Click for next page ( 34


The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 33
4 Dispersion Modeling: Application to C/B/N Releases The dispersion of an effluent plume in the atmosphere is the result of transport by the wind field and distortion and mixing by turbulence. Figure 4.1a, a snapshot of a plume downwind of a continuous point source in a turbulent flow illustrates these turbulence effects. Before the availability of modern computers, treatments of atmospheric dispersion focused on a time-average plume that varies smoothly in space, as illustrated in Figure 4.1b. In a flow where the time-averaged velocity and the turbulence properties are spatially uniform, this plume has a Gaussian concentration profile, and the downwind evolution of the plume width is related to statistical parameters of the turbulence. These concepts are the basis of the Gaussian-plume models that have long been used to predict dispersion from continuous point sources in air quality applications. Today's computa- tional fluid flow models can use a numerical grid with several hundred points in each of the three coordinate directions, and numerical techniques allow tailored grids that are finer near the source and coarser farther downwind. Such computational advances have led to a proliferation of the number and types of atmospheric dispersion models. ~ Gaussian-plume models assume that the concentration of the agent downwind of the source (averaged over a large number of realizations of the given dispersion problem) has the form of the Gaussian, or "normal", probability distribution in the vertical and lateral directions. The amplitude and width of this "bell curve" are determined analytically by the rate of emission, mean wind speed and direction, atmospheric stability, release height, and distance from the release. Such models assume continuous and constant emission of agent, and they also generally assume flat terrain, no chemical reactions or absorption, and constant mean wind speed and direction with time and height. 33

OCR for page 33
34 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES FIGURE 4.1 (a) A snapshot of the instantaneous plume downwind of a continuous source in a turbulent flow (horizontal cross section). (b) The time-averaged plume. (SOURCE: EPA Fluid Modeling Facility). CATEGORIES OF DISPERSION MODELS Atmospheric dispersion models can be broadly placed into categories using three distinguishing characteristics: (1) their coordinate systems, (2) their windfie1/~ds, and (3) the type of averaging used in developing the models from the underlying conservation equations. Two coordinate systems, Eulerian and Lagrangian, are used. In a Eulerian system, the flow variables depend on time and on position in Earth-based coordinates. The Lagrangian system follows individual "fluid parcels" whose locations depend only on time. The Eulerian system is used in the vast majority of today's numerical flow models, including weather forecasting and climate models, but the Lagrangian system is naturally suited to dispersion problems. The winclfie11~1 in a dispersion model is, in some cases, defined by only a single value of the average wind at a specified height, such as in the simplest Gaussian-plume models. A step toward higher resolution is the incorporation of time-varying winds measured at several points within the domain. The highest-resolution dispersion models use a three-dimensional grid of winds calculated from a meteorological model.

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 35 The averaging used in the model development process is required for fluid- mechanical reasons. The equations governing flow and dispersion in the lower atmo- sphere have turbulent solutions with a range of spatial and temporal scales far wider than can be resolved on today's computers. Before the equations can be solved numerically in dispersion models, it is necessary that the range of resolved eddies be limited. This is done by ensemble or spatial averaging of the equations on which the models are based (as explained below). Of all the sorting criteria for dispersion models, the type of averaging applied to these governing equations has the broadest and deepest implications. Ensemble averaging transforms the equations from a set describing a single episode of a turbulent dispersion problem to one describing the average of a large number of episodes (formally called realizations) of the problem. Figure 4.1a is a snapshot of a single realization. The bottom panel shows the time-average plume, which in this case is identical to the average of a large ensemble of such snapshots. The difference between the single-realization and ensemble-average effluent concentration fields is profound. Gaussian-plume models for continuous releases are the oldest and simplest examples of ensemble-average dispersion models. They require a minimum of input information (average wind speed and direction, plus rudimentary information on whether the wind and temperature conditions favor turbulence and hence mixing, which allows diagnosis of the downstream growth of the Gaussian plume). There also are Gaussian models for finite-duration releases (called instantaneous releases) that can use an en- semble-average wind field derived from observations or computed through the dynamical equations. in contrast to ensemble averaging, spatial averaging has quite a different effect; it produces an equation set describing a coarser-grained version of a realization of the problem. The solution fields retain their turbulent character at scales larger than that of the spatial averaging. One could visualize a snapshot of such a solution by removing the finer-scale detail from Figure 4.1a. Examples of spatial-average models include coarse- mesh meteorological models having a horizontal grid scale of tens of kilometers. Finer- mesh examples include mesoscale models with grid size on the order of 1-10 km, and large-eddy simulation (LES)2 codes with grid size on the order of 100 m or less. INTERPRETING AND EVALUATING DISPERSION MODEL OUTPUTS In turbulent flow, the effects of slightly different initial conditions grow with time. As a result, two flows with nominally the same initial conditions eventually become quite different. This dependence on initial conditions has been found to be so sensitive that the initial conditions of a specific realization of a turbulent flow are unlikely to be known well enough to allow its reliable prediction. Thus, in the turbulent dispersion of effluent from a source, the downwind concentration patterns in two realizations of a given event will differ, the variation being more pronounced farther downwind of the source. For this reason. the output of a spatial-average dispersion model is properly interpreted not as a 2 Large-eddy simulation is the term used for the numerical calculation of three-dimensional, time- dependent turbulent flows using spatial resolution sufficient to resolve the largest turbulent eddies.

