Development of a large eddy simulation model for the study of pollutant dispersion in urban areas

School of Doctorate in Industrial and Environmental Fluid Dynamics University of Trieste Development of a large eddy simulation model for the study of pollutant dispersion in urban areas Supervisors by V. Stocca Assistant supervisors Prof. V. Armenio, UT Prof. K. R. Sreenivasan, NYU Dr. D. B. Giaiotti, ARPA-FVG Dr. S. Del Frate, ARPA-FVG Final Presentation - April 29 2010, Trieste

Research project partially supported by: Agenzia Regionale per la Protezione dell Ambiente del Friuli Venezia Giulia ARPA-FGV University of Trieste Abdus Salam International Centre for Theoretical Physics

Outline 1) Motivations 2) LES-AIR code: a) mathematical & numerical model; b) improved wall model modulus. 3) Validation of the wall models for the body-fitted grid: a) channel flow; b) Ekman layer; c) atmospheric boundary layer. 4) Validation with the immersed boundaries: a) CEDVAL wind tunnel experiment; b) MUST field campaign. 5) Flow & dispersion in the Servola-Valmaura area of Trieste. 6) Conclusions.

1) MOTIVATIONS 2) LES-AIR code: a) mathematical & numerical model; b) improved wall model modulus. 3) Validation of the wall models for the body-fitted grid: a) channel flow; b) Ekman layer; c) atmospheric boundary layer. 4) Validation with the immersed boundaries: a) CEDVAL wind tunnel experiment; b) MUST field campaign. 5) Flow & dispersion in the Servola-Valmaura area of Trieste. 6) Conclusions.

URBAN AREAS PROBLEMS OF AIR & WATER POLLUTION NEW MULTIDISCIPLINARY FOCUS AREA URBAN FLUID MECHANICS (Fernando et al. E. F. M., 01) flow patterns around buildings transport of pollutants study of pollution sources from traffic, heating systems.. air quality studies

NUMERICAL SIMULATIONS VERY HELPFUL IN AIR QUALITY STUDIES Some examples Air Quality Report ARPA-FVG 09 Brown et al., ASCE 01 Tseng et al., E. S. T. 06

Approach to urban simulations Dispersion models in order of increasing computer cost: Gaussian empirical, semi-empirical RANS Large Eddy Simulations (LES) LES: big scales solved directly, small, more isotropic, scales parametrized with a subgrid model TURBULENT, THREE DIMENSIONAL, NON STATIONARY FLOWS WITHOUT NEED OF A-PRIORI PARAMERIZATIONS LES computationally expensive, in the past applied only to idealized flow configurations but with increased CP power URBAN LES SIMULATION SUBMESO, COAMPS, HIGRAD-LES, FEM3MP, FAST3DCT-MILES

LARGE EDDY SIMULATION IN AIR QUALITY STUDIES For regulatory purposes usually two kind of studies required: 1. Long term 2. Short term MICROSCALE SIMULATIONS: Domain of some kilometers squares Period of 1 day Estimation of dry-wet deposition LES SIMULATIONS

GOAL OF THE THESIS To develop a LES model for high resolution air quality studies to predict flow and dispersion in urban areas LES-AIR model derived from LES-COAST model (Roman et al., I. J. H. F. F. 10) adapting it to atmospheric flows LES-AIR expected to be used by ARPA-FVG for short term analysis of pollutant dispersion

1) motivations 2) LES-AIR code: a) mathematical & numerical model; b) improved wall model modulus. 3) Validation of the wall models for the body-fitted grid: a) channel flow; b) Ekman layer; c) atmospheric boundary layer. 4) Validation with the immersed boundaries: a) CEDVAL wind tunnel experiment; b) MUST field campaign. 5) Flow & dispersion in the Servola-Valmaura area of Trieste. 6) Conclusions.

