- Ground-motion model is: where Y is in g, a = -3.27, b = 1.95, c = -0.202, d = -3.11, e = 0.00751, h = 8.9km and σ = 0.399
for horizontal PGA and repi, a = -3.37, b = 1.93, c = -0.203, d = -3.02, e = 0.00744, h = 7.3km
and σ = 0.358 for horizontal PGA and rjb, a = -2.96, b = 1.79, c = -0.184, d = -3.26, e = 0.00708,
h = 11.3km and σ = 0.354 for vertical PGA and repi and a = -3.18, b = 1.80, c = -0.188, d = -3.13,
e = 0.00706, h = 9.1km and σ = 0.313 for vertical PGA and rjb.
- Believe relation valid for rather rigid soil.
- Use data from the Seismometric Network of Friuli-Venezia Giulia (SENF) (converted to acceleration),
the Friuli Accelerometric Network (RAF), data from the 1976 Friuli sequence and data from temporary
seismometric (converted to acceleration) and accelerometric stations of Uprava RS za Geofiziko (URSG)
of the 1998 Bovec sequence.
- Data from 1976 Friuli sequence is taken from ISESD. Records have been bandpass filtered with cut-offs of
0.25 and 25Hz. No instrument correction has been applied. Data from other networks has been instrument
corrected and high-pass filtered at 0.4Hz.
- Hypocentral locations and ML values adopted from local bulletins and studies.
- Use running vectorial composition of horizontal time series because horizontal vector is the actual motion
that intersects seismic hazard. Find that on average running vectorial composition is 8% larger than
the larger horizontal peak and 27% larger than the geometric mean. Find that using other methods to
combine horizontal components simply changes a by about 0.1 downwards and does not change the other
- Use data from 19 earthquakes with ML ≥ 4.5 (161 vertical records, 130 horizontal records).
- Note that distribution w.r.t. magnitude of earthquakes used roughly follows log-linear Gutenberg-Richter
distribution up to about ML ≥ 4.5.
- Few records available for d < 10km and ML > 3.
- Focal depths between 1.0 and 21.6km. Average depth is 11.4 ± 3.6km.
- Apply multi-linear multi-threshold truncated regression analysis (TRA) of Bragato (2004) to handle
the effect of nontriggering stations using the simplification that for SENF and URSG data the random
truncation level can be approximated by the lowest value available in the data set for that station. For
data from the 1976 Friuli sequence use a unique truncation level equal to the minimum ground motion for
that entire network in the dataset. Use same technique for RAF data.
- Develop separate equations for repi and rjb (available for 48 records in total including all from ML > 5.8).
Note that physically rjb is a better choice but that repi is more similar to geometric distance used for
seismic hazard assessment.
- Use ML because available for regional earthquakes eastern Alps since 1972.
- Conduct preliminary tests and find that weak-motion data shows higher attenuation than strong-motion
data. Investigate horizontal PGA using entire data set and data for 0.5-wide magnitude classes. Find that
attenuation is dependent on magnitude and it is not useful to include a coefficient to model anelastic
- Since data is not uniformly distributed with magnitude, inversely weight data by number of records within
intervals of 0.1 magnitude units wide.
- Because correlation between magnitude and distance is very low (0.03 and 0.02 for vertical and horizontal
components, respectively) apply one-stage method.
- Note that large differences between results for repi and rjb are due to magnitude-dependent weighting
- Plot predicted and observed ground motions binned into 0.3 magnitude intervals and find close match.
- Plot residuals w.r.t. focal depth, rjb and ML. Find that it appears equation over-estimates horizontal PGA
for df > 80km, ML < 3 and focal depths > 15km but note that this is due to the truncation of low
amplitude data. Check apparent trend using TRA and find no significant trend.
- Note that difficult to investigate importance of focal depth on attenuation due to unreliability of depths
particularly for small earthquakes. Find that focal depths seem to be correlated to magnitude but believe
that this is an artifact due to poor location of small earthquakes. Try regression using rhypo and find larger
σ hence conclude that depth estimates are not accurate enough to investigate effect of depth on ground
- Investigate methods for incorporation of site effect information using their ability to reduce σ as a criteria.
- Note that largest possible reduction is obtained using individual average station residuals for each site
but that this is not practical because this method cannot be used to predict ground motions at arbitrary
site and that it requires sufficient number of observations for each station. Using just those stations that
recorded at least five earthquakes obtain estimate of lowest possible σ by adopting this method.
- Try using a classification of stations into three site categories: rock (16 stations, 1020 records), stiff soil (9
stations, 117 records) and soft soil (4 stations, 27 records) and find no reduction in σ, which believe is due
to the uneven distribution w.r.t. site class. Find that the strong site effects at Tolmezzo has a significant
effect on the obtained site coefficients.
- Use Nakamura (H/V) ratios from ambient noise for a selection of stations by including a term g(S) =
cHVN(S), where N(S) is the Nakamura ratio at the period of interest (0.125–1s for PGA), in the equation.
Find large reductions in σ and high correlations between Nakamura ratios and station residuals.
- Use receiver functions from earthquake recordings in a similar way to Nakamura ratios. Find that it is
reduces σ more than site classification technique but less than using the Nakamura ratios, which note
could be because the geometry of the source affects the computed receiver functions so that they are not
representative of the average site effects.
- Believe equation is more appropriate than previous equations for ML < 5.8 and equivalent to the others
up to ML6.3. Discourage extrapolation for ML > 6.3 because it overestimates PGA in the far-field from