Global sensitivity analysis of a 3D street canyon model—Part II: Application and physical insight using sensitivity analysis

Abstract

In this work global sensitivity studies using Monte Carlo sampling and high dimensional model representations (HDMR) have been carried out on the k– ε closure computational fluid dynamic (CFD) model MISKAM, allowing detailed representation of the effects of changing input parameters on the model outputs. The scenario studied is that of a complex street canyon in the city of York, UK. The sensitivity of the turbulence and mean flow fields to the input parameters is detailed both at specific measurement points and in the associated canyon cross-section to aid comparison with field data. This analysis gives insight into how model parameters can influence the predicted outputs. It also shows the relative strength of each parameter in its influence. Four main input parameters are addressed. Three parameters are surface roughness lengths, determining the flow over a surface, and the fourth is the background wind direction. In order to determine the relative importance of each parameter, sensitivity indices are calculated for the canyon cross-section. The sensitivity of the flow structures in and above the canyon to each parameter is found to be very location dependant. In general, at a particular measurement point, it is the closest wall surface that is most influential on the model output. However, due to the complexity of the flow at different wind angles this is not always the case, for example when a re-circulating canyon flow pattern is present. The background wind direction is shown to be an important parameter as it determines the surface features encountered by the flow. The accuracy with which this is specified when modelling a full-scale situation is therefore an important consideration when considering model uncertainty. Overall, the uncertainty due to roughness lengths is small in comparison to the mean outputs, indicating that the model is well defined even with large ranges of input parameter uncertainty.