bams-87-12-1713

8/13/2019 BAMS-87-12-1713

1/14

BY STEVEN R. HANNA , MICHAEL J. BROWN , FERNANDO E. CAMELLI, STEVENS T. C HAN , W ILLIAM J. COIRIER,

O LAV R. HANSEN, ALAN H. H UBER, SURA KIM, AND R. MICHAEL REYNOLDS

DETAILED SIMULATIONS OFATMOSPHERIC FLOW ANDDISPERSION IN DOWNTOWNMANHATTANAn Application of Five Computational Fluid Dynamics Models

Using the same urban atmospheric bound-ary layer scenario in New York City, thefive CFD models produce similar windflow patterns, as well as good agree-

ment with winds observedduring a field experiment.

There are increased concerns about releases of chemical or biologicalagents or toxic industrial chemicals by terrorist activities or acci- dents in downtown urban areas. For planning purposes and for real-time emergency response, decision makers want to know whether eitherevacuation or shelter-in-place is required and what areas are impactedand for how long a time. City dwellers are familiar with the swirling,nonuniform wind patterns in downtown street canyons, which causestandard straight-line atmospheric transport and dispersion models tobe inappropriate. Many papers describe the variability that characterizesflow and turbulence in urban

1713DECEMBER 2006AMERICAN METEOROLOGICAL SOCIETY |

8/13/2019 BAMS-87-12-1713

2/14

areas (e.g., Oke 1987; Rotach 1997; Roth 2000; Britterand Hanna 2003). To address this problem, a groupof scientists and engineers has been using compu-tational fluid dynamics (CFD) models to estimateairflow and dispersion patterns in the street canyonsof large cities.

New York, New York, is the focus of a set of recent

field experiments sponsored by the Urban DispersionProgram of the Department of Homeland Security(DHS). The wind data from the March 2005 MadisonSquare Garden (MSG05) experiment are used in thecurrent paper. In addition, a second field experimenttook place in August 2005 in the Midtown area. TheManhattan experiments are part of a sequence ofintensive urban field experiments that have takenplace over the past five years, sponsored collabora-tively by a number of agencies. Others include theSalt Lake City Urban 2000 experiment (Allwine et al.2002), the Oklahoma City Joint Urban 2003 experi-ment (Allwine et al. 2004; Dugway Proving Ground2005), and the London Dispersion of Air Pollutantsand their Penetration into the Local Environment(DAPPLE) experiment (Britter 2005). These experi-ments make use of dense networks of fast-responsesonic anemometers sited at street level and onbuilding tops, as well as remote sounders such asminisodars. The experiments also include tracer gasreleases and sampling at many locations. The avail-ability of these extensive urban databases provides anopportunity for further development and evaluationof many types of urban flow and dispersion models,including CFD models.

CFD MODELS. For urban applications, the CFDmodels solve the basic equations of motion on a high-

resolution (110 m) three-dimensional grid system,within a domain with sides of a few kilometers atmost and with typical depths from 0.5 to 1 km.Detailed three-dimensional (3D) building data arealso needed for the simulations. Now that comput-ers are faster and have more storage, it has becomepossible to run CFD models on an urban domain

within a reasonable time frame (say less than a fewhours). Multiple sensitivity studies are now possible.However, most early applications were to scenariosthat were strongly forced by obstacles, such as a singlecube, and only the near-field results were analyzed.Modelers (e.g., Hanna et al. 2002) found that the CFDmodel-simulated turbulence was reasonable near theobstacle but, in many cases, died away too quicklyonce the flow passed the influence of the obstacle.Some models had a difficult time maintaining suf-ficient turbulence over, say, a uniform grassy field.The atmosphere is naturally quite turbulent, withturbulence intensities of 0.1 or more. The turbulence-maintenance question and other related scientificquestions were the subject of a workshop held in July2004 at George Mason University in Fairfax, Vir-ginia, where the current authors were in attendanceand agreed to proceed with collaborative studies. Amethodology for overcoming the turbulence-mainte-nance problem was suggested by the EnvironmentalProtection Agency (EPA) CFD modeling group (Tanget al. 2006) at the American Meteorological Society(AMS) Annual Meeting.

One aspect of the collaborative studies discussedat the workshop was to use some standard field da-tabases to advance the development and evaluationof the CFD models. These databases included theKit Fox, Mock Urban Setting Test (MUST), PrairieGrass, and Evaluation of Model Uncertainty (EMU)data (e.g., Hanna et al. 2004). A key database was theclassical 1955 Prairie Grass field experiment, whichtook place over a grassy field, allowing model perfor-mance to be tested for a scenario with no buildingsor other obstacles to force the flow. Although there

is not space in this paper to describe the details,some modelers were able to improve their turbulenceparameterizations so as to produce good agreementwith the Prairie Grass data.

