboratory or water flume. Work at LLNL has concentrated on the first two of these areas. The dominant physical processes which occur during dense gas dispersion are very different than those recognized for trace gas releases. Some dense gas dispersion models now include these processes and can be evaluated against good quality data from a variety of well-instrumented dense gas dispersion experiments.
CAUSES OF DENSE GAS BEHAVIOR

Research into the atmospheric dispersion of denserthan-air gases was begun in the early 197Os,prompted by accidents such as the cyclohexane explosion at Flixborough (U.K.) in 1974, the pesticide release in Seveso (Italy) and the ammonia tanker truck spill in Houston (U.S.), both in 1976. Since then, the tragic methyl isocynate (MIC) release in Bhopal, India (1984) killed over 2000 people and the LPG explosions in Mexico City (1984) killed over 400 people. These tragic accidents indicate that further research into the possible consequences of accidental release of hazardous materials coupled with the development of prediction, mitigation and emergency response capabilities is still needed. The purpose of this paper is to review the major advances that have been made over the last 10 years in our understanding of the atmospheric dispersion of large-scale spills and our ability to predict the consequences of an accidental release. Dense gas dispersion phenomena, which are often characteristic of these spills, have been investigated intensively both through theory and experiment (Koopman, 1987). The emphasis of the research at Lawrence Livermore National Laboratory (LLNL) has been on gaining a physical understanding of the many processes involved such that a physically detailed and accurate quantitative predictive capability

(1) mathematical

The accidental release of hazardous gases into the environment often results in what has come to be known as dense gas behavior, even when the gases themselves may be nominally less dense than the atmosphere into which they are released. This dense gas behavior can dominate the consequences of the accidental release making them either worse or better

132

RONALD P. KOOPMAN et ~1.

than might otherwise be expected and making prediction of the consequences of the release more difficult. Materials released during an accident can be denserthan-air due to one or a combination of the following: (1) they are cold or cool from evaporation when released, (2) their molecular weight is higher than that of air; (3) they are contained at elevated pressure and flash to aerosol and vapor upon release; (4) chemical reactions. The first two categories, low temperature and high molecular weight, are largely obvious and will not be discussed further. The consequences of flashing, on the other hand, are not so obvious. Many materials, such as NH, and HF are kept at elevated pressure so that they may be efficiently stored as liquids or efficiently participate in a chemical process. If these materials are accidentally released, flashing will result in a mixture of cooled vapor and cooled liquid aerosol. The amount of vapor produced by flashing, the flash fraction, is a function of the initial storage temperature and pressure and the thermodynamic properties of the released material, and can be calculated from basic thermodynamics. The amount and size distribution of the liquid aerosol entrained in this vapor plume is not easily calculated and is currently the subject of research. Several competing theories are being examined but it appears to this author that the small droplet size predicted by the boiling breakup theory (Crowe and Comfort, 1978) is most consistent with the pm-size particles observed in the sparse existing data. Better data on HF aerosol size should be available soon and will help identify the correct approach to modeling aerosols. The presence of aerosols causes several dense-gas effects. First, the liquid droplets make the cloud denser than air even though the vapor may be lighter. Second, the evaporation of the liquid droplets cools the surrounding air/vapor mixture, which could further increase the cloud density. Third, turbulence is suppressed within the dense cloud, reducing mixing with ambient air. The importance of this effect was demonstrated during large-scale NH, releases performed at the Nevada Test Site in 1983 for the U.S. Coast Guard and The Fertilizer Institute (Goldwire, 1985,1986) and during HF tests (Goldfish series) performed in 1986 for Amoco Oil Co. (Blewitt, 1987a). In both cases approximately 20% of the liquid flashed to vapor and the remaining liquid (80%) formed a fine aerosol which remained airborne. Figure 1 shows the temperature depression measured 20 m and 60 m downwind for the first three Goldfish HF spill tests (Blewitt, 1987a). Flashing has created a cold, aerosol-laden cloud, and the aerosols evaporate as they travel downwind, causing the cloud to remain cold and dense. For the largest spill rate, Test 1, the cloud temperature is nearly the same at 60 m (Fig. 1b) downwind as it is at 20 m downwind (Fig. la). The HF was stored at approximately lOYF, close to ambient temperature. Chemical reactions can occur which increase or decrease cloud density. For example, when N204 (a

-10 -20 -30 OTest2 hTul3 _

-a01
0

200

400 do0 owl Tlm# from rplll (8%)

low

1200

5 0

(b)

-501
0

I ml

I 1200

400

1M w

Tlmo from s@II (m)

Fig. 1. Temperature differences for Goldfish HF Tests 1,2 and 3 measured at (a) 20 m and (b) 60 m downwind of the spill point.

rocket fuel oxidizer) is released, it dissociates rapidly to NO2 which undergoes a further rapid endothermic reaction with atmospheric water vapor to produce an HNO, mist (McRae, 1984, 1985). This mist behaves differently than does the ambient temperature NO,, cooling the cloud and increasing its density which reduces vertical mixing As another example, cold HF droplets cause water vapor to condense on them and react, releasing heat. This increases the temperature which could be expected to increase the evaporation rate and further cool the droplet. However, the water-HF solution has lower volatility allowing the droplets to persist longer downwind. The net result is that cloud travel distances to concentration levels of concern are expected to increase with increasing humidity up to about 70% for HF releases even though the HF-water reaction is exothermic. Work is planned to include these effects in dense-gas dispersion models so that parameter studies and analysis can be performed.
Dense

gas effects

The important dense-gas effects which are not observed in the dispersion of trace emissions include turbulence damping and gravity spreading due to

