Run-Time GPU Computing and Rendering of Earthquake Ground-Motion Data

Abstract

Earthquake simulations generate large-scale ground-motion velocity data. However, in addition to the ground-motion velocity data, seismologists also need to study both ground-motion acceleration and ground-motion displacement data. Therefore, data post-processing is required to differentiate velocity data into acceleration data and to integrate velocity data into displacement data. This would require additional storage space. In this study, we use GPU to perform near real-time computing of acceleration and displacement data so that the data storage space can be reduced. This is because data post-processing is no longer required to produce acceleration and displacement data. The evaluation of earthquake simulation data is a complex task that requires significant training. Thus, we propose to use volume rendering to better understand and analyze time-varying earthquake ground-motion data, including acceleration, velocity, and displacement. A case study of the 1999 Chi-Chi earthquake in Taiwan is presented in this paper. We used CUDA to implement volume rendering (ray casting), numerical integration (trapezoidal quadrature), and numerical differentiation (centered finite difference) and achieved a speedup of approximately 100 times faster than on a CPU.