at SciVerse ScienceDirect

Atmospheric Environment 64 (2013) 25e39

Contents lists available

Atmospheric Environment

journal homepage: www.elsevier .com/locate/atmosenv

Urban transport and dispersion model sensitivity to wind direction uncertaintyand source location

Luna M. Rodriguez*, Paul E. Bieringer 1, Tom WarnerResearch Applications Laboratory, National Center for Atmospheric Research, 3450 Mitchell Lane, Boulder, CO 80301, USA

h i g h l i g h t s

< We characterize the impact of wind direction uncertainties on dispersion solutions.< The sensitivity is inversely related to the complexity in the topography.< There is a broad range of model sensitivities to wind direction uncertainty.

a r t i c l e i n f o

Article history:Received 12 April 2012Received in revised form10 August 2012Accepted 15 August 2012

Keywords:Urban transport and dispersion modelingEmergency responseMeteorological uncertainty

* Corresponding author. Tel.: þ1 303 497 2745; faxE-mail addresses: lunar@ucar.edu (L.M. R

(P.E. Bieringer).1 Tel.: þ1 303 497 8391; fax: þ1 303 497 8386.

1352-2310/$ e see front matter Published by Elsevierhttp://dx.doi.org/10.1016/j.atmosenv.2012.08.037

a b s t r a c t

The transport and dispersion (T&D) models used for air-quality and defense applications require infor-mation describing the source parameters and meteorological conditions to forecast concentration anddosage fields. In many cases the source parameters are known and the meteorological conditions arebased on observational data or mesoscale-model-generated forecast conditions. This research examineshow errors in the input wind fields translate into uncertainty in the contaminant concentrationpredictions. In particular, this study focuses on street-level errors in the dispersion patterns that occur in“building aware” T&D models that are sensitive to urban designs (e.g. road and building patterns) andrelease locations relative to the buildings. This problem was evaluated by first creating a “truth” plumefor a given release location and wind direction. Then the T&D model uncertainty associated with inputwind errors were determined by comparing plumes calculated using wind directions varied at 2�

increments to the truth plume. The uncertainty is quantified as fraction of overlap (FOO). The results areevaluated in a control simulation with no buildings, and in two commonly observed city designs(e.g. a regular grid, and hub and spoke configuration). The analysis examines both idealized buildingconfigurations along with the urban topography from cities that represent the regular grid and hub andspoke city designs. Results show that the relative impact of the uncertainty in the meteorologicalconditions and the corresponding sensitivity of the model to variations in the wind direction varysignificantly with the release location and city designs. This suggests that some source locations are less(more) sensitive to uncertainty in meteorological conditions and that this information can be factoredinto the confidence that is placed in emergency response decisions based on this information.

Published by Elsevier Ltd.

1. Introduction

Since September 11, 2001, the United States (US) governmenthas made significant investments in sensing and modeling tech-nologies designed to protect the US armed services and homelandagainst the threats posed by weapons of mass destruction (WMD).These technologies include the development of fast response

: þ1 303 497 8386.odriguez), paulb@ucar.edu

Ltd.

transport and dispersion (T&D) models that can account for thedispersion of chemical and biological (CB) agents released in urbanareas. The need for accurate atmospheric T&D forecasting tech-niques has become increasingly important because of the threat ofan intentional release of hazardous material into the atmosphere,particularly in areas of complex local surface forcing and for longertransport distances (Rife et al., 2004). Although extensive invest-ments have been made to improve the accuracy of these T&Dmodels in urban settings, the accuracy of the solutions are stillhighly dependent upon the meteorological conditions used. Changet al. (2003) determined that in cases wheremeteorological modelswere coupled with T&D models, the T&D models were stronglyinfluenced by the diagnostic wind model that was used to generate

Fig. 1. Modern city grid domain with dimensions of 4000 � 4000 � 100 m3. Fig. 2. Colonial Era city domain with dimensions of 4000 � 4000 � 100 m3.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3926

gridded wind fields from observed winds. Brown et al. (2008)showed that the sensitivities of plume transport in cities to winddirection, including how the street-level flow patterns in cities canbe very robust (i.e., unchanging) as the upper-level wind directionchanges, and then suddenly shift 180� at critical upper-level winddirections. The research presented here enhances these findings byquantitatively characterizing how errors in the input wind direc-tion translates into street level T&D uncertainty, specifically, intouncertainty in downwind hazard zones.

