Poster #259, Earthquake Forecasting and Predictability (EFP)

Towards a distributed seismicity model for the New Zealand national seismic hazard update

Sepideh J Rastin, David A. Rhoades, Matthew C. Gerstenberger, Chris Rollins, & Annemarie Christophersen
Poster Image: 

Poster Presentation

2021 SCEC Annual Meeting, Poster #259, SCEC Contribution #11266 VIEW PDF
We aim to develop a distributed seismicity model to forecast the long-term (e.g., 100 years) spatial distribution of seismicity for the New Zealand National Seismic Hazard Model (NZNSHM) update. The distributed spatial rates combined with the rates on known fault sources should provide a comprehensive representation of seismic sources for the NZNSHM. Multiplicative hybrid models help us to assess the relative value of different data inputs, including information from fault studies, tectonics, the earthquake catalogue and strain rate models for forecasting the long-term spatial distribution of earthquake rates. We have used time-invariant and time-variable inputs to fit the hybrid models to 7...0 years of NZ earthquakes (1951-2020) with magnitudes M > 4.95, using a revised magnitude scale. The inputs and hybrid models are defined on a spatial grid with 0.1 degree spacing.

Time-invariant inputs (covariates) include Proximity to the Plate Interface (PPI), Proximity to Mapped Faults weighted by slip rate (PMF), the Haines and Wallace (2020) maximum Shear Strain rate (HWS) and the presence or absence of a mapped fault in each cell (FLT). Time-dependent inputs are smoothed seismicity using a variety of spatial kernels and declustering methods. These are designed for fitting to seven 10-year windows with strict separation of contributing and fitting data.

We aim to optimize the information gain of hybrid models with respect to a spatially uniform baseline model. The most informative time-invariant covariate over the 70 years is PMF, followed by PPI, HWS and FLT. This contrasts with previous findings that strain rates are more informative than all other covariates for forecasting ahead over one or two decades. The information value of smoothed seismicity covariates depends on details of how they are defined. When extracted from declustered learning catalogues and combined in additive and multiplicative mixtures they are more informative than all other covariates.

So far, the best performing hybrid is a multiplicative combination of PMF, HWS and a mixture of two smoothed seismicity models. The latter models have Gaussian 50 km and power law spatial kernels, respectively. The final selection of the distributed seismicity model(s) for the NZNSHM will depend on a range of considerations, only one of which is the information gain.