Ground-motion prediction equations for subduction zones based on strong-motion records from Japan

Executive Summary

In this report, a set of ground motion prediction equations (GMPEs) were developed for subduction zones, based on strong-motion data from Japan. GMPEs are a critical part of both probabilistic seismic hazard analyses (e.g., to determine the earthquake design loads for engineering structures, such as hydro-power stations), and earthquake loss modelling. The ground-motion parameters used in the this study are the peak ground acceleration and the 5% damped acceleration response spectrum. 

A GMPE is an empirical model derived from strong-ground motion records that are obtained from earthquakes with a moment magnitude about 5.0 or larger and at stations within a distance to the hypocentre or to the fault rupture plane less than 200–300km. As for all empirical models, natural variability dictates that a large number of strong-motion records should be used. For engineering design purposes ground-motions at distances over 150km are not as important as those from shorter distances. However, the number of records for a given moderately large earthquake increases with increasing distance, and the distant records are important for deriving parameters that represent site effects and parameters that control the attenuation of the ground-motion with distance. We used records up to a maximum distance about 130km for a magnitude 5 earthquake and up to 300km for magnitude 7 or larger events.

Earthquake ground motions at a soil site can be larger than those at a nearby rock site subjected to moderate ground shaking, and this is usually referred to as site amplification. This part of the GMPE has to be modelled appropriately, and requires detailed site information, such as the site class based on measured soil material properties, site period or shear-wave velocity profiles down to bedrock. Unfortunately, these high-quality site parameters are very currently limited in many countries, such as New Zealand.

The modelling of ground motions in an area with a subduction tectonic setting, such as in New Zealand and Japan, is more complex than that in an area which has essentially shallow crustal earthquakes only, such as in the California, the United States. The material properties of the crust usually differ significantly from those of the upper mantle, the mantle wedge and the subducting plate. The material properties close to the subducting interface, either between the crust and the subducting slab, or between the mantle wedge and the subducting plate differ from those of the subducting slab down to a depth about 50km. Thus the ground motions from earthquakes in different tectonic locations often have very different characteristics and require different GMPEs.

Japan has a large database of strong-motion records, has high-quality site information for a large number of recording stations, and has a reasonably reliable geometric model of the subducting plate, all of which are essential for developing a set of reliable GMPEs. Also
Japan is also the only country in the world that has a large number of strong-motion records and that has a tectonic setting similar to that of New Zealand. We have assembled nearly 15,000 strong-motion records from earthquakes with moment magnitudes of 5.0 or larger, which this study aims to develop a set of GMPEs that may be suitable for many subduction zones around the globe. The GMPEs will be optimised for New Zealand data in a future study.

The first step in developing GMPEs from records in Japan or other subduction zones is to classify the earthquakes according to their tectonic locations. We compared the earthquake information, including moment magnitude, focal mechanism and hypocentral location derived from three organizations, and designed four classification schemes. Next, we developed a set of statistically optimized GMPEs (without a requirement of having a broadly smoothed spectrum across the site period of 0.0–5.0s) using records from each scheme and then we compared the goodness-of-fit between the four models. We successfully derived the best possible tectonic category for 331 earthquakes. Soil sites usually develop nonlinear response during strong ground shaking. The nonlinear response reduces the amplification of soil sites and sometime leads to deamplification, i.e., the ground motion at a soil site being less than that at a nearby rock site. For engineering applications, the effect of nonlinear soil response is important. However, the nonlinear terms cannot be derived from the existing strong-motion records from Japan, and so we used numerical modelling of soil sites based on the measured site parameters of a number of selected Kik-net stations. We derived the terms that control the nonlinear site effects by numerical modelling and we adjusted the weak-motion amplification terms from the numerical modelling by applying amplification ratios derived from the strong-motion records.

Next we divided the strong-motion records into three groups according to the earthquake categories and developed a GMPE for each group. These GMPEs share most parameters for the nonlinear soil response part. We attempted to use relatively simple functions for the attenuation effect, and we also modelled the effect of volcanic paths over which the strong ground motion is further attenuated by the rock magma underneath the volcanoes. An important feature of the GMPEs is that the increase in the ground motion with increasing earthquake magnitude over 7.1 is much smaller than that from events with smaller magnitudes. This feature reduced the extent of over-predicting the response spectrum from large earthquakes, such as the MW=9 2011 Tohoku earthquake in Japan. For large crustal earthquakes, we used world-wide records to constrain the extent of increase in response spectrum with increasing magnitude.