The urban settings evaluated in this study are designed after twocommonly occurring city design characteristics. The first is based

Fig. 3. Modern city, grid design with mixed building heights us

on a rectangular grid design and the other is a mix of rectangulargrids and a hub-spoke or web-like design. Due to the historicsignificance, related to its use in the USA and the ready availabilityof lethal dosage (LD) values, this study uses anthrax and its corre-sponding lethal exposure concentrations values to define hazardthresholds. Detailed information on the domain characteristics,source placement, and source characteristics can be found inSections 2.1e2.3 respectively.

This study uses the Röckle (1990) based Quick Urban IndustrialComplex (QUIC) Dispersion Modeling System developed at theLos Alamos National Laboratory (LANL) to evaluate the dispersion

ed in this study and a zoomed view of the downtown area.

Fig. 4. Colonial city, hub spoke design with mixed building heights used in this study and a zoomed view of the downtown area.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 27

pattern variability associated with wind direction errors. QUIC isa fast-response, urban dispersion modeling system capable ofcomputing three-dimensional wind patterns and dispersion ofairborne contaminants around clusters of buildings. The systemused is comprised of a wind model (QUIC-URB), a Lagrangiandispersion model (QUIC-PLUME), and a graphical user interface(GUI), QUIC-GUI (LANL, 2007). Hanna and Coauthors (2007)determined that the performance of QUIC-URB/PLUME from LANLwas comparable to the performance of other Röckle based modelslike MicroSWIFT/SPRAY (MSS) from the Science ApplicationsInternational Corporation (SAIC) & ARIA Technologies, Three-Dimensional Wind Field Model (3DWF) from the Army ResearchLaboratory (ARL), and the Israel Institute for Biological Research(IIBR) Kaplan and Dinar Model. The QUIC URB/PLUME modeling

Fig. 5. Urban topography from parts of Denver, CO. The area shown was selecte

system is used because its performance was representative of theComputational Fluid Dynamics (CFD) models used when evaluatedwith the New York City (NYC) Midtown tracer measurements(Allwine et al., 2008) and it is representative of the Röckle classbuilding-aware models and relevant for urban T&D applicationslike those developed for the Pentagon and surrounding facilities(Warner and Coauthors, 2007).

2. Methods

2.1. Domain characteristics

To evaluate the effect of uncertainty in T&D solutions associatedwith the input winds; this study uses domains that vary from

d so that the domain contained mixed building heights with a grid design.

Fig. 6. Urban topography from parts of Washington, DC. The area shownwas selected so that the domain contained mixed building heights with both grid and hub and spoke urbandesigns.

Fig. 7. Example source locations for the generalized Modern city domain of uniformbuilding heights.

Table 2LCt values for Anthrax using a probit slope, a spore ratio of3� 107 spores mg�1, and a light breathing rate of 0.02 m3 mg�1.These values are calculated using the Lawrence BerkeleyNational Laboratory database of physical, chemical and toxi-cological properties of chemical and biological (CB) warfareagents for modeling airborne dispersion in and aroundbuildings.

LCt percent LCt value (g s m�3)

50 0.0007810 0.000012

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3928

a simple non-urban domain that serves as a control simulation tothe complex urban building configurations. For the complex citydesigns, this study handles both generalized urban environmentsand actual building locations and heights from central Denver, COand Washington, DC. The generalized urban designs diagnose theresponse of the T&D model to wind direction errors relative to thenon-urban control simulation, and are used to infer the modelsresponse to a basic category of urban building/road network design.The generalized urban designs are chosen because they are repre-sentative of two types of city designs found in North America,a modern city grid design (Fig. 1) and the colonial era, hub andspoke design (Fig. 2). The red outlines in both Figs. 1 and 2 repre-sent individual city blocks where the buildings are 30 m high and

Table 1Anthrax characteristics from the Lawrence Berkeley National Laboratory database of phyagents for modeling airborne dispersion in and around buildings.

Biological class Spore forming Persiste

Acillus anthracis 2 h YearsDissemination/route of entry Incubation/onset ContagSpore inhalation, ingestion (rare), broken skin 1e2 h, 1e7 days No

cover a spatial footprint of 100 � 100 m. Houston, TX, Portland, OR,and Sacramento, CA are examples of cities with a grid-basedbuilding and road structure with city blocks on the order of100 � 100 m. The generalized urban design illustrated in Fig. 2 wasdeveloped to emulate a common characteristic of a hub and spokeurban design where several major streets converge at a city centeror square and are narrower in comparison to a modern city street.This design is common in colonial era cities of the northeasternUnited States like Boston, MA, Philadelphia, PA, and WashingtonDC. The city blocks in the colonial era design have a spatial footprintof 100 � 100 m and have a building height of 30 m.

