Impact of Coupled CFD (urban Scale) and WRF model (MESO-GAMMA scale)SIMULATIONS ON Urban Scale Contaminant Transport and Dispersion

Mukul Tewari[1]

NationalCenter for Atmospheric Research, Boulder, Colorado

William J. Coirier

CFD Research Corporation, Alabama

Fei Chen

NationalCenter for Atmospheric Research, Boulder, Colorado

Sura Kim

CFD Research Corporation, Alabama

SuBmitted to

GRL

FINALDraft

7 May 2008

AbSTRACT

We present the results from a study designed to evaluate the impact uponurban area transport and dispersion modeling accuracyby a CFD (Computational Fluid Dynamics) model whenit is coupledtoaNumerical Weather Prediction (NWP) model. The CFD model taking part in the evaluation was the CFD-Urban model, while the NWP model was the Weather Research and Forecasting (WRF) model. Three different approaches of supplying boundary condition data to the CFD model were evaluated by comparing the resulting tracer gas transport fields to field data; (i) using data from a single sounding location, (ii) using the data from WRF/WRF_UCM model while performing the CFD in unsteady mode and (iii), using the data from WRF/WRF_UCM model while performing the CFD in quasi-steady mode. The WRF/WRF_UCM model and the CFD model results were evaluated against the data from VTMX 2000 and URBAN2000 field experiments. It was found that CFD model shows significant improvement when using wind fields produced by downscaling boundary condition data from the WRF/WRF_UCM model while operating in a quasi-steady mode.

1. Introduction

In the past two decades, flow and dispersion studies in urban regions have madesignificant improvements in both accuracy and speed. A common problem encountered when using CFD or any other microscale model in an urban area is the proper specification of boundary and initial conditions.Chan and Leach (2007) developed a CFD model called Finite Element Model in 3-Dimensions (FEM3MP) to simulate airflow and dispersion of chemical/biological agents released in urban areas. They evaluated the model with observations from the Intensive Operating Period (IOP) 3 and 9 of the URBAN2003 field study conducted in Oklahoma city, Oklahoma. The results from the model compared reasonably well with the observations. Warner et al (2004) compared the transport and dispersion using HPAC model and used the URBAN 2000 data for SF6 release. Zhong and Fast (2002) conducted an inter-comparison of MM5, RAMS and Meso-Eta model using VTMX data for the SaltLakeValley. They found that large forecast errors exist even after using fine spatial resolution and indicated that for accurate local forecasting, improved model parameterization of longwave radiation and turbulent mixing is needed.Bacon et al 2000 in their study described an OMEGA model and it’s Atmospheric Dispersion Model which can adapt its gridboth statically and dynamically to different phenomenon suchas fronts, clouds, hurricanes, and plumes. Chin et al (2005) evaluated the performance of an Urban Canopy Parameterization (UCP) in the COAMPS model and compared the results obtained with and without using UCP with the IOP10 of URBAN2000 and VTMX data occurring from 25-26 Oct 2000. They found a close agreement between predicted and observed tracer concentration. Holt and Pullen (2007) coupled two different UCMs (W-UCM which is based on Kusaka’s UCM (Urban Canopy Model) and BW-UCM which is based on Brown and Williams (1998) multilayer urban parameterization) with the COAMPS model for a study over the New York City region and compared their performances. Miao et al. (2007) has used the WRF/UCM to study the urban heat island over the Beijing (China) region.

Modeling of atmospheric dispersion involves atmospheric motion of all scales.It is required that the transport and dispersion model should be initialized with more accurate initial conditions and use appropriate boundary conditions supplied on the lateral extents of the computational domain.. None of the earlier studies have attempted to explore the coupling of a mesoscale model with the urban scale model although it is well known that each of these modeling scales lack the information the other needs. In one of the first effort of its kind, to improve the initial conditions supplied to the Transport and Dispersion (T&D) models, the Weather Research and Forecasting (WRF) model was run at a sub kilometer resolution and the meteorological fields from WRF model were used to supply boundary conditions to the CFD model. The purpose of this study isto evaluate the improvement in the simulation of urban area contaminant transport and dispersion modeling accuracy using a combined mesoscale and microscale (CFD) modeling approach. By statistically comparing the computed transport and dispersion fields using different methods of applying boundary conditions, we are able to quantify potential improvements.

The coupling of urban area and mesoscale models offers the advantage of improving the performance of both models. Mesoscale models can provide more accurate prevailing conditions to the urban scale model, while the latter can provide detailed parameterization of urban processes as a feedback to the mesoscale model. In the present study, WRF and WRF_UCM modelswere applied at the meso-gamma scale (with grid spacing of 500 m) to the complex urban environment over the Salt Lake City (SLC) region. The objectives of the study were to:1) evaluate the simulation of wind fields by WRF/WRF_UCM over SLC region , 2) test the feasibility of providing the WRF-simulated meteorological fields as input to the Computational Fluid Dynamics (CFD)-based urban area transport and dispersion model (CFD-Urban), and 3) explore ways to couple WRF/WRF_UCM with CFD-urban in order to improve the performance of CFD-urban.

