CSEP Minutes 03-14-2018

Csep call notes (03/14/2018):

Reviews on GEAR1 Global Experiement

[as] questions about the reviews of the GEAR paper. more about development of model than results.

[dj] should we define the magnitude or the moment magnitude? should use equation for the magnitude and moment. discrepancies between the conversion between moment and moment magnitude definition.

[mw] reviewer asked that the strain-rate forecast depends greatly on the upper end of the GR distribution where the relationship is clipped.

[dj] the numbers are quite sensitive in this upper clipping. 1 Mw unit = factor of 3. It was worked out by determining corner magnitudes in 5 different tectonic regions (e.g., subduction zone, continental transform faults) would match these corner magnitudes to compute the earthquake-rate from the strain-

rate. This requires some smoothing between the different boundaries to remove step-like changes.

[mw] summary review comments

1. difference between CSEP Mw threshold and gear1 mw threshold. This causes a difference in eq. rates but does not affect the results. [why?]
2. sensitivity of seismic rates that come from strain-rate models. Fitting corner magnitudes of tectonic region fixes this.
3. number of large eqs. insufficient to make statements about utility of strain-rate maps in seismicity models.

   [mw] we would like to challenge this criticism by stating that we want to show models how we will be using in future, and that strain-rates help prospective forecasting. this was confirmed by retrospective testing. reviewer is pessimistic about using only 2 years of data for a global model.

   [dj] issues: N-test and spatial distribution of the test. 2 years of data is not sufficient for N-test, could be rectified by using probabilities directly as opposed to Poisson assumption. All work indicates that spatial distribution is stable.

   [mw] is spatial distribution stable at a 2yr time scale? The relative likelihood scores are stable.

   [am] question about spatial distribution: huge difference between rates at plate boundaries vs everywhere else. Should get good scores at boundaries, but not elsewhere. Do these tests give information about performance of model in lower-rate areas? should talk about the stability of the tests.

   [mw] I think that we should be more precise about the timeframes and develop some ideas of how much data is absolutely necessary.

   [dj] something could be learned over 2 years, so maybe rephrase it instead of making it a binary

Canterbury Experiment Issue with STEPCoulomb

[mw] STEPCoulomb == STEPJava. but fixing might take some time. Might not happen before SRL reviews. Identified a fix and working toward implementing. Might have to remove some of the elec. supp. But opens the door for a new paper.

Special Issue

[mw] we have 10 papers, with some being resubmitted and reviewed. 1 accepted. Should be a good set of papers coming out of this. July/august issue.

Webpage

[mw] need a separate call for feedback for the scec.org/research/csep landing page. Need feedback from group.

5 year plan

[mw] to group: please review the 5-year plan and make modifications as necessary to the working group wiki page. A request has been made to develop plan and would take into account the commitments to the USGS.

[mw] Should we continue on with the old CSEP model? Or take advantage of man-power? How does CSEP IT fit into this problem?

[pm] have some responsibility to USGS to uphold our promises, so we don't have complete freedom to start over.

[am] the usgs needs the simulation-based testing, declustering issues, rupture association, ground motion testing, non poisson behavior problems. Could be a focus on the newer activities. could talk with mike blanpied. Want to test operational aftershock forecasting, testing of sequences, but need CSEP1 : style Poisson tests. Need to define what we need by simulation based testing.

[mw] learned from csep1 that the poisson assumption is too restrictive and need new developments scientifically to implement these new and useful testing.

[dj] ultimate goal to assign probabilities to things, but the question is what stage to we assign probabilities.

[mw] csep provides the methods to conduct scientifically interesting experiments. CSEP should continue adhere to prospective testing where we can along with reproducibility and transparency.