OCR for page 33
36 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES prediction of the dispersion under the specified conditions, but rather as one of a range of possible outcomes under those conditions. The timely availability of f~ne-scale wind-f~eld measurements (i.e., with spatial resolution finer than a dispersing plume's local crosswind dimension and temporal resolution finer than the scale of its local time changes) could change this situation. Such data used in a spatial-average dispersion model could substantially reduce the realization- to-realization variability that now accompanies the prediction of the atmospheric dispersion of a short-term release. Currently, such measurements are not feasible except in special circumstances. Radar and lidar have high potential for enabling such appli- cations in the future, although the time required to collect and assimilate high-resolution wind field data may continue to limit applicability to immediate emergency response needs. As Figure 4.1 suggests, concentrations at any one point in a given realization can differ substantially from the ensemble average at that point. This further suggests that neither an individual realization nor the ensemble average of realizations is sufficient in general for assessing the detailed, short-term dispersion characteristics of hazardous materials. Both a prediction of the ensemble-average field (interpretable as the most likely outcome) and a measure of the realization-to-realization variations about this average field are needed. Some models (e.g., the Second-order Closure Integrated Puff, or SCIPUFF, model) predict the ensemble-average dispersion plus a measure of the variability of the concentration field from realization to realization (such as the variance or the probability density functions) (Appendix E). Spatial-average models also allow probabilistic concentration forecasts. For in- stance, it is possible to vary the initial and boundary conditions and subgrid-scale physics of the dispersion model in order to generate an ensemble of forecasts of a given dispersion problem. This allows estimates of the spatially smoothed ensemble-average dosage field resulting from an instantaneous release as well as estimates of the pro- bability that the dosage for any area will exceed given thresholds. The concept of ensemble averaging need not focus only on the uncertainties in the turbulent dispersion process itself. One possible strategy for obtaining "end-to-end" uncertainties in a dispersion forecast is to create an ensemble average (and associated confidence levels) that includes a defined range of source and wind input variation by running multiple independent LES or physical simulations. This "brute force" approach cannot be applied directly to a real-time prediction, but it can be used to estimate uncertainties for a wide range of potential scenarios, and such a scenario database could provide an immediate first prediction for emergency responders. The scenario possi- bilities then could be updated with real-time model results as event-specific source and wind information become available. 3 The probability density function gives the probability of occurrence of values of the function. In mathematical terms, the probability that a random function (C) lies in an interval (AC) around CO is ~B(Co) AC' and the integral from -oo to CO is the probability distribution of C.

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 37 Because of the long use of ensemble-average dispersion models in the air quality community, their evaluation techniques (Weil et al., 1992) are more advanced than those for spatial-average models. Time-average concentrations or dosages measured at individual points downwind of a source typically differ substantially from the predictions of ensemble-average models. However, it has also been found that concentrations averaged over one hour, say, retain a good deal of random variability (Figure 4.2~. The interpretation is that downwind of a continuous source in the lower atmosphere, the time required for the convergence of a time-average concentration to the ensemble average can be much more than one hour. If so, one-hour-average observations would scatter sub- stantially around predictions of even a perfect ensemble-average model, but the models are not perfect, and model physics errors also contribute to the observed differences. It can be difficult to apportion these differences between errors in model physics and the inherent statistical scatter, or "inherent uncertainty" as it is called in the dispersion- modeling community. Improved models have been found to have decreased scatter, as illustrated in Figure 4.3, and it is now evident that much of the scatter between pre- dictions of the CASTER, a standard Gaussian-plume model, and observations at the Kincaid site (Figure 4.2) was due to inadequate model physics. The comparisons of model predictions and observations in Figures 4.2 and 4.3 are "paired in time and space," meaning that the observation and prediction associated with a given data point are for the same position in space and the same time period. Such comparisons typically lead to very large scatter, even with models having improved physics. For that reason it is common today to use quantile-quantile (Q-Q) comparisons instead. These are made by ordering the entire set of predictions by magnitude (from highest to lowest, say) and ordering the corresponding observations in the same way. Then the ordered predictions and observations are paired, the first with the first, the second with the second, and so forth, and the new pairs are plotted. Because of the ordering process, the observation and prediction associated with a given data point in general now do not correspond to the same position in space or the same time period. For this reason Q-Q comparisons are referred to as "unpaired in time and space." Figure 4.4 shows a Q-Q plot for the model at the Kincaid site. Its space-time unpairing greatly reduces its scatter from that in the conventional plot (Figure 4.3~. This decoupling of the predicted and observed points can mislead the reader into thinking the model performs better than it actually does. What this technique does show is the ability of the model to predict the probability distribution of the time-averaged concentrations downwind of a continuous release. Some unpairing of points also is done in testing other types of models. Figure 4.5 shows a scatter plot of observations downwind of a three-hour point release of sulfur hexaflouride (SF6) versus the predictions of the VLSTRACK (Vapor, Liquid, and Solid Tracking) model. The observations are of the maximum dosage along the sampler lines 5-20 km downwind for each run, and the predictions are of the maximum dosage along the sampler lines for that run. In general, the predicted and observed maxima occur at different points on the line. If a dispersion model used in an air quality application yields a 1:1 line on a Q-Q plot, the probability distribution of its predictions over a downwind region agrees with that of the observations in that region. Often, more spatial specificity is not needed, and in such cases the Q-Q plot can be an effective model evaluation tool.

