Access to additional variance components of GMM standard deviation models

[mmoschetti-usgs]

I don't think there is a straightforward way to access additional variance components (specifically, phiSS and phiS2S) from the GMM standard deviation models that compute these values. Are the OQ developers considering ways to evaluate and return these values in the GMM calculations?

[micheles]

Not at the moment. Saving such quantities would be hard and would require a lot of data storage, possibly slowing down continental scale calculations. If you are interested only in single site calculations, then there are no performance constraints; but then, you can just implement what you want at the level of hazardlib, without using the engine proper.
Maybe you can share some light on the use case you have in mind?

[emthompson-usgs]

I could be wrong, but I don't think that @mmoschetti-usgs is asking for anything that would require additional storage. For example, the Parker et al model already has the variance components in COEFFS attribute:
https://github.com/gem/oq-engine/blob/master/openquake/hazardlib/gsim/parker_2020.py#L386
So I think it is just a matter of providing a method to access that and potentially interpolate to an arbitrary period like what is done for computing the mean and the "traditional" three standard deviations.

[mmpagani]

Hi @emthompson-usgs and @mmoschetti-usgs. Sorry, I am seeing this thread only now. Is this something like this of any help?

from openquake.hazardlib.gsim.parker_2020 import ParkerEtAl2020SInter
from openquake.hazardlib.imt import PGA, SA
gmm = ParkerEtAl2020SInter()

coeffs = gmm.COEFFS[PGA()]
print(coeffs['phi2S2S,0'])

coeffs = gmm.COEFFS[SA(0.045)]
print(coeffs['phi2S2S,0'])

Unfortunately, we do not have standard methods to extract these parameters as the variability across various models is still too large. Note that in the example, the value for SA(0.045) is already the result of interpolation as this period is not available in the original table of Parker et al.

[emthompson-usgs]

@mmpagani Thanks, I didn't realize that COEFFS does interpolation. It would be nice to have a standard interface to these other uncertainty parameters eventually, but it makes sense that how they are reported by the different models is still too variable.

[mmpagani]

@emthompson-usgs Thanks. Let's think about this, it would be nice indeed to find a way to standardise this part. One initial simple but challenging thing would be standardising all the labels used to define the various explanatory variables and outputs.

[mmoschetti-usgs]

Hi Marco, Thanks for thinking about this. I will test your suggested approach to extracting the coefficients from Parker et al. For the moment, we have found a solution for computing the phi_{SS} and phi_{S2S} from Parker et al, but I think this may increasing be an issue for people wanting to compute partially nonergodic hazard. I understand that not all models provide these partially nonergodic variance components, but I suspect that this will increasingly be important. We would like to test some partially nonergodic hazard calculations, which will presumably make use of multiple models, so finding a solution that standardizes output for variance components would be valuable. Best, Morgan From: Marco Pagani ***@***.***> Date: Wednesday, August 7, 2024 at 9:06 AM To: gem/oq-engine ***@***.***> Cc: Moschetti, Morgan P ***@***.***>, Mention ***@***.***> Subject: [EXTERNAL] Re: [gem/oq-engine] Access to additional variance components of GMM standard deviation models (Issue #9783) This email has been received from outside of DOI - Use caution before clicking on links, opening attachments, or responding. @emthompson-usgs<https://github.com/emthompson-usgs> Thanks. Let's think about this, it would be nice indeed to find a way to standardise this part. One initial simple but challenging thing would be standardising all the labels used to define the various explanatory variables and outputs. — Reply to this email directly, view it on GitHub<#9783 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ABX5IHV5KHR5X2DUJP23YIDZQIZWPAVCNFSM6AAAAABJLEKYKWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDENZTGY4TMNZXGA>. You are receiving this because you were mentioned.Message ID: ***@***.***>