The sensitivity of the T&D solution to wind uncertainties inurban environments with mixed building heights is also examined.Many urban centers have a mixture of building heights with tallcommercial buildings at the core of the city surrounded by a ring ofshorter multi-story buildings and an outer ring of single storybuildings. Two generalized urban domains, which incorporate theeffects of mixed building heights along with grid vs. hub and spokeurban design, were created for this study. Fig. 3 illustrates themodern city grid design and Fig. 4 illustrates the colonial era urbandesignwith mixed building heights used in this study. The buildingheights and road networks of the domain in Figs. 3 and 4 respec-tively were inspired by the buildings from the central businessdistricts of Denver, CO and Boston, MA. The results of the sensitivity

sical, chemical, and toxicological properties of chemical and biological (CB) warfare

nce Size (mm) Shape

w1 diameter � w1.5 length Rodious 50% infective dose (organisms/person) Untreated lethality (%)

8000e20,000 100

Fig. 8. The schematic of methodology depicted here first determines the intersection of the plume generated by the true wind direction and the plume generated by the error wind.Then the intersection is normalized by the plume generated by the true wind direction as described in Section 2.3.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 29

studies using the generalized urban domains described above arecontrasted with results from a comparable experiment using theactual urban topography from parts of Denver, CO andWashington,DC. The areas of the cities used were selected so that the domainscontain mixed building heights with both grid and hub and spokeurban designs. Figs. 5 and 6 illustrate the buildings used in the T&Dsensitivity simulations for these two cities. It is important to notethat in these case studies we evaluated one area in each domain todemonstrate how a plume in different urban environments, ina general sense, are affected by errors in the input winds.

2.2. Source placement

The impact of wind uncertainties on an urban T&D solution isalso affected by the location of the release location relative to

Fig. 9. Example of Methodology calculations, the FOO quantifies how well the plume generwind (Truth Plume).

a downwind and upwind building obstruction. To examine thiseffect, this study uses sources located in three locations: (1) in themiddle of the street (or urban canyon), (2) directly upwind ofa building obstruction, and (3) upwind of the corner of the buildingobstruction. Examples illustrating these source locations for thegeneralized Modern city domain of uniform building heights areshown in Fig. 7.

2.3. Source characteristics

To define a hazard area that could be used in the sensitivityanalysis it was necessary to set a threshold that would define theextent of the hazard area from the T&D simulation. Realism wasadded to the experiment by choosing anthrax as the material beingreleased; its human response characteristics determine the

ated from error winds (Error Plume) overlaps with the plume generated from the true

Fig. 10. The Fraction Of Overlap (FOO) for the non-urban domain with a function fitthat shows a spread of 4.7596.

Fig. 11. Plan view of smoke dispersal through an array of staggered cubes (left) andunobstructed fetch (right). Taken from Brown (2004) but originally published inDavidson et al. (1995).

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3930

downwind hazard area. The anthrax characteristics (Table 1) andthe lethal concentration toxicity (LCt) calculations are derived fromthe Lawrence Berkeley National Laboratory database of physical,chemical, and toxicological properties of chemical and biological(CB) warfare agents for modeling airborne dispersion in and aroundbuildings (Thatcher et al., 2000).

To quantify anthrax exposure mortality we use different dosagethresholds to identify the hazard areas. These areas are definedwithin the 50 and 10 LCt (Table 2). The LCt is defined as

LCt ¼ LDSR � BR

where: LD is the lethal dosage (spores), SR is the spore ratio(spores per mass of contaminant released), and BR is the breathingrate (m3 s�1). The thresholds represent the minimumvalue used todefine a hazard zone with an anticipated level of health responsewithin a given population. Any dosage above that value isconsidered hazardous and any dosage below that value is stillhazardous but has a lower probability of lethality relative to thispopulation health response. In this study we deem any dosagebelow that LCt value to be non-hazardous. Since concentrationscales linearly and for simplicity, only results that used an LCt10threshold (1.2 � 10�5 g s m�3) of anthrax are shown.

