CSEP Minutes 05-09-2018
Participants: W. Savran; A. Michael; A. Strader; D. Jackson; D. Rhoades; J. Gilchrist; M. Werner; P. Maechiling; T. Jordan
Minutes
Simulation Based Testing and Association Problem
- Need clarification and examples for simulation based testing methods. Some issues of doing these is storing long catalogs, so there are some computations challenges associated with dealing with these large files.
- USGS rolling out some software to separate the spatial kernel from the temporal kernel in a space-time ETAS model. Would save on computations required to generate forecasts with large (~Mw 7.0 eq). Outputs would be hazard be rate and hazard maps, but stochastic event sets could be buried.
- Tests should be directed towards interesting/useful scientific questions: comparing UCERF3-ETAS to non-fault ETAS... should include Bruce and Ned about discussions on these types of tests.
- Previous idea was to use pseudo-likelihood scores derived from stochastic event sets using Poisson likelihoods. There could be some concern with power and severity of these tests. Could be a worthwhile exercise with some simple examples to make clear.
- Some example catalogs from physics-based simulations would be useful as well, and how could RSQ-Sim be conditioned on certain sequences and events. How can RSQSim be involved in the association problem?
- With respect to the UCERF3-ETAS how can we associate the ruptures of real-events to faults/subsections in UCERF3. This is challenged by no authoritative data source for fault slip.
- Association could be posed in terms of probability distribution instead of central estimate?
- Propose challenge to Kevin and Ned to calibrate UCERF3 to some event (chosen at random from the number of >Mw 6.5 in UCERF3 over last 50 years). This could be used as a starting point for some discussion. Was this done before in BSSA [2017] - UCERF3-ETAS?
- Max: contract Kevin and Ned about the association problem in UCERF3-ETAS and how they would solve the problem.
Curated CSEP1 Data
- Need to think carefully about how to provide a curated CSEP 1.0 data set with respect to reproducibility. Most important to have a complete set of forecasts, a catalog, and a set of results.
- List of most important forecasts (and evaluations, catalogs) to include with this data set. (Call scheduled 05.21 to discuss this information)
SSA Annual Meeting
Tentative informal meeting on Tuesday evening to chat about CSEP things.