OCR for page 33
38 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES :~ ~ nit .t IF . . .~ . /,... :,,,~F' .' :~ :' : cRST,~R ~~:~.! .~ *,* * ~ ~ ~. :~ * ~ ~ :~: - : - . ~ -in- . _' : - A:'. ~ _. - . in. ~ ~ ~% - ~ - - .o ~ . .~ ~ low ~ ~ ~ -: 'a ~' ~ ~ % id- ~ ~ ~ ~ : ,* . . ... .. ;~ ..55 :~ :~ 'AR no :;4} KIn~d S I. ~ is, 0,~. 10~0 ~ .. .... .. 10 10~ 1,.~:2 ~ ;590:) FIGURE 4.2 Observed versus predicted ground-level sulfur hexaflouride (SF6) concentrations for the CRSTER Gaussian plume model at the Kincaid power plant. The observations are one-hour SOURCE: From Well et al. (1997~. averages. The diagonal line corresponds to Cobs = Cpre~. Reprinted with permission from the American Meteorological Society. :~ ~ . At.. $ ~ ~ :~1 I.. .. O.~. 0.~. .' Ad: if' --~L ~ ~ OF.: ~ t:4 n~ ~ : /~ L. ~ ~ ~4~ ' J if'= fit ', A*. ~ an! if' :' id.. .' _ : : - ~ hi-- 038 .: . ~ .~ ~~ Hi. FIGURE 4.3 Observed versus predicted ground-level sulfur hexaflouride (SF6) concentrations, normalized with the emission rate, for the PDF model at the Kincaid power plant. PDF is an ensemble-average model with improved physics. The observations are one-hour averages. SOURCE: From Well et al. (1997~. Reprinted with permission from the American Meteorologi- cal Society.

OCR for page 33
DISPERSIONMODELING:APPLICATIONTOC/B/NRELEASES _; ~ . 1~ ~ A ~ IMP '~E it, ~ 0,01 an' . ~ 00:1 O:~:1 a_ 'a .,iF: .;* an' _' i_,iF' An' .~ ~- : ~ 'a: : .~F . an' :;,~' .' ~..'.~. .....,j' ~ ,~ . ~ :~r: ilr- . O^0'1 0~1,. FIGURE 4.4 A Q-Q plot of the data in Figure 4.3. SOURCE: From Well et al. (1997~. Reprinted with permission from the American Meteorological Society. . 10 ~ ~ 7 ~- ~ .* ~ ~ / ~ -. ~ * * - ~~ ~7 .' 1 '~1 -a - 10 16130 1000 10000 1~00000 Ott 39 FIGURE 4.5 A dosage scatter plot for three-hour releases of sulfur hexaflouride (SF~. SOURCE: From Chang et al. (2003~. Reprinted with permission from the American Meteoro- logical Society.

OCR for page 33
40 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES It is not clear that this is the case for the episodic models needed in emergency response applications, however. It appears that a more useful test of episodic dispersion models would involve the probability density function (PDF) of the short-term-average concentration or dosage. This can be calculated from a sufficiently large set of observations at field or laboratory scale and is also provided by some advanced ensemble-average models (Well et al., 1992~. Ideally, any prediction of an episodic concentration field from a dispersion model should be accompanied by a prediction of its episode-to-episode variability quantified by the variance or, better yet, its PDF. OVERVIEW OF C/11/N DISPERSION MODELING SYSTEMS The workshop discussions and presentations addressed many facets of how trans- port and dispersion models can benefit homeland security. These models have the poten- tial to greatly assist emergency management personnel in the: meet. preparedness stage of predicting the outcome of C/B/N release scenarios; response stage of evaluating and containing the hazard zone; and recovery anc! analysis stage of assessing impacts on health and the environ- Different dispersion modeling capabilities are required for each of these stages. The preparedness stage may include site-specific meteorological data coupled with probability-based dispersion model predictions and/or wind-tunnel simulations for typical scenarios. During the response stage, short execution time dispersion models are essential for providing emergency personnel with event-specific forecast data. During the recovery stage, all available data can be incorporated into a dispersion model designed to reconstruct the plume's space/time concentration distribution. Dispersion models, particularly in the response and recovery stages, require meteorological observations to initialize the local wind field and contaminant data and, in turn, to initialize the source characteristics. Surface characteristics (e.g., topography, vegetation, built environment) for the area upwind and encompassing the C/B/N impact zone are also an important model input. All of these model inputs may vary from simplistic to highly complex, depending on the sophistication of the dispersion model. An additional challenge is that atmospheric dispersion models must be capable of assimilating measurements that come from an assortment of data collection networks, with information of uneven quality and quantity, collected over irregular time periods. For a model to be useful in the response stage of C/B/N events, input data must be available in real time and the model must have a short execution time. The predicted concentration field from a dispersion model is combined with source toxicity, persistence, human and environmental sensitivity factors, and geographical information data to create maps of the event impacts. These maps are critical in the efficient allocation of emergency resources in the preparedness, response, and recovery and analysis stages of C/B/N events. To meet the needs of emergency response per-

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 41 sonnet, a dispersion model should map the hazard zone and provide an estimate of the concentration or dosage PDF at locations throughout the plume's domain. The accuracy of a dispersion model's output (a statistical description of concen- tration in space and time) will depend on the quality of model inputs, the model's analytical methodology, and the inherent random nature of turbulent processes in the atmosphere. As discussed earlier in this chapter, the "true" concentration field of a specific C/B/N event cannot be predicted. However, a probabilistic description of the concentration field can be estimated via dispersion modeling, even with an incomplete wind field input. Hazard Source Characterization The C/B/N source characteristics (Iocation, release rate, timing, buoyancy, momen- tum, toxicity, persistence, etc.) are critical in defining the ultimate event impact. This potential source variability requires that dispersion models include scales of motion ranging from meters to thousands of kilometers and account for chemical reactions and particle deposition physics. Figure 4.6 depicts a typical decision process for C/B/N source characterization. If the source is a known hazard, then remote detection of source character can be automatic with proper pre-event planning. If an unknown source is imported into a likely target area, then remote real-time instrumentation may yield sufficient data to initialize the dispersion model. In the case of an unknown source released in an area with little real- time instrumentation, trained first-response personnel with portable sensors must define the source character (hospitals and local or regional poison centers may be able to provide additional information through symptom identification) for subsequent dispersion modeling. Defining the source quickly and accurately is extremely important for the successful application of a dispersion model in the response stage of C/B/N events. To respond effectively to a weapon release, remote observational instrumentation coupled with source prediction algorithms must have been implemented previously. Decisions about the source toxicity and persistence will determine the type of transport and dispersion model most suitable for C/B/N events. Significantly different transport and dispersion methodologies and observational data requirements will be employed based upon the anticipated extent of the hazard zone. Wind Field Characterization Depending on the sophistication of the dispersion model, the availability of real- time data, and the horizontal scale of the area over which the dispersion must be calculated, the windfie1/~d may be defined by a variety of methods. These include using: a mean wind vector with an estimate of atmospheric stability in the vicinity of the release;

