GPU Accelerated Surface Energy Balance Computations for Urban Environment Simulation
Many challenges arise with solving small scale surface energy budgets across urban centers larger than a few blocks in size. The efficiency of computation quickly plays a role in determining the size and resolution of the urban spaces that can be simulated. Our surface energy budget computations utilize GPU-based resources to accelerate the calculations associated with radiation transport, turbulent advection, and land surface interactions. The computational framework is designed around the need to couple multiple models together into a heterogeneous computing substrate. Our system, called QUIC EnvSim, utilizes both CPU cores and GPU cores to handle these computations. In particular, we utilize NVIDIA's CUDA and OptiX frameworks to support the computations. For instance, with regard to radiation transport, our approach uses GPU-based ray tracing with NVIDIA's OptiX API to sample direct and diffuse shortwave irradiation, diffuse longwave irradiation, and longwave exchange between terrestrial objects. Simulated wavelengths have the potential to be absorbed, scattered, and reflected by objects in the domain, including participating media or vegetative volumes. Models implemented within the QUIC EnvSim framework are quite modular and can be linked with other surface energy budget components running on either the GPU or CPU.
In this work, we present the techniques and approaches we have developed to support GPU acceleration for solving the surface energy balance in urban domains. In particular, several optimizations and algorithm changes are necessary to afford efficient computation on the GPU. Moreover, QUIC EnvSim supports computation across multiple GPUs transparently by abstracting data buffer manipulation. Current results will also be presented demonstrating the capabilities of QUIC EnvSim and its ability to simulate the surface energy balance within urban domains.