Quantifying inflow uncertainties for CFD simulations of dispersion in the Atmospheric Boundary Layer
15 September 2017
Campus Groenenborger, U0.25 - Groenenborgerlaan 171 - 2020 Antwerpen (route: UAntwerpen, Campus Groenenborger
Clara Garcia Sanchez
Gustaaf Van Tendeloo & Catherine Gorlé
PhD defence Clara GARCÍA SÁNCHEZ - Faculty of Science, Department of Physics
To support design and policy decisions for prevention and mitigation of pollution in the cities of the future, accurate predictions of wind and dispersion will be of fundamental importance. Present computational fluid dynamic predictions of atmospheric boundary layer (ABL) flow and dispersion represent a major challenge. The full complexity of the ABL flow and dispersion phenomena cannot be incorporated in a single numerical simulation, which results in several sources of uncertainty in our predictions. These uncertainties range from geometrical and physics model uncertainties, to uncertainties in the boundary conditions for the simulation. The goal of this thesis is to improve the realism of our predictions of flow and dispersion in the ABL by targeting three objectives: 1) establish a method to quantify the effect of uncertainty in the inflow boundary conditions on the simulation results, using either field experiment or mesoscale simulation data; 2) analyze the effect of turbulence form uncertainties and their importance compared to inflow uncertainties; and 3) validate the results with full scale measurements from two different field experiments, Joint Urban 2003 in Oklahoma City to represent an urban area, and Askervein Hill to represent a natural terrain.
Inflow uncertainty is irreducible and naturally present in the system, largely affecting the flow pattern in rural and urban canopies. The proposed method quantifies the effect of inflow uncertainty on the simulation results based on three main steps: 1) characterize the uncertain inflow parameters, using either field experiment or mesoscale simulation data; 2) propagate inflow uncertainties to the quantities of interest using a polynomial chaos expansion, which requires a finite set of RANS simulations; and 3) construct and sample the resulting response surfaces to determine the 95% confidence intervals for the results. In addition, we evaluated the effects of turbulence model form uncertainties with two different studies. First, the uncertainties introduced by the turbulence model in the flow simulations in Oklahoma downtown are quantified by introducing perturbations in the Reynolds stress tensor. Second, a large-eddy simulation of the flow in Oklahoma City was performed for a single wind direction and compared to the corresponding RANS simulation and the field measurement data. The results demonstrated the capabilities of the uncertainty quantification methodology to assess the effect of inflow and turbulence model form uncertainties in RANS simulations. The approach allows identifying areas in the flow that are sensitive to inflow and turbulence model uncertainties, providing valuable information compared to deterministic forecasts.