Fig. 12. Schematic of two plume dispersion patterns

2.4. Uncertainty calculation procedure

An identical twin setup using the QUIC system is applied toassess the sensitivity of the downwind hazard area from the T&Dmodel to wind direction uncertainties. To assess this sensitivity wefirst create a “building aware” wind field using QUIC-URB. Theprescribed “truth” wind on the upwind side of the release andbuildings is set at 3 m s�1 at all levels in the model. QUIC-PLUME isthen used to compute the dispersion of 10 kg of anthrax throughthe urban environment. For the purposes of this identical twinexperiment, this plume is now considered as truth against which allof the others are compared. In this way, truth plumes are created foreach of the urban configurations and release locations beingexamined. Next, plumes from the same release location are createdwith the QUIC system using winds that are varied to emulate winddirection uncertainties. These winds depart from the “true” winddirection value in 2� increments to a maximum of 40� in both thecounter-clockwise and clockwise directions from the direction usedto produce the truth plumes. The process is repeated for “true”wind directions in each urban design to evaluate the sensitivity inthe hazard areas for scenarios where the source is located upwindof building obstructions, urban canyons, and building corners of thebuilding/road network. The plume sensitivity to wind direction at2 m above ground level at LCt10 are then quantified using thefraction of overlap (FOO), figure of merit in space (FMS), measure ofeffectiveness (MOE), and normalized absolute difference (NAD)metrics (Warner et al., 2004). The FMS, MOE, and NADmetrics havebeen tested and evaluated with success against field observations;

expected using small wind direction variations.

Fig. 13. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction that transports a plume down an urban canyon in a Modern ( ) and the Colonial (C)domain for (a.) equal block heights, (b.) varying building heights, and (c.) the case studies. The Non-Urban domain (dotted line) is also shown for comparison.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 31

however, in this study, given that what we wish to quantify is howwell the modeled error plumes line up with respect to the modeltruth plume, the simple FOO suffices as ametric. The FOO quantifieshow well the plume generated from error winds (Error Plume)

overlaps with the plume generated from the true wind (TruthPlume). It calculates this by determining the intersection of theTruth Plume and Error Plume and normalizes by the Truth Plume, itis defined as

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3932

FOO ¼ areaðTruth PlumeXError PlumeÞareaðTruth PlumeÞ

Table 3The spread values normalized by the non-urban spread (4.7596) that were fit to eachFOO scenario.

Spread

Source location Urban canyon Buildingcorner

Buildingobstruction

Similarbuildingheight

Varyingbuildingheight

Constant buildingheight

Modern 2.78 e 2.09 2.21Colonial 2.35 2.48 2.14 2.20

Varying buildingheight

Modern 3.45 2.77 4.74 3.86Colonial 3.12 3.18 3.95 3.98

Case studies Denver 3.14 3.06 e 3.54DC 6.61 3.41 e 6.50

Thus, we will focus our results in terms of the FOO. Fig. 8 isa schematic of this methodology using the FOO and Fig. 9 shows anexample that illustrates this approach using results for the moderngrid domain from this study. When using the FOO metric, a highervalue of FOO corresponds to a higher tolerance to wind directionuncertainty.

Finally, given that we calculate the FOO at 2� increments toa maximum of 40� in both the counter-clockwise and clockwisedirections from the direction used to produce the truth plume, weexpect an exponential increase and then decrease in the FOO fora non-urban domain. We created a function that is fit to our non-urban FOO so that we can quantify the spread. This function isdefined as

Fit ¼ A

ð2pÞ12exp

��jqj2s

�;

where A=ð2pÞ1=2 is the amplitude, q are the wind angle increments,and s is the spread. Then we fit this function to each FOO scenarioand optimize the values of the A and s to compare the spread ineach scenario to the spread of the non-urban FOO. For the purposeof this analysis the spread corresponds to the relative sensitivity ofthe hazard zone to uncertainty in wind direction. Larger spreadvalues correspond with lower relative sensitivities to uncertaintiesin wind direction and smaller spread values correspond withhigher relative sensitivities to uncertainties in wind direction. Anexample of the fit to the non-urban FOO is plotted in Fig. 10.