OCR for page 33
42 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES , Location tn~ ~ 8~,)~. f or Symptom 'tdentIflCatlon : ~ . ~ R.~e Of Unkn~n / ~9~e \. nD Agent: tmm unkno~f#1 '~ q18~s user W~ Decree I- ~ ~ d - . n (~-~S,{~ - ~ Re~) "I"' HE ~~\ if- -a _~ . ,, TO ~~ Eme T~~ ~ . '~iO5O~ ~ ~~:~inn:~l ' M08~r''0 A50~t , ~~e $ OL~:~n ~ . E~ ~~n a. am.: tnIt:j~! ' c'\~33~:~i ~r M~Y3t ~ _. :~_E .~, I' -I' ~ '., ~ Ft2nd the SOUme' F~~ H~ b COf~ (Purim .. ~ ,., ~ E$ten3816 ~6 I, ~ Ch~.~.~.~.11 _~J1 FIGURE 4.6 Hazard source characterization time line. a spatial array of winds provided by an analysis system that employs mass- continuity constraints; and a spatial array of winds and other meteorological parameters provided by a low-resolution Computational Fluid Dynamics (CFD) model or a mesoscale model that ingests observations. As higher speed computer processing becomes available, high-resolution model simulations of the wind and turbulence will be available on the fine scales of urban canyons. However, presently full-physics simulations are limited to the large scales of metropolitan areas. Finer scales must be estimated using systems with more limited physics. These observational wind data are processed in some dispersion models with local surface characteristics (topography, vegetation, structures) to form an estimate of the spatial and temporal wind field over the domain of C/B/N events. Lagrangian models use this wind field to transport and disperse either particles or Gaussian puffs to form con-

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 43 centration predictions. Figure 4.7 is an example using this type of system. CFD4 models and wind-tunnel models require the wind field and the surface characteristics as initial and boundary conditions prior to simulation. The CFD model may calculate hazard concentrations with each time step (coupled solution) or it may solve for only the wind field, which subsequently is used in a Lagrangian tracking model (uncoupled solution). CFD models for most turbulent flow problems fall into two general categories: Reynolds-averaged Navier Stokes (RANS) and LES. RANS uses the ensemble-mean equations of motion, and LES uses the spatially averaged equations with spatial resolution adequate to resolve the largest-scale turbulent eddies in the flow field. The typical spatial ranges of several dispersion modeling methods are depicted in Figure 4.~. The geographic extent of the wind data used for dispersion modeling should be several times greater than the anticipated maximum extent of the hazard (i.e., if the level of concern will stay within an urban area, then wind data in the surrounding area .f,,;~,..~..'::::::::: ^,,~ fin / ~~n ~~ Hi Rem.~:~e ~ g] ~ I'd' -A ;75 w~ i' .( - ~~m .) 'i ~quLre ' i~ ~ t`~' ~ ~ `~ - >~ ~ Lo~! W4~d or; S' R~iO~t W1~d2 if'' ,:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:.:. ::::::::::::: :: ::::::::::::::::::::::::: :::::::::::::::::::::::::: .....~........... :.:.:.:.:.:.:.:.: :.:: :.... :.:: :.:.:: :.:.:.: :.:.:.:.:.:.:.:.:.:.:. .................. : ::::::::::::::::::: ::::::::::::: ::::::::::::::: :::::::::::::: ~~T.~.~:.:.:.:.::::::::: 2.2 222""""""",2,:,:,:,:.:.2.2,2,",2: 2,2,2,2,''' """""""""""""2~ ~.~"""""""""""""2 ,:,:,:,:,:,:,:,:,:,:,:,:,:,:,:,:,:,:::: :,:,:,:,:,:,:,:,:,:,:,:: :,: :,: . .,:,:,:,:,:,:,:,:,:,:,:,:,:,:,:, ................... i .2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.2.""""""""""::::: :' '':. :,:' ,~2 2.2.2..2 2 2 2.2.2.".2 2 2 2.2.2.~.~....,.,.,.,, ,~,, ,:~:;t a' 3~ ,; I('\ i.." _ - I.... . ~ ....~. ~ S ~ S i .,. At; FIGURE 4.7 Wind field characterization time line. -A ..................................... :::::: :: ::: ::::: ::::::::::::::: :::::::::::::::::::::::::: 1 . ~ ~ :::::::::: ::: :.:: ::: :::::::::::::::::::::::::::::::::::::: . ~ ........................................ .......... ...................................... ...................................... ............ :"2"""2222222222222 , <~.,.,.r~ I'd ... 1"f' . . 4 CFD is a numerically based solution technique that solves the governing conservation equations for fluid transport physics. The solution provides flow values (velocity, pressure, temperature, concentration, etc.) at a large number of grid points within a predefined domain.