Review of recent field tests density gradients in the horizontal direction. In some situations, the dense cloud actually displaces the ambient wind field (Koopman, 1982a, 1987) in much the same way that wind flows over a solid body. This displacement results in a stably stratified dense gas layer and an interface through which it is difficult for external turbulence to penetrate (Hunt et al., 1983). Further, in many situations, the kinetic energy of the turbulence within the cloud is not sufficient to give fluid elements the large, vertical displacements needed to mix upward with the ambient atmosphere. Thus, stable stratification damps the vertical component of turbulence at the interface. Displacement of the ambient wind field also causes the cloud to become nearly decoupled from the ambient flow and to move downwind at a rate slower than the ambient wind speed. All of these effects are most pronounced when the ambient wind speed is low and the atmospheric conditions are stable. Figure 2 shows data from bivane anemometers both upwind and downwind of the 1980 Burro 8 liquefied natural gas (LNG) test. The wind speed recorded by the anemometer in the cloud drops to near zero when the cloud is present indicating that the cloud is displacing the ambient wind flow. Figure 3 shows vertical turbulence data from a bivane anemometer as it experienced a cold dense LNG cloud during the 1987 Falcon series tests (Brown et al., 1989). The effect of the cloud on turbulence is marked. Turbulence damping is accounted for in the various dense gas dispersion models in a number of different ways (Ermak, 1988a). Currently, FEM3 (Chan, 1988) uses a K-theory model to parameterlze turbulence which is dependent on cloud Richardson number, a

733

m0 (88~) Fig. 3. Bivane anemometer data from Falcon test series showing modification of vertical fluctuations due to presence of LNG vapor cloud. measure of the stable stratification in the cloud. As the

Fig. 2. Bivane anemometer data upwind and downwind of the Burro 8 LNG test. Significant modification of wind speed due to LNG cloud displacing atmospheric flow was observed.
AE 23:bB

cloud density approaches that of the ambient air, the Richardson number also approaches that of the ambient atmosphere. Gravity flow is driven by the excess hydrostatic pressure caused by the density difference between the cloud and the ambient atmosphere (Hunt, 1983). The fluid motion is generally horizontal except near the front where there is a recirculating vortex. Most of the mixing occurs just behind the front due to Kelvin-Helmholtz instability. For a continuous spill into a steady wind, gravity spreading also produces vortices in the crosswind direction which entrain air into the cloud at the edges. This can be seen in Figs 4 and 5. Figure 4 is a crosswind vertical cross section showing the flow field reproduced by a FEM3 computer simulation of the Burro 8 LNG test both with and without terrain compared to contours constructed from data (Chan and Ermak, 1983). Gravity flow in the crosswind direction produces the bifurcation and lobe structure. Terrain effects enhance the lobe on the left. Figure 5 shows horizontal contours for a FEM3 simulation of Burro 8, with terrain effects, compared to contours constructed from data. The gravity-flow-produced cloud bifurcation is even more apparent in this view. Instantaneous gravitydriven releases have been studied in detail both experimentally and theoretically during the Thomey Island Trials (McQuaid, 1987). Gravity flow in the near field usually increases cloud surface area and produces intense outward moving vortices, both of which tend to enhance the entrainment of air into the cloud. In the far field, gravity flow is much less significant, thus entrainment is reduced due to the low cloud profile and stable stratification, Gravity flow causes shear at the ground/cloud interface and at the air/cloud interface. In addition, for cryogenic clouds, heat transfer from the ground into the cloud, is enhanced, increasing mixing with air due to increased buoyancy. Mathematical dispersion models Over the last 10 years, many models of the dispersion of gases from accidental releases have been

134

RONALD P. KOOPMAN er ul

Crosswinddii

(m)

Fig. 4. Burro 8 crosswind contour plots of concentration 140 m downwind at t = 180 s: (a) experiment; (b) flat terrain simulation; (c) variable terrain simulation.

created.

These models fall into three categories of varying physical completeness and numerical complexity: three-dimensional conservation equation, similarity profile and Gaussian-based. There have been at least four reviews of 6 dense-gas dispersion models during this time (Ermak, 1988~ Blackmore et al., 1982; Havens, 1982; Wheatley and Webber, 1984). Much of the information presented here has been summarized from Ermak et 01. (1988a,b). The models which provide the most physically complete description of dense gas dispersion are those that are based on the three-dimensional, timedependent conservation equations. Examples of this type of model include FEM3, SIGMET, MARIAH and ZEPHYR, which have recently been evaluated by Havens et al. (1987). At the intermediate level of completeness and complexity are the similarity-profile models. These models use simplified forms of the conservation equations that are obtained by averaging the cloud properties over the crosswind plane. Quasi-three-dimensional solutions are obtained by assuming a functional form for the crosswind profile of concentration and other cloud properties. Examples of this type of model include SLAB, HEGADAS and DEGADIS. At the simpkst level are the modified Gaussian plume models. These models were designed

to simulate trace-gas releases into a normal atmosphere. They are usually used to simulate continuous releases and employ a variety of modifications which attempt to include the effects of dense gas dispersion within the Gaussian framework. Box or top-hat models, which are used to simulate instantaneous and continuous releases, fall into either the intermediate or simple category depending upon the complexity of the model regarding the number of conservation equations to be solved. The three types of models differ considerably in their approach to simulating the atmospheric dispersion of a dense gas release. Perhaps the most obvious differences are related to the degree to which each model type incorporates the basic conservation laws and three-dimensional effects. The modified Gaussian plume model is based on the single conservation of species equation and either neglects momentum and energy transfer or attempts to include them in some ad hoc manner. On the other hand, the intermediate similarity profile models include the conservation equations of mass, momentum and energy, in addition to the species equation, but only in an average way. In SLAB, variations in the properties of the vapor cloud in the crosswind plane are described by functions (Gaussian, for example) and treated by the code as

Review of recent field tests 200.

735

I Cl Expuinmnt

I
/ ,

releases under low wind speed, stable, ambient conditions (Figs 4 and 5), (3) cloud deflection caused by sloping terrain. FEM3 is the only known 3D conservation equation model with an optional submodel for phase change. Two versions of this submodel have been developed. The first version (Leone et al., 1985) is for atmospheric water (ambient humidity-fog) and the second (Rodean et al., 1986) is generalized so it can be used for a variety of hazardous liquids. It is assumed in both versions that the mixture components (air in the gas phase, liquid material and material vapor) are always in thermodynamic equilibrium. Therefore, there is never any supercooled vapor or superheated liquid, and there is no thermal lag in the system. In principle, three-dimensional models are capable of modeling complex time-dependent phenomena and taking into account complex boundary conditions imposed by terrain and structures. While these models generally provide the most detailed and complete description of heavy-gas flow, there are limitations to their use, and often simplifications and compromises are made due to the complexity of these models. Wheatley and Webber (1984) discussed several of these aspects including the following.
400