3. Results and discussion

The goal of this study is to characterize the sensitivity of urbandispersion simulations to uncertainties in the winds used to drivethese simulations, and to identify scenarios where urban buildingand road configurations are more (less) susceptible to theseuncertainties. The results provide a means to more accuratelycharacterize simulation uncertainty that is critical to guidingemergency response decisions. The presentation of the results ispartitioned to illustrate patterns that correspond with: (3.1) thefidelity/complexity/symmetry of the urban building and roadnetwork, (3.2) the impact of the source location relative to thedownwind building, and (3.3) how these results depart from thegeneral behaviors discussed in 3.1 and 3.2, when the release loca-tion is moved from a major street intersection to a smaller streetfurther from the urban center. Implications of model sensitivity arediscussed for each result and are then summarized to providegeneralized guidance that can be used by emergency responsepersonnel that rely on these tools.

3.1. Impact of fidelity/complexity/symmetry of the urbantopography

Simulation speed is critical for emergency response applica-tions. In many cases simplified urban topographies are used toreduce the number of building elements, thereby reducing theamount of time required to compute a solution. While the reduc-tion of the fidelity of the building data clearly results in a lessaccurate solution since detailed geospatial information is notavailable, the impact on the solution sensitivity to wind directionuncertainties is not as clear. Key questions addressed here are: (1)how does the tradeoff between the need for a faster simulation andsimulation fidelity influence the dispersion solutions’ sensitivity toerrors in the wind fields, and (2) does this vary for building and

road configurations in cities with modern vs. colonial layouts.Numerous studies of contaminant dispersion in an urban envi-ronment have used uniform grids of building obstructions tocharacterize the properties of urban dispersion. Fig. 11 is an imageof a smoke plume from one such study conducted by Davidson et al.(1995). This figure contrasts smoke plume releases in an urban vs.non-urban domain and illustrates that the plume tends to dispersemore in the urban environment of “staggered buildings”. Thisenhanced dispersion due to the building obstructions also tends tomake the staggered scenario more tolerant to wind directionuncertainties because the downwind plume is much wider unlikein the scenario with no “staggered buildings”. These qualitativefindings were also documented by Brown et al. (2008). In that studyBrown demonstrated that downwind dispersion in urban envi-ronments can be highly variable given small errors or uncertaintiesin the wind direction. This behavior as depicted in Fig. 12, results insituations where small variations (uncertainty in the wind direc-tion) at times result in small variations in the downwind dispersionpattern while other situations with small wind direction uncer-tainties result in large differences to the downwind dispersionpattern. This property of reduced sensitivity of the dispersedsolution to wind direction uncertainties for locations whenbuilding obstructions are present is also shown quantitatively inFig. 13. Here, all of the results for scenarios with buildings indicatea higher FOO score than a corresponding non-urban scenario. Thisfigure also provides an example of how the dispersion solutionsensitivity to wind direction uncertainty can vary for differenturban topographies. In this example the contaminant releaseoccurs in the center of an urban canyon and the winds are parallelto the street canyon. Fig. 13a shows the FOO results for simulationswhere simplified urban terrain representing city blocks witha uniform building height are used. The FOO of the contaminantplumes for cities with both uniform grids and grids with largediagonal boulevards (common in the North America colonial eracities) have similar sensitivities, and are a factor of 2.78 and 2.35respectively, larger than the spread when no buildings are presentimplying that the size/location of the plume footprint is lesssensitive to wind direction. This difference in sensitivity is due tothe presence of more narrow “roads” in the colonial domain. Whenthe complexity of the urban topography is increased by addingbuildings of differing heights, the dispersion solution becomesslightly less sensitive to the wind uncertainties than the moresimplistic urban topography used in Fig. 13a. This point is illus-trated in Fig. 13b where FOO results indicate that the simulationsfor both the modern grid and colonial city designs are a factor of3.45 and 3.12 respectively, larger than the spread when no build-ings are present implying that the size/location of the plumefootprint is less sensitive to wind direction. The addition ofnon-uniform buildings also results in differences in the error

Fig. 14. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction that advects a plume towards a building obstruction in a Modern ( ) and the Colonial (C)domain for (a.) equal block heights, (b.) varying building heights, and (c.) the case studies. The Non-Urban domain (dotted line) is also shown for comparison.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 33

sensitivity between the positive and negative wind directiondifferences. Fig. 13b which shows the modern grid indicates moretolerance to wind direction uncertainty because the larger build-ings in the downtown area appear to channel the plume moreefficiently than in a colonial grid smaller buildings and narrower

streets. This Fig. 13c shows the results for the comparable FOOanalysis for the urban topographies of the cities of Denver, CO andWashington, DC where the urban canyon formed in each scenarioare between buildings of similar heights. In this example the urbantopography is substantially more complex than was used in the