Another aspect of the collaborative study was torun the CFD models as part of ongoing major studiessuch as the Manhattan (MSG05) study. The modelswere used to plan the experiment and are now beingused to analyze the results, as discussed in the cur-rent paper. It should be mentioned that, while someof the CFD work [Finite Element Model in 3D andMassively-Parallel version (FEM3MP) and FLU-

AFFILIATIONS : HANNA Harvard School of Public Health ,Boston, Massachusetts; B ROWN Los Alamos NationalLaboratory, Los Alamos, New Mexico; C AMELLI George MasonUniversity, Fairfax, Virginia; CHAN Lawrence Livermore NationalLaboratory, Livermore, California; C OIRIER AND KIM CFD

Research Corporation, Huntsville, Alabama; H ANSEN GexCon,Bergen, Norway; H UBER Air Resources Laboratory, NOAA,Research Triangle Park, North Carolina; R EYNOLDS BrookhavenNational Laboratory, Upton, New York CORRESPONDING AUTHOR : Steven R. Hanna, 7 CrescentAve., Kennebunkport, ME 04046E-mail: [email protected]

The abstract for this article can be found in this issue, following thetable of contents.DOI:10.1175/BAMS-87-12-1713

In final form 19 July 20062006 American Meteorological Society

1714 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

3/14

8/13/2019 BAMS-87-12-1713

4/14

M. Brown at the 2006 AMS Annual Meeting (Camelliet al. 2006), and a brief comparison of FEM3MP withobservations for the MSG05 west-northwest case waspresented by M. Leach at the same meeting (Leachet al. 2006). The conclusions by Camelli et al. (2006)concerning model-to-model comparisons weresimilar to what is found here (i.e., good agreement

concerning general flow patterns), although therewere no observations from south-southwest winddirections to aid in the evaluations.

Summaries of the CFD model characteristics andassumptions for the current MSG05 exercise (for thewest-northwest wind directions observed during thefield experiment) are given in Table 1. Readers inter-ested in more details can consult the references. Allmodels except for FEFLO-Urban were run in Reyn-olds-averaged NavierStokes (RANS) mode. The

RANS models outputs have 3D variability but rep-resent an average over time and are therefore steadystate. FEFLO-Urban was run in large-eddy simulation(LES) mode, which requires more computer time butproduces time-variable flow fields.

All CFD models used the three-dimensionalbuilding database for Manhattan licensed by the

Vexcel Corporation. These licensed building datahave a resolution of about 1 m, and support visualiza-tions that look like real photographs. However, wepoint out that, because buildings in large cities suchas New York are razed and rebuilt with surprisingfrequency, it is necessary to update the 3D file forapplications at any particular time.

To allow for comparisons with the observed windsat street level, the CFD model simulations were madefor average rooftop wind conditions observed during

TABLE 2. Summary of wind observations during t wo MSG05 experiment days (10 and 14 Mar 2005).

Sitelabel

Name z (m)

AGL

10 March wind speed

(m s 1)

10 March winddirection ()

14 March wind speed

(m s 1)

14 March wind

direction ()Comment

R1 One Penn Plaza 229 7.3 286 7.0 327 Tall rooftop

R2 Two Penn Plaza 153 5.8 306 3.8 318 Tall rooftop

R3 Farley Post Office 34 3.6 281 3.9 269 On broad, flatbuilding

CCNY* City College ofNew York 58 5.2 266 5.2 309 Open rooftop

SIT* Stevens Institute ofTechnology 52 5.7 297 6.9 335 Open rooftop

EML*Environmental

MonitorLaboratory

82 3.3 286 4.3 323 Open rooftop

LBR* Lehmann Bros.Building 160 4.7 286 3.6 308 Open rooftop

JFK* Airpor t 3.4 6.2 290 6.5 320 Flat airpor t

S1 Northwest MSG 3.0 3.0 212 2.7 187 See figure

S2 Southwest MSG 3.0 1.7 27 steady 1.2 80 variable See figure

S3 Southeast MSG 3.0 3.3 76 steady 2.6Variable

westeast See figure

S4 Northeast MSG 3.0 1.6Variable, north-

northwest tosouth-southeast

3.6 165 steady See figure

S5 NorthwestOne Penn Plaza 3.0 2.6 238 1.7 292 See figure

S6 Front of NewYorker Hotel 5.0 1.2 162 Channeled

S7 8th Ave Side ofMSG 3.0 1.2 17 2.0 28 Channeled

* These sites are outside of the MSG area shown in Fig. 1.

