Minutes from 4 February 2009

2009/02/04 CSEP meeting

Attendees: Yan K., Phil M., Danijel S., John Y., Masha L.

1. CSEP V9.1.1 release has been installed on csep-op Thursday, January 29, 2009:

   • Forecasts are generated for new groups that were introduced in CSEP V9.1.0:
     
     1. New forecast group for Global testing region that includes Anna Maria's model is still running - approximately 6 more days to finish
   • Yan will take a look at generated forecasts on the web
2. Masha circulated a document with requirements for model submission to the testing center
3. Matt and Masha upgraded New Zealand testing center to V9.1.0
4. Fabian sent tar ball with Japanese testing center codes
   • Masha submitted to SVN

5. csep-op is running out of disk space (75Gb left)

   • John will work on it
   • John will upgrade existing 30Tb storage for use by CSEP operational server
6. Tasks for V9.4 release:

   • RELM S (space) and M (magnitude) evaluation tests are available on csep-cert for RELM 5-year forecasts groups

   • Staging of existing raw catalog data to be used by re-processing - work in progress
     

     1. Masha discovered lost reproducibility for first 3 operational months of raw one-day forecasts files due to:

        • V1.0 didn't create metadata files for forecasts files as they were generated by the model codes
        • V1.0 did create metadata files for forecasts files converted to XML format, but these files were re-generated due to discovered bug (Trac ticket #91)
        • Possible solutions:
          
          1. Can manually re-store metadata files for forecasts based on easy isolation of run-time Dispatcher directories that store input catalogs used by these forecasts
          2. Can use metadata of XML format forecast to stage existing raw catalog - downloaded at time of re-creating XML format forecasts with CSEP V8.7.0
          3. Masha will talk to Danijel to decide which approach to use
   • New web representation of test results:
     
     1. Danijel suggested to create plots for all results on the web-server site
     2. Acceptance tests for web viewing code
     3. Will need to publish already existing test results in XML format, but extra care must be taken in posting re-processed test dates - only latest results should be posted
   • New web-page for the CSEP:
     
     1. Phil suggested to change color theme and to have a meeting with Tran and John M. to review existing themes
   • Modular functionality of the CSEP Toolkit:
     
     1. Most of the modules can be invoked in standalone mode
     2. There is a lack of documentation on how to invoke modules:
        • Use Doxygen?
   • Distribution of CSEP and dependent packages to other testing centers
     
     1. Phil, Danijel, John and Masha met on Monday, February 2, to discuss implementation design
     2. Might be time-consuming to set up rpm distribution, but beneficial not to provide support in a future to other testing centers when they upgrade CSEP
   • New models for Global testing region from Yan, Danijel:
     

     1. Yan has global forecast models using CMT catalog (M>=5.8) and PDE catalog (M>=5.0)

        • Danijel suggested to install CMT version of the model
        • Masha will send email to Yan with model requirements for the submission
   • Re-processing due to Trac ticket fixes:
     
     1. #110: RELM Mainshock forecasts groups should use declustered catalog for evaluation tests
     2. #128: N-test delta should be computed as poisscdf(N, lambda)
     3. #114: Cutting catalog to geographical region eliminates valid events if their longitude or latitude values fall on the area cell borders
        • Should stage existing raw catalog, re-create observation catalog, and time-dependent forecasts
     4. #113: ANSS events with missing depth value have their time set to 00:00:00 by Matlab import routine
     5. #121: r-test results flipped
   • OS upgrade to Fedora Core 10