Select data with Mw> 4, rhypo< 230km, focal depth ≤ 30km and a value of Vs,30.
Use regression approach to investigate regional dependency.
Modelled derived using 1261 records from 437 earthquakes, separated thus: north Turkey (324), Appenines
(303), south Turkey (250), east Turkey (122), east Alps (91), south Greece (87), north Greece (58), Iran
(19) and Sicily (7 records). Coefficients can be regionally-dependent but they are constrained to fall on a
common low-dimension manifold.
Compare results with manifold-aligned approach to grouping all data together (global) or keeping data
from each region separate (regional). Manifold-aligned approach generally works better than global or
regional approaches for regions with limited (about 50–250 records); when fewer records recommend global
approach and when more records prefer regional approach.
Find coefficients using neural networks.
Use two faulting mechanisms:
ϕ = 0
ϕ = 1
Note only a few records for Mw> 7 and limited data from Mw> 6 and hence magnitude scaling would
be poorly constrained if using regional datasets individually.
43Mixture oflnandlogpresent in original formulation.