-2001

200

I I I (b) FEM3 (variable terrain)

loo 200 300 Downwind distance(m)

Fig. 5. Burro 8 contour plots of gas concentration at 1 m above ground at 180s. FEM3 compared to experiment (dashed lines show edge of instrument array).

average values which vary in the downwind direction only. The conservation equation models, such as FEM3, include the most complete description of the conservation laws by treating them explicitly in three dimensions. Both the three-dimensional conservation equation models and the intermediate similarity profile models incorporate mathematical descriptions of these physics that cause the major heavy gas dispersion effects. Of the two types, the three-dimensional conservation equation models provide the more detailed and complete description of the physics involved in dense gas flows. An example of this type of model is the FEM3 model. FEM3 simulates the dispersion of a released gas by solving the time-dependent, three-dimensional, conservation equations of mass, momentum, energy and species along with the ideal gas law for the equation of state. In addition, it can treat flow over variable terrain and around obstructions such as cylinders and cubes. Turbulence is treated by using a K-theory submodel. Since it is fully three-dimensional, FEM3 can simulate complicated cloud structures such as:

crosswind

(1) The modeling of turbulence is possible only by making assumptions and approximations. The necessity of using some form of turbulence closure in order to solve the system of equations for the mean fields of velocity, temperature, and concentration is the weakest part of 3D models. (2) Other approximations (e.g. hydrostatic, Boussinesq and anelastic) are made in order to simplify the computations and reduce their cost. (3) The numerical solutions necessarily involve discrete approximations (finite-difference or finiteelement) to continuous functions. The design of a three-dimensional mesh for a numerical simulation involves compromises among several parameters including the largest and smallest scales of the phenomena and environment to be represented, the capacity of the computer, and the cost of machine time. (4) As indicated above in (a)-(c), these models are expensive and time-consuming to develop and use. An example of an intermediate similarity profile model is Steady State SLAB. The SLAB model solves the crosswind averaged equations for the conservation of mass, species, downwind and horizontal crosswind momenta and energy along with an additional equation for cloud width and the ideal gas law equation of state. The current version of SLAB includes the steady-state assumption for continuous releases, but a puff version (finite duration release) is under development. Thus, the code is one-dimensional with downwind distance being the independent variable; however, since cloud width and cloud height are also

(1) the vortices that are typical of dense gas flows, (2) cloud bifurcation observed during heavy gas

736

RONU D P. KOOPMAN CI al

calculated, the model is, in this sense, quasi-threedimensional. The crosswind concentration distribution is determined by using similarity profiles based upon the calculated crosswind height and width. Mixing of the cloud with the ambient atmosphere is treated by using the entrainment concept. The main advantage of the SLAB code over a 3-D code is its low computing cost. Typical simulations require only a few seconds on a CDC 7600 computer or a few minutes on an IBM micro-computer. A unique feature of the SLAB model is that it calculates only crosswind-averaged properties, and characterizes the cloud shape by the height, h, and half-width, B. The parameters B and h do not correspond to any particular concentration level. Rather, they can be considered to describe a surface which encloses the bulk of the cloud, for example 90%. Consequently, the crosswind concentration distribution is not uniquely defined. This causes difficulties in attempting to compare the predicted cloud shape from this model with contour plots obtained from experiments. To overcome this difficulty, one generally defines B and h in terms of an assumed distribution (such as Gaussian, exponential or quadratic) for the crosswind horizontal and vertical vapor cloud concentration. There are other important differences and these are related to the manner in which each model type treats the effects of gravity and turbulence. As indicated above, the modified Gaussian plume models use ad hoc formulas with empirical coefficients to describe the gravity spread and turbulent dispersion of the cloud. In contrast to this, the SLAB and FEM3 models use conservation principles to treat the effects of gravity. This is done in the FEM3 model by solving the three momentum conservation equations, including the buoyancy term and variable density, while the SLAB model solves two layer-averaged momentum equations and uses the hydrostatic approximation. These two models differ considerably in their approach to turbulence. The SLAB model uses the somewhat artificial concept of entrainment across the cloud-air interface and essentially neglects any explicit treatment of turbulence within the vapor cloud. Air is entrained into the cloud at the surface and then is assumed to mix rapidly in the cloud creating a nearly uniform layer in the crosswind plane. Thus, there are two separate regions: the cloud and the ambient atmosphere. Mixing between the two is assumed to occur at the interface and is governed by an entrainment velocity which depends on the local properties of both the cloud and the surrounding atmosphere. The FEM3 model assumes that turbulence can be described as a diffusion process and uses a continuous diffusion coefficient which depends on the local properties of the dense gas cloud. While the entrainment and diffusion concepts are peculiar to the SLAB and FEM3 models, respectively, the choice of a particular entrainment or diffusion submodel is not an essential aspect of the models. Several submodels have

been proposed in the literature including higher order turbulence submodels and these could be used without changing the whole model. This occurred recently when FEM3 was used to model the Thorney Island Phase 1 trials (Chan er ul.. 1987a). which revealed that the dense gas Richardson number determined in a previous K-theory turbulence submodel was not adequate for regions where flow velocities were significantly perturbed, either by the presence of obstacles or a heavy-gas cloud. A new K-theory model was developed based on the gradient Richardson number concept.

with the assumptions

P-P~=(P-P.)I----xpC-(zlhcrl
u = ? In(z/z,
k

).

This resulted in a local Richardson number defined


by

&u2--!ia.

l (u:+w:)

+nk2(z/h

where u+ = friction velocity = u,+, where a designates ambient, w2 = in cloud convection velocity (Chan et al., 1987a), k = von Karmans constant = 0.4, z = height above ground, h, =characteristic cloud height, p = density, Ri, = ambient Richardson number = z/L, g = acceleration of gravity, n =2 based on comparison to experiments (McQuaid, 1976). z0 = surface roughness, L = Monin-Obukhov length scale. Thus the Richardson number vertical profile has a maximum value near the top of the cloud where the density gradient is largest and approaches zero at ground level. As cloud density approaches ambient density, the ambient value of Richardson number is recovered. Although this new K-theory model performs much better, a higher-order turbulence model will probably be necessary for adequate general treatment of complex situations. Finally, a three-dimensional, time-dependent, conservation equation model is required in order to describe the distribution of gas concentration in space and time from a heavy gas release into an atmospheric boundary layer with speed and directional wind shear in the presence of complex terrain and man-made structures. This versatility of the three-dimensional conservation equation model in treating more realistic situations and providing a more detailed description of the flow is somewhat balanced by the increased

Review of recent field tests

737

computer costs in running these models. Conversely, similarity models, while giving up some degree of realism and detail, are much faster to run on computers, Consequently, many researchers tend to view the three-dimensional conservation equation models more as research tools that help them discover new things about the flow and to view the intermediate, or similarity models, as more of an operational model for situations where computing costs and time are of the essence. The 3-D models are also very useful for doing assessments at fixed sites where dense gas and terrain or structure effects are im~rtant.

