- Ground-motion model is:
where y is in g, C1 = -5.60, C2 = 1.63, C3 = -1.70, C4 = 0.51552, C5 = 0.63255, C6 = 0.0075,
C7 = -0.27 and σ = 0.61.
- Use V s,30 to characterise sites. V s,30 are from PS logging at or near the station.
- Focal depths 11 ≤ H ≤ 45km, with only one earthquake with H > 27km.
- Use data from Taiwan Strong Motion Instrumentation Program from 1995–2013.
- Only use data from normal-faulting events, which are defined as those with rake between -60 and -90∘.
- Only include earthquakes with Mw ≥ because focus is on events that can cause damage. Only one
earthquake has Mw > 5.1.
- Only retained earthquakes recorded by ≥ 20 stations.
- Baseline correct records after removing instrument response. High-pass filter (Butterworth) data using
cut-offs determined by the signal-to-noise ratio in respective displacement waveform. Signal amplitude
determined by averaging the absolute displacement within a 10s time window starting from P arrival.
Noise amplitude estimated using 10s time window before P arrival. Use automatic procedure to determine
filtering band to obtain signal-to-noise ratio of ≥ 14 by incrementally trying high-pass filters with cut-offs
increasing by 0.01Hz. Finally visually check data.
- Find that C4, C5 and C6 are fairly consistent across periods so fix them to avoid trade-offs between
- Plot residuals as histogram, w.r.t. Mw, V s,30, H and R and find no trends and good match to lognormal
- Compare predictions and observations for 201102010816 earthquake (Mw4.9, H = 23km), for a Mw4.1
event not used because fewer than 20 records, and two earthquakes in the USA (Borah Peak, Mw5.1, and
SW Nevada, Mw5.7). Find most observations within one standard deviation of prediction even when data
not used in regression.