Adaptively smoothed background seismicity rates in the Intermountain West, United States
Abstract
Spatially smoothed seismicity rates are an important seismic source for seismic hazard calculations across much of the Intermountain West (IMW). The U.S. national seismic hazard maps have historically used smoothed seismicity rate models generated with fixed-bandwidth smoothing methods (Frankel, 1996; Petersen et al., 2008); however, recent tests using the California earthquake catalog indicate that adapting the smoothing bandwidth to the local seismicity density (e.g., Helmstetter et al., 2007; Werner et al., 2011) produces improved seismic source models relative to models with fixed smoothing bandwidths (Schorlemmer et al., 2010). To test the ability of adaptively smoothed seismicity models to match epicenter locations from later parts of the IMW earthquake catalog, I generate time-independent maps of smoothed seismicity rates by spatially smoothing the seismicity rates of M4+ earthquake epicenters using fixed-radius and adaptive smoothing methods. I evaluate the 'forecast' smoothed seismicity models generated from the early part of the earthquake catalog by comparing the locations of earthquakes that occur in the later times of the catalog with the forecast seismicity rates. Forecasts are generated from a de-clustered catalog (Gardner and Knopoff, 1974) with completeness levels ranging from M4-6. The forecasts assume that the Gutenberg-Richter relation describes the magnitude-frequency distribution and that the locations of smaller earthquakes (M4+) can identify the locations of future large, and damaging, earthquakes. Spatially smoothed seismicity rate models are generated with isotropic Gaussian and power-law smoothing kernels using fixed and adaptive bandwidths; the adaptive smoothing bandwidths are calculated with the method of Helmstetter et al. (2007). To identify optimal smoothing methods for long-term earthquake rates, I calculate likelihood values for all smoothed seismicity models by using a Poisson distribution for earthquake occurrence and select the smoothed seismicity model that maximizes an information gain parameter. Adaptive smoothing methods produce higher information gains than fixed-bandwidth smoothing methods. I also investigate the use of different magnitude thresholds for the smoothed seismicity models and find that the incorporation of smaller earthquakes (M4+) improves the forecast of M5+ earthquakes compared to the use of only larger magnitude (M5+) earthquakes.
- Publication:
-
AGU Spring Meeting Abstracts
- Pub Date:
- May 2013
- Bibcode:
- 2013AGUSM.S53A..05M
- Keywords:
-
- 7223 SEISMOLOGY / Earthquake interaction;
- forecasting;
- and prediction;
- 9350 GEOGRAPHIC LOCATION / North America