Section 2 describes the models used in the present study followed by a brief description of the URBAN2000 field experiment. Section 4 discusses the numerical experiments performed in the study. The next section describes the results followed by the conclusions.

2. Numerical Modeling System

2.1Numerical Models

The numerical models used for the current study were the WRF and CFD-Urban models. A brief description of these models and their configurations are as follows:

2.1.1WRF model

WRF is a joint development effort among NCAR (NationalCenter for Atmospheric Research), AFWA (Air Force Weather Agency), NCEP (NationalCenter for Environment Prediction) and other government agencies and university community. The document of the WRF can be found at WRF is a completely redesigned code, has superior numerics, targeted for the 1-10 km grid-scale and intended for operational weather forecasting, regional climate prediction, air-quality simulation, and idealized dynamical studies. In this project, we used the research-quality version of the WRF model with nesting capability (WRF V2.0) released in May 2004. The following WRF configurations were used for this study: Non-hydrostatic dynamics, two-way interactive nesting procedure, radiative upper-boundary condition, time-dependent lateral-boundary conditions, relaxed toward large-scale model forecasts, new Kain-Fritsch (Kain and Fritsch 1993) cumulus parameterization on 10 km grid-increment, or larger, grids, Liu et al. (1993) mixed-phase ice scheme, Mellor-Yamada-Janjic TKE level 2.5 (Janjic, 2002) planetary boundary-layer parameterization, cloud-radiation scheme of Dudhia (1989) for shortwave and the Rapid Radiative Transfer Model (Mlawer et al. 1997) for longwave, Noah land-surface model (a modified Oregon State University land-surface model, Chen and Dudhia 2001).

2.1.2Urban Modeling in WRF

Two types of urban parameterization schemes were used in the WRF/Noah coupled system. : 1) a simple urban treatment in the Noah LSM (Liu et al., 2005), and 2) a coupled Noah/UCM. The UCM is well documented in Kusaka et al (2001, 2004). The UCM is implemented in WRF/Noah (Chen et al. 2004, 2006) and released in Dec 2006.One of the main feature of the UCM is that in this model all the urban effects in the vertical are considered to be subgrid scale. The UCM assumed that all the urban processes in the vertical are happening below the lowest model level. Some of the features of the UCM are: it has user defined canyon orientation, shadowing from buildings and radiation reflection, multi layer heat transfer equation for roof, wall and road surfaces, and wind profile in the canopy layer.

2.1.3CFD model

CFD-Urban is a suite of Computational Fluid Dynamics modeling software that is being used to simulate the wind, turbulence and dispersion fields in urban areas (Coirier, et al., 2002, 2003, 2005, 2006.a,b). CFD-Urban has been developed under a program sponsored by the Defense Threat Reduction Agency, and has been built using parts of a commercially available software suite, CFD-ACE+. It solves the Reynolds-Averaged Navier-Stokes (RANS) equations using a collocated, Finite-Volume method implemented upon structured, unstructured and adaptively-refined grids using a pressure-based approach based upon the SIMPLE algorithm. Turbulence closure is found by solving a variant of the standard k- model. Buildings are modeled either explicitly, by resolving the buildings themselves, and/or implicitly, by modeling the effects of the buildings upon the flow by the introduction of source terms in the momentum and turbulence model equations. CFD-Urban solves the steady-state and unsteady RANS equations. Since CFD-Urban solves the governing mass and momentum conservation laws at scales smaller than the buildings themselves, important urban aerodynamic features are naturally accounted for, including effects such as channeling, enchanced vertical mixing, downwash and street level flow energization.

2.2Coupling of WRF and CFD models

WRF forecast data was made available at 15-minute intervals on disk for subsequent use by the CFD model. In this study, we evaluate loosely coupling the two models by using a file-based data transfer mechanism. Two different strategies of coupling the CFD and WRF models have been addressed: An unsteady approach and a “quasi-steady” approach. Both of these approaches spatially interpolate data from the WRF data files and apply the data as boundary conditions to the CFD model. For the unsteady approach, the CFD model is run in an unsteady (ie: time-accurate) fashion, while in the quasi-steady approach, wind fields are computed at 15 minute intervals, stored to disk, and used in the transport and dispersion solution via the frozen hydrodynamics approach described in Coirier et al., 2006.a.

2.2.1Spatial interpolation of WRF Data

WRF uses a staggered grid approach (Arakawa C-staggering), where thermodynamic data are stored at the cell-centers of the mesh, and the velocity fields are stored at the face-centers of the mesh. In order to simplify the interpolation procedures, we average the velocity components to the cell centers. Furthermore, we store the mesh dual (cell-centers) as nodes in the on-disk data representation, which simplifies both visualization and processing of the data, since all data stored at the same location. The field data is spatially interpolated from the hexahedral WRF cells to the individual face centroids in the CFD mesh using a continuous, linear interpolant. The particular data read in from WRF includes the velocity components, pressure base state and perturbation potential, temperature, turbulence kinetic energy from the Mellor-Yamada-Janjic model and the momentum diffusion coefficient. The turbulence kinetic energy dissipation rate for the k-e model is found from the definition of the diffusion coefficient:

In this formula, the density is computed from a perfect gas (dry air) equation of state, k is from the Mellor-Yamada-Janjic model, and is the momentum diffusion coefficient. We have noticed that the dissipation rates determined from this relation are higher than what is predicted from equilibrium theory, and typically see unrealistic turbulence kinetic energy and dissipation rate fields from the CFD solution. We defer to future studies to improve this behavior.

