The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27,

2010

The Best Practise Guideline for the CFD simulation of flows in the urban environment : an outcome of COST 732
Jrg FRANKE a, Antti HELLSTEN b, Heinke SCHLNZEN c and Bertrand CARISSIMO d
Department of Fluid- and Thermodynamics, University of Siegen, Germany E-mail: joerg.franke@uni-siegen.de b Department of Applied Mechanics, Helsinki University of Technology, Helsinki, Finland E-mail: antti.hellsten@tkk.fi c Meteorological Institute, University of Hamburg, Hamburg, Germany E-mail: heinke.schluenzen@zmaw.de d CEREA, Teaching and Research Centre in Atmospheric Environment, Chatou, France E-mail: carissim@cerea.enpc.fr
a

ABSTRACT: This paper is a summary of the best practice guideline document (Franke et al., 2007) produced in the framework of the European COST Action 732 Quality assurance and improvement of micro-scale meteorological models, available from the site given in the reference section. The full document provides guidelines for undertaking simulations that are used to evaluate micro-scale obstacle-accommodating meteorological models. This paper provides an overview of the topics covered in the full document without reproducing the specific recommendations. In addition, just after the end of the action we give a first feedback on the usage made of this guide by the participants. 1 INTRODUCTION The main objective of the European COST Action 732 was the improvement and quality assurance of micro-scale obstacle-accommodating meteorological models and their application to the prediction of flow and transport processes in urban or industrial environments (Britter and Schatzmann, 2007). The name micro-scale obstacle-accommodating meteorological models is used to discern them from cloud resolving models which are called micro-scale models in meteorology. Micro-scale obstacle-accommodating meteorological models are in the following also often called Computational Fluid Dynamics (CFD) codes. The quality assurance of the application is closely related to the users' knowledge of the models. Actually, numerical simulation is mainly a knowledge based activity (Hutton, 2005; Coirier, 2005). The knowledge is, in general, most effectively transferred by the formulation of a best practice guideline (BPG) for the intended application, which is the prediction of dispersion in urban areas at neighbourhood and street scale within this COST Action. However, even for this well-defined application the formulation of BPGs faces the problem of giving general advice for specific problems that may vary substantially although belonging to the same field. The BPG of

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

the COST action 732 (Franke et al., 2007), which is summarised in the following, is therefore also not exhaustive but tries to cover as many aspects of the proper usage of CFD for the prediction of urban flows as possible. The full document provides guidelines for undertaking simulations that are used to evaluate CFD codes. This paper provides only an overview of the topics covered in the full document without reproducing the specific recommendations. In the full document those sources of error and uncertainty in CFD simulations, that can be controlled and quantified by the user are discussed in detail and best practice guidelines for their reduction and quantification are given. These BPGs are based on previous available guidelines as far as possible. For topics that have not yet been covered by existing guidelines further needs for research within the COST action 732 are indicated. The most general discrimination of these errors and uncertainties divides them into two broad categories (Coleman and Stern, 1997) with corresponding subdivisions. errors and uncertainties in modelling the physics: simplification of physical complexity usage of previous experimental data geometric boundary conditions physical boundary conditions initialisation numerical errors and uncertainties : computer programming computer round-off spatial discretisation temporal discretisation iterative convergence When performing validation simulations it is mandatory to quantify and reduce the different errors and uncertainties originating from these sources. In the full document (Franke et al., 2007) the errors or uncertainties listed above are further defined and explained. In this short summary best practice advice on how to avoid errors and where this is not possible how to estimate and reduce errors and uncertainties in the numerical solutions is provided partly. The BPG is meant to avoid or at least reduce what is known as user errors (Casey and Wintergerste, 2000). 2 BEST PRACTICE GUIDELINE

2.1 Review of existing guidelines There are several previous initiatives to establish best practice guidelines in the field of flow simulation in general and also for the application to the built environment.