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3934

simulations shown in Fig. 13a and b. Of the two “real-world” citiesexamined, Washington, DC has smaller but more buildings. Theportion of Denver used in the analysis has larger and taller build-ings. The inverse relationship between complexity of the urbantopography and the sensitivity of the atmospheric dispersionsolution to wind direction uncertainties is also evident in Fig. 13c(Table 3 under Urban Canyon, Similar Building Height). The FOOvalues for both Denver and Washington, DC are larger than thoseseen in Fig. 13b with Washington DC environment having the mostcomplex urban topography and was a factor of 6.61 larger than thespread when no buildings are present implying that the size/loca-tion of the plume footprint is less sensitive to wind direction. A fulllist of all of the normalized spread values for each of the buildingscenarios can be found in Table 3.

Fig. 14aec depict results for a comparable experiment to thatshown in Fig. 13 except that in this case the base wind directionand release location are confined so that a building obstructs thecontaminant dispersion directly downwind of the release. In thiscase an inverse relationship between complexity of the urbantopography and the sensitivity of the atmospheric dispersionsolution towind direction uncertainties is again evident. While theresults between the modern and colonial era cities tend to besimilar to each other for the less complex specifications of thebuildings/roads, the FOO results shown in Fig. 14c show a reversalof the results from 13c for errors less than 15� from the true wind.For errors larger than 15� from the true wind we see that Denver isless tolerant to errors because the plume does not interact withand is not channeled by as many buildings as for WashingtonD.C. This finding suggests that while the results presented hererepresent the general behavior of the model, there can be specificinstances of building configurations where this behavior does notapply.

This inverse relationship between the complexity of the urbantopography and solution sensitivity to wind uncertainties can bemore clearly seen in Fig. 15, which shows the spread of the FOOresults plot for the scenarios of increasing complexity in thebuilding/street configurations. The abscissa in this plot ranges fromlow complexity of configuration, non-urban on the left to the largercomplexity configuration, Washington DC results on the right. Thisfigure also includes the results from Figs. 13a/14a, 13b/14b, andthe Denver and Washington DC FOO results from both sets ofsimulations. The results imply that cities with complex buildingtopographies typically see a broader pattern of dispersion, and

Fig. 15. Function spread for modern, colonial, and case studies for urban canyon andbuilding obstruction scenarios.

consequently, changes to the input winds have less of an impact onthe solution. There is only a slight difference between the resultswhen there was and was not a building obstruction downwind ofthe release location.

In summary, these results for our specific release locationindicate that the urban dispersion model solutions for NorthAmerican colonial era cities (common on the US East Coast) areless sensitive to wind direction uncertainties than the cities builtmore recently that typically follow a relatively uniform grid. Thisfinding is due to the greater complexity in the urban building/roadnetwork. However, in order to fully capitalize on this behavior ofthe urban dispersion model, it is necessary to have sufficientfidelity in the urban building/road data sets used in the models.

3.2. Impact of the source location relative to the downwind building

The location of a contaminant source coupled with the winddirection and complexity of the urban environment results ina large number of possible release scenarios. This portion of thestudy attempts to consolidate these scenarios into three basiccategories. The first is the situation where the release occurs in thestreet canyon and the nominal wind direction is such that thecontaminant is carried downwind without any short-rangebuilding obstructions. The second is a situation where the releaseagain occurs in the street canyon, but this time the nominal winddirection is such that the contaminant is carried downwind directlyinto the face of a nearby building. The third category of scenarioexamined is the situation where the contaminant release occurs inan intersection of streets and the nominal wind direction is suchthat the contaminant is carried downwind into the corner of thebuilding. The question being addressed here is, does the winddirection and release location relative to down-wind buildingobstacles influence the solutions sensitivity to errors in the pre-vailing wind direction?

