5:15 PM - 5:30 PM
[S10-1-04] On what time scales can strain rates contribute to earthquake likelihood models?
Recently a multiplicative modelling technique was developed for combining earthquake likelihood models and gridded covariates into hybrid earthquake likelihood models. Multiplicative hybrids often have larger information gains than additive hybrids produced from the same inputs. We have applied the technique to evaluate the information gains from incorporating strain rate maps into existing earthquake likelihood models on two timescales.
In a long timescale study, we use three components of the strain rate in New Zealand estimated over the period 1991-2011 – the shear (SSR), rotational (RSR) and dilatational strain rate (DSR). We fit multiplicative hybrid models using selections of covariates, constructed from earthquake catalogue and fault data as well as the strain rate components, to the period 1987-2006 and test them over the period 2012-2016. The SSR provides the greatest single information gains in forecasting M 5 and greater earthquakes in the fitting and testing periods. Most models including strain rates are more informative than the best models excluding strain rates. A hybrid that combines SSR and DSR with a smoothed seismicity covariate is the most informative in the fitting period, and a simpler hybrid without DSR is the most informative in the testing period.
In a short timescale study, we use daily estimates of strain rate on the shallow Hikurangi subduction megathrust as a multiplicative covariate. We average daily strain rates over four time windows (1 day, 30 days, 100 days, and 460 days) to test the link between duration of strain and triggered earthquakes of magnitude 3.5 and greater in subsequent time windows of the same length. We find that the 30-day window provides the largest information gain, followed by the 460-day window. These results may be related to the rapid triggering of earthquakes during slow slip events (SSE). Shallow SSE have typical durations of weeks and deep SSE have typical durations ranging from months to over a year.
In a long timescale study, we use three components of the strain rate in New Zealand estimated over the period 1991-2011 – the shear (SSR), rotational (RSR) and dilatational strain rate (DSR). We fit multiplicative hybrid models using selections of covariates, constructed from earthquake catalogue and fault data as well as the strain rate components, to the period 1987-2006 and test them over the period 2012-2016. The SSR provides the greatest single information gains in forecasting M 5 and greater earthquakes in the fitting and testing periods. Most models including strain rates are more informative than the best models excluding strain rates. A hybrid that combines SSR and DSR with a smoothed seismicity covariate is the most informative in the fitting period, and a simpler hybrid without DSR is the most informative in the testing period.
In a short timescale study, we use daily estimates of strain rate on the shallow Hikurangi subduction megathrust as a multiplicative covariate. We average daily strain rates over four time windows (1 day, 30 days, 100 days, and 460 days) to test the link between duration of strain and triggered earthquakes of magnitude 3.5 and greater in subsequent time windows of the same length. We find that the 30-day window provides the largest information gain, followed by the 460-day window. These results may be related to the rapid triggering of earthquakes during slow slip events (SSE). Shallow SSE have typical durations of weeks and deep SSE have typical durations ranging from months to over a year.