1716 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

5/14

the two time periods (from0900 to 1400 LST 10 and 14March 2005) during whichtime the MSG05 f ield experi-ment took place. These twodays were both characterizedby fairly steady moderate west-

northwest to northwest windflows, well-mixed near-neu-tral conditions, and cold tem-peratures (near 0C). Hanna etal. (2006) provide an overviewof the MSG05 field experimentand a summary of the windand turbulence observations.As seen in Table 1, the fiveCFD models assumed similarinflow conditions, based onaveraged wind observationsat rooftop and other exposedsites on 10 March, as listedin Table 2. Although the fivemodels used slightly differentassumptions for inflow (up-wind) wind speeds, the resultsare expected to be relativelyunaffected because the build-ings have such a strong ef-fect on the wind patterns andthe incoming flow has a fewblocks to adjust to the underly-ing built-up urban area.

To illustrate the magnitudesand variability of the observedwind speeds and directions,Table 2 (prepared by Hannaet al. 2006) contains averageobserved winds during t hefive-hour (09001400 LST)experiment periods on 10 and14 March. Data are given for

the anemometers shown onFig. 1, as well as for anemom-eters from other Manhattanbuilding tops, from John F.

F IG . 2. Observed wind vectors(red near street level and blueat rooftop) at 0900 LST 10 Mar2005. At the S site on the Postoffice, the sodar wind vectors atz = 20 and 120 m above the roofare shown.

F IG . 1. Aerial photograph of area (of approximate dimensions 500 m 500 m)around MSG in Manhattan, where MSG is the round building and has 130-mdiameter and 50-m height. The 229-m-tall One Penn Plaza building is tothe northeast of MSG and the 153-m-tall Two Penn Plaza building is tothe east-southeast of MSG. At the R3 site (on the Farley Post Office), Mrefers to the fixed anemometer and S refers to the sodar.


8/13/2019 BAMS-87-12-1713

6/14

Kennedy International (JFK)Airport, and from the topof a building on the StevensInstitute of Technology (SIT)campus in Hoboken, NewJersey, on the west bank of theHudson River.

The sonic anemometerslisted in Table 2 measuredwinds at eight locations nearstreet level and on severalrooftops, such as the OnePenn Plaza (229 m) and TwoPenn Plaza (153 m) buildings,which are adjacent to Madi-son Square Garden (MSG).Figure 1 shows the MSG do-main and buildings, and givesthe positions of the anemom-eters at street level (S) and atrooftop (R). Figures 2 and 3use the same geographic do-main and include, as an exam-ple, the observed 30-minute-averaged wind vectors from0900 to 0930 LST 10 Marchand 14 March, respectively.The observed rooftop winds

have speeds of about 6 m s 1 and arefrom the west-northwest direction on10 March and the northwest direc-tion on 14 March, while the observedstreet-level winds (with an averagescalar speed of about 2 m s -1) havemany directions, depending on nearbybuildings. The figures show that, withthe exception of the two sonic ane-mometers close to the windward (east)side of Two Penn Plaza, the observedstreet-level wind patterns do roughlyagree on the two days.

As intuition would suggest, therelative influence of the One PennPlaza (229 m) and Two Penn Plaza(153 m) buildings switches for thesouth-southwest wind direction usedin the CFD planning runs (Camelli etal. 2006) and the west-northwest winddirections observed during the fieldexperiment and used in the currentcomparisons. Because the broad sideof One Penn Plaza faces the south-southwest, it dominates the flow for

F IG . 3. Observed wind vectors (red near street level and blue at rooftop) at0900 LST 14 Mar 2005. At the S site on the Post office, the sodar windvectors at z = 20 and 120 m above the roof are shown.

F IG . 4. Simulations of horizontal wind vectors (m s 1) at z = 5 m byCFD-Urban model for 10 Mar 2005 upstream wind inputs (flow fromthe west-northwest).

1718 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

7/14

the south-southwest wind direction.And because the broad side of TwoPenn Plaza faces the west-northwest,it dominates for the west-northwestwind direct ion, as seen in the resultsin the figures. Although we did notcarry out any CFD model runs for

light winds with variable directions,it is obvious that the flow patternswould flip back and forth from be-ing dominated by one building orthe other if the wind directions are varying back and forth between thesouthwest and northwest.