FIELD EXPERIMENTS AND COMPARISON TO MODELS

Field experiments in dense gas dispersion have been performed since the early 1970s. Initial experiments were mostly with LNG with some early work also with LPG and NH,. The quantity and quality of data obtained from these experiments has varied greatly. Most of the good quality data have been obtained since 1980 when, essentially simultaneously, Shell/ NMI conducted a series of LNG and LPG trials at Maplin Sands in England and DOE/LLNL conducted the Burro series of LNG tests at China Lake, CA. There have been several recent reviews of field programs (Koopman, 1987; McQuaid, 1983a, 1983b; Puttock and Colenbrander, 1985; Vilain, 1986; Puttock et a/., 1982). There are a number of reasons for doing field-scale

dispersion model evaluation. This requires extensive, carefully verified, quantitative data from well documented and well instrumented experiments. The need for such data was demonstrated by McQuaid (1976) of the British Health and Safety Executive (HSE), when he invited the mathemati~l modeling immunity to predict in advance the results of an instantaneous release of 2000 m3 of gas with an initial density 2 times that of air, Variations between models of two orders of magnitude were present even for this relatively small, isothermal release onto flat terrain without chemical reactions, the~~yn~ic effects or other complications. Models have improved since that time, but the tasks they are being given are much more difficult and include many more complicated effects. Thus, model evaluation continues to be one of the major reasons for the conduct of field experiments. In spite of the number of tests which have been conducted, data for model evaluation are still in short supply. Table 1 gives an edited summary of recent large scale dispersion tests. The emphasis is on the observation of dense gas effects as defined in this paper. Consequently, small tests which do not show dense gas behavior clearly and tests which were not well instrumented have been left off. Each major set of field experiments has had a particular emphasis and particular goals. This section provides a brief summary of the major achievements of each of the test series listed in Table 1 and selected examples of model comparisons to the data.

experiments. The relative importance of the many effects which occur upon accidental release is often
unknown. Much of the physics dominant in dense gas dispersion is highly nonlinear and may not even be observable in small scale tests in the atmosphere. Stable atmospheric conditions allow dense gas effects such as turbulence damping and gravity spreading to dominate cloud behavior. Nonisothermal effects (i.e. heat transfer, phase-change, etc.) can also be important and are not easily scaled. Thus, one of the major reasons for performing these tests is the determination of the important effects and their scaling laws. In addition, the well instrumented tests allow quantitative measurement of these effects so that correct descriptions can be incorporated into the mathematical dispersion models. This is necessary if accurate predictions are to be made for circumstances different than those under which tests were performed. Other reasons for conducting field tests include accident simulation or evaluation of mitigation equipment such as water or steam curtains. These may be situations which are simply too complicated, with too many unknown contributions, to yield to mathematical or physical model simulation. Other complicating effects which are difficult to model or scale include chemical reactions and certain irreversible thermodynamic effects such as flashing two-phase flow. Perhaps the most common reason for the conduct of field experiments has been to obtain basic data for

The Maplin trials were sponsored by Shell and conducted at Maplin Sands, England, by the National Maritime Institute (NMI) (Puttock et al., 1982). This was a very ambitious well-instrumented experimental program involving some 20 spills of 5-20 m3 of LNG and 14 spills of 13-31 m3 of propane onto water. Some of these trials were instantaneous releases from a submersible barge, some were continuous and some were ignited. Good data on dispersion over water under various meteorological conditions, flame propagation in un~on~ned gas clouds, heat transfer and humidity effects were obtained. Large-scale rapid phase transition (RPT) explosions occurred on one of these tests.
Burro

The Burro series of tests were sponsored by the U.S. DOE and conducted by LLNL and Naval Weapons Center (NWC) personnel at China Lake, California, (Koopman, 1982a,b). This series involved eight large LNG releases onto water, conducted under a variety of meteorological conditions. Good quality data from a large array of instruments were obtained. Very revealing dense gas e&c&, showing that the cloud caused displacement of the ambient atmospheric flow and modification of turbulence, were observed and measured on the Burro 8 test. An example of this effect is shown in Fig. 2. Good data on dispersion over land

738

under a variety of met~#rolo~ca~ conditions and data on heat transfer, turbulence, and humidity effects were obtained. Large-scale RPT explosions with chamcteristics unobserved before occurred on one of these tests.

The Coyote test series was a continuation of the Burro series into the areas of &ombustion and RPTs (Goldwire et al., 1983; Rodean et al., 1984; Morgan et al., 1984; McRae, 1984b). Five combustion tests were conducted involving 8-28 m3 of LNG spilled under a variety of meteoro~o~~a~ conditions with hardened instrumentation so that dispersion data, prior to ignition, could also be obtained. Flame propagation was measured for free-doud combustion with both soft (flare) and high-vel~ity (Same jet) ignition sources. Thirteen spills of 3-14 m3 with varying compositions were performed to get basic data on the new type of RPT observed during the Burro tests.

