- Ground-motion model is (Model 1):
where pre is in cm∕s2, M

_{w,01}= 8.2, M_{w,1}= 16.0, a_{1}= -0.0321, b_{1,I}= -0.005315, b_{1,II}= -0.005042, b_{1,III}= -0.005605, c_{1,I}= 7.0830, c_{1,II}= 7.1181, c_{1,III}= 7.5035, d_{1}= 0.011641 and σ = 0.3761.Ground-motion model is (Model 2):

where pre is in cm∕s2, M

_{w,02}= 8.1, a_{2}= 0.5507, b_{2,I}= -0.004531, b_{2,II}= -0.004716, b_{2,III}= -0.005273, c_{2,I}= 0.4631, c_{2,II}= 0.5418, c_{2,III}= 0.9338, d_{2}= 0.006875 and σ = 0.377556.Both models include these adjustment factors (add to log pre):

_{d}= 0.0663, D_{l,min}= 100.00, D_{0}= 250, p_{s}= -0.3709, V_{s,max}= 1950.00, V_{0}= 350, γ_{NEJapan}= 0.00007602 and γ = 0.00006327. - Use V
_{s,30}and D_{l}(the depth to the layer whose V_{s}is l in m∕s) to characterise sites. - Use 3 types of earthquake:
- 1.
- Crustal. Use coefficients b
_{,I}and c_{,I}. - 2.
- Interface. Use coefficients b
_{,II}and c_{,II}. - 3.
- Intra-slab. Use coefficients b
_{,III}and c_{,III}.

- Focal depths, H, from 5 to 108km
^{42}. - Use data of Kanno et al. (2006) extended with data from K-Net, KiK-Net, JMA and Port and Airport Research Institute to the end of 2011.
- Data selection criteria are: M
_{w}≥ 5.5, record from ground surface, two orthogonal horizontal components available, ≥ 5 stations triggered by earthquake and r_{rup}< 200km. Truncate data at r_{rup}where PGA predicted by model of Kanno et al. (2006) < 10cm∕s2. - Few earthquakes for M
_{w}> 8. Lack of data for r_{rup}< 50km and M_{w}> 7. - Apply distance-dependent weight in regression to increase statistical power of near-source data (note
that no physical meaning of weights). Weights are: 8 for r
_{rup}≤ 10km, 4 for 10 ≤ r_{rup}≤ 20km, 2 for 20 ≤ r_{rup}≤ 40km and 1 for r_{rup}> 40km. - Based on the first step of a two-step analysis, find evidence for saturation beyond M
_{w}8. This analysis is basis of functional forms adopted. - Find σ of model 1 is slightly lower than that of model 2 but that this difference is not statistically significant. Hence cannot conclude which model is better.
- Adjustment factors based on analysis of residuals from model 1, which are assumed to apply also for model 2.
- To find G
_{d}use data with H < 30km (to avoid anomalous results from deeper events) and PGA< 100cm∕s2 (to avoid nonlinear site response). Use model of deep sedimentary layers in Japan to define D_{l}. Choose D_{1400}as D_{l}based on residual analyses and trial-and-error fitting of p_{d}for different D_{l,min}and D_{0}. Next fix D_{0}= 250m as average value for D_{1400}from first step and obtain D_{l,min}and p_{d}. - To find G
_{s}use same data as for G_{d}but with the additional criterion that the V_{s}profile down to 20m or more is known. V_{s,30}is estimated from V_{s,20}using previously-published conversion formula. Use residuals after correcting for G_{d}. Find V_{0}and p_{s}by trial-and-error analysis of residuals. Fix V_{0}= 350m∕s and find V_{s,max}and p_{s}. Note that this correection cannot account for differences in predominant periods between sites. - Use X
_{v,f}, the distance from a volcanic front to an observation site, to model anomalous motions for deep earthquakes. Use residuals after correction for shallow and deep site response (G_{s}and G_{d}) from data with H > 30km and PGA< 100cm∕s2. Use data from earthquakes in NE Japan from Pacific Plate (excluding those from stations south of 36N) and in SW Japan occuring in Philippine Sea Plate (setting absolute value of X_{v,f}to 75km or less) separately to find γ via weighted regression in three steps.

^{42}It is not clear if this is the entire depth range.