OCR for page 33
44 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES I. ~ ''-' ~ ~~-.~:Ubh '''' .- , ~'K~.~.~. . :::::::::::::':':':':'T'' 22:: :.. ...... -.:::::::::::::.'.' ............. ,,: ,,: .. , ~ \"""""""""""""""""""~ ~"""2 22.2.2.2.2 2 ~"""""22 22.2.2.2.2... :::::::::::::::::::::::::::::::: :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::':'::: :::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::: ntt2ne~t ol~! ~ tstance t~m S-~0 tom OK-1 ~ 10 100 I ::::::::::::::::::::::::::::::::::::::::::::::::::: ~ _:::::::::::::: ]: ~ ~~'~ ~ ~ ~ ~ ~ ~ ~ ~:~ : ,::::::::::::::::::::::~ : : . ~ -'-:' .: ~ E.:.:.:.:.:.:.:.:.:.:.:.:.:.:,:,:,:,:,:,:,:,:~ ~ . ~ l, ~ - ~ I: ~ .......................................... m.'.' ~ . . ~ : 3 - ::::::::::::::::::::::::::::::::e~::::::::::::::::::::::::::::::::::::::::::::::::::~.~-:::::::::::::::::::::: ~ ~ ~ her ~ : : Mar A:: . ~ ~ ~.~ ~~ ~~.~ ~ ~d ~ m~o ~ .~ . ~ c ^::::::::::::~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~ : ~~ :~ 44 i~:~` ~ ~ O~ :~ Io ~ ~ ::::::::::::::::::::::::::::::::::::::::::w :::::::::::::::::::--- ---I Am, ~ ~ 8'~&~'~ ~:~7rf" - ~~w ~ ~ ~ Fed : :,:,:,:,:,:,:,:,:,:,:~,:,:,:,:,:,:,:,:2:':':':':':':':':':':':':':':':':':'":~:':':':':':':::::::::::::':':'::':':'.~::::::::::::: ::::::::::: :::::::: : ~ ~:: : . : ::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::-:-:-::-:-:-:-:-:-:-.-:-:-:-:-:-:-:-:-:-:-:-:::::::::::::::::: ::::::: :: ~: : : ~ _. ~~e m.,n,~n ~~e ~ ,:,:,:,~,:':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':':'::: :':::::::::: :':':':':':':':':':':':':':':':':':'~: :':':~:~:':~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~: :::: .:::::::::::::::::::::e - ::::::::::::::::::::::::::~::::::::::::-:-:-:-:-:-:-:-:-:-:-:-:-:-:-:-:-~:~::-:-:: :-::::::::::::::::: ::::: ,:,:,:,~,:,:,:,:,:,:,:,:,:,:,:,:,:':':':':':':':: ~:':':':':':':':':':':':':':':':':':':':':':':':':' I':::::::::: :':':':':':':':':':':':':':':':':':'=:-:':':~:~:':~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~ ~:~:~:~: ,,,, Am; ,,,,,,,,,,,,,,,,,,,, C44~P,,,,,,,,,,,,,,,,,,,,,,,, ,,,,,,,,,,,,, ye : .:::::::::::::::: $,~ ,:,:,:, :,:,:,:,:,:,:,:,:,:,:,:,:,:':':':':':':':'~:':':':':':':':':':':':':':':':':':':':':':':':':':~':::::::::: :':':':':':':':':':':':':':':':':':'::::: :':':~:~:':~:~:~:~:~:~:~:~:~:~:~:~:~:~:~:~ ,,,,, ,,,,,,,,,,,,,,,,,,,,,,, ,,~ :.:.:.:.:.:.:.:.:::: I::::::::: :.:-:-:-:::::::::::::::::::::::::::::F - i:::::: : : : : :: : : : : : ::::::::::::: I:::: -' - i'- - ,:: m::::::::::::::::::::::::::::::::::::::::::::::::: :::::::::::::::::: :::::::::::::::::::::::::::::~:::::::::::::::::::::::::::::::::::::::::::: :::. ::::::::: : : :~: ... ,,,c~... ~ 'm" ~~ ~ ~ ................................................................................................... _ -A ~ . red A ~ l FIGURE 4.8 Spatial range of various dispersion model types. upwind of the urban area are required). Figure 4.8 also shows the relationship between plume arrival time and the distance from the source for mean wind velocities of 2 and 5 ms~~. In the case of urban C/B/N events, the plume likely will have passed through the entire urban zone in the early stages of event response (i.e., the first hour after the release). REVIEW OF SELECTED C/B/N DISPERSION MODELING SYSTEMS During the workshop proceedings, several presentations discussed the current status of selected dispersion modeling systems that were applicable to C/B/N events. A summary of these modeling systems is presented in Table 4.1, and all acronyms are defined in a list at the end of the report. CAMEO, HPAC, and NARAC are operational quick-response systems that are in use today, each serving a separate user base. All three modeling systems have modules for source identification (e.g., chemical, biological, and/or nuclear databases), meteoro-