The Thorney Island trials were organized by the British HSE and conducted at Thorney Island (U.K.) by NM1 (M&&aid, 1987). They invotved large instantaneous isothermal dense gas (Freon+N2) reieases from a cohapsibie tent-like structure. Although the quantity of gas released was relatively small (2000 m3) compared with most of the other test series (23,000 m3 of gas for Burro) the release was instantaneous, and the density was high, producing large Richardson numbers for some of the trials. These Wials had a strong experimental emphasis on creation of a we&defined model source and gravity driven flow. In most cases gravity flow dominated cloud behavior over the entire measurement array, The triafs were well instrumented, and high quality data were obtained. The first series (Phase If consisted of f 6 releases over unobstructed terrain, with gas densities varying between neutral (air) and 4 times air, with most done at 2 times air. The second series (Phase II) consisted of ten releases into an obstacle field featuring an impermeable wail, a permeable screen simulating fohage, and a cube simulating a building. A third series (Phase III) consisted of 17 continuous releases, some 14 of which were for the U.S. Department of Transportation and involved releases into a vapor fence containment structure. FEM3 was used to model Phase I trials 913 and 17 (Chan, 1987a). Simuiat~on of the dispersion was performed in two stages. Initiafly, the flow field over and around the source was calculated. This is shown in Fig. 6 for Trial No. 9. A vertical plane of symmetry is assumed and only half of the structure and flow field is shown in Fig. 6(b). Then the constraints were removed and the gas cloud was allowed to slump and mix with the complex ambient air Row caused by the structure. Figure 7 shows the resulting gas concentration contours and velocity vectors at 4 s after the release. To demonstrate the effects of the different K-theory (Richardson number) models considered, Trial No. 13

Review of recent field tests

(b) lWkontalpb~atz=O.4m

-40

-20

20

40

60

100

200

300

400

M)

DownwInddbtubm (m) Fig. 6. FEM3 velocity field prior to Thomey Island Trial No. 9.

Downwind diatma (m)

Fig. 8. Peak concentrations from Thomey Island Trial No. 13 compared to FEM3 with various turbulence models.

ICI nmst4s

Desert Tortoise

40

-20

20

40

DownwInd dhtanw (m)

Fig. 7. FEM3 predicted concentration contours and velocities along the vertical symmetry plane for Thomey Island Trial No. 9 at 4 s after release.

was simulated with the previous model (Ri peaks at ground level), with ambient turbulence only (Ri = Ri, = Z/L where L = ambient Monin-Gbukhov length scale) and with the present model where Ri peaks near the top of the cloud. The result is shown in Fig. 8. Without the density stratification term to damp out turbulence in the cloud, the ambient K model underpredicted the peak concentrations by a factor of nearly 2 for the entire curve. The previous K model improved agreement close to the source but actually caused poorer agreement beyond about 1OOm downwind. The present K model produced the best agreement, but further improvement is needed to better simulate the early phases of the release. This K model still does not adequately account for significantly perturbed velocities due to the presence of obstacles or heavy gas clouds.

These large-scale NH, releases (Goldwire, 1985, 1986) were performed by LLNL at the Nevada Test Site under the sponsorship of the U.S. Coast Guard and the The Fertilizer Institute (TFI). The goal of the tests was to obtain basic dispersion data for two-phase (flashing) releases of pressurized ambient temperature ammonia under simulated worst case accident conditions. The instrument array is shown in Fig. 9. Detailed measurements of cloud shape, composition and aerosol density were made at 100 m and 800 m downwind (measurements at 1,3,8 m heights), with an arc of single level (1 m) measurements of gas concentration at 2800 m downwind. Four tests under Pasquill-Gifford category D and E stability conditions were performed with good quality data obtained on all of them. Aerosol effects dominated cloud behavior on these tests with approximately 80% of the mass released being in the form of an aerosol, most of which was transported downwind. This created strong dense gas effects which persisted for long distances downwind. The latest version of FEM3, with improved turbulence and phase change models, was used to simulate the fourth Desert Tortoise test, a release of 60 m3 in the form of a strong horizontal jet pointing downwind. Since a large pool of NH, was observed immediately after the test and since mass flux measurements at 100 m and 800 m downwind recorded only 70% of the mass released in the main cloud, the source strength was reduced to 70% of the amount released. Figure 10 shows FEMZpredicted cloud centerline temperature compared to measurements closest to the center. Since the current phase-change model does not include

740
N

RONALD P. KOOPMAN et al

225

: I

Dispwaion

array -

. Gas sensor station A Anmwmotw station

Fig. 9. Diagnostic instrument array for Desert Tortoise and Eagle series experiments.

40

M-

O
-20 -

-80 0

1. 200

I 400

I 6a

I a00

1000

Fig. 10. Temperature along downwind distance at an elevation of 1 m for Ammonia Spill Experiment No. 4 at t = 300 s. The experimental data are measurements closest to the center of the vapor cloud. The FEM3 curve is based on predicted values along the cloud centerline.

Review of recent field tests

741

water condensation and reaction with the NH, aerosol, the aerosol evaporates too fast causing the predicted temperature to be too low at IOOm. Thii also results in too little aerosol being present at 800 m in the model cloud and a predicted temperature which is too high. Figure 11 shows the predicted cloud centerline gas concentration from both FEM3 and a Gaussian plume model compared to the peak gas concentration data from the various arcs of gas sensons. FEM3 concentrations are too high at 100 m and too low at 800 m because not including reactions with water vapor causes the aerosol to evaporate too fast. The high-speed jet source, which was not modeled by FEM3, may also account for some of the discrepanties. The Gaussian prediction is high by nearly a factor

of ten at 100 m and iow by a factor of ten at 2800 m. Presumably these differences are due mainly to the absence of aerosol and dense gas effects in the Gaussian plume model. Differences in concentration averaging times between the data and the Gaussian model may also account for some of this difference. The aerosol evaporated within a few hundred meters, but dense gas effects persisted and the cloud remained cold and compact. gqfe The Eagle series of N,O, experiments (I&Rae, 1984a, 1985), was performed by LLNL under the sponsorship of the U.S. Air Force for the purpose of determining source. conditions for confined and un-

loo1

10%

111

0.11

Fig. 11. Comparison of measured peak ammonia concentrations as a function of downwind distance at an elevation of 1 m for Desert Tortoise Test 4 with FEM3 and Gaussian plume calculations.

