S31D-01
Design, Implementation, and Status of the CSEP Testing Center at SCEC
SCEC began development of the Collaboratory for the Study of Earthquake Predictability (CSEP) in January of 2006 with funding provided by the W. M. Keck Foundation. Since that time, a large group of scientists and software engineers have translated the concepts of CSEP into an operational testing center. The initial implementation of the W. M. Keck Foundation Testing Center at SCEC for the California natural laboratory became operational on September 1, 2007. The design and implementation of the SCEC Testing Center have been guided by four design goals that were originally identified as objectives for the RELM testing center which are: (1) Controlled Environment, (2) Transparency, (3) Comparability, and (4) Reproducibility. By meeting these goals, the CSEP Testing Center can provide clear descriptions of how all registered earthquake forecasts are produced and how each of the forecasts are evaluated. In this presentation, we describe how the currently operational CSEP Testing Center at SCEC has been constructed to meet these design goals and we describe the status of the Testing Center and our experiences operating the center since its inception. http://www.scec.org/csep
S31D-02 INVITED
A Double Branching Model for Earthquake Forecasting
Earthquake forecasting is one of the most relevant scientific contributions for society, being one of the primary ingredients for a well-founded seismic hazard assessment. Despite the importance of the issue, we are still far from a commonly accepted methodology to accomplish this goal. The epistemic uncertainty on this issue is well depicted by the simultaneous use, in practical cases, of different and sometimes antithetical models. One of the major problems in achieving a general consensus on this issue is due to the difficulty in testing and comparing the forecasting capability of different models; in practice, several models are not testable at all. Here, we propose an Earthquake Forecasting Model that can be tested in a forward perspective; this is the most straightforward way to evaluate the reliability of any forecasting model. Basically, the model is a double stochastic branching process that is set up through two sequential steps. In the first step, we estimate the well-established short time clustering. Then, we analyze and characterize the "declustered" catalog through a "second order" branching process that accounts for the statistically significant departures from a stationary Poisson model. The model is applied to different time-magnitude window in Italy and at worldwide scale in order to test the forecasting performance for moderate and large earthquakes in different tectonic domains. In general, the model shows a good fit to the data, and it reveals a better forecasting performance compared to similar models so far proposed, like stationary Poisson and ETAS models.
S31D-03
A method for forecasting the locations of future large earthquakes
The objective of this work is to quantify the use of past seismicity to forecast the locations of future large earthquakes. To achieve this the binary forecast approach is used where the surface of the Earth is divided into 2°× 2° cells. The cumulative Benioff strain of m≥ 5.5 earthquakes that occurred during the training period 1/1/1976 to 12/31/2000 is used to retrospectively forecast the locations of large earthquakes with magnitudes ≥ 7.0 during the forecast period 1/1/2001 to 12/31/2006. The success of a forecast is measured in terms of hit rates (fraction of earthquakes forecast) and false alarm rates (fraction of alarms that do not forecast earthquakes). This binary forecast approach is quantified using a relative operating characteristic (ROC) diagram and an error diagram (ED). An optimal forecast is obtained by taking the maximum value of Pierce's skill score. The result for this optimal choice is that 64% of the locations of m≥ 7.0 earthquakes occurred in 1.8% of the earth surface area.
S31D-04
Earthquake Prediction From Foreshocks
Foreshock sequences are characterized by some common properties which are of great interest for the mainshock prediction. The first is that the activity rate, r, increases as the inverse of time in a time interval from hours to a few months before the mainshock, and the second is that usually the b-value is lower than in background seismicity and aftershocks. However, foreshocks precede only some mainshocks and not others. Understanding better why this ¡°preferential" incidence of foreshocks happens would imply a significant step towards the achievement of the mainshocks prediction. In this paper we examine the incidence of foreshock activity before mainshocks in Greece. We investigated foreshock activity for mainshocks of Ms ¡Ý 5.5 occurring in the Greek region from 1986 up to December 2006 inclusive. It appears that about 50% of the mainshocks are preceded by foreshocks. However, due to constraints related to the seismic monitoring capabilities some foreshocks pass without notice in the standard earthquake catalogues, such as the sequence of East Aegean Sea, October 2005. Therefore, the real percentage of foreshock sequences should be higher. We indicate how empirical probability approaches along with the variation of the parameters b and r could be utilized for the short- term prediction of the mainshock. In another paper we show how such variations can be detected by an automatic system performing statistical tests by updating the data sets produced on a daily basis by the national seismograph system. This is a contribution to the EU research project SAFER, contract n. 036935, FP6-2005- Global-4, Reduction of seismic risks.
