SCEC Award Number 15102 View PDF
Proposal Category Collaborative Proposal (Integration and Theory)
Proposal Title Testing the Global Earthquake Activity Rates (GEAR) Model: A Pathway to High-Resolution Global Testing
Investigator(s)
Name Organization
Danijel Schorlemmer University of Southern California Maximilian Werner University of Bristol (United Kingdom)
Other Participants Maria Liukis, Thomas Beutin
SCEC Priorities 2e SCEC Groups WGCEP, EFP, CSEP
Report Due Date 03/15/2016 Date Report Submitted 10/09/2016
Project Abstract
The Collaboratory for the Study of Earthquake Predictability (CSEP) joined forces with the Global Earth- quake Model (GEM) Foundation for the purpose of evaluating GEM’s recently developed global earth- quake forecast using retrospective and prospective observations. Bird et al. (2015) produced GEM’s high-resolution global long-term forecast, named the Global Earthquake Activity Rate model version 1.0 (GEAR1). GEAR1 is now under retrospective and prospective testing within CSEP and first results are available. To be able to begin testing, we needed to develop substantially more efficient CSEP formats and testing procedures, because of the high (0.1 by 0.1 degree) spatial resolution of the global forecast. We implemented several efficiencies that, in the most extreme case, cut processing time from weeks to minutes. The two most important changes are (i) a magic index that allows calculation (rather than a search) of the index of a bin, and (ii) making simulations of model likelihood scores more efficient. We evaluated three global high-resolution models: GEAR1 and two strain-rate based forecasts, namely GSRM (Global Strain Rate Map, Bird et al., 2010) and an updated GSRM2.1 (Bird and Kreemer, 2015). Over the prospective testing period between October 2015 and April 2016, 59 earthquakes Mw> 5.767 occurred, very close to the expected number of all three models. All models pass the space, magnitude and conditional likelihood tests, i.e. the observations are consistent with the forecasts. In a head-to-head comparison, GEAR1 achieves an information gain of 0.5 over both strain-rate based forecasts, indicating that combined smoothed seismicity and strain-rate information provide the greater predictive skill.
Intellectual Merit Global forecasting experiments are crucial because data of large earthquakes accumulate much faster than in southern California. Many important earthquake science questions can only be addressed on a global scale in a reasonable amount of time. This project contributed to a global testing facility. Our results suggest that combining information from strain-rate maps and past earthquake locations provides the greatest predictive skill.
Broader Impacts GEAR1 is an important GEM product that is likely to be used for a range of purposes, including seismic hazard and loss estimation. Validation of such an important forecast is a high priority for both GEM and CSEP. This pro- ject has strengthened ties between SCEC, GFZ Potsdam and the GEM Foundation. CSEP software was significant- ly improved, which will facilitate testing elsewhere. A female postdoc participated in the testing.
Exemplary Figure Figure 3: Average probability gains per earthquake of the GEAR1 and GSRM2.1 models over the GSRM model for the prospective 2015/10/1-2016/4/7 testing period. Displayed are gains of the models on the y-axis over the model in the title. Horizontal lines indicate 95% confidence intervals; numbers above circles show the number of earthquakes. Both strain-rate based models can be rejected in favor of the GEAR1 model.