Figs. 16 and 17 provide a comparison of FOO results that addressthis question and illustrate the impact of source location relative todownwind buildings for both the idealized modern and colonialcity designs. The urban topography used in the analysis displayed inFig. 16 is the more simplistic representation of the buildings wherethe building obstacles represent entire city blocks. The urbantopography in Fig. 17 represents a more complex representation ofthe buildings where individual buildings of varying height areprescribed. The FOO analysis results in both cases suggest thatsituations where the corner of the building obstructing is directlydownwind of the release is the least sensitive to wind directionuncertainties. In Figs. 16 and 17, the results are mixed for the situ-ations where the release is carried downwind through the urbancanyon and the face of a building is directly downwind from therelease. In each of these scenarios the model solutions are moresensitive to the wind direction uncertainties than the buildingcorner case but roughly comparable to each other and similarbetween modern and colonial city designs. In Fig. 16b, both theModern and Colonial domains have symmetric FOOs and in Fig. 16c,they both become less symmetric. The release location being ina “main” avenue for the Colonial domain, better channeled for thecorner release (Fig. 16c) as well as, the release location being ina larger center plaza area. Fig. 18 summarizes this finding by plot-ting the spread of the FOO results for the scenarios illustrated inFigs. 16 and 17 where the release locations are varied relative to thedownwind building obstructions. The abscissa in this plot ranges onthe left from the scenario where there is not a building directlydownwind of the release in the truth simulation, to a scenariowhere a building is directly downwind of the release, to a scenarioswhere the release occurs directly upwind of the corner of a buildingon the right.

Fig. 16. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction that advects a plume towards an urban canyon (a.), building obstruction (b.), and a buildingcorner (c.) in a Modern ( ) and the Colonial (C) domain with equal block heights. The Non-Urban domain (dotted line) is also shown for comparison.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 35

3.3. Results that depart from the generalized wind-error sensitivitybehaviors

Due to the complexity of many urban landscapes it is difficult tobreak a release scenario down to a single category. In reality, the

real-world urban landscapes combined with the release scenario islikely to be a combination of several categories and their corre-sponding effect on the sensitivity of the solution to wind directionerrors. An example of this response can be seen in Fig. 19. Thisfigure shows the FOO analysis results for scenarios where the

Fig. 17. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction that advects a plume towards an urban canyon (a.), building obstruction (b.), and a buildingcorner (c.) in a Modern ( ) and the Colonial (C) domain with varying building heights. The Non-Urban domain (dotted line) is also shown for comparison.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3936

contaminant flows toward a nearby building corner. Here the FOOresults for the colonial city layout are skewed higher to the left of0� and the modern city layout FOO results are skewed much higherto the right of 0�. This illustrates the channeling effects that largebuildings near the source location can have on the dispersionpatterns and corresponding sensitivities to wind direction

uncertainties. A similar example can be seen in Fig. 20, whichillustrates the FOO analysis solutions for scenarios where thecontaminant flows down urban canyons in Denver and Wash-ington, DC without any direct obstruction from a nearby building,but here the canyon includes significantly varying building heights(Table 3 under Urban Canyon, Varying Building Heights). In Fig. 20a

Fig. 18. Function spread of plume when transported toward modern/colonial blocksand buildings for building obstruction, urban canyon, and building corner scenarios.

Fig. 19. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction thdomain for (a.) equal block heights and (b.) varying building heights. The Non-Urban doma

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 37

we see that the modern city layout is more sensitive to winddirection uncertainties than the colonial city layout, while 20bshows the opposite result when a different baseline wind direc-tion that carries the contaminant down alternative urban canyon isselected. In this situation there is little difference in the Denverresults when the baseline wind direction is changed, but a largerchange in the Washington DC results. This is largely due to size ofthe urban canyon. The baseline wind direction carries thecontaminant from Dupont Circle down Massachusetts Avenue inFig. 13a, which is a much wider urban canyon than the resultsshown in Fig. 20b where the baseline winds carry the contaminantfrom Dupont Circle down the narrower 19th Street.

4. Conclusions

Modeling contaminant dispersion in urban environments isa critical capability for the emergency response community. Thiscommunity relies on the accuracy of the results from these tools tomake decisions that impact the health and safety of both thegeneral public and first responders during both intentional andaccidental airborne releases of hazardous materials. In spite of thefact that during these crisis situations, high quality weather infor-mation is sometimes not available, critical decisions still need to be

at advects the plume towards a building corner for a Modern ( ) and the Colonial (C)in (dotted line) is also shown for comparison.

Fig. 20. Fraction Of Overlap (FOO) statistic and spread value when using a wind direction that advects a plume towards an alternate type of urban canyon in a Modern ( ) and theColonial (C) domain for (a.) varying building heights and (b.) the case studies. The Non-Urban domain (dotted line) is also shown for comparison.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e3938

made. In these situations, emergency response managers need tohave a measure of confidence in the dispersion model results. Thiswork addresses this need by characterizing the impact of winddirection uncertainties on the corresponding dispersion solutions.

