HR: 11:35h
AN: S32B-06    [Abstracts]
TI: Testing the Daily Predictive Power of Clustered Seismicity Models on the 1992 Landers Aftershock Sequence
AU: * Woessner, J
EM: j.woessner@sed.ethz.ch
AF: ETH Zurich, Swiss Seismological Service, ETH Hoenggerberg HPP P7.2, Schafmattstr. 30, Zurich, 8093, Switzerland
AU: Hainzl, S
EM: hainzl@gfz-potsdam.de
AF: GeoForschungsZentrum Potsdam, Telegrafenberg, Potsdam, 14473, Germany
AU: Werner, M J
EM: werner@moho.ess.ucla.edu
AF: University of California, Los Angeles, Institute of Geophysics and Planetary Physics, Department of Earth and Space Sciences, 595 Charles Young Dr E, Los Angeles, 90095, United States
AU: Catalli, F
EM: catalli@ingv.it
AF: Istituto Nazionale di Geofisica e Vulcanologia, Via di Vigna Murata 605, Rome, 00142, Italy
AU: Gerstenberger, M C
EM: m.gerstenberger@gns.cri.nz
AF: GNS Science, PO Box 30368, Lower Hutt, 6007, New Zealand
AB: Aftershock hazard is a significant and strongly time-dependent contribution to seismic hazard. Large aftershock sequences can serve as natural laboratories for earthquake predictability experiments, allowing for retro- and prospective testing of seismicity forecasts on various spatial and temporal scales. The 1992 Landers earthquake and the resulting highly clustered earthquake sequence is one of the best recorded and best studied earthquake sequence; however, a comparative experiment to retrospectively forecast earthquakes applying established statistical and physical models has not yet been pursued. We analyze the performance of (1) the Short-Term Earthquake Probability (STEP) model and STEP model elements, (2) a suite of Epidemic Type Aftershock Sequence (ETAS) models with and without parameter dependence on time and space and various spatial triggering kernels, and (3) a model deriving seismicity rates from a rate and state model incorporating multiple stress changes due to large and moderate earthquakes in the aftershock sequence. Comparing these models allows us to address the questions: Which models perform well on short time and small spatial scales? Do physical models lead to an information gain over purely statistical models on the scale of an aftershock sequence? Where with respect to the faulting do the models perform well? For the experiment, we define rules similar to the RELM testing approach. Forecasts are computed for 90 days starting on June 28,.1992, forecasting the seismicity on a predefined grid in the magnitude range 4 ≤ M ≤ 8 each day for 24 hours. The forecasts are evaluated on a daily basis using the RELM likelihood tests that test data consistency and relative performance of the models. Preliminary results show that the statistical models perform well in the long run of the earthquake sequence but not at the onset of the sequence, due to the lack of sequence specific information. We plan to investigate additional well recorded aftershocks sequences from a range of tectonic settings, such as the Colfiorito earthquake sequence in Italy or earthquake sequences in Iceland. The ultimate goal of the retrospective experiments in the framework of the EU project Seismic Early Warning For EuRope (SAFER) is to improve the ability of the models to forecast in real-time short-term earthquake probabilities for earthquake sequences.
DE: 7200 SEISMOLOGY
DE: 7223 Earthquake interaction, forecasting, and prediction (1217, 1242)
DE: 7230 Seismicity and tectonics (1207, 1217, 1240, 1242)
SC: Seismology [S]
MN: 2007 Fall Meeting