Abstract

In recent years, due to the rapid population growth and the preference to live in urban areas, urbanization has intensely increased. Currently, based on a United Nation report, 55% of the population live in the cities and the number is expected to reach about 68% by 2050. Urban microclimate has significant impacts on human life and health, and building energy performance. Urban microclimate information, such as wind velocity, temperature, humidity, pollutant dispersion levels, and local precipitation, are often important for accurate evaluations of building energy performance, indoor and outdoor human comfort, extreme events, and emergency situations. For example, it was reported that indoor temperature estimated with the microclimate information could be at least 5 °C different from that without it, which could be significant for the evaluations of indoor thermal comfort. The study of urban microclimate includes both observational and numerical approaches. The observational study is often related to field measurements, satellite imagery, and laboratory tests, e.g. in wind tunnels. The numerical approach is often based on computer models, such as CFD (computational fluid dynamics), for high-resolution and relatively small computing domains, compared to larger scale regional climate models, such as WRF and GEM-SURF. The latter two models are mostly used to model the domain size of 1~10 km with the resolution more than 100 m so they are not developed for urban microclimate and building-level simulations. In comparison, CFD has been applied to the urban microclimate of less than 1 km with a resolution less than 10 m down to the building level. However, conventional CFD solvers often perform unsatisfactorily for microscale and complicated problems because of numerical constraints such as stability issues associated with CFL condition, which is a necessary condition for convergence while solving certain partial differential equations (usually hyperbolic PDEs) numerically. Thus, conventional tools are often computationally expensive for modeling microclimates and consequently impractical for urban-scale problems. Recently, there are an increasing amount of efforts focusing on developing faster and accurate CFD techniques such as based on Fast Fluid Dynamics (FFD) methods. A FFD method relies on semi-Lagrangian and fractional step methods. FFD methods is fundamentally an explicit method without the CFL constraint so it is unconditionally stable even under large time steps and coarse grid resolutions, which are common for urban microclimate problems. In the meantime, the conventional FFD methods are often dependent on low-order interpolation schemes and thus with high numerical errors, which are the main drawbacks of this approach.
The main objective of this thesis is to develop a fast and accurate CFD solver with a series of new computing algorithms based on semi-Lagrangian approach for modeling urban/city scale microclimates. The new solver with the name of CityFFD (city Fast Fluid Dynamics), is designed for tackling the challenges of large domain, coarse grid, and/or large time step, which are typical for urban microclimate simulations, without a heavy reliance on computer resources, such as the possibility of running on personal computers. First, a novel high-order interpolation scheme is proposed to significantly reduce the numerical errors of conventional semi-Lagrangian solvers. The new interpolation scheme enables the possibility of obtaining fast and accurate results even on coarse grids. The second algorithm focuses on the simulation accuracy associated with the time step of the semi-Lagrangian method. A new scheme of an adaptive time step is developed to adjust the time step dynamically according to local truncation errors. To improve the estimation of the characteristic curves, a new algorithm is proposed by considering the acceleration of the flow particles inside the computational domain which can provide highly accurate results and capture the complicated flow fields even by using a large time step. The fourth algorithm is to speed up the simulation by eliminating the need for solving the Poisson equation, which is often the most time-consuming operation of conventional semi-Lagrangian models. The new scheme is based on the concept of the artificial compressibility of solving incompressible flows and makes it easier to implement parallel computing techniques, such as the NVIDIA GPU CUDA and the OpenMP. The last feature of CityFFD is adding Large Eddy Simulation (LES) model to capture the turbulence behavior of the flow in urban environments. In this section, a parallel OpenMP geometry reader is developed to read the city scale geometries in a fast manner. At the end, the proposed CityFFD model is demonstrated by a case study: the modeling of an extreme weather event, the snow-storm of the century in Montreal, for evaluating building resilience during the storm, to show the importance of urban microclimate and its impact on human health and indoor environment.