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

For general CFD applications the ERCOFTAC Best Practice Guidelines (Casey and Wintergerste, 2000) is still the most complete document. Special problems of micro-scale meteorological applications are however deliberately not addressed. Best practice guidelines on CFD for wind engineering problems have been published by the Thematic Network for Quality and Trust in the Industrial Application of CFD (QNET-CFD) (Bartzis et al., 2004; Scaperdas and Gilham, 2004). Besides these European activities the Architectural Institute of Japan has conducted a cooperative project for CFD prediction of the pedestrian wind environment (Tominaga et al., 2008). For the same application a working group of the European COST action C14 Impact of Wind and Storms on City Life and Built Environment has compiled recommendations for conducting CFD simulations from a comprehensive literature review (Franke et al., 2004). Panskus (2000) suggests and applies test cases to evaluate micro-scale obstacle-accommodating meteorology models. The closely related guideline of the VDI (the German Association of Engineers) concentrates on evaluation and validation of these models for flow around buildings and obstacles (VDI, 2005). The guideline is structured according to the general steps of conducting a numerical simulation (Franke et al., 2004; Casey and Wintergerste, 2000; Menter et al., 2002). For each of these steps, which are shortly described in the paragraphs below, more detail can be found in the full document. This guideline differs from most of the guidelines and recommendations mentioned above by specifically addressing validation simulations, i.e. the careful comparison of the simulation results with experimental measurements. 2.2 Choice of target variables As proposed by Schlnzen (1997) and Menter et al. (2002) the first step in a validation simulation should be the definition of the target variables. These should include the variables that are representative of the goals of the simulation and those that can be compared with the corresponding experiments. 2.3 Choice of approximate equations describing the physics of the flow The choice of the basic equations has the largest impact on the modelling errors and uncertainties referring to the physics. The turbulent flow within urban or industrial environments is in general modelled by the Navier-Stokes equations using one of the following closures for the turbulence: Steady Reynolds Averaged Navier-Stokes (RANS) Unsteady RANS (URANS) Large Eddy Simulation (LES) and hybrid RANS-LES approaches The guidelines focus on steady RANS simulations as they are the most common approach to date. The recommendations should however also be considered when using URANS or LES. 2.4 Choice of the geometrical representation of obstacles Normally the distribution of buildings has the greatest impact on wind flow patterns. Secondary influence factors in the urban area include vegetation, orography and surface characteristics (e.g. roads, grass, sand). The central area of interest should be reproduced with as much detail as possible. This naturally increases the number of cells that are necessary to resolve the details. The available resources often limit the details which can be reproduced.

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

2.5 Choice of the computational domain The size of the entire computational domain in the vertical, lateral and flow directions depends on the area that shall be represented and on the boundary conditions that will be used. More specific guidance is given in the full document for the Vertical extension of the domain. Lateral extension of the domain. Extension of the domain in flow direction, divided in the region in front and behind the built area. 2.6 Choice of boundary conditions The boundary conditions represent the influence of the surroundings that have been cut off by the computational domain. As they determine to a large extent the solution inside the computational domain, their proper choice is very important. Often, however, these boundary conditions are not fully known. Therefore the boundaries of the computational domain should be far enough away from the region of interest to not contaminate the solution there with the approximate boundary conditions. Inflow boundary conditions At the inflow an equilibrium boundary layer is usually prescribed with certain requirements that need to be fulfilled. Wall boundary conditions At solid walls a special treatment of the no-slip boundary condition by using wall functions is normally applied for the velocities if the resolution is not sufficient. In case of commercial CFD software attention must be paid to the correct definition of the roughness height. Top boundary conditions The choice of the top boundary condition is very important for sustaining equilibrium boundary layer profiles. Lateral boundary conditions In commercial CFD codes symmetry boundary conditions are normally used at the lateral boundaries when the approach flow direction is parallel to them. In micro-scale obstacleaccommodating meteorological models open lateral radiation boundaries are frequently used at the lateral boundaries. With these every horizontal boundary grid point can allow for inflow and outflow and this might also change in time (e.g. URANS applications). In both cases there are requirements for the minimum distance between the boundary and the area of interest. Outflow boundary conditions At the boundary behind the obstacles (where all or most of the fluid leaves the computational domain) open boundary conditions are used in commercial CFD and micro-scale obstacleaccommodating meteorology models. This boundary should be ideally sufficiently far away from the built area to not have any fluid entering into the computational domain through this boundary.