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 45 logical data input, dispersion modeling, and consequence analysis with graphic output. The dispersion models in each of these modeling systems are unique. The Computer-Aided Management of Emergency Operations (CAMEO) system uses the Areal Locations of Hazardous Atmospheres (ALOHA) dispersion model, which is a modified Gaussian-plume formulation that predicts ensemble-averaged concen- trations (also time averaged to several minutes) out to a distance of 10 km. Wind data from only one meteorological station are used in this model. For wind speeds of less than 1 men, it draws a wind-direction-independent envelope around the source. If the source gas is heavier than air, it uses a modification of the Dense Gas Dispersion Model (DEGADIS), a freeware PC program that executes rapidly but has no provisions for topography or individual building geometry. ~ , . ~ ~ . The Hazard Prediction and Assessment Capability (HPAC) system has chemical, biological, and nuclear databases for source identification purposes, and it accesses weather data from in-house, NWS, and military providers. HPAC uses the SCIPUFF dispersion model, which uses a collection of Gaussian puffs to predict both the ensemble average concentration and the concentration variance out to regional scales. For dis- persion distances less than approximately 10 km, its output characteristics are similar to the ALOHA model described above. The local topography is incorporated into the model via generation of an interpolated wind field with data from any number of surface and upper-air measurements, but the surface roughness is required to be constant over the entire domain. . If the source gas is heavier than air, it uses a modification of the DEGADIS dense gas dispersion model. It is a registered user freeware PC program that has moderate execution times, but it has no provisions for individual building geometry. Similar to HPAC, the National Atmospheric Release Advisory Center (NARAC) system has all three databases for source identification, and it also receives weather data from in-house, NWS, and military providers. NARAC uses a suite of dispersion models to custom tailor event predictions to a subscribing client's needs. The system runs 24 hours a day, 7 days a week so that near-real-time mockups of release events are possible via a network link. The dispersion models in NARAC range from a simple Gaussian puff model (INPUFF), to Lagrangian particle methods (LODI), to CFD approaches (FEM). Specifically for nuclear and chemical applications, NARAC has the stand-alone (non- reachback) Gaussian-plume models HOTSPOT and EPICode. The only building-aware5 models in the NARAC system are CFD-based and have slow computation times. Both HPAC and NARAC models have been tested in the URBAN 2000 field experiment (Allwine et al., 2002~. Los Alamos National Laboratory (LANL) is in the final stages of testing two dispersion models that are designed for predicting hazardous concentrations in the urban environment. One is a Lagrangian particle dispersion model (QWIC-PLUME) coupled with a diagnostic wind field model (QWIC-URB). The other is a CFD-LES model named HIGRAD. Both models are building aware and were compared to URBAN 2000 field data. 5 Building aware means that the model does consider individual building geometry.

OCR for page 33
A, A> :::: . - %o ~ ~ Id :~: :: :: I: ~~ - A} o ~ ~ % ~e : : : : I: ~ ~~ I:: : :: : :: ::~:::~: ::: :: A) : ~ ~ a.) : ~ : : ::: c) : ~ - uO o c: :: 'e ::: ~ :: ::: t :~ :~ I: : ~ ~~ it: C:i ~ ~ Ct. ~ I: 53 =~ 53~:~: ~ ~ ~ : I: :~ ~ ~~ ~ ~ 5,` = ~ ~ =~ : ~ ~~ of: : :~ LIZ ~ ~ m~= : ~ ._ : ~ ~ O V) ~ :S o e : t S ~ ~ m:: CC : :: ~ _ 40'' ~ 04~ C<~ : hi ~ ~ : ::: .= Cd o o ~ ;> 3 ~ I: ~ ~:~ ~O ~~:~ 'cap: :~e ~ ~ ,~ CO: ~ o: ~ : :~::~: en ~ ._ . ~ m Cat ~:~ = i;, ~ 3~ ~~ o t<~ 4~ Ct o Ct o ._ lo: 3 o - U) :: 4_ ~ no ~ ~ A:' on >~ ~ ;^ ~ C\' ~ o Ce c: o : :e : Aces in: :: ~ : : : - : ~ ~ :~t, = :~: Hi ~ : ~ : : <~: ._ CO :: U): ::::: : KICK: ~c,: : :: Hi.; =

OCR for page 33
a, o a, a, cd O ~ ~ .5 . - cO 50 cd .cd o cd ~ o to a, o = at I) r a, a ~ to Ct . - C) . o ~ D 5 E E O Y ~ m 5 ~ ~ h ~ ~ 3 ~ ~ 11 ~ ~ ~ ~ D a b ' ~ ~ ~ a .5 ~ .g ~ ~ ~ 3 ' ~ hi. 5 ~ a a a ~ ~ ~ E ~ a if O . ~ , . E ~ ~ ~ a . ~ ~ D 5 a ~ a Y ~ ~~ ~~ ~ a C;3 ~ Cal O cx3 0 ~ ~ ~ 0 5 D ~~ O O A, =~ a.= an ~ ~ E ~ ~ ~ <; ~ ct ~ == c, ~ ~ ~ ~ ~ ~ ~ O

OCR for page 33
48 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES The LANE presentation (Appendix I) provided a concentration comparison between the URBAN 2000 field measurement program and the United Kingdom Defense Science and Technology Laboratory urban dispersion model (UDM). This Gaussian-puff models was specifically designed for dispersion in an urban environment and has been evaluated in many field and wind-tunnel experiments. The Environmental Protection Agency (EPA) presentation (Appendix F) demon- strated the use of several dispersion models to study the plume from the World Trade Center towers fire. EPA is developing a Gaussian puff dispersion model (CALPUFF), coupled with the diagnostic wind field model (CALMET); a CFD model (FLUENT); and a wind-tunnel simulation approach. The CALPUFF model produces one-hour averaged concentrations, thus it should be used only for steady sources. The CFD and wind-tunnel models are building aware, but the CALPUFF model is not. The Science Applications International Corporation presented the Operational Multiscale Environmental model with Grid Adaptivity (OMEGA), which is capable of using either an Eulerian or a Lagrangian particle approach. In OMEGA, the dispersion is fully coupled to a high-resolution NWP system with extra emphasis on surface and boundary layer processes. OMEGA has been used to support short-range dispersion in complex terrain (e.g., White Sands Missile Range) and long-range dispersion at continental scale (e.g., the European Tracer Experiment, ETEX). The OMEGA pre- sentation included comparisons of predicted and observed concentration for ETEX. DISCUSSION OF C/11/N MODELING SYSTEMS Dispersion model systems applied to C/B/N event scenarios can be divided into those that are useful for pre-event planning and training, those that are immediately available for response tactics, and those that will be used for post-event evaluation and recovery7. Dispersion models used for C/B/N event planning and response should provide emergency personnel with a common impact mapping format. In particular, given a dosage level of concern (LOC) for a toxin and a prediction confidence level, the dispersion model should provide a spatial contour defining the three-dimensional hazard zone. For examnie. if the confidence level was set at 99 percent, the dosage LOC would ~ , occur outside of the hazard zone contour only one time in 100 independent release events. To define the hazard zone in this format, some estimate of the spatial distribution of the concentration PDF is required. Several of the models discussed at the workshop do not currently provide sufficient statistical information to estimate PDFs. Models that cannot provide hazard zone confidence levels are of limited usefulness in emergency management; they are better suited to the chronic release situation of air pollution model- ~ng. CFD-LES models and laboratory simulations (i.e., reduced-scale models in a wind tunnel) of urban dispersion are the only tools currently available that can create the PDF 6 This type of model divides emissions into a series of overlapping volumes (puffs), so that it is not necessary to assume horizontally homogeneous emissions or to require steady-state conditions. 7 Note that post-event concentration data, if dense enough, could recreate the concentration field of an event.