142

RONALD P. KOOPMAN et al.

confined spills, making basic dispersion measurements, and providing emergency response training and equipment evaluation. The instrumentation array was similar to Desert Tortoise, Fig. 9. In addition, extensive reaction of NO* with atmospheric water vapor caused approximately half of the cloud to be transported downwind as an HNO, mist which damaged gas sensors and was not detectable. The net result is that maximum gas concentrations had to be estimated from analysis of these data. Good data on cloud width and height at 800 m downwind were obtained which show significant dense gas effects and have been useful for model data comparison. Data from two of these tests are summarized in Table 2 where they are also compared with simple Gaussian-type dispersion models in use by the USAF at the time of the tests. The data indicate that gas concentrations 5 to 10 times higher than those predicted by the simple models were measured, that measured cloud heights were l/5 to 1/lO those predicted by the models, and that measured cloud widths were about half those predicted. Thus, the cloud remained low and compact due to its high density.
Goldjish

One of the major findings of the test series was that all of the material released, both vapor and liquid aerosol, was transported downwind as a heavy cloud. Simulations of these tests were performed using the FEM3, (Chan, 1987b) SLAB and DEGADIS (Blewitt et al., 1987~) models. Figure 12 shows SLAB calculations of the first three tests compared to peak concentration data. The major differences between the.tests were spill rate, which varied from 1.8 m3 min- on Test 1 to 0.6 m3 min _ on Tests 2 and 3. The measured and SLAB-predicted plume crosswind concentration profiles at 1000 m downwind are compared in Fig. 13 for Test 1.
Falcon

The Goldfish test series was performed in 1986 by LLNL and Amoco Oil Co. under Amoco sponsorship (Blewitt et al., 1987a,b,c; Chan, 1987b). A series of six tests, four with releases of about 4 m3 (1000 gal) and two of about 2 m3 (500 gal) of HF were performed at various spill rates and with varying meteorological conditions. The first three tests were designed to obtain data on source characteristics and dispersion and the last three were to evaluate the effectiveness of various water spray curtains. An extensive instrument array was fielded with rows of gas sensors at 300 m, 1OOOm and 3000m. Prior to the tests, there was considerable uncertainty about how much HF aerosol would form and be transported downwind for HF released suddenly through an orifice at 40C and 115 psig. Under these conditions, approximately 20% was expected to flash to vapor, and the remaining 80% to remain as liquid droplets, either raining out on the ground or being transported downwind as an aerosol.

The Falcon test series was performed by LLNL for the Gas Research Institute and the U.S. Department of Transportation in 1987 (Brown et al., 1988). This test series involved five large-scale (20-66 m3) releases of liquefied natural gas (LNG) into a 44 m x 88 m x 10 m-tall vapor containment fence. The tests were a full-scale follow-on to the Thorney Island Phase III tests and were designed to provide the detailed data necessary to validate wind tunnel and mathematical models of dispersion from vapor fences and evaluate their effectiveness. Data reduction and reporting are currently underway for these tests with analysis using FEM3 to begin soon.
Model evaluation

Model evaluation, always an important part of dispersion modeling, has received more attention in the past few years as industries and regulating agencies begin to rely on models for risk assessment, siting, emergency planning and other regulatory decisions. This has led to an increasing need to know how accurate models really are and what are the limits of their applicability. Various approaches have been developed to air quality model evaluation in recent years, but the emphasis has been on statistical methods (Fox, 1981, 1983). This approach has recently come under criticism by several authors (Venkatram, 1986; Dennis, 1985) because it does not promote understanding of

Table 2. Comparison of eagle test results at 800 m downwind with simple dispersion model predictions
Eagle

3 Peak concentration (ppm)

Eagle 6

Peak concentration (ppm) Test results Recorded Estimated


Model predictions

(3)

(2)

(2)

(G)

>500 2275 165 163 199 220

35

3.8

315 575 73 89 112 127

35

1.6

Ocean breeze/dry gulch Gaussian plume Shell Charm (McRae, 1985)

_ 60 56 60

_ 32 27 24

60 56 59

-. 32 27 23

Review of recent field tests

143

TEST1

TEiT 2

TEST 3

Fig. 12. Comparison of observed and SLAB-predicted plume centerline concentrations at a height of 1 m for the Goldfish HF test series.

(2) Examination of model structure including the accuracy of the mathematical framework, the realism of the model representation of important physical processes, and the appropriateness of assumptions used in the model when applied to real situations. (3) Sensitivity analysis of the model to uncertainties in the input meteorological and source data. (4) Testing of the model predictions against observations including both laboratory and field-scale experiments. Since dense gas dispersion is a relatively new field, little had been done to develop a methodology for model evaluation until recently. Mercer (1986) conducted a review of the methods used to validate dense gas dispersion models and concluded that there have been few, if any, attempts to objectively compare a number of dispersion models using a common set of criteria. Authors have often verified their own models and in essentially all cases made only subjective, qualitative assessments on how well the model prediction compares with the data. Sufficient good quality data from a variety of dense gas releases now exist (Ermak et al., 1987) such that systematic model evaluation is possible. A methodology for evaluating dense gas dispersion models has recently been developed (Ermak et al., 1988b) which relies heavily on physical understanding of the processes characteristic of dense gas releases in the atmosphere. Dense gases act as intrusions, with different physical properties, which significantly displace and alter ambient atmospheric flow. Dense gas dispersion is dominated by effects which are not observed in the dispersion of trace emissions; there is a reduction in turbulent mixing due to stable stratification of the dense layer and gravity flow keeps this dense layer low and wide. It is important that the tests and comparisons which make up the evaluation address those

Distance (m)

Fig. 13. Comparison of observed and SLAB predicted plume crosswind concentrations for Goldfish HF Test 1 (1 km downwind and 1 m height).

the underlying physical processes and often the statistics generated are meaningless. The importance of developing a physical understanding of a models performance has been well-recognized within the atmospheric sciences community (Knox and Walton, 1984; McNaughton et al., 1986). In order to gain this insight, the evaluation must be tailored to the specific model under investigation, the available experimental data, and the specific application for which the model is to be used. The steps usually followed in this process are given below. (1) Evaluation of the accuracy of the input meteorological data and the gas concentration data used for comparison.