LES-AIR GOVERNING EQUATIONS Continuity, momentum and density eqs. under Boussinesq approx.: SGS CONTRIBUTIONS LES FILTERING SMAGORINSKY MODEL j Pr EDDY DIFFUSIVITY MODEL K SGS T T x Pr K T T j T x const j

LES-AIR NUMERICAL METHOD CURVILINEAR COORDINATES Collocated grid, fractional step formulation by Zang et al., J. C. P. 94 SPATIAL INTEGRATION second order central finite differences in computational space (ξ, η, ζ). Optionally, quadratic upwind interpolation (QUICK) for convective fluxes. TEMPORAL INTEGRATION second order explicit Adams-Bashfort implicit Crank-Nicolson viscous diagonal terms OVERALL SECOND ORDER ACCURACY

BODY-FITTED GRID Curvilinear grid base to reproduce terrain slopes like hills, valleys How to take into account the smaller obstacles that are usually the anthropic constructions? IMMERSED BOUNDARY (IB) METHOD Identification of solid regions inside the computational domains separated from the fluid through an interface surface ψ

LES-AIR: IMMERSED BOUNDARY METHOD IB nodes identified during pre-processing phase Additional forcing term in momentum equation to account for solid region presence (Fadlun et al., J. C. P. 00, Roman et al., C. & F. 09) PP velocity found from V velocity through interpolation IB velocity found from PP velocity considering no slip-condition at ψ, through linear interpolation fitting PP velocity to a log profiles and deriving the IB velocity value Besides, intensification of SGS eddy viscosity at IB to account for right wall stress (Roman et al., P. o. F. 10)

EULERIAN TREATMENT of DISPERSED PHASES Density equation ha been extended to compute the dispersion of a set of Ns scalars Different sets of BC for each pollutant SGS contribution same as density equation Important now possible to compute dispersion of more than one pollutant considering at the same time also the air thermal stratification

NEED FOR WALL MODELING LES: FLOW INTEGRAL SCALE HAS TO BE LARGER THAN GRID SIZE Near solid walls grids with resolution of the order of the wall unit This resolution impossible in atmosphere 1) too computationally expensive 2) useless: in practical applications wall roughness In the last decades many wall-layer models for LES to overcome wall layer resolution problem (off-wall, equilibrium wall stress, hybrid RANS-LES) Each one suited for a particular configuration still lack of a general model (Piomelli & Balaras An. Rev. F. M. 02, Piomelli, P. A. S. 08)

LES-AIR WALL MODELS AT BODY-FITTED GRID VERY IMPORTANT PROPER DESCRIPTION OF MOMENTUM & HEAT FLUXES AT THE GROUND GOAL: TO FIND WALL MODELS TO BE USED IN ATHMOSPHERIC FLOWS FOR MOMENTUM AND HEAT FLUXES MOMENTUM WM stress imposition at the wall modification of eddy viscosity HEAT WM Heat flux imposition at the wall modification of eddy diffusivity

EQUILIBRIUM STRESS MOMENTUM WALL MODEL Hypothesis equilibrium layer assumption (true in atmospheric flows) very coarse grids Instantaneous velocity value Vp at first off-wall node is assumed to belong to logarithmic velocity profile: SMOOTH WALLS ROUGH WALLS 3- z- w 2- y- v 1- x- u 2 W / u the stress * to be imposed as boundary condition is derived and directed along the tangential velocity direction:

EQUILIBRIUM STRESS MOMENTUM WALL MODEL A correction is made also on the SGS Smagorinsky model The dominant term in the S evaluation is derived analytically from log-law: z 11 z 11 SMOOTH WALLS Near wall RANS-like eddy viscosity model: gives more correct near wall integral scale; it avoids the usage of the wall velocity value (very wrong for very coarse grids)

TEMPERATURE WALL MODEL temperature as passive scalar representative for dispersed non reacting pollutants In analogy with the momentum wall model the temperature value Tw at the first off-wall node is assumed to belong to the analytic logarithmic Kader law (Kader I. J. H. M. T., 81) 1 / 2.12 ln( 1 z ) C e z e T T W T Pr * T q / w * u BC at walls: Tw fixed qw derived or qw fixed Tw derived Pr const. SGS k T T correction on the νt calculus influences also SGS heat flux

