Bridging the Gap Between Statistics and Engineering Statistical calibration of CFD simulations in Urban street canyons with Experimental data Liora Malki-Epshtein and Serge Guillas With Nina Glover, Stella Karra

Outline Background: The challenges – measuring and modelling urban airflow and pollution dispersion Simple Urban streets Complex Urban streets Our study Our methods What we can achieve

Challenges of Measuring Urban Air Flows  Airflow, meteorological variables and pollution are difficult and expensive to measure.  Few monitoring stations, equipment is normally installed on rooftops high above the ground  Urban geometry is very complex  Large and dense population combined with many sources of pollution in a relatively small geographical area.  Result: Low resolution measurements in the urban environment, capturing mainly the background Numerical models produce detailed three dimensional outputs that can be explored in depth.

*Some* Challenges in CFD Modelling of Urban Airflows – turbulence models  Direct Numerical Simulation of turbulence is still impossible at this scale. Simplifications are needed – turbulence models  The standard k-ε model most commonly used for urban flow and dispersion, cheap and fast to run  The default parameters of the model are based on best fit to a wide range of applications in mechanical engineering, not necessarily suitable for urban flows  Weakness: lack of universality - unreliable for flows with different geometry than those used to develop the model.  Poor performance compared with more complex models such as LES (Large Eddy Simulation)  Performance improved by adjusting the default model parameters Even the most basic, idealised urban streets are a challenge to model

Urban Airflow and Dispersion  Previous research: simple models for street canyons with a simplified geometry  Street canyons classified by the ratio of Height to Width  Deeper street canyons are poorly ventilated  Accumulation of pollution and heat Airflow over building arrays with increasing H/W. (Oke, 1988)

But: Real Streets are More Complex London Nicosia CO data at 1.5, 2.5 m height – higher exposure on the ground Wind speed profiles

Our Project To develop a technique to improve models of air flow throughout complex urban spaces, based on a combination of CFD simulation and field and laboratory observations, integrated using Bayesian statistical methods. Calibration of the numerical model parameters in CFD by data from lab and field measurements. Better understanding of where to position monitoring equipment in the field based on laboratory models.

A Day in the Life - CFD Research ANSYS CFX software

Field Measurements Nina on the roof of a church in South London 2-D and 3-D sonic anemometers to measure wind speed and direction CO monitors to measure pollution levels, as a passive (chemically inert) tracer following the airflow

Experimental Setup Stella setting up her experiment PIV and PLIF measure velocity fields and dye concentrations Low turbulence flume in CEGE Fluids lab Laser system

Comparing Different Street Geometries Symmetrical street canyon Cross section of the street

Comparing Different Street Geometries Step-down street canyon Cross section of the street

Comparing Different Street Geometries “Real” street canyon Cross section of the street

Airflow and Pollution Dispersion in a Complex, “Real” Street Canyon Dye concentration (in colour) and velocity arrows, calculated from PLIF and PIV Fluid flow visualised with fluorescent dye and laser

CFD Model Testing and Validation Different turbulence models and boundary conditions yield different results Difficult to match model outputs to experiments even for a simple flow Difficult to reproduce turbulence patterns within street canyons

Model Calibration Identify the parameters that give the best model outputs Known parameters of the experiment set up: geometry and typical length of the street canyon Unknown calibration parameters: turbulent kinetic energy, velocity profiles – tested in the pilot study last year Serge Guillas, Department of Statistical Science The next step: Calibration of the model coefficients - the parameters that are the building blocks of the numerical model An iterative process between the collaborators …

Evaluation of Model Errors The statistical calibration results in estimates of uncertainties of the model and of the calibration parameters.

Where is all this going? Our immediate goal: to help end users make informed choices about which numerical CFD model to use in which situation and where more accurate models, at greater cost, need to be embedded. The Urban environment requires a different approach than that adopted by the Meteorology community. We are integrating a variety of modelling and measuring techniques, in order to represent accurately the Urban micro-climate.

Conclusion Ultimately, modelling air flow and pollution dispersion should lead to better design of urban spaces – to be better ventilated, accumulate less heat, use energy more efficiently and be better observed and monitored on a regular basis. We aim to develop fundamental building blocks towards achieving this.