744

RONALD P. KOOPMA~ et ul.

processes which are dominant for dense gas releases. Based on analysis of these processes, four key plume parameters emerge for comparison. (1) Maximum gas concentration as a function of downwind distance from the source. (2) Average ground-level plume centerline gas concentration as a function of downwind distance. (3) Plume half-width. (4) Plume height. The piume height and width comparisons allow assessment of the combined effects of gravity flow and mixing with the atmosphere. The recommended comparison technique is based on developing ratios between the four model-predicted parameters and the experimentally-observed values. Ratios allow comparison over a wide range of values from high concentrations where gravity effects dominate to trace levels, where toxicity is still an issue. Ratio methods also aliow determination of the bounds of model error and the limits of model appli~bility. In addition, the characteristic averaging time of the model must match that of the data. The model averaging time must be known and if possible adjusted to match that of the sensors, Ifmodel adjustments cannot be made, then the data must be averaged to match the model. A maximum concentration has meaning only if referred back to the time constant characteristic of either the measurement or the model.

and the 1D cross-wind averaged SLAB model have been improved with the addition of phase-change submodels and evaluated against dense-gas data sets. Sufficient good quality dense-gas data now exist such that the evaluation of dense-gas models can be done in a systematic, quantitative and scientific fashion. A methodology teristics of dense-gas systematic evaluation based on the physical ckaracreleases has been developed. A

of all models in common use has not yet been performed.


Acknowledgement -Work performed under the auspices of the U.S. Department of Energy by the Lawrence Livermore National Laboratory under contract No. W-7405-ENG-48.

Disclaimer--This document was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor the University of California nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commer~i~ products, process, or service by trade name, trademark, manufacturer, or otherwise, does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or the Universitv ofcalifornia. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government thereof, and shall not be used for advertising or product

endorsement purposes.

REFERENCES
SUMMARY

Certain key dense-gas effects, such as stable stratification and gravity-driven flow have been identified as the factors making dense gases behave differently when released into the atmosphere. Dense-gas ROWS displace the ambient atmospheric flow and modify ambient turbulent mixing. Chemical and physical processes, such as flashing to aerosol and vapor, which occur for many commonly used hazardous chemicals contribute strongly to dense-gas effects by creating heavy, cold, liquid- laden clouds which behave very differently from trace emissions of the same substances. Chemical and thermodynamic effects are unique to each material and can have a significant effect on dispersion distance. Gaussian and other trace-gas based models were intended to describe the dispersion of a tracer in a normal atmosphere and are not adequate for describing dense-gas releases. Comparison to experimental results for several materials show that errors of up to a factor of ten can occur if trace gas models are used for dense-gas problems. Two different approaches to dense-gas modding have produced models capable of predicting cloud behavior in three-dimensional, time-dependent detail in complex terrain or averaged over both time and space. The 3D conservation equation model FEM3

Blackmore D. R., Herman M. N. and Woodward J. L. (1982) Heavy gas dispersion models. J. Haz. Mat. 6, 106-128. Blewitt D. N., Yohn J. F., Koopman R. P. and Brown T. C. (1987a) Conduct of anhydrous hydrofluoric acid spill experiments. AIChE lnternational Conference on Vapor Cloud Modeling Boston, Massachusetts. Blewitt D. N., Yohn J. F., Koopman R. P., Brown T. C. and Hague W. J. (1987b) Effectiveness of water sprays on mitigating anhydrous hydrofluoric acid releases. AIChE International Conference on Vapor Cloud Modeling, Boston, Massachusetts. Blewitt D. N.. Yohn J. F. and Ermak D. L. (1987~) An evaluation of SLAB and DEGADIS heavy gasdiskion models using the HF spill test data. AIChE International Conference on Vapor Cloud Modeling, Boston, Massachusetts. Brown T. C., Cederwall R. T., Ermak D. L., Koopman R. P., McClure J. W. and Morris L. K. (19891F&on Series Data Report, 1987 LNG Vapor Barrier Verffication Field Trials, UCID- , Lawrence Livermore National Laboratory, Livermore, Caliiomia (in preparation). Ghan S. T. (1988) FEM3A-A Fin&e Element Model/or the Simulation qfGps Transport und Dispersion: Users Manual. UCitL-21043. Lawrence Livermore Nationat Laboratory, Livermore, California. Chart S. T. and Ermak D. L. (1983) Recent progress in modeling the atmospheric dirpersion of heavy gases over variable terrain using the threedimensional conservation equations. Proceedings of the IUTAM Symposbn on Atmospheric Dispersion of Heavy Gases and Srrcall Particles, Ddft Univ. of Technology, The Hague, The Netherlands.

Chart S. T, Ermak D. L. and Morris L. K. (1987a) FEM3 model simulations of sekctsd Thomey I&and Phase I Trials. J. Huz. Mat. 16.

Review of recent field tests Chan S. T., Rodean H. C. and Blewitt D. N. (1987b) FEM3 modeling of ammonia and hydrofluoric acid dispersion. AIChE International Conference on Vapor Cloud Modeling, Boston, Massachusetts. Crowe C. T. and Comfort W. J. III (1978) Atomization methods in single-component two-phase flows. 1st International Conference on Liquid Atomization and Spray Systems, 45-50. Dennis R. L. (1985) Issues, design and interpretation of performance evaluations: ensuring the Emperor has clothes. In Air Pollution Modeling and Its Application (edited by C. De Wispelaere, F. A. Schiermeier and N. V. Gillani), pp. 41 l-424: Plenum Press, New York. Ermak D. L.. R.. Goldwire H. C., Jr. Gouveia F. J. , Chaoman . and Rodean H. C. (1987) Heavy Gas Dispersion Test Summary Report, UCID-21102, Lawrence Livermore National Laboratory, Livermore, California. Ermak D. L. and Merry M. H. (1988b) A Methodology for Evaluating Heavy Gas Dispersion Models, UCRL-21025, Lawrence Livermore National Laboratory, Livermore, California. Ermak D. L., Rodean H. C., Lange R. and Chan S. T. (1988a)
A Survey of Denser-than-Air Atmospheric Dispersion Mo-

745

Koopman R. P. (1987) Atmospheric dispersion of large-scale spills. Proceedings from the AICHE Symposium Series, 1986, Cryo. Prop., Processes Appl. Vol. 82, No. 251, pp. 141-159. Leone J. M., Jr., Rodean H. C. and Chan S. T. (1985) FEM3 Phase Change Model, UCID-20353, Lawrence Livermore National Laboratory, Livermore, California. McNaughton D. J., Atwater M. A., Bodner P. M. and Worley G. G. (1986) Evaluation and Assessment of Models for Emergency Response Planning, TRC Environmental Consultants, Inc., TRC Project No. 3088-R31, report prepared for the Chemical Manufacturers Association. McQuaid J. (1976) Some experiments on the structure of stably stratified shear flows. Technical Paper, p. 21, Safety in Mines Research Establishment, Sheffield, U.K. McQuaid J. (1983a) Large-scale experiments on the dispersion of heavy gas clouds. Presented at the IUTAM Symposium, Delft. McQuaid J. (1983b) Observations on the current status of field experimentation of heavy gas dispersion. Presented at the IUTAM Symposium, Delft. McQuaid I. (1987) Proceedings of the Symposium on Heavy
Gas Dispersion Trials at Thorney Island (J. Haz. Mat. 16).