The results of this study indicate that there are some generalrules of thumb that can be applied to the problem of how sensitivethe dispersion solution is to wind direction errors. First, thesensitivity of dispersion solutions to uncertainties inwind directionis inversely related to the complexity and dynamic variability in theurban topography. In cases where the complexity of the urbantopography is artificially reduced by simplifying the buildingdatabase to improve the speed of the solution, the resulting solu-tion becomes more sensitive to wind direction uncertainty. Second,simulation results from releases that occur directly upwind ofa corner of a building are less sensitive to wind direction uncer-tainties than those scenarios where there is no obstruction directlydownwind or the face of a building is directly downwind. Third, thepresence of large buildings near the release location can act as largebarriers to the flow and corresponding material dispersion andstrongly influence the sensitivity of the solutions to wind direction.Fourth, the presence or lack of major (wide) streets, which act aslarge urban canyons, can result in significant differences in thesensitivity of the model solutions to the wind uncertainty. Finally,this analysis indicates that there is a broad range of model sensi-tivities to wind direction uncertainty. While this may make it

challenging to identify generalized solutions to the wind directionuncertainty problem affecting the emergency response community,it does mean that there are release scenarios and locations that arevery tolerant to wind direction uncertainties. This finding suggeststhat for important locations it may be beneficial to pre-compute thewind direction error sensitivities so that they will be readilyavailable should the need unfortunately arise.

References

Allwine, K.J., Flaherty, J.E., Brown, M., Coirier, W., Hansen, O., Huber, A., Leach, M.,Patnaik, G., 2008. Urban Dispersion Program: Evaluation of Six Building-resolved Urban Dispersion Models, Official Use Only PNNL-17321 Report, 88 pp.

Brown, M.J., 2004. Urban dispersion e challenges for fast response modeling. In:5th AMS Symposium on the Urban Environment, Vancouver, B.C., Amer. Meteor.Soc., pp. 1e13.

Brown, M.J., Zajic, D., Gowardhan, A., Nelson, M., 2008. Limits of fidelity in urbanplume dispersion modeling: sensitivities to the prevailing wind direction. In:15th Joint Conference on the Applications of Air Pollution Meteorology with theA&WMA. Amer. Meteor. Soc., New Orleans, LA, pp. 1e11.

Chang, J.C., Franzese, P., Chayantrakom, K., Hanna, S.R., 2003. Evaluations of CAL-PUFF, HPAC, and VLSTRACK with two mesoscale field datasets. J. Appl. Meteor.42, 453e466.

Davidson, M., Mylne, K., Jones, C., Phillips, J., Perkins, R., Fung, J., Hunt, J., 1995.Plume dispersion through large groups of obstacles e a field investigation.Atmos. Environ. 29, 3245e3256.

Hanna, S., et al., 2007. An intercomparison of diagnostic urban wind flow modelsbased on the Röckle methodology using the Joint urban 2003 field data. In:Proceedings of the 6th International Conference On Urban Air Quality, Limassol,Cyprus.

L.M. Rodriguez et al. / Atmospheric Environment 64 (2013) 25e39 39

Los Alamos National Laboratory, cited 2007. QUIC-URB & QUIC-PLUME: FastResponse Urban Dispersion Modeling System. Available online at: http://www.lanl.gov/orgs/d/d4/atmosphere/docs/LA_UR_02_2559.pdf.

Rife, D.L., Davis, C.A., Liu, Y., Warner, T.T., 2004. Predictability of low-levelwinds by mesoscale meteorological models. Mon. Weather. Rev. 132,2553e2569.

Röckle, R., 1990. Bestimmung der Stromungsverhaltnisse i m Bereich komplexerBebauungsstrukturen, Ph. D. dissertation, Vom Fachbereich Mechanik, derTechnischen Hochschule Darmstadt, Germany.

Thatcher, Tracy, Sextro, Rich, Ermak, Don, 2000. Database of Physical, Chemical andToxicological Properties of Chemical and Biological (CB) Warfare Agents forModeling Airborne Dispersion in and Around Buildings. Lawrence BerkeleyNational Laboratory. LBNL Paper LBNL-45475.

Warner, S., Platt, N., Heagy, J.F., 2004. User-oriented two-dimensional measure ofeffectiveness for the evaluation of transport and dispersion models. J. Appl.Meteor. 43, 58e73.

Warner, T., et al., 2007. The Pentagon Shield field program: toward critical infra-structure protection. Bull. Amer. Meteorol. Soc. 88, 167.