T-RELM: Testing Center for probabilistic earthquake forecasts
D. Schorlemmer, M. C. Gerstenberger, S. Wiemer, E. Field, L. Jones, and D. D. Jackson
T-RELM: Testing Center for probabilistic earthquake forecasts
AGU Fall Meeting 2005, San Francisco, USA.
Eos Trans. AGU, Vol. 86, No. 52, Fall Meet. Suppl., Abstract S43B-1074, 2005.
Abstract
Testing probabilistic earthquake models is becoming an increasingly important task in seismology. Numerous forecast models exist, but lack rigorous testing against each other in order to determine their comparative performances. In general, models have only been tested pseudo-prospectively against either an assumed null hypotheses or the data. Different characteristics of models make model comparisons difficult. As part of the regional earthquake likelihood model (RELM) initiative of SCEC, a group of modelers has agreed to test their models prospectively using a community agreed upon set of rules. We are currently establishing a central testing center at ETH and we have implemented all necessary procedures for truly prospective tests of different sets of models. We distinguish between short-term models, updating their forecasts daily, and quasi-stationary models, issuing their forecast on a yearly basis. These testing classes have been defined based on both scientific and public interests. In establishing a testing center, a number of decisions needed to be taken, such as: What area needs to be covered by the forecast? Which requirements has a viable model to meet? Do modelers have access to their code? What data is a forecast tested against: Real time, or final corrected data? Which authorized data sources can a model use?
In this presentation, we describe the scientific, administrative and societal implications of setting up T-RELM, the testing effort within the RELM framework. The prescribed testing region is California, and we still highly encourage new models to participate in the upcoming real time test.
