Scaling And Machine Learning Analysis Of Turbulent Fluxes Of Momentum And Heat In The Microclimate Of An Urban Canyon

Turbulent flow inside the urban roughness sublayer, despite its complexities, plays a crucial role in the microclimate of the built environment. The parameterization of flow in the urban roughness sublayer provides a better understanding of turbulent exchange process leading to accurate weather forecasting. This study focused on developing relationships between turbulent quantities, including momentum and heat fluxes, and mean quantities such as mean wind speeds. Field data, including wind directions, wind speeds, and thermal stability conditions, were collected from an urban canopy in Guelph, Ontario, Canada during the summer 2017. Comparative data was obtained from a nearby rural station. A systematic scaling analysis was performed to identify a range of quantities highly related to turbulent fluxes. All combinations of quantities leading to dimensionless groups were evaluated. Linear and nonlinear correlation coefficients between different groups of variables identified when mean and turbulent quantities were related. Significant improvement in correlation coefficients was observed using high order polynomial regression, revealing the challenge of developing a robust model for predicting nonlinear behavior of turbulence. This study also used artificial neural networks (ANNs) to find nonlinear relationships between turbulent and mean quantities. As used here, an ANN is a multivariable function which attempts to approach the exact value of turbulent flux based on independent variables, properly chosen from dimensionless groups. Results showed that these approaches can successfully relate most, but not all, turbulent quantities to mean quantities.