Although the MSG05 CFD modelcomparison exercise is nearly fin-ished, these same CFD models arebeing run for the August 2005 Mid-town field experiment (MID05) inManhattan. MID05 was the secondexperiment in the series begun byMSG05 and involved more experi-ment days (six instead of two) andmany more meteorological and sam-pling instruments. Wind speeds werelighter, by a factor of 3, in MID05than in MSG05. The MID05 CFDcomparisons are being planned to bemore quantitative than the currentqualitative comparisons for MSG05.

There are 19 figures presented thatillustrate the degree of agreement (ordisagreement) among the five modelsfor MSG05. The first set of figurespresents the horizontal wind vectorsnear the ground. The second set offigures presents the vertical veloci-ties near the ground. The third set offigures shows along-wind (x z ) crosssections of wind vectors. The final setof figures shows some predicted nor-

malized concentration patterns forthe tracer-release locations aroundMSG. With much more planningand support, we perhaps could haveproduced the figures in the sameformat and color scheme for eachmodel. However, because several ofthe modelers are participating ona volunteer basis, we were satisfiedonce the domains, inputs, heightsof outputs, and other details werereasonably close for each model. Per-

F IG . 5. Simulations of horizontal wind vectors (m s1

) atz

= 5 m byFLACS model for 10 Mar 2005 upstream wind inputs (flow from the west-northwest).

F IG . 6. Simulations of horizontal wind vectors (m s 1) at z = 4 m by theFEM3MP model for 10 Mar 2005 upstream wind inputs (flow fromthe west-northwest).


8/13/2019 BAMS-87-12-1713

8/14

haps future comparison studies could impose precisecriteria for these detai ls beforehand.

RESULTS. Examples of simulated wind vectors nearstreet level, at 0900 LST 10 March of the MSG05 fieldexperiment, are given in Figs. 4 through 8 for the five

CFD models (CFD-Urban, FLACS, FEM3MP, FEFLO-Urban, and FLUENT-EPA, respectively). Side-by-sidecomparisons with the 30-minute-averaged observedwind vectors (in Fig. 2) show reasonable agreementin speed (within a factor of 2) and direction (withinabout 30) for most street-level sites. For example,the models capture the diverging flow toward the

upwind and crosswind directions onthe windward side of MSG and TwoPenn Plaza (just east of MSG).

In the planning run compari-sons for the south-southwest winddirection, Camelli et al. (2006) showthat there are a few locations on the do-main that show significant differencesamong the models, and these warrantfurther investigations. Usually the dif-ferences occur where the wakes of twoadjacent buildings are battling eachother for dominance. Careful com-parisons of Figs. 4 through 8 revealsimilar discrepancies in certain partsof the domain.

The CFD model outputs and

the observations are comparedqualitatively in the current paper.The CFD team is proceeding withlimited quantitative comparisonsfor the MSG05 experiment, andmore extensive comparisons for the

MID05 experiment, and results will be shown in afuture paper. For example, the 30-minute-averagedwind speed and direction simulated by the five mod-els at each anemometer location will be comparedusing standard statistics. Vertical profiles and crosssections of model outputs such as turbulent kinetic

F IG . 9. Simulations of vertical velocity w (m s 1) atz = 5 m, by CFD-Urban for 10 Mar 2005 upstream windinputs (flow from the west-northwest).

F IG . 7. Simulations of horizontal wind vectors (m s 1)at z = 5 m by FEFLO-Urban model for 10 Mar 2005 up-stream wind inputs (flow from the west-northwest).

F IG . 8. Simulations of horizontal wind vectors (m s 1) at z = 2 m byFLUENT-EPA model for 10 Mar 2005 upstream wind inputs (flowfrom the northwest).

1720 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

9/14

8/13/2019 BAMS-87-12-1713

10/14

F IG . 13 ( ABOVE ). Simulations of vertical velocity w (m s 1)at z = 2 m, by FLUENT-EPA for 10 Mar 2005 upstream

wind inputs (flow from the nor thwest) .

F IG . 12 ( LEFT ). Simulations of vertical velocity w (m s 1)at z = 5 m, by FEFLO-Urban for 10 Mar 2005 upstream

wind inputs (flow from the west-nor thwest) .

F IG . 15. Simulations of wind vectors (m s 1) on x z cross section through MSG for the west-northwest direction,for FLACS. The view is toward the north-northeast.

F IG . 14. Simulations of wind vectors(m s 1) on x z cross section throughMSG for the west-northwest direc-tion, for CFD-Urban. The view istoward the north-northeast.

1722 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

11/14

F IG . 18. Simulations of wind vectors on x z cross section through MSG for thenorthwest direction, for FLUENT-EPA.The view is toward the north-northeast.

