monte carlo calculation using a catalog of past events

[violet]

Hi OQ users,

I have a catalog of past eq events (data: lat, lon, depth, mag). Through ebil & kafka algorithm for example we can produce a synthetic catalog with a specified duration.

Now I have two questions:

1) How can we produce such synthetic catalog using OQ?

    the only way I could find was through source defining. which I don't want to define all sources since there are lots of them and I want to use existing catalog.

2) If I make a gmf file myself how can I use it as an input for event based risk calculator ?

    the --load-gmf command doesn't work anymore.

Best,

[Parisa]

I'm using oq engine 2.0.0

[Michele Simionato]

Il giorno giovedì 13 ottobre 2016 23:05:21 UTC+2, Parisa ha scritto:

Hi OQ users,

I have a catalog of past eq events (data: lat, lon, depth, mag). Through ebil & kafka algorithm for example we can produce a synthetic catalog with a specified duration.

Now I have two questions:

1) How can we produce such synthetic catalog using OQ?

    the only way I could find was through source defining. which I don't want to define all sources since there are lots of them and I want to use existing catalog.

I will leave this question to the scientific staff

 

2) If I make a gmf file myself how can I use it as an input for event based risk calculator ?

    the --load-gmf command doesn't work anymore.

The `--load-gmf` functionality is still there, but in order to use it now you have the change the job.ini and add a line like this

gmfs_file = gmf.xml

where gmf.xml is a NRML file containing the GMFs. However, notice that this feature is restricted to the scenario calculators.

It is not implemented for the event based calculators and was never there.

HTH,

       Michele Simionato

[Graeme Weatherill]

Unfortunately OpenQuake doesn't have a means of producing synthetic catalogues from the observed catalogue in the manner described by Ebel and Kafka. OpenQuake is currently capable of building synthetic catalogues from seismogenic source models, so as to ensure consistency between the event-based approach and the Cornell-McGuire approach.

There are certain facets of the Ebel & Kafka approach that could actually be replicated if you were to simply define your catalogue as a large set of point sources, with each earthquake location in the corresponding period of completeness representing the location of the point source. Aside from addressing issues such as focal mechanism, scaling relation, seismogenic depth etc., which are not considered by Ebel & Kafka, the main difference would be in the definition of the magnitude frequency distribution. Ebel & Kafka sample directly from the magnitudes in the catalogue whereas we assume a source-dependent magnitude frequency model. Again, one suggestion here might be to define, by virtue of a histogram, the observed number of earthquakes in the catalogue in a set of magnitude bins, then normalise these rates by the observed catalogue duration (adjusting for completeness) to determine an equivalent annual rate of occurrence in each bin. This would give you a total catalogue "EvenlyDiscretizedMFD", which is effectively non-parametric. As magnitude and location are sampled independently in the Ebel & Kafka approach, it may be the case that for each point source the assigned occurrence rates should simply be the total catalogue EvenlyDiscretizedMFD divided by the number of complete events. The temporal occurrence model should be Poissonian.

Whilst I cannot say for certain without implementing this approach, it may give results that are consistent with the assumption made by Ebel & Kafka. I invite other users who have experience in this area to provide feedback if they have it. If you were to try this approach, it would be interesting to look at the synthetic catalogues generated by your "equivalent catalogue-to-source model" in OpenQuake and to verify that the behaviour is consistent with that of the Ebel & Kafka algorithm in terms of the recurrence properties in the synthetic catalogue. If you, or anyone else on the mailing list, wish to try to implement the approach I have described then we here at GEM would be very interested to see the results!

Thanks,

Graeme

--
You received this message because you are subscribed to the Google Groups "OpenQuake Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to openquake-use...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Parisa]

Dear Graeme,

Thanks for your detailed answer

If I got your point right, this will fix the problem of magnitude selection, but remains location selection unsolved.

Through Ebel & Kafka method, the location of future events is determined by a distribution function and occurrences closer to the past events are more probable. 

Through Han & Choi method, the region under study is gridded and the number of events in each grid determines the seismicity of that grid. The probability of occurrence in each grid is determined after smoothing and we can use its cdf to pick up locations of future events randomly and again more probable in grids with more past occurrences. 

I think OQ picks up locations with uniform distribution, if it is so, the smaller are the sources area, more precise are the results. since I'm going to calculate risk in a country-wide scale this is not a time-friendly choice for me. 

 

[surajg...@gmail.com]

Dear Parisa

Have you managed to find a solution to this yet? (I want to achieve exactly the same). 

Any tools/software out there that facilitates this?

Any feedback would be much appreciated. 

Regards
Suraj