TURBULENT PLANE CHANNEL FLOW Constant pressure gradient, smooth walls, Domain 2π/h x 2 x (2/3)π/h solved with 32³ equi-spaced nodes Re*= 4000 Δx+= 785 Δy+= 262 Δz+= 250 Re*=20000 Δx+=3927 Δy+= 1309 Δz+=1250 Re * u h INVESTIGATION OF the performance of the improved momentum wall model the best performing Smagorinsky model constant the best performing interpolation for convective terms (2 nd order central vs. QUICK) the dependence of the solution on the time step

TURBULENT PLANE CHANNEL FLOW Re =4000 * Re =20000 * rms velocities mean velocities

TURBULENT PLANE CHANNEL FLOW ALSO TESTED: ROUGH WALLS Z0 = const PrSGS = 0.5, 0.8, FORCED CONVECTION 0.9, 1.1 PrSGS = 0.5, 0.9

TURBULENT PLANE CHANNEL FLOW CONCLUSIONS very bad velocity prediction with QUICK scheme good prediction of velocity profiles and first order statistics with central scheme and CFL<0.3 momentum wall model worked well both for smooth and rough walls spurious merging region between near wall RANS-like model and outer LES model due to unphysical streaks formation promising results for temperature wall model when PrSGS = 0.5 good results both a Re*=4000 & Re*=20000

LOW REYNOLDS EKMAN LAYER Idealized atmospheric boundary layer G GOAL to test momentum wall model in case of rotation Ω Results compared with DNS of Spalart et al., P. o. F 09 α CHARACTERISTICS z y x PRO comparison with DNS CON wall model tested at very low Re GD ReE 2000 Neutral stratification Polar case Smooth walls TESTED DIFFERENT grid resolutions Aspect Ratio (AR) values of Smagorinsky constant (Cs)

LOW REYNOLDS EKMAN LAYER SIMILARITY LAWS (Coleman, J. A. S. 99) Reference values: 16.97, 14.21, u* / G 0. 0461 EQ Resulting values Best results lowest horizontal resolution, Cs=0.1 cases CD2 CD5 In this situation rotation angle α and the friction velocity are well reproduced Not strong influence from aspect ratio

LOW REYNOLDS EKMAN LAYER MEAN VELOCITIES Peak of fluctuations shifted away from the wall like lower Re flow Bagget, A. R. B. 98 Outer diagonal Re stresses well reproduced with SGS correction DIAGONAL STRESSES OUTER DIAGONAL STRESSES

FULL SCALE ATMOSPHERIC BOUNDARY LAYER (ABL) G Ω Reproduces the full scale ABL LES simulation of Moeng & Sullivan, J. A. S. 94 α T(z) Ekman layer whose depth is contained by a capping inversion at Re typical of atmospheric flows z y Rough momentum wall model at grid base x Mid-latitude case

FULL SCALE ABL Only qualitative comparison because of difficulty to average on the same stage of the transient Good agreement of both first and second order statistics Main differences in the upper part of the BL where there is the capping inversion both for velocities & fluctuations

CEDVAL WIND TUNNEL EXPERIMENT flow & dispersion around isolated building-like obstacle reproduced with IB (1:200) rough wall model for momentum at grid base, grid clustered near obstacle difficult case for IB: obstacle presents sharp edges Tested three ways to find velocity at IB nodes: linear interpolation (LIN cases) tangential velocity logarithmic, normal velocity parabolic plus eddy viscosity enhancement (Roman et al., P. o. F. 09) (WM cases) LIN WM modified version of (Roman et al., P. o. F. 09): normal velocity scaled of the same factor of tangential velocity (WM-MOD cases)

When flow invests a bluff body reproduced with IB spurious numerical oscillations are generated from the upwind edge. Two strategies: QUICK dissipative scheme Centered 2 nd order interpolation with 4 th order implicit filtering in x,z and 6 th order explicit filtering (Lele, J. C. P. 92) in the y (parallelization direction).

