New hybrid modelling methods improve earthquake rate forecasts
Statistical seismologists Annemarie Christophersen and David Rhoades stand in front of a map of hybrid earthquake rate models for California. Photo: GNS Science
Working with colleagues from the USA, Europe and Japan in the Collaboratory for the Study of Earthquake Predictability (CSEP), we strive for better forecasts of earthquake rates on time scales of days, months and years. Models are developed and characterised, adapted to the testing centre requirements so others can use it, and installed by linking the model into the testing centre software. Each of these steps involves a substantial amount of work and models are tested and compared using agreed metrics. Currently, New Zealand has contributed 16 models to the Collaboratory, with considerable outside interest in our new hybrid methods.
Hybrid earthquake rate models were first used in New Zealand in response to the Canterbury earthquakes where there was a demand for estimates of the expected rates of future earthquakes. Our forecasts supported decisions on refinement of building codes, red-zoning of suburbs, and timing of the Christchurch rebuild. Late in 2011, an international expert panel was convened at GNS Science to discuss short-, medium- and long-term modelling options. The outcome is the Expert Elicitation (EE) hybrid earthquake rate model for Canterbury which provides estimated rates for the next 50 years. Since its development, the EE model has been retrospectively tested* on the whole of New Zealand using high-quality earthquake records since 1986. The EE model statistically outperforms all of the individual component models.
Did you know ?
We have more than 10 GNS Science models in the California earthquake forecast testing centre; 4 models in the Japan testing centre, and 2 models in the European testing centre.
In our long-term National Seismic Hazard Model, the data from mapped faults and earthquake catalogues are combined following a standard practice established more than 20 years ago. Our new hybrid modelling techniques allow us to consider other ways of combining the same data, or new types of information into the earthquake rate models; for example, the maps of crustal deformation derived from GPS data can now be included. Also, proposed earthquake precursors can be more readily assessed by including them in a hybrid with what we already know about time-varying earthquake occurrence - an improvement over assessing them in isolation. Our methods have proved useful in California, where a five-year prospective test of twelve different long-term earthquake rate models was conducted from 2006-2010.
It is expected that a slew of new hybrid models will be developed in the next few years using our methods. As a result, it is hard to know what earthquake rate models will be in use ten years from now. However, one thing is certain: the models of the future – whether short-, medium- or long-term – will include more types of information, and will be better than the models we use today.
As our models are implemented, New Zealand will be in a better position to cope with future earthquake crises. More dependable forecasts of future earthquake expectation on several timescales will be available, of value to public authorities, planners and the insurance sector.
*In retrospective testing we compare the total number of earthquakes expected under each model with the number observed over the test period, as well as determine the overall information gain of one model over another.
Back to