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

2.7 Choice of initial data In RANS, URANS and LES models a boundary and initial value problem has to be solved numerically. The larger the model domain or the smaller the wind speed, the more relevant the initial data become. For steady RANS stationary solutions are searched thus the iteration is stopped as soon as the solution is not changing any more or the iterative solution converges. In these cases mainly the boundary values influence the model solution and the impact of the initial data is small. Initialising with a flow field that is close to the final solution will reduce the computational efforts needed to reach stationary solutions. For URANS and LES, the initial data determine the time dependent development in the beginning of the simulation. As a rule of thumb the impact time can be estimated with a relation including the domain size and wind speed. During this initial period the model results are very dependent on the initial data and should not be interpreted as solution which reflects the final flow. Initial data and inflow data are very often used as one and the same. Therefore the current best practice advice is to keep initial data uncertainty as little as possible and keep in mind that the initial data influence the model results in unsteady simulations. 2.8 Choice of the computational grid When referring to the computational grid one first has to define the discretisation method that shall be used for the basic equations. The following discussion is restricted to the closely related Finite Volume and Finite Difference methods with a strong bias towards the Finite Volume method as this is widely used in commercial CFD codes and micro-scale obstacleaccommodating meteorological models. For the Finite Element discretisation method different requirements exist for the quality of the computational grid (Casey and Wintergerste, 2000). With the Finite Volume and Finite Difference methods the computational results depend crucially on the grid that is used to discretise the computational domain. The grid has to be designed in such a manner that it does not introduce errors that are too large. This means that the resolution of the grid should be fine enough to capture the important physical phenomena like shear layers and vortices with sufficient resolution. Methods to quantify the influence of the grid resolution are provided in the full document (Franke et al., 2007). In addition the grid should not be too much skewed and grid stretching/compression should be small in regions of high gradients to keep the truncation error small (expansion ratio between two consecutive cells should be typically below 1.3). 2.9 Choice of numerical approximations To render the basic equations solvable on the computer they have to be discretised and transformed into algebraic equations. The most important numerical approximation is the one used for the non-linear advective1 terms in the basic equations (see e.g. Cowan et al., 1997; Menter et al., 2002).

In engineering sciences advection is named convection (transport caused by the flow field). Since convection is dedicated in meteorology to describe a mostly unresolved vertical atmospheric movement forming in an unstable (convective) atmosphere, we use advection in this text.

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

2.10 Choice of the time step size When performing unsteady simulations, the size of the time step is another important parameter for the accuracy of the results. If the relevant frequency range can be estimated, then the highest frequency should be resolved with at least 10 20 time steps per period (Menter et al., 2002). Another method to estimate the time step in advection dominated problems is the relation t = CFL xmin / Umax, where xmin is the minimum grid width, Umax is the maximum velocity and CFL is the Courant-Friedrichs-Lewy number. 2.11 Choice of iterative convergence criteria Most of the computer programs use iterative methods to solve the entire algebraic system of equations or part of it (e.g. the equation for the pressure). Starting from an initial guess the flow variables are recalculated in each of the iterations until the equations are solved up to a userspecified error. The termination criterion is usually based on the residuals of the corresponding equations. These residuals should tend towards zero. Especially in complex geometries like urban areas the residuals may not drop below a user-specified error. Therefore it is also recommended to monitor the target variables as a function of iteration or time step to have additional information on the steadiness of the computed solution. This procedure should also be followed when unsteady simulations are to be performed. Implicit time integration methods require iterations within the time steps so the above should be applied within each time step. 3 USER FEEDBACK During the COST action no special funding was provided to perform the simulations. However a number of participants performed quality assurance steps on the simulations of the two cases for which detailed information was provided. They both were wind tunnel physical models of real experiments performed in the US, firstly the Mock Urban Setting Test (dispersion around a regular array of container in the Utah desert) and secondly the Oklahoma Joint Urban 2003 field campaign. On the first case, a number of users did grid sensitivity analyses with different codes and some went as far as Richardson extrapolation. The influence of boundary conditions, especially upstream was studied by others. A team did a detailed study of the influence of the turbulent Schmidt number. Fort the second case, for which the simulations were started later in the action, fewer teams were involved. The mesh of this very complex case was already a big undertaking and therefore little was done, within the action, which required grid modifications. Rather the analyses performed by the different teams involved change of turbulence modelling, change of the modelling of the source, the influence of having open or closed garages in the city (especially around the release point) and the influence of the mean wind direction, for which a difference of 10 was shown to have quite a large influence (see Fig. 1). 4 CONCLUSIONS The best practice guideline, which we have briefly described here is a collection of results from former initiatives in the field of CFD in general and for its application to urban flows. The guideline focuses on applications of the statistically steady RANS equations for situations with neutral stratification without dispersion modelling. However, users of other models like unsteady RANS (URANS) and LES models should consider the same suggestions. Differences and some more

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

but not extensive information for URANS and LES applications are also given. The guideline provides general advice that should be taken into account when performing the simulations for model validation and has been tested within the COST Action 732. From the results of these validation simulations specific guidelines for the validation test cases and refined general guidelines have been produced in the course of the action. These include advice on pollution modelling within CFD codes and on the proper use of non-CFD codes. During the action, several groups applied these practise on two cases involving very complex urban geometries.