We carried out extensive statistical tests to evaluate whether the site terms and model prediction uncertainty from the three sets of models were statistically similar. We also approximately separated the uncertainty associated with site effects from that associated with attenuation. These separated uncertainties can be used in probabilistic seismic analyses for different earthquake categories and different site classes.

In our study, the measured site parameters proved to be important and the overall variability as well as the average predicted spectrum are improved by excluding the records from sites with inferred site conditions. This demonstrated that the measured site information is critically important. For Geonet, the strong-motion recording network in New Zealand, the measured site conditions are still rare outside of Christchurch city.

For developing “next generation attenuation models” for New Zealand, where there are far fewer high-quality strong-motion records than in Japan, the present study may provide some useful clues, lessons and insight on the different geometry of the subducting interface and the different relative positions of the strong-motion recording stations with respect to the subducting plate or subduction trench. Some of the features from the present study can be adopted directly while the other features can be adopted with appropriate modifications.

This report contains three sections, with each section having its own abstract, introduction etc. and conclusions. This format leads to some repetition of material, but has the advantage that each part is almost completely self-contained, and so readers who are interested in a particular section do not have to read the whole report.

---

Technical Abstract

Establishing a set of ground-motion prediction equations (GMPEs) for Japan requires
earthquake source categories in the dataset. Earthquakes are typically divided into three groups: shallow crustal events that occur in the Earth’s crust, subduction interface events that occur at the interface between the crust or mantle and the subducting plate, and the subduction slab events that occur in the subducting plate.

In the present study, we compared the hypocentral locations published in the catalogues of the International Seismological Centre (ISC-EHB), the Japan Meteorological Agency (JMA) and the National Earthquake Information Centre (NEIC). The hypocentral locations for the same earthquakes vary significantly from one catalogue to another. We used the subduction interface model from the US Geological Survey, Slab 1.0, to help guide the classification. We designed four classification schemes using locations from these three catalogues. We then fitted a random effects model to the strong-motion dataset from these earthquakes to assess the merits of the classification schemes. Our results showed that using ISC-EHB locations for events before 2005, and then using the preference order of catalogues as: 1) JMA locations with high precision levels, 2) ISC-EHB, and 3) NEIC (excluding the events with a fixed depth) for events since 2005, together with some conditions for subduction interface events, produced the best GMPEs in terms of the maximum log-likelihood. We also found that having a separate group for the earthquakes above the subduction interface, but with a depth over 25km, improved the goodness-of-fit of the GMPEs.

Nonlinear site models are an important part of ground-motion prediction equations (GMPEs) and can be constructed in a number of ways. If a numerous soil site strong-motion records contain the effect of strong nonlinear soil response, the parameters for the nonlinear model can be a part of the regression parameters for GMPEs. However, the number of strongmotion records from Japan that contain the effect of strong nonlinear soil response is still too small to derive nonlinear site terms. It is also possible to derive nonlinear site models by numerical simulation. We present a model of nonlinear site terms using site class as the site parameter in GMPEs based on a 1-D equivalent linear model. The 1-D model was constructed based on the shear-wave velocity profiles from the Kik-net strong-motion stations with a wide range of site periods, soil depth and impedance ratios. The rock site strong-motion records were from different earthquake categories in Japan and the PEER dataset. Those records had a wide range of earthquake magnitudes, source distances and peak ground accelerations. A random effects regression model was fitted to the calculated spectral amplification ratios accounting for the effect of site impedance ratios, earthquake magnitudes and source distances of the rock site records. We also designed a method to adjust the 1-D model so that it can be used in a GMPE accounting for the fact that a 1-D model is an overly simplistic assumption for many real strong-motion recording stations in many parts of the world.

Ground motion prediction equations (GMPEs) derived from strong-motion records in Japan are presented. We assembled a large dataset from earthquakes with a moment magnitude (MW) over 4.9 and a reliable earthquake category, up to the end of 2012. The earthquakes were divided into four tectonic categories: shallow crustal, upper mantle, subduction interface and subduction slab events. The shallow crustal and the upper mantle (SC-UM) events were combined as one group, and a set of three GMPEs were derived for the SC-UM, subduction interface and subduction slab events, respectively.