2.298 Lin and Lee (2008)

• Ground-motion model is:

where y is in g, C1 = -2.5, C2 = 1.205, C3 = -1.905, C4 = 0.516, C5 = 0.6325, C6 = 0.0075, C7 = 0.275 and σ = 0.5268 for rock sites and C1 = -0.9, C2 = 1.00, C3 = -1.90, C4 = 0.9918, C5 = 0.5263, C6 = 0.004, C7 = 0.31 and σ = 0.6277 for soil sites.

• Use two site categories (separate equations for each):
Rock
B and C type sites
Soil
D and E type sites
• Use two earthquake types:
Interface
Shallow angle thrust events occurring at interface between subducting and over-riding plates. Classified events using 50km maximum focal depth for interface events. 12 events from Taiwan (819 records) and 5 from elsewhere (54 records). Zt = 0.
Intraslab
Typically high-angle normal-faulting events within the subducting oceanic plate. 32 events from Taiwan (3865 records) and 5 from elsewhere (85 records). Zt = 1.
• Focal depths, H, between 3.94 and 30km (for interface) and 43.39 and 161km (for intraslab).
• Develop separate ML-Mw conversion formulae for deep (H > 50km) and shallow events.
• Use data from TSMIP and the SMART-1 array.
• Lack data from large Taiwanese earthquake (especially interface events). Therefore, add data from foreign subduction events (Mexico, western USA and New Zealand). Note that future study should examine suitability of adding these data.
• Exclude poor-quality records by visual screening of available data. Baseline correct records.
• Weight data given the number of records from different sources (Taiwan or elsewhere). Focus on data from foreign events since results using only Taiwanese data are not reliable for large magnitudes. Note that should use maximum-likelihood regression method.
• Compare predicted and observed PGAs for the two best recorded events (interface Mw6.3 H = 6km and intraslab Mw5.9 H = 39km) and find good fit.
• Examine residuals and find that a normal distribution fits them very well using histograms.
• From limited analysis find evidence for magnitude-dependent σ but do not give details.
• Note that some events could be mislocated but that due to large distances of most data this should not have big impact on results.