Time varying hazards research programme (the NZ Earthquake Forecast Testing Centre: Prospective Testing Phase 1)

Abstract

Validation of seismic hazard models and forecasts has entered a new era in recent years with a focus on establishing rigorous community developed and accepted testing procedures. These efforts were originally led by the Regional Earthquake Likelihood Models group of the Southern California Earthquake Centre and are now firmly in place with the work of the Collaboratory for the Study of Earthquake Predictability (CSEP), which includes the New Zealand Earthquake Forecast Testing Centre (NZTC).

The work presented in this report was largely focused on the technical and operational aspects of the NZTC. Important developments have been fully automating all testing procedures, publication of testing-results on a secured web page, and the installation of forecast models into the NZTC. The NZTC system has been kept in step with CSEP code releases and is currently operating using CSEP v9.1.1. Eleven models, including their variations, have been submitted to the testing centre and seven of these have been installed and are being unofficially tested. 

The following models have been submitted: NZ national seismic hazard model (Stirling, et al, 2002), EEPAS (Rhoades & Evison 2004), PPE (Rhoades & Evison 2004), SUP (Rhoades & Evison, 2004), STEP (Gerstenberger, et al, 2005) and Double branching model (Marzocchi and Lombardi, 2008). Of these, the national seismic hazard model, PPE, SUP and STEP are installed and are currently being unofficially tested. The STEP model calculates a new forecast every day which is tested against the daily observations. The other models have provided 1 forecast for a five year period; the tests are updated daily, but the final result awaits five years of data. The five year models are tested against a declustered and a non-declustered catalogue. Official testing is dependent on a complete historical catalogue; currently, the GeoNet catalogue has a completeness gap from April 2007 to July 2008. With current practice, this gap will be filled in about 2 years, at which point official 5-year tests can begin.