Difference between revisions of "CSEP Minutes 05-15-2019"
From SCECpedia
Jump to navigationJump to search (-) |
|||
| (10 intermediate revisions by the same user not shown) | |||
| Line 1: | Line 1: | ||
'''[[CSEP_Working_Group|CSEP Working Group Home Page]]<br>''' | '''[[CSEP_Working_Group|CSEP Working Group Home Page]]<br>''' | ||
| − | + | ''Participants: Andy Michael, Max Werner, David Rhoades, Giuseppe, Phil Maechling and William Savran | |
| − | + | ''<br> | |
| − | |||
| − | + | <b>Italian OEF</b> | |
| − | + | * Operational, but only available to civil protection. | |
| − | + | * Working on publicly available system, but needs work with different levels of society. | |
| − | |||
| − | |||
| − | |||
| − | |||
| − | |||
| − | + | <b>Overlapping Windows</b> | |
| − | How well can you predict the number of earthquakes immediately after an event versus at some time after an event? | + | * Using simulation-based tests the correlations are preserved so can compute cumulative statistics. |
| − | When do forecasts start becoming useful after a large earthquake? | + | * Testing all of the windows at once. |
| + | * Simulated likelihood distribution based on set of parameters within each forecast window. | ||
| + | * Compute likelihoods within each simulated distribution. | ||
| + | * Aggregate log-likelihoods from likelihood distribution. | ||
| + | * Worth investigating meteorology literature for their solutions. | ||
| + | * Separating testing windows useful for understanding model performance but not required mathematically. | ||
| + | * Marzocchi et al., 2019 applied method to estimate model performance after some period using power-law model. | ||
| + | * Usage must be considered to guide evaluations. | ||
| + | * For example, multiple time-windows from same start time assume that decision maker needs to make a decision now. | ||
| + | |||
| + | <br><b>CSEP Workshop Funded!!</b> | ||
| + | |||
| + | |||
| + | <b>Relevant questions:</b><br> | ||
| + | * How well can you predict the number of earthquakes immediately after an event versus at some time after an event? | ||
| + | * When do forecasts start becoming useful after a large earthquake? | ||
| + | * ETAS without parameter updating versus RJ with updating versus FastETAS. | ||
Latest revision as of 19:57, 15 May 2019
Participants: Andy Michael, Max Werner, David Rhoades, Giuseppe, Phil Maechling and William Savran
Italian OEF
- Operational, but only available to civil protection.
- Working on publicly available system, but needs work with different levels of society.
Overlapping Windows
- Using simulation-based tests the correlations are preserved so can compute cumulative statistics.
- Testing all of the windows at once.
- Simulated likelihood distribution based on set of parameters within each forecast window.
- Compute likelihoods within each simulated distribution.
- Aggregate log-likelihoods from likelihood distribution.
- Worth investigating meteorology literature for their solutions.
- Separating testing windows useful for understanding model performance but not required mathematically.
- Marzocchi et al., 2019 applied method to estimate model performance after some period using power-law model.
- Usage must be considered to guide evaluations.
- For example, multiple time-windows from same start time assume that decision maker needs to make a decision now.
CSEP Workshop Funded!!
Relevant questions:
- How well can you predict the number of earthquakes immediately after an event versus at some time after an event?
- When do forecasts start becoming useful after a large earthquake?
- ETAS without parameter updating versus RJ with updating versus FastETAS.
