[Doctor] Matthew Gerstenberger: Earthquake Clustering and Time-Dependent Probabilistic Seismic Hazard Analysis for California

Matthew Gerstenberger

Earthquake Clustering and Time-Dependent Probabilistic Seismic Hazard Analysis for California

Abstract

California accounts for more than one half of the entire U.S. annual earthquake risk of $4.4 billion. A necessary step in the mitigation of this risk is the continued development of seismic hazard mapping techniques. Using an established aftershock forecasting model based on earthquake clustering and two fundamental laws in seismology (the Gutenberg-Richter relationship and the modified Omori law) I have developed a short term probabilistic seismic hazard mapping routine. The maps, exhibiting the probability of exceeding Modified Mercalli Index VI in the next 24 hours, are currently calculated every 30 minutes and are available on a United States Geological Survey web site. Using a multimodel approach based on the Akaike Information Criterion, I allow for three levels of complexity in the calculations: a generic California model; an isotropic model based on an overall aftershock sequence; and a model including spatial heterogeneities within an aftershock sequence.

After large mainshocks (>~ Magnitude 6) sufficient aftershocks are generally available to map differences in aftershock productivity, decay rate and frequency-magnitude distribution. I have investigated these differences in four large California aftershock sequences and one from Alaska. The ability to obtain results is clearly dependent on data quality. While quality is variable throughout the sequences, heterogeneities large enough to effect the subsequent hazard calculations appear to be common. I have demonstrated that by including these spatial variations the forecasting ability of our seismicity based model can be significantly improved. Regions, such as the northern end of the 1999 M_w 7.1 Hector Mine, California aftershock zone that produced anomalous numbers of large aftershocks clearly represent a larger hazard than regions producing only smaller aftershocks. Further extending these ideas, I have examined the change in frequency-magnitude distribution in depth for all of California. A clear and significant trend of a decrease with depth was observed.

To validate the model as well as to investigate and improve its overall performance, I have tested the earthquake forecasts using various statistical tests; a necessary step before the acceptance and implementation of any forecasting routine. Using likelihood based testing I have shown that standard long term hazard maps, the generic California model and the isotropic model can be rejected with a 5% significance when compared to the added complexity of including spatial variations. Additionally, and as expected, I have shown that probabilistic seismic hazard mapping still has a long way to go before it can be described as consistent with the data; even though our model significantly out performs less complex models, it can still be rejected over the long term when compared to observed earthquakes. Using these same statistical testing procedures, I have examined assumptions made in the model. Unfortunately, due to computational expense the results were largely inconclusive. However, observations include that in the initial months of an aftershock sequence, including spatial variability on a local scale (~10km) performs better than smoothing over a larger region and the use of moving time windows in estimating seismicity parameters may be provide more accurate forecasts that using a time period of the entire aftershock sequence. I have also discovered how weaknesses in the magnitude of completeness estimation and aftershock zone definition can be improved for future implementations.

Download