- Ground-motion model is:
where Y is in m∕s2, a = -3.07±0.14, b = 0.73±0.03, c = -0.76±0.06, h = 1.7±2.3, d = -0.0029±0.0002, e = 0.326 ± 0.006, τ = 0.17 (inter-event), σ = 0.34 (intra-event) and σ = 0.38 (total).

- Derive station terms for each site because insufficient geological information to use, e.g., V
_{s,30}. Firstly derive reference model not including site effects. Then compute first-order station corrections from residuals w.r.t. reference. Use Z-test at 95% confidence level to check null hypothesis of a zero mean. Stations where null hypothesis was rejected are assumed to have a site effect. Derive site correction for each station by assigning s = 0 to all sites with zero-mean residual, s = 1 to stations with positive deviation and s = -1 to stations with a negative deviation. Prefer adjusted model based on the total σ and R^{2}statistics. - Data from 132 stations of Korean Seismic Network from March 2007 to March 2012. Data from both accelerometers and weak-motion (broadband and short-period) sensors. Also data from 24-bit digital recorders. Analyse about 30000 records in terms of signal-to-noise ratios and by visual inspection. Use only records with signal-to-noise ratios > 5. Data from over almost all over South Korea, both inland and offshore. Remove mean and trend from records and bandpass filter using a zero-phase 4-pole Butterworth filter with cut-offs of 0.1 and 20Hz. Select motions by taking portion from 2s before P-wave arrival up to time corresponding to 98% of total energy. Signals tapered by cosine function.
- Only two earthquakes with M
_{L}> 4. - Compare predictions and observations for the 1/4/2014 M5.1 earthquake that was not used to derive the model. Find quite a good match.
- Compare reference model to data in magnitude bins around 2.5, 3.5 and 4.5.
- Examine residuals w.r.t. M and r
_{epi}and find no trends.