PRVI GMMs
@mmoschetti and @kwithers have articulated the details of PRVI GMM requirements via email. At a high level the GMMs will use techniques and approaches used in past regional NSHMs, but combined in new ways.
Backbone Models (for crustal, interface, and intraslab)
- Backbone models will use adjusted and unadjusted forms of NGA-West2 and NGA-Sub GMMs plus MA05 (for both crustal, interface and intraslab). See documentation for weights
- Models of epistemic uncertainty will be magnitude, distance, and period-dependent with asymmetric upper and lower values.
- Models of aleatory variability will be decoupled from median models.
Implementation
- PRVI adjusted versions of NGA-WEST2 (4) and NGA-Sub (3) GMMs (as inner classes)
-
Factory class implementation of aleatory variability models (2)(inlined in backbone class) -
New PRVI backbone model class with inner classes for active crust, subduction interface and subduction intraslab.
- Follow implementation of the 'combined' GMMs that we used to provide weighted averages of GMM medians.
- Logic tree for each will include epistemic uncertainty branches on median ground motion and multiple sigmas.
- Combined PRVI models. These will support display in online tools of the underlying median models used in the backbone
Other Issues
- Ensure that implementations being used for Sammon's mapping and development of epistemic uncertainty model are consistent with nshmp-lib implementations (especially MA05). For example, check that T-interpolation being done by OpenQuake is consistent with nshmp-lib.
- Support for PGV. Add Abrahamson conditional model to python codes being used for Sammon's mapping.
- Support for GMM versions that allow separation of components per attached requested sensitivities
Adjusted GMM and Other Development
Per meeting with @mmoschetti @baagaard @ashumway we are going to implement versions of the NGA GMMs with alternate c0 and c1 (or equivalents) coefficients along with PRVI specific Vs30 scaling. These would replace the GMMs in the current data adjusted backbone model.
- Add PRVI coefficient tables and implement subclasses
- Update epistemic uncertainty branches with new NGA-West2 and NGA-Sub derived values