2.2.2Imposition of WRF pressure gradient

We impose the WRF pressure gradient onto the CFD model by finding the difference in the imposed WRF pressure from the local pressure that would be present in an ideal atmosphere, and supply this pressure difference on the boundaries of the CFD model. Furthermore, we operate the CFD model in a constant density mode and do not apply a gravitational source term to the z-momentum equation. This approach was chosen after a considerable effort was made running the CFD solver with the gravitational source term included,, since it was found that directly coupling it to the thermodynamic field from the mesoscale model was problemmatic.. If the direct, hydrostatic (ie: gravitational source term include) coupling is made, both models must have consistent thermodynamic models (including humidity transport and equations of state), as well as having similar air to ground heat transfer models. Small differences in these thermodynamic quantities can produce unrealistic flow behavior in the CFD model. After a series of computations, we found that the best approach is to apply the difference of the local WRF pressure to that of an ideal atmosphere, which the CFD model uses as the pressure difference from the (constant) CFD reference pressure. That is, impose on the face-centers of the CFD mesh:

2.3Unsteady coupling

The unsteady coupling mode operates the CFD solver in an unsteady fashion, and linearly interpolates the WRF data in time from 15 minute storage intervals onto the boundary faces of the CFD mesh. This approach is quite costly, as the CFL restrictions limit the allowable time step to be much lower than the mesoscale model time step, and the CFD solver must solve the mass constraint, momentum conservation and turbulence model equations at each time step. The contaminant transport equation is solved in an unsteady mode, along with the other model equations.

2.4Quasi-steady coupling

The quasi-steady mode first computes the steady state, equilibrium, flow fields at 15 minute intervals, using the WRF data as boundary conditions. The unsteady, contaminant transport evolution equation is then solved using the quasi-steady velocity and turbulence field that is found by linearly interpolating the appropriate steady state velocity and turbulence fields in time. We call this collection of steady-state wind fields a “wind field library”, and the blending of these wind fields in time to solve the contaminant transport equation, the “unified frozen hydrodynamic solver”, as noted in Coirier et al., 2006.a.

3URBAN 2000 field experiment

The URBAN 2000 field experiment was sponsored by U.S. Department of Energy’s Chemical and Biological National Security Program. The URBAN 2000 meteorological and tracer field campaign was conducted during October 2000 and was highlighted by seven night-long intensive operations periods. The objective of this Urban 2000 Field Test was to investigate transport and diffusion around a single downtown building, within and through the downtown area and into the greater Salt Lake City urban area. Doran et al (2002) described in detail about the field campaign, instrumentations and measurment strategies, weather conditions during the IOPs of the URBAN 2000 experiment. The data from this field experiments have been used by many researchers to understand the meteorological and fluid dynamical processes governing dispersion in urban areas e.g.Hanna et al (2003a), Zhong and Fast (2003).

IOP10 was performed on October 25 to October 26, 2000.Sixteen MET data files are available and the data nominally covers the period 1200 MST October 25, 2000 (DOY 299) through 1200 MST October 26, 2000 (DOY 300). SF6 was released, as a point source at 1 rate of 1g/s, three times over the entire IOP, with the release “on” for 1 hour, and then “off” for one hour, for a total IOP of 6 hours.

We have used the URBAN 2000 Field Test data from Intensive Operating Period (IOP) 10 to quantifiably measure the accuracy of the transport and dispersion modeling using different types of downscale/input data from WRF. The WRF/WRF_UCM model results are verified with the MET observations.

4Numerical Experiments

The integrated WRF_UCM and WRF models were configured with five two-way interactive nested grids having grid spacing of 40.5,13.5,4.5,1.5, and 0.5 Km, as shown in Figure 1. It has 154 grid points in the east-west direction and 250 grid points in the north-south direction for the finer grid of 0.5Km horizontal resolution. There were 31 vertical levels with 16 levels within the lowest 2 km. The WRF model was initialized at 00 UTC 26 October 2000 (IOP-10 period) and applied to perform a 24-h simulation. The two WRF runs which were used to initialize the CFD model are: 1) WRF with simple urban treatment identified as WRF in this text, and 2) Coupled WRF/UCM run called WRF_UCM. The initial and lateral boundary conditions for WRF/WRF_UCMruns were supplied by analyses and forecast from the NCEP Eta data assimilation system (EDAS). For this study, a combination of USGS 24-category 1-km dataset and a gridded high resolution (30-meter) urban land-use data with detailed classifications for the SLC urban zones (low-intensity residential, high-intensity residential, and the industrial/commercial zone) which was aggregated into the WRF nested domain-5 with 0.5 Km grid spacing (fig 1b) was used.