Action 5.3.2 : Weights for GMPEs

Technical Issue

How should data (empirical or simulated) be used to choose an exhaustive set of models, set the weights for GMPEs and exclude redundant models.

Proposed Approach

  • Literature review (many recent work), compare the SWUS method based on Sammon’s maps with the LLH method (Current methods do not use consistent treatment of correlations in the data when computing likelihood).
  • Assess whether selected GMPE span the space of possible models by visualization and mapping of GMPE’s: Sammons map, k-means
  • Model reduction: detect and delete redundant information
  • Model ranking by classical information criteria (AIC, BIC) & Bayesian Model Averaging (BMA) methodology to assign weights (no need for mutually exclusive of models) and to address epistemic uncertainty

Deliverable

  • Methodolog
  • application to case study