Opportunities and Challenges for Urban Land-Use Change Modeling Using High-Performance Computing

Abstract

Simulating urban land-use changes involves both high modeling and computational complexities. This paper focuses on a typical spatio-temporal modeling method that has been commonly used in urban land-use change studies—Cellular Automata (CA). After reviewing the recent development of utilizing various parallel computing technologies (e.g., computer clusters and Graphics Processing Unit [GPU]) in CA-based urban models, this paper presents a pilot study, in which a classical CA model, the Game of Life, was implemented as a parallel program over the GPU/CPU heterogeneous cluster architecture, and 300+ speed-up was achieved using 20 GPUs. In conclusion, emerging high-performance computing technologies, such as GPU/CPU heterogeneous cluster architecture, provide promising potentials to overcome the computing obstacle of urban land-use change models, and enable researchers to examine, validate and advance urban land-use change theories and derive sound urban planning strategies. To efficiently utilize the computing power of the GPU/CPU clusters, hybrid parallelism must be implemented to coordinate the computing among GPU/CPU nodes, as well as among the threads on each GPU. However, implementing such hybrid parallelism is challenging for its high development complexity.