S31D-05 INVITED
Optimization of smoothed-seismicity earthquake forecasts using the area skill score
Rigorous evaluation of 5-year earthquake forecasts has begun at Collaboratory for the Study of Earthquake Predictability (CSEP) testing centers. For the California natural laboratory, initial forecasts are stated in terms of expected seismicity rate in latitude/longitude/magnitude bins. Three tests based on likelihood criteria are used to evaluate the forecasts with respect to observations of M≥5 earthquakes. These forecasts are of a general form that provides a relative ordering of estimated hazard, i.e., alarm-based forecasts. Some forecasts of the more general form cannot be tested and compared under the CSEP likelihood framework. We have developed a testing procedure based on what we call the "area skill score" of the Molchan diagram that can accommodate forecasts derived from alarm functions, including the CSEP forecasts. We understand the area skill score distribution and we use it for hypothesis testing of alarm-based forecasts. In this work, we discuss the use of the area skill score for forecast optimization. Many earthquake occurrence models, and most seismic hazard analyses, employ some form of epicentral smoothing to estimate the spatial distribution of earthquakes. In most cases, an implicit assumption is that the spatial distribution of past small earthquakes can be used to forecast the locations of future large earthquakes. We demonstrate the use of the area skill score to optimize a variety of so-called smoothed seismicity forecasts, varying the form of the smoothing kernel and the minimum magnitude of target earthquakes. We consider a series of retrospective predictability experiments in California and compare our optimization results with published results based on maximum likelihood. We also show attempts to invert for smoothing parameters using synthetic catalogs where the smoothing kernel is explicitly specified. Based on these experiments, we have constructed a set of time-invariant 5-year forecasts for California and submitted them to CSEP for testing. http://www.cseptesting.org
S31D-06
Evaluating Thermoelastic Strain as an Earthquake Trigger
Spatio-temporal variations of temperature at the Earth's surface cause strains and stresses below the surface in two ways. The diffusion of heat into the ground leads directly to thermal deformation that may have a significant amplitude several meters below the Earth's surface. More importantly, variations of surface temperatures produce surface tractions that propagate stress-strain variations many kilometers into the crust. Under the latter mechanism, seasonal temperature variations with spatial wavelengths of 10-30 km can produce thermoelastic strains with amplitudes of ~10-7-10-8over the seismogenic depth range of 1-10 km in the crust. Since the threshold for triggering seismicity is expected to decrease with increasing period of excitation, these strain levels and associated stresses (~7x102-7x103 Pa using a nominal rigidity of 70 GPa) may trigger seismicity. We suggest that thermoelastic strains may explain the growing evidence for seasonal variations in the earthquake activity rates at a number of places in the western U.S. Seasonal variations of seismicity with increased activity in the summer and autumn were observed in California for earthquakes less than M1.5 during the several years following the 1992 Landers earthquake, for earthquakes with M>=2.0 in California and Nevada from 1990-2005, and for earthquakes with M>=4.0 in California and Nevada from 1932- 2005. A similar seasonality with increased seismicity during the autumn and winter has been documented at some of the Cascades volcanoes and in the Himalaya. Others have explained observed seasonal seismicity patterns by annual air pressure changes, rainfall effects or snow loads; however, some such explanations require unrealistic values of material properties. We argue that an annual thermoelastic effect appears to be the best explanation for the observed seasonal seismicity patterns in the dry, warm California climate.
S31D-07
The Possible Asperities on the Xianshuihe-Anninghe-Zemuhe Fault Zone: Evidence From Microseismicity
The Xianshuihe-Anninghe-Zemuhe fault zone, which forms the eastern margin of the Tibetan plateau, is a 700km long, left-lateral seismically active fault zone in western Sichuan, China. We mapped b value and local recurrence time TL along this fault zone using the relocated microseismicity that projected on the cross section to identify the possible locations of the asperities. Strong variations in b values are observed along the fault zone, which are consistent in geometry with the surface faulting segmentation. The anomalously short local recurrence time TL and low b value appear on the Luhuo-Daofu segment of the Xianshuihe fault zone, Shimian-Mianning segment of the Anninghe fault zone, and Xichang-Puge segment of the Zemuhe fault zone, indicating that these segments are the relatively active parts that produce more large microearthquakes than average. With the hypothesis of b value being in inverse proportion to the stress and the method of mapping asperities by using TL proposed by Wiemer and Wyss (1997), we interpret these segments as the asperities of the fault zone which have accumulated high stress and locked. The Luhuo-Daofu segment, which has generated 4 historic earthquakes greater than Ms6.0 since 1900, is structurally simple to be the most active segment amongst the whole fault zone with the slip rate up to 14mm/yr, while the Shimian-Mianning segment is well known for void of historic earthquakes. They are all easy to accumulate stress and become locked. The anomalously high b value and long TL estimates as well as high seismicity rate of a value in the southeastern segment of the Xianshuihe fault zone along Qianning-Kangding-Shimian suggest that this segment is creeping at present and not likely to produce a major earthquake, although both the historic earthquakes and microseismicity are active there.
S31D-08
Earthquake probability at the Kashiwazaki Kariwa nuclear power plant, Japan, assessed using bandwidth optimization
On July 16, 2007, a strong 6.8 magnitude earthquake occurred on Japan's west coast, rocking the nearby Kashiwazaki Kariwa nuclear power plant, the largest nuclear power station on Earth. Shaking during this event produced ground accelerations of ~680 gal, exceeding the plant seismic design specification of 273 gal. This occurrence renews concerns regarding seismic hazards at nuclear facilities located in regions with persistent earthquake activity. Seismic hazard assessments depend upon an understanding of the spatial distribution of earthquakes to effectively assess future earthquake hazards. Earthquake spatial density is best estimated using kernel density functions based on the locations of past seismic events. Two longstanding problems encountered when using kernel density estimation are the selection of an optimal smoothing bandwidth and the quantification of the uncertainty inherent in these estimates. Currently, kernel bandwidths are often selected subjectively and the uncertainty in spatial density estimation is not calculated. As a result, hazards with potentially large consequences for society are poorly estimated. We solve these two problems by employing an optimal bandwidth selector algorithm to objectively identify an appropriately sized kernel bandwidth based on earthquake locations from catalog databases and by assessing uncertainty in the spatial density estimate using a modified smoothed bootstrap technique. After applying these methods to the Kashiwazaki Kariwa site, the calculated probability of one or more Mw 6-7 earthquakes within 10 km of the site during a 40 yr facility lifetime is between 0.005 and 0.02 with 95 percent confidence. This result is made more robust by calculating similar probabilities using alternative databases of earthquake locations and magnitudes. The objectivity and quantitative robustness of these techniques make them extremely beneficial for seismic hazard assessment.