dels, UCRL-21024, Lawrence Livermore National Laboratory, Livermore, California. Fox D. G. (1981) Judging air quality model performance, a summary of the AMS workshop on dispersion model performance, Woods Hole, Massachusetts, 8-11 September 1980. Bull. Am. met. Sot. 62(5), 599609. Fox D. G. (1983) Uncertainty in air quality modeling. Bull
Am. met. Sot. 65(l), 27-36.

McRae T. G., Cederwall R. T., Goldwire H. C., Jr., Hipple D. L., Johnson G. W., Koopman R. P., McClure J. W. and Morris L. K. (1984a) Eagle Series Data Report: 1983 Nitrogen Tetroxide Spills, UCID-20063, Lawrence Livermore National Laboratory, Livermore, California. McRae T. G., Goldwire H. C., Jr. and Koopman R. P. (1984b)
Analysis of Large-Scale LNG/ Water RPT Explosions, UCRL-91832, Lawrence Livermore National Laboratory,

Goldwire H. C., Jr., Rodean H. C., Cederwall R. T., Kansa D. J., Koopman R. P., McClure J. W., McRae T. G., Morris L. K., Kamppinen L., Kiefer R. D., Urtiew P. A. and Lind C. D. (1983) Coyote Series Data Report,
LLNLINWC 1981 LNG Spill Tests Dispersion, Vapor Burn, and Rapid-Phase Transition, Vols. land Z,UdD19953 Lawrence Livermore National Laboratory, Liver-

Livermore, California. McRae T. G. (1985) Analysis and Model/Data Comparisons of


Large-Scale Releases of Nitrogen Tetroxide, UCID-20388,

more, California. Goldwire H. C., Jr. (1985) Desert Tortoise Series Data Report: 1983 Pressurized Ammonia Spills, UCID-20526, Lawrence Livermore. National Laboratory, Livermore, California. Goldwire H. C.. Jr. (1986) Lame-scale ammonia soil1 tests. In Chemical Engineering Progiess, pp. 3541. _ Havens J. A. (1982) A review of mathematical models for prediction of heavy gas atmospheric dispersion. J. Chem.
E. Symposium Series No. 71.

Lawrence Livermore National Laboratory, Livermore, California. Mercer A. (1986) Methods of Validating Models of DenseGas Dispersion: A Review. Health and Safety Executive, Research and Laboratory Services Division, Safety Engineering Laboratory, Sheffield, U.K. Morgan D. L., Jr., Morris L. K., Chan S. T., Ermak D. L., McRae T. G., Cederwall R. T., Koopman R. P., Goldwire H. C.. Jr.. McClure J. W. and Hoaan W. J. (1984) Phenomenology and Modeling of Liquefied Natural Gas Vapor Dispersion, UCRL-53581, Lawrence Livermore National

Havens J. A., Spicer T. 0. and Schreurs P. J. (1987) Eoaluation


of 3-D hydrodynamic computer models for prediction of LNG vapor dispersion in the atmosphere. Final Report

Laboratory, Livermore, California. Puttock J. S., Blackmore D. R. and Colenbrander G. W. (1982) Field experiments on dense gas dispersion. J. Haz.
Mat. 6, 1341.

GRI-87-0173. Hunt J. C. R., Rothman J. W. and Britter R. E. (1983) Some physical processes involved in the dispersion of dense gases. IUTAM Symposium on Atmospheric Dispersion of Heavv Gases and Small Particles. Delft. The Netherlands. Knox J: B. and Walton J. J. (1984) Modeling regional to global air pollution. In Atmospheric Science and Power Production (edited by Randerson D.). Technical Information Center, Office of Scientific and Technical Information, U.S. Dept. of Energy. Koopman R. P., Cederwall R. T., Ermak D. L., Goldwire H.-C., Jr., Hogan W. J., McClure J. W., Mckae T. G., Moraan D. L.. Rodean H. C. and Shinn J. H. (1982a) Analysis of Burro series 40-m3 LNG spill experiments. JI Haz. Mat. 6, 43-83, also published as Dense Gas Dispersion. Elsevier Scientific, Amsterdam. Koopman R. P., Baker J., Cederwall R. T., Goldwire H. C., Jr., Hogan W. J., Kamppinen L. M., Kiefer R. D., McClure J. W., McRae T. G., Morgan D. L., Morris L. K., Spann M. W., Jr. and Lind C. D. (1982b) Burro Series Data Report LLNL/NWC 1980 LNG Spill Tests, UCID-19075, Lawrence Livermore National Laboratory, Livermore, California.

Puttock J. S. and Colenbrander G. W. (1985) Dense-gas dispersion-experimental research. Presented at the Heavy Gas Workshop, Toronto. Rodean H. C., Hogan W. J., Urtiew P. A., Goldwire H. C., Jr., McRae T. G. and Morgan D. L., Jr. (1984) Vapor Burn Analysis for the Coyote Series LNG Spill Experiments, UCRL-53530, Lawrence Livermore National Laboratory, Livermore, California. Rodean H. C. and Chan S. T. (1986) Generalized PhaseChange Submodels in the FEM3 Model for Gas Dispersion, UCID-20817, Lawrence Livermore National Laboratory,

Livermore, California. Venkatram A. (1986) Model Evaluation. In A Short Course on Air Pollution Modeling. American Meteorological Society, Boston, Massachusetts. Vilain J. (1986) Two-phase flows in major technological hazards. Presented at the von Karman Institute for Fluid Dynamics Lecture Series. Wheatley C. J. and Webber D. M. (1984) Aspects of the
Dispersion of Denser-Than-Air Vapours Relevant to Gas Cloud Explosions, Contract Report SR/OO7/80/UK/H/

EAEC/UKAEA No. X11/829/84-EN.

You might also like