10120 8433 7028 5856 4880 4067 3389 2824 2354 1961 1634 1362 1135 946 788 657 547 456 380 317 264 220 183 153 127 106

WD=170o; flow z=3m; dispersion z=10m

WD=180o; flow z=3m; dispersion z=10m

WD=190o; flow z=3; dispersion z=10

Figure 1. Example of the influence of small mean wind direction changes on the flow and dispersion in Park Avenue of Oklahoma City (from wind tunnel measurements) and plotted with the same tool as used for all model comparison.

The Fifth International Symposium on Computational Wind Engineering (CWE2010) Chapel Hill, North Carolina, USA May 23-27, 2010

5 ACKNOWLEDGEMENTS The templates for analysis (visualization and statistics) of the CFD results were provided by Ruwim Berkowicz 6 REFERENCES
Bartzis, J.G., Vlachogiannis, D. and Sfetsos, A., 2004. Thematic area 5: Best practice advice for environmental flows. The QNET-CFD Network Newsletter, 2 (4), 34-39. Britter, R. and Schatzmann, M., eds., 2007. Background and justification document to support the model evaluation and guidance protocol, Brussels: COST office. http://www.mi.uni-hamburg.de/Official-Documents.5849.0.html Casey, M. and Wintergerste, T., eds., 2000. ERCOFTAC SIG "Quality and Trust in Industrial CFD": Best Practice Guidelines. ERCOFTAC. Coirier, W. J. (2005) Evaluation of CFD codes, US perspective, in Schatzmann, M. and Britter, R., Eds., Proceedings of the International Workshop on Quality Assurance of micro-scale meteorological models, Hamburg, Germany, July 28-29, pp. 15-20. Coleman, H. W. and Stern, F., 1997. Uncertainties and CFD Code Validation. Journal of Fluids Engineering, 119, 795-803. Cowan, I.R., Castro, I.P. and Robins, A.G., 1997. Numerical considerations for simulations of flow and dispersion around buildings. Journal of Wind Engineering and Industrial Aerodynamics, 67 & 68, 535-545. Franke, J., Hirsch, C., Jensen, A.G., Krs, H.W., Schatzmann, M., Westbury, P.S., Miles, S.D., Wisse, J.A. and Wright, N.G., 2004. Recommendations on the Use of CFD in Wind Engineering. In: J.P.A.J van Beeck, ed. Proceedings of the International Conference on Urban Wind Engineering and Building Aerodynamics: COST Action C14 - Impact of Wind and Storm on City Life and Built Environment, Rhode-Saint-Gense, Belgium, May 5 - 7, C.1.1 - C.1.11. http://www.costc14.bham.ac.uk/documents/Wg2/FinalDocument.pdf Franke, J., Hellsten, A., Schlnzen, H. and Carissimo, B., eds., 2007. Best practice guideline for the CFD simulation of flows in the urban environment, Brussels: COST office. http://www.mi.uni-hamburg.de/OfficialDocuments.5849.0.html Hutton, A., 2005. Evaluation of CFD codes The European Project QNET-CFD. In: M. Schatzmann and R. Britter, eds. Proceedings of the International Workshop on Quality Assurance of micro-scale meteorological models, Hamburg, Germany, July 28/29, pp. 11-14. Menter, F., Hemstrom, B., Henrikkson, M., Karlsson, R., Latrobe, A., Martin, A., Muhlbauer, P., Scheuerer, M., Smith, B., Takacs, T. and Willemsen, S., 2002. CFD Best Practice Guidelines for CFD Code Validation for Reactor-Safety Applications. Report EVOL-ECORA-D01, Contract No. FIKS-CT-2001-00154. Panskus, H., 2000. Konzept zur Evaluation hindernisauflsender mikroskaliger Modelle und seine Anwendung auf das Modell MITRAS. VDI-Fortschrittsberichte Reihe 7, Nr. 389, Dsseldorf:VDI. Scaperdas, A. and Gilham, S., 2004. Thematic Area 4: Best practice advice for civil construction and HVAC. The QNET-CFD Network Newsletter, 2 (4), 28-33. Schlnzen, K.H., 1997. On the validation of high-resolution atmospheric meso-scale models. Journal of Wind Engineering and Industrial Aerodynamics, 67&68, 479-492. Tominaga, Y., Mochida, A., Yoshie, R., Kataoka, H., Nozu, T., Yoshikawa, M. and Shirasawa, T., 2008. AIJ guidelines for practical applications of CFD to pedestrian wind environment around buildings. Journal of Wind Engineering and Industrial Aerodynamics, 96, in press. doi:10.1016/j.jweia.2008.02.058 VDI, 2005. VDI guideline 3783 Part 9: 2005-11, Environmental meteorology Prognostic micro-scale wind field models Evaluation for flow around buildings and obstacles. Berlin: Beuth. http://www.vdi.de/en/7636.0.html