CEDVAL WIND TUNNEL EXPERIMENT F-LIN F-WM F-WM-MOD Height of downwind cavity larger in LIN No big differences between WM and WM-MOD streamlines In WM-MOD less oscillations

CEDVAL WIND TUNNEL EXPERIMENT STREAMWISE VEL. VERTICAL VEL. ORIZZONTAL PLANE XY VERTICAL PLANE XZ Recirculation on the roof not predicted General good agreement with experimental data

CEDVAL WIND TUNNEL EXPERIMENT Emission of a scalar from the building downwind edge like emission from a parking garage TOP VIEWS SIDE VIEWS tracer emitted from four IB nodes placed near ground on the downwind side qualitative comparison because of different tracer concentrations at source with filtering high level of concentration in a broader region Some differences on the lateral building side

MUST FIELD CAMPAIGN: case α=-41 low rising neighborhood neutral stratification H/W=0.19 isolated roughness flow regime wind α IB rough momentum WM at grid base + WM at IB (Roman et al., P. o. F. 09) full scale simulation Forcing: ΔP=const with ΔPx, ΔPy having intensities derived from measured friction velocity to reproduce α=-41 Both QUICK and CS+FILTER tested

M U S T α=-41 CASE DIMENSIONAL NON DIMENSIONAL TKE MEAN VELOCITY

M U S T α=-41 CASE CS+FILTER QUICK Flow channeling in y direction near the ground like in Yee & Bitolft, B. L. M. 04 QUICK destroys the smaller turbulent structures CS+FILTER should be preferred because reproduces a broader range of turbulent structures important for stratification

SERVOLA-VALMAURA SIMULATION WIND outflow (prerun) Domain 1.5km x 1km x 0.6 km neighborhood scale simulation Full-scale, neutral stratification, idealized wind from SOUTH-WEST buildings IB (lin interp.), topography curvilinear grid base grid resolution Δx=5.9m, Δy=3.9m, Δz min=1m Δz max=30m studied flow and dispersion from a line source archetypal of emission from vehicular traffic

SERVOLA-VALMAURA SIMULATION AVERAGED VELOCITY IN THREE AREAS CANOPY LAYER SURFACE & INERTIAL LAYER A B C

SERVOLA-VALMAURA SIMULATION TWO ANIMATIONS OF VERTICAL VELOCITY 2) vertical plane 1) curvilinear surface ~10m from ground

SERVOLA-VALMAURA SIMULATION emission from line source: instantaneous concentration after 1.4h z=0.5m z~10m vertical x-z plane in correspondence of the emission source

CONCLUSIONS GOAL: to develop a predictive LES model, LES-AIR, for the study of flow and dispersion in urban areas at neighborhood scale (short term air quality studies) Starting from LES-COAST a model suited for atmospheric flows was built improved wall models for momentum and heat at body-fitted grid suited for atmospheric flows extension of transport equation to the study of Ns scalars combined usage of curvilinear grid and IB to reproduce urban geometries An extensive set of validation test cases has been presented wall model at body-fitted grid: channel flow, Ekman layer, ABL IB test case: CEDVAL small scale wind tunnel experiment, MUST field campaign Best performing numerical set up was detected: CFL, Smagorinsky constant, IB velocity treatment Also different spatial interpolation schemes tested to damp spurious numerical oscillations : QUICK and CS+ filter QUICK too dissipative, to be used only at constant flow rate CS+filter retains also smaller turbulence structures, better for dispersion

CONCLUSIONS In the final part of the thesis the flow in the Servola-Valmaura suburban area of the city of Trieste was reproduced under idealized wind conditions LES-AIR proved to be a powerful tool in the determination of the bulk quantities of the flow over sloped terrains taking into account the exact building shapes LES-AIR was able also to reproduce in detail the dispersion path of a scalar emitted from a line source archetypal of emission from vehicular traffic Although under idealized conditions LES-AIR demonstrated its ability in dealing with very complex urban environment PERSPECTIVES Nesting from larger scales models Urban simulation with thermal stratification Inclusion of humidity

Thanks for your attention Special thanks to S. Dal Gesso who ran cases F-LIN & Q-LIN of CEDVAL simulations in the framework of the her master thesis