- Ground-motion model is:
_{1}= -0.033598, a_{2}= 0.49784, a_{3}= -0.14873, a_{4}= 0.22496, M_{h}= 5.6, b_{1}= -0.96495, b_{2}= 0.20938, b_{3}= -0.014, h = 1.36, c_{1}= -0.34393, ϕ = 0.65621 (intra-event), τ = 0.53412 (inter-event) and σ = 0.84611 (total) [take M_{h}, h and b_{3}from Rodriguez-Marek et al. (2011)]. - Use V
_{s,30}to characterise sites. V_{s,30}from 500 to 2000m∕s but only 94 records with V_{s,30}> 1000m∕s and majority of data from V_{s,30}< 700m∕s. - Use data from surface stations of KiK-Net (240 sites, mainly on weathered rock or thin sediments) and K-Net (165 sites, mainly on sediments) up to 2009. Use only earthquakes in the F-net catalogue to have consistent metadata. Data similar to that used by Rodriguez-Marek et al. (2011).
- Only use data from events with focal depth < 25km to consider only crustal earthquakes.
- Exclude offshore earthquakes except those in Sea of Japan with M
_{w}≥ 5.5. - Apply magnitude-distance cut-off using threshold of 2.5cm∕s2 and predicted median from model of Kanno et al. (2006) to exclude higher than average distant records.
- Correct records for linear baseline trends. Visually inspect records and remove faulty records (e.g. S-wave triggers) and shorten those with multi-events. Note that this does not cause significant loss of data. Add Tukey-windowing taper to last 2s and add zeros to obtain homogeneous durations. Re-sample records to 100Hz to remove difference between KiK-Net and K-Net records.
- Only limited near-source data (< 10km). Magnitude-distance coverage roughly uniform.
- Derive model to investigate site amplification at stiff-soil and rock sites and effect of κ not to present a new model for use in hazard assessments.
- Constrain some coefficients based on previous study because of interdependence between coefficients.
- Compare predictions and observations and inter- and intra-event residuals w.r.t. M
_{w}, r_{rup}and V_{s,30}. Find residuals well distributed except for possible slight underestimation for V_{s,30}≥ 1300m∕s but this is based on very limited data. - Believe higher σ due to consider variability in sites.
- Plot intra-event residuals w.r.t. κ computed using technique of Anderson and Hough (1984) for short periods. Find no trends. Note that this may be due to response of KiK-Net and K-Net instruments, which could lead to overestimated κ.
- Also consider various definitions of κ computed from response spectral shape based on stochastic
simulations. Compute κ based on one of these definitions for 53 sites with three or more records
(701 records and 123 events). Find that intra-event residuals w.r.t. κ computed using this approach
show clear trend. Derive new site function incorporating κ for T ≤ 0.2s based on residuals: F
_{S}= a_{1}ln(V_{s,30}∕800) + c_{2}κ_{0,RESP1}, where a_{1}= 0.42164 and c_{2}= -18.3175 (ϕ = 0.6194, τ = 0.54418 and σ = 0.82449) for PGA. Find that this new site term removes trend in residuals and reduces ϕ.