Evaluating New Zealand models for short-term earthquake probability

Abstract

The possibility that a moderate earthquake may be followed by a larger one (foreshock probability) increases the hazard in its immediate vicinity for a short time by an order of magnitude or more. Similarly, damaging aftershocks often occur after large events. Yet previously, these fluctuations in hazard have not been included in the New Zealand seismic hazard model.

We combine the current average background New Zealand seismic hazard model with a new model for foreshock probability and an established model of aftershock probability to create a dynamic seismic hazard model for New Zealand. This model calculates the chances of a damaging earthquake with peak ground acceleration over 0.05g, and changes daily to reflect the increased or decreased hazards depending on recent events.

We develop a method to test these ground motion forecasts against 40 years of strong motion observations in New Zealand. The standard background model underestimates the hazard at 0.05g level for the time period and locations in the dataset by a factor of two. The new dynamic seismic hazard model provides a closer match to the strong motion records, but with some overestimation (factor of three) of hazard on days of high probability and underestimation (factor of six) on days of low probability forecasts.

---

Technical Abstract

A new dynamic seismic hazard model for New Zealand is created to add the effects of short-term hazard fluctuations due to earthquake triggering to the current Poissonian probability from the national probabilistic seismic hazard model.

Prospective foreshock probability decay is modeled as a function of origin time and epicentral distance from the potential foreshock, and the magnitude difference between foreshock-mainshock pairs. We calculate the probability of an initial earthquake (a foreshock) being followed by a mainshock in New Zealand, considering the parameters of elapsed time and distance and magnitude differences between foreshock and mainshock. We use non-aftershock events between 1964 and 2003, with magnitude ≥  3.8 and shallower than 40 km, separating the catalogue into events within and outside the Taupo Volcanic Zone (TVZ).  We provide a model for the probability P’ that at time t after a potential foreshock of magnitude Mfs, and at distance r, a mainshock with magnitude Mfs+δM will occur: P’(t,r,δM) = P(t=1,r=10,δM=0) * 10(-B*δM) * 1/(t+ct)texp * 10rexp/(r+cr)rexp

We find (1) foreshock probabilities are independent of foreshock magnitude, (2) foreshock probabilities decrease with increasing inter-event time with texp significantly larger than one (2.5±0.5 (TVZ) and 1.7±0.2 elsewhere), (3) foreshock probabilities decrease with increasing epicentral distance with rexp of 2.6±0.2 (non-TVZ) and 3.66±0.2 (TVZ) and (4) the mainshock magnitude distribution follows the Gutenberg-Richter relationship with a significantly higher than normal b-value (B=1.4±0.1 (Non-TVZ) and 1.8±0.2 (TVZ)).

The behaviour of these foreshock probabilities with time and distance differs from the behaviour of aftershock probabilities as modeled by previous workers, suggesting that there may be a different triggering mechanism for foreshocks than for aftershocks.

Combining a previous, similar foreshock model with a generic aftershock model, fluctuating daily hazard maps for New Zealand are calculated, showing the regional probability distribution for peak ground accelerations of 0.05g or more being observed.

A methodology has been developed to test these ground motion forecasts against 40 years of strong motion observations in New Zealand. The Poissonian model underestimates the hazard at 0.05g level for the time period and locations in the dataset by a factor of two. The new dynamic seismic hazard model provides a closer match to the strong motion records, but with some overestimation (factor of three) of hazard on days of high probability and underestimation (factor of six) on days of low probability forecasts.