F IG . 16. Simulations of wind vectors (m s 1)on x z cross section through MSG for the

west-northwest direction, for FEM3MP.The view is toward the north-northeast.

F IG . 17. Simulations of wind vectors (m s 1)on x z cross section through MSG for the

west-nor thwest direct ion, for FEFLO -Urban. The view is toward the north-northeast.


8/13/2019 BAMS-87-12-1713

12/14

F IG . 20. FLACS simulationof tracer gas dispersion fora point release near streetlevel on the southwest side ofMSG for the west-northwest

wi nd di re ct io n. Th is is on e

of the five source locationsused during the MSG05 fieldexperiment. The figure is for900 s after the release wasinitiated.

F IG . 21. FEM3MP simulation of tracergas dispersion for the west-northwest

wind direction and a point release nearstreet level on the southwest side ofMSG. This is one of the five sourcelocations used during the MSG05 fieldexperiment.

F IG . 19. CFD-Urban simulation oftracer gas dispersion for a pointrelease near street level on thesouthwest side of MSG, for the

west - no rth west wi nd di rect io n.This is one of the five source loca-tions used during the MSG05 fieldexperiment.

1724 DECEMBER 2006|

8/13/2019 BAMS-87-12-1713

13/14

is because the x z cross section passes through themiddle of a narrow street.

Four of the models were also used to simulatetracer dispersion patterns for eventual comparisonwith observations during MSG05. Although thetracer studies are not the main emphasis of the cur-rent paper, it is found that the models agree that thetracer initially spreads a block or two upwind andlaterally while it is still near street level, and thenspreads downwind as a broad plume after it mixes

vertically to the building tops. Examples of CFD mod-el simulations of tracer dispersion from the releasepositions around MSG are shown in Figs. 1922 forCFD-Urban, FLACS, FEM3MP, and FEFLO-Urban,respectively. The simulations are presented in nor-malized mode, and comparisons are not given withthe actual observations because of security reasons.

When time series of CFD model concentrationplots are studied, they show the hold up of tracermaterial in recirculating zones behind buildings orin blocked regions with very low velocities. Thesezones are very important for emergency response

decisions, and further analysis of theCFD model outputs and the tracerdata should aid in devising decisionstrategies.

It is seen from these examplefigures that the simulations by thefive models are qualitatively simi-

lar. They agree fairly well with eachother and with the MSG05 flowobservations, at least concerninggeneral patterns and flow magni-tudes. Although more analysis isclearly needed, these preliminaryCFD results suggest that they holdpromise for aiding in increasingour understanding of wind flow andtracer dispersion in urban areas.

ACKNOWLEDGMENTS. The re-search has been sponsored by the U.S.Department of Homeland Security,the Defense Threat Reduction Agency,the National Science Foundation, theDepartment of Energy, the NationalOceanic and Atmospheric Administra-tion, and the Environmental ProtectionAgency. FLACS runs were supportedby GexCon internal research funds andFEFLO-Urban runs were supportedby the George Mason University CFD

Laboratory internal research funds.The MSG05 field experiment and the current research

are part of the multiyear Urban Dispersion Program (UDP),whose primary sponsor is the Department of HomelandSecurity. The March 2005 MSG05 field experiment and theAugust 2005 Midtown field experiment (MID05) datasetsare still undergoing quality assurance/quality control butwill soon be placed in a permanent data archive.

REFERENCES

Allwine, K. J., J. H. Shinn, G. E. Streit, K. L. Clawson,and M. Brown, 2002: Overview of Urban 2000. Bull. Amer. Meteor. Soc., 83, 521536. , M. Leach, L. Stockham, J. Shinn, R. Hosker, J.Bowers, and J. Pace, 2004: Overview of Joint Urban2003An atmospheric dispersion study in Okla-homa City. Preprints, Symp. on Planning, Nowcast-ing and Forecasting in the Urban Zone, Seattle, WA,Amer. Meteor. Soc., CD-ROM, J7.1.

Britter, R. E., cited 2005: DAPPLE: Dispersion of AirPollutants and their Penetraiton into the Local Envi-ronment. [Available online at www.dapple.org.uk.]

F IG . 22. FEFLO-Urban simulations of tracer gas dispersion for a west-northwest wind direction. There is a continuous release from fivepoint sources near street l evel (on sidewalk off four corners of MSGand on sidewalk north of One Penn Plaza) as used during MSG05.The figure presents the plume concentrations at a time of 1000 safter the release is initiated.


8/13/2019 BAMS-87-12-1713

14/14

bams-87-12-1713

Documents