- Ground-motion model is not given here since it requires evaluation of a matrix equation that cannot be summarised. Study similar to Derras et al. (2014) (see Section 2.380). Model derived using an artificial neural network.
- Investigate the effect of using different site-condition proxies (SCPs) on predictions and particularly on σ
(separated into τ and ϕ). Consider 4 SCPs, which use in models in terms of their logarithms as find they are
lognormally distributed:
- 1.
- V
_{s,30}: with values 152.94–1432.8m∕s. - 2.
- Topographic slope: with values 0.0025–0.3748m∕m.
- 3.
- Fundamental resonance frequency, f
_{0}: with values 0.22–22.72Hz. - 4.
- Depth at which V
_{s}> 800m, H_{800}: with values 1–550m.

- Derive models using no SCPs, each SCP individually, the 6 unique pairs of SCPs, the four unique triples of SCPs and, finally, a model using all 4 SCPs. Find some correlation between SCP pairs but sufficient scatter to consider them as almost independent.
- Find choice of SCP has little impact on predictions of the median but it does affect σ.
- Use data from KiK-net database of Dawood et al. (2016) because all considered SCPs are available for many sites. Use data from 199 different sites.
- Focal depths from 0 to 30km.
- Few records for < 10km and when only considering stiff-to-rock sites few records for < 30km.
- Provide ranges of applicability of models w.r.t. independent parameters by considering 5th and 95th fractiles of observed distributions.
- Find evidence of nonlinearity in site amplification.