- Ground-motion model is:
where Y is in g, a = -3.19±0.02, b = 0.87±0.01, c = -0.042±0.002, d = -1.92±0.01, e = 0.249±0.005
and σ = 0.316.
- Use two site classes, because lack local geological information (e.g. average V s):
S = 0
- Rock, eight stations, 3790 records.
S = 1
- Soil, seven stations, 3109 records.
Classify station using geological reports, ML station corrections and H/V spectral ratios computed over a
30s wide time window of S waves for entire waveform data set.
- Data from Regional Seismic Network of Northwestern Italy and Regional Seismic Network of
Lunigiana-Garfagnana (ten Lennartz LE3D-5s and five Guralp CMG-40 sensors with Lennartz Mars88/MC
recording systems). Sampling rate either 62.5 or 125samples∕s. Records from broadband and enlarged band
seismometers converted to acceleration by: correcting for instrument response, bandpass filtering between
1 and 20Hz and then differentiating. Accuracy of conversion verified by comparing observed and derived
PGA values at one station (STV2), which was equipped with both a Kinemetrics K2 accelerometer and a
Guralp CMG-40 broadband sensor.
- Find strong attenuation for short distances (< 50km) and small magnitudes (ML < 3.0).
- ML calculated using a calibration formula derived for northwestern Italy using a similar dataset.
- Compute signal-to-noise (S/N) ratio for the S phase using windows of 3s wide and find that data is good
quality (85% of windows have S/N ratio greater than 10dB. Only use records with S/N ratio > 20dB.
- Most earthquakes are from SW Alps and NW Apennines.
- Most records from earthquakes with 1 ≤ ML ≤ 3, small number from larger earthquakes particularly those
with ML > 4. ML < 1: 1285 records, 1 ≤ ML < 2: 2902 records, 2 ≤ ML < 3: 1737 records, 3 ≤ ML < 4:
693 records and ML ≥ 4: 282 records.
- Data shows strong magnitude-distance correlation, e.g. records from earthquakes with ML < 1 are from
0 ≤ R ≤ 100km and those from earthquakes with ML > 4 are mainly from R > 50km. Distribution is
uniform for 2 ≤ ML ≤ 4 and 0 ≤ R ≤ 200km.
- Originally include an anelastic decay term (d1R) in addition but the value of d1 was positive and not
statistically significantly different than zero so it was removed.
- Regression in two-steps: firstly without site effect coefficient (e) and then with e added.
- Compare data to estimated decay within one magnitude unit intervals and find predictions are good up
to ML = 4.0.
- Find no systematic trends in the residuals.