where y is in cm∕s2, c1= 8.5851, c2= 1.4863, c3= -0.4758, c5= -0.00138, c6= 0.00484, σ = 0.491
(intra-event), τ = 0.550 (inter-event) and σT= 0.738 (total).
Data from 4 site classes based on Eurocode 8:
360 < Vs,30≤ 800m∕s.
180 < Vs,30≤ 360m∕s.
Vast majority of data from B, C and D except some class-A records from epicentral zone of Vrancea events.
Do not consider effect of site in model. Believe that model predicts motions on soil sites.
Use records of intermediate-depth earthquakes in Vrancea (recorded in Romania, Moldova, Bulgaria and
Serbia) and elsewhere (Japan, New Zealand, Mexico, Chile, India and Myanmar).
Focal depths 69 ≤ h ≤ 173km. Believe can be used for 60 ≤ h ≤ 200km
Data mainly from 100 ≤ repi≤ 200km.
Similar data distributions w.r.t. repi, Mw, h and site class for Vrancea and foreign events.
Compare predictions and observations from 3 most-well-recorded Vrancea events (1986, Mw7.1; 1990,
Mw6.9; 2004, Mw6.0). Find good match with most data within ±1σ.
Examine normalised residuals w.r.t. Mw, repi and h (binned into Mw ranges). Find no significant trends.
In particular find no evidence for magnitude dependency of τ.
Plot histograms of normalised total, inter-event and intra-event residuals and likelihoods and find that fit
normal distribution closely.
Mean, median and standard deviations of normalised residuals considering only Vrancea data are -0.06,
-0.03 and 0.82 and hence conclude that model can be used for Vrancea events.
Examine azimuthal dependency for Vrancea data by plotting normalised residuals on map. Find no evidence
for azimuthal variations nor w.r.t. site class but find slight underestimation in area in front of Carpathians
(fore-arc) and overestimation behind mountains (back-arc).
Cap magnitudes for Mw> 7.6 (T ≤ 1s) or Mw> 8 (T > 1s) to avoid decrease in predictions due to
quadratic M term. Note that this capping could be avoided by assuming linear M dependency but find
quadratic dependency fits observations better.