- Ground-motion model is:
_{1}= 9.6231, c_{2}= 1.4232, c_{3}= -0.1555, c_{4}= -1.1316, c_{5}= -0.0114, c_{6}= -0.0024, c_{7}= -0.0007, c_{8}= -0.0835, c_{9}= 0.1589, c_{10}= 0.0488, τ = 0.406 (inter-event), σ = 0.568 (intra-event) and σ_{T }= 0.698 (total). - Use 3 site classes (2 are based on Eurocode 8):
- B
- 360 < V
_{s,30}≤ 800m∕s. Includes a few class A (V_{s,30}> 800m∕s) sites, which are included here because of scarcity of data. S_{b}= 1 and S_{c}= S_{s}= 0. - C
- 180 < V
_{s,30}≤ 360m∕s. S_{c}= 1 and S_{b}= S_{s}= 0. - Soil
- Sites that could not be assigned into B or C. Generally stations in Mexico, India and Martinique.
S
_{s}= 1 and S_{b}= S_{c}= 0. B, C and soil sites are grouped within regression to find c_{10}.

- Classify records into 2 classes on location w.r.t. arc:
- Fore-arc
- ARC = 1.
- Back-arc
- ARC = 0.

For Romania this is based on location w.r.t. Carpathian Mountains.

- Focal depths, h, between 87 and 151km for Vrancea events and 60 and 173km for others.
- Investigate the impact of the fore-arc and back-arc regions on ground motions, in particular w.r.t. the Vrancea intermediate-depth source. Aim to improve model of Vacareanu et al. (2014) (see Section 2.385).
- Use data from NIEP, INCERC, CNRRS and GEOTEC analogue and digital networks. Use some already-processed records so processing procedures are not uniform. Filter raw analogue records using Ormsby bandpass filter with cut-offs 0.15–0.25 and 25–28Hz. Filter raw digital records using 4th-order Butterworth bandpass with cut-offs 0.05 and 50Hz.
- Include data from foreign intermediate-depth in-slab earthquakes in Japan (fore-arc data with r
_{epi}≤ 80km to limit number of records), New Zealand, Mexico, Chile, India, Martinique (France) and Peru to fill in gap in magnitude-depth-distance distribution. - Fore-arc data from 8 countries. Back-arc data from only Romania and Japan.
- Limited Romanian data for r
_{epi}≤ 80km and larger number for 100 ≤ r_{epi}≤ 200km. - Try fixing c
_{4}to -1 but find that this is not valid at all periods. - Find average misfits (offset) (using total residuals) for entire set is around 5%, with peaks of 10%, and for Vrancea records only it is around 10%, with peaks of 18%.
- Derive model without site terms but based on analysis of variance and the F test on intra-event residuals w.r.t. site classes find that site effects are significant at the 5% level for all periods except 0.1s.
- Examine inter- and intra-event residuals and the average misfits. Find that ground motions from Vrancea and foreign events are similar and justifies inclusion of foreign data.
- Find that by examining inter-event residuals w.r.t. M
_{w}that obtained magnitude scaling is appropriate and that there is no difference between Vrancea and foreign data. - Examine inter-event residuals w.r.t. number of records per event. Find weak correlations (correlation coefficient in range 0.1 to 0.2) and hence conclude that no undue influence from well-recorded events.
- Examine intra-event residuals classified by site category w.r.t. r
_{epi}. Find no trends in residuals and hence conclude that distance scaling is appropriate, at least for B, C and average soil classes (there are insufficient records from Eurocode 8 A and D classes to conclude). - Examine intra-event residuals grouped into back-arc and fore-arc classes w.r.t. r
_{epi}. Find no clear trends. - Compare observed (corrected to soil class) and predicted motions for the M
_{w}6 27/10/2004 Vrancea earthquake. Find good match for both fore-arc and back-arc. - Do not advise using model for soil classes A or D because of lack of data from these categories and hence model is not calibrated for these classes.