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 49 of concentrations or dosages downwind of a transient C/B/N event. In both the wind- tunnel and the LES approaches, an ensemble of independent simulations covering a range of wind direction and wind speed may need to be formed in order to include scales of motion greater than that present in the domain of the model. This decoupling of turbulent motion scales will introduce errors in the model's predictive capabilities. No model in existence today can precisely deal with the full range of motions present in the urban dispersion problem; hence, there is a need for new computational tools to address this issue. Laboratory simulations are important tools for creating site-specific databases of C/B/N event scenarios and for the development and evaluation of both fast-response urban dispersion models and CFD-based dispersion models. Laboratory simulations provide better resolution of turbulent motions than current CFD models in the urban setting. CFD models have the potential to predict dispersive events in flow regimes that laboratory simulations find difficult, such as low wind speed and thermally dominated flows. CFD-LES modeling approaches also could potentially be used to study where in an urban area a plume of hazardous material likely is to be most heavily deposited. CFD models interface more interactively with meteorological data systems, even though execution times on the order of several days are common. With today's technology, a wind-tunnel urban simulation with enough data to define the probability density function throughout the plume's domain would take about a week to perform. The time required to build site-specific urban boundary conditions for both the CFD and the laboratory simulations would be substantially reduced if each urban area of concern had three-dimensional databases of buildings and topography that were compatible with the dispersion modelers' needs. The National Ima~erv and Mapping Agency of the U.S. Geological SurveY already has work underway in this area as n art of its National Mapping Program (see http://mapping.usgs.gov/~. These databases should be flexible in the amount of detail they provide so as to not overwhelm the computational model. Such databases would be useful for many other purposes, for example, air pollution modeling and urban planning for extreme winds. C7 -, To simplify dispersion models and reduce their error, it is desirable to define a minimum spatial and temporal scale at which averaged concentrations (over specific spatial and/or temporal scales) will be sufficient to determine C/B/N event impacts on health and the environment. This minimum scale will depend on the toxin released; thus, prior to designing a dispersion model, the potential range of toxin dosage levels of concern should be explored. Because of the importance of proper preparation before sending emergency personnel into harm's way, it is prudent to conservatively predict the extent of hazard zones. This might be accomplished by using statistical data obtained through an en- semble of model runs and by setting zone thresholds conservatively, based on the statistics of the ensemble. Introducing real-time observations into the process through data assimilation or the comparison of model with observational data would increase confidence levels.

OCR for page 33
so ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES It is clear that the existing suite of dispersion models currently in operational use by various government agencies has room for improvement. Both fast-execution response models and slower (but more accurate) preparedness and recovery models need further development and evaluation. Once a viable set of dispersion models capable of C/B/N event predictions is established, an independent quantitative review of these models should be initiated, and the results should be used to improve model performance. It ultimately may be determined that an ensemble of outputs from different models would yield better dispersion estimates than those of any one model alone. Many of the intercomparison studies carried out to date have been qualitative in nature and lacking in carefully controlled ground rules. Simplistic attempts to compare models against one another may serve to validate those models that have been prefer- entially designed for the conditions prescribed by the given competition, but this does not necessarily indicate superior modeling approaches for other, more arbitrary conditions. What are needed are carefully designed intercomparison studies that allow quantitative evaluation of models under the same controlled conditions. Procedures will need to be formulated so that experiments and models with significantly different output formats (e.g., field experiments producing a single realization and model outputs of ensemble- average statistics) are properly evaluated. Such rigorous intercomparisons have never been done adequately to date and will require careful experimental design. Proper evaluation would be aided by full documentation of each model's range of applicability, typical setup and execution times, forms of output (e.g., ensemble or spatial averaging),

