- Ground-motion model is:
_{cr}but without F_{R}and F_{N}. h is depth to top of rupture for events with rupture geometry model and focal depth otherwise. x^{v}is horizontal distance of path through volcanic zones (straight line). S_{kNL}is previously-published nonlinear site response term. g_{um}is geometric attenuation rate for upper-mantle events. g_{L}is geometric attenuation rate for distances > 40km. cr is for crustal events, um is for upper-mantle events, intS is for shallow subduction interface events, intD is for deep suduction events with depth h > 25km and SL is for subuction slab events. C_{max}= 7.1 as determined by goodness-of-fit parameter. h_{1}= 25km and h_{intD}= 50km. - Use site classes of Zhao et al. (2006).
- Use three faulting mechanisms using definitions of Boore and Atkinson (2008):
- 1.
- Reverse. Plunge of T axis > 40
^{∘}. Use F_{R}. - 2.
- Normal. Plunge of P axis > 40
^{∘}. Use F_{N}. - 3.
- Strike-slip. Other plunge angles.

- Develop functional form because of previous observations on effect of mantle wedge etc. on attenuation rate.
- Purpose of model is to classify earthquakes into types rather than for ground-motion prediction. Hence coefficients not reported.
- Use Slab1.0 subduction interface geometry model (Hayes et al., 2012) for classification schemes.
- Use some data from Alaska, California, Turkey and Iran, Taiwan and Wenchuan (China) to supplement Japanese data for large crustal earthquakes and near-source distances.
- 31 Japanese earthquakes from before 1996 have M
_{w}≥ 6.5. - Use maximum log-likelihood as goodness-of-fit measure because σ not very sensitive to the selection of model parameters when they are close to the best solutions using a random effects model and maximum log-likelihood sensitive to biased residual distribution and it is important to eliminate any bias in residuals because models often extrapolated. Use Akaike information criterion to check if model parameter should be retained.
- Try 4 classification schemes and various combinations of earthquake catalogues. Find using locations from
the International Seismological Centre (ISC)/Engdahl et al. (1998) (EHB) for events before 2005 then
using locations of 1) Japan Meteological Agency (only those with high precision level), 2) ISC/EHB and
3) National Earthquake Information Center (excluding events with a fixed depth) for more recent events
produced best models in terms of maximum log-likelihoods. The best event classification is the following.
Classify reverse events, depth within 5km of interface, depth < 50km and dip angle for one of nodal planes
within 15
^{∘}from interface dip angle as interface earthquakes. Classify events above the interface that are not classified as interface and have depth ≤ 25km as shallow crustal earthquakes. Classify events above subduction interface but are not shallow crustal earthquakes are upper-mantle events. Classify all other earthquakes as subduction slab earthquakes. Note that this leads to highest maximum log-likelihood but little reduction in σ. Also note that this classification does not guarantee correct classification and that misclassification of a small number of events would not affect the goodness-of-fit parameters.