Calculating CyberShake Map 1.0 on Shared Open Science High Performance Computing Resources

Current Probabilistic Seismic Hazard Analysis (PSHA) calculations produce predictive seismic hazard curves using an earthquake rupture forecast (ERF) and a ground motion prediction equation (GMPE) that defines how ground motions decay with distance from an earthquake. Traditionally, GMPEs are empirically-based attenuation models. However, these GMPEs have important limitations. The observational data used to develop the attenuation relationships do not cover the full range of possible earthquake magnitudes. These GMPEs predict only peak ground motions, and do not produce ground motion time series. Phenomena such as rupture directivity and basin effects may not be well captured with these GMPEs. To improve the accuracy of PSHA calculations, researchers at the Southern California Earthquake Center (SCEC) are performing physics-based PSHA using the CyberShake project. CyberShake utilizes full 3D waveform modeling as the GMPE to compute PSHA hazard curves for various sites in Southern California. For each rupture in the Uniform California Earthquake Rupture Forecast (UCERF) 2.0, we capture rupture variability by varying the hypocenter and slip distribution to produce about 410,000 different events (“rupture variations”) per site of interest. We calculate strain Green’s tensors for each site and use seismic reciprocity to compute the intensity measure of interest for each rupture variation. These intensity measures are then synthesized into a hazard curve, relating shaking levels to probability of exceedance. The goal of CyberShake is to calculate more accurate hazard curves using high performance computing techniques. A set of hazard curves, computed for many sites in a region, can be integrated to produce a regional hazard map. In 2009, a series of CyberShake calculations, called the CyberShake 1.0 Map calculation, were performed using distributed resources on Texas Advanced Computing Center’s Ranger system, part of the NSF-funded TeraGrid, and the University of Southern California’s HPC cluster. Over 54 days, 190 million tasks were executed, producing 175 TB of data and creating hazard maps for 223 sites in Southern California. The hazard curves were then interpolated to produce the world’s first physics-based PSHA map for Southern California. In our poster, we will present statistics, metrics, and results from the CyberShake 1.0 map calculation. We will also discuss the tools used in bringing high performance computing to bear on complex, meaningful seismological problems such as CyberShake. These tools help address challenges such as submission of jobs, data management, provisioning of computational resources, and status monitoring. Additionally, we will discuss plans for future hazard maps.