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 51 analytical methods used for dealing with plume advection and growth and with different scales of motion, and other relevant factors. KEY FINDINGS AND RECOMMENDATIONS For purposes of threat assessment, preparation, and training, existing dispersion models meet some needs of the emergency response community. In the case of actual emergencies, the needs of emergency management may not be well satisfied by existing models. In particular, single-event uncertainties in atmospheric dispersion models are not well bounded, and current models are not well designed for complex natural topographies or built urban environments. Most available atmospheric dispersion models predict only the ensemble-average concentration (that is, the average over a large number of realizations of a given dis- persion situation). New approaches are needed for modeling a single hazardous release. Dispersion models used for emergency planning and response should provide confidence estimates that prescribed concentrations will not be exceeded outside of pre- dicted hazard zones. This requires that models provide some measure of the possible variability in a given situation. Different dispersion modeling methodologies are required in the preparedness, response, and recovery stages of C/B/N events. For the preparedness stage, an accurate model capable of providing confidence-level estimates is desired, but model execution time is not important. For the response stage, accuracy can be compromised to obtain timely predictions, but the dispersion model must still provide confidence-level estimates. For the recovery stage, model execution time is not important, but accurate model reconstruction of the plume concentration distribution over time is desired. In order to use a dispersion model's predictions effectively during the early response phase, the wind field and other conditions at the site of the release must be available in near real time and a short model execution time is essential. The most appropriate dispersion model for any given scenario may depend on the quantity, toxicity, and persistence of the hazardous agent; thus, it is critical that source identification be as rapid as possible. The committee's review of selected existing dispersion modeling systems determined that no one system had all the features that the committee deemed critical: confidence estimates for the predicted dosages, accommodation of urban and complex topography, short execution time urban models for the response phase, and accurate though slower models for the preparedness and recovery phases. Better integration between existing and future modeling systems could supply all of these critical features. The "unpairing" of concentration predictions and observations in time and space (commonly done with continuous sources in air quality applications) is inappropriate when evaluating dispersion model performance in episodic releases. Evaluation tech- niques based on more advanced probabilistic methods need to be developed. Toward that end, existing dispersion models should identify the type of averaging (ensemble, time and space) inherent in their modeling methodology, both in the wind field formulation and in

OCR for page 33
52 A TMOSPHERIC DISPERSION OF HAZARDOUS Al4 TERIAL RELEASES the treatment of dispersion. The reliability of existing and future dispersion modeling systems should be evaluated against field and laboratory measurements for potential C/B/N event scenarios. If predicted confidence limits are found to be unacceptable, then empirical corrections should be applied to model outputs so as not to place emergency personnel in harm's way. Increasing the density of the wind measurements in a plume's domain will potentially reduce uncertainty, thus reducing the predicted extent of the hazard without compromising confidence. Meteorological observations are a critical element of dispersion modeling. Observational technologies have been evolving rapidly in recent decades, and the committee has identified many existing measurement technologies that have not been fully exploited through data assimilation. Model operators and developers would benefit from broader interaction with the meteorological community, to take advantage of leading-edge research in data assimilation, quantitative precipitation forecasting, short- range numerical weather prediction, and high-resolution forecasting initialized with radar data. Likewise, observational research programs studying issues such as weather prediction, properties of boundary layer turbulence, and air pollution transport should be viewed as targets of opportunity for testing and evaluating dispersion models. A nationally coordinated effort should be established to foster support and systematic evaluation of existing models and research and development of new modeling approaches, undertaken in collaboration with the broader meteorological community. The Office of the Federal Coordinator for Meteorology, which recently organized a review of U.S. dispersion modeling capabilities, could provide valuable input as to which agencyties) is best suited to oversee this coordinated effort. Among the issues that should be addressed through this coordinated program are the following: . . New dispersion modeling constructs need to be further explored and possibly adapted for operational use in urban settings. This includes advanced, short execution time models; slower but more accurate computational fluid dynamics and large-eddy simulation models; and models with adaptive grids. Techniques must be developed for constructing ensembles of model solutions on the urban scale so that probabilistic rather than deterministic infor- mation can be provided to emergency managers. It will be necessary to quantify the level of confidence as a function of the number of ensemble members, which in turn, will have implications for the computational power required. It is necessary to learn how to more effectively assimilate into models an appropriate range of meteorological data (e.g., wind, temperature, and moisture data) from observing systems as well as real-time data from C/B/N sensors, espe- cially as the quality and availability of these data increase. It also is important to effecttively couple dispersion models with appropriate source characterization models. Urban field programs and wind-tunnel simulations should be conducted to allow for the testing, evaluation, and development of existing and new modeling systems (both meteorological and dispersion models). Developing an appropriate experimental design for such studies is a critical task that itself will require careful evaluation.

OCR for page 33
DISPERSION MODELING: APPLICATIONTOC/B/NRELEASES 53 The bulk effects of urban surfaces on the surface energy, moisture, and momentum are not well accounted for in most meteorological models. Existing development work in this area should be enhanced, and the improved modeling techniques adopted more widely. Urban building and topography three-dimensional databases need to be developed and maintained for use in numerical and wind-tunnel dispersion simu- lations. In at least one large urban area, a fully operational dispersion tracking and forecasting system should be established that is, a comprehensive system for collecting relevant meteorological and C/11/N sensor data, assimilating this informa- tion into a dispersion model, and maintaining the expertise and organizational capacity to provide immediate model forecasts on a full-time basis. If possible, a few such systems should be established and evaluated for different types of urban areas (e.g., coastal versus continental cities, low-latitude versus high-latitude cities). Such systems can be used as test beds for gaining understanding of model capabilities and limitations, and their use should not be limited to emergency situations. These observational and modeling tools could have multiple applications, which would help justify costs and ensure that the systems are frequently used, maintained, evaluated, and quality controlled. There is a wealth of knowledge about meteorological and dispersion models residing in universities, NWS Weather Forecast Offices, and private sector facilities throughout the nation. These sources of expertise, together with the existing programs in several national laboratories and military facilities, should be integral components of the coordinated national effort recommended above, to assist with developing local and regional models that are optimized for the topography and seasonal weather patterns in vulnerable areas. At the most basic level, this integration can be implemented via collaborative research and development efforts.

OCR for page 33
54 ATMOSPHERIC DISPERSION OF HAZARDOUS MATERIAL RELEASES