- Ground-motion model is:
where PGA is in cm∕s2, b1 = 2.479 ± 0.308, b2 = 0.513 ± 0.062, b5 = -0.800 ± 0.242, b9 = -0.003 ± 0.001
and σ = .
- Categorise 29 stations into 3 classes:
- Rock: Granite, sandstone, bedrock, siltstone and conglomerate. 17 stations, 54 records.
- Soil: alluvium, diluvium and weathered conglomerate. 5 stations, 61 records.
- Soft soil: clay and subclay. 7 stations, 17 records.
but site terms are not included in the preferred model.
- Use Heterogeneous Bayesian Learning (HERBAL) with an Automatic Relevance Determination (ARD)
prior to find simultaneously: the optimum functional form and the model of the ground-motion variability.
Approach explores many possible functional forms with different magnitude and distance scaling as well
as different homo- and hetero-scedastic σ models.
- Report coefficients for top eight models. Coefficients for only the top model are reported here.
- Plot residuals w.r.t. PGA. and compute variances of data binned into PGA intervals. Find that variance
from binned residuals match those from σ model derived using HERBAL closely.