New Zealand NSHM Disagg

[Claudio Schill]

Hi all, 

I have successfully managed to compute single-site hazard using OQ for the 2022 NZ NSHM, and when compared with the published results GNS (https://nshm.gns.cri.nz/Hazardcurves) these match. However, when I try and do the same for disaggregation, results are different (see attached plots), and I was wondering if anyone has any insights/thoughts as to what the issue might be? Or if anyone has managed to run full LT disagg for the 2022 NZ NSHM?

Steps I took are:
1) Prepare inputs, see attached job ini and site file
2) Run disagg with "oq engine --run disagg_job.ini -p use_rates=true -p max_potential_paths=1000000"
3) Extract individual LT realisations (with "rlzs-traditional"), and then apply the realisation weights to compute the final disagg (https://github.com/ucgmsim/seismic_hazard_analysis/blob/nshm2022/seismic_hazard_analysis/nshm_2022/oq_interface.py#L200). 

Used OQ version 3.23.2

Thanks,

Claudio

(First plot is mine, second is based on csv from NSHM website)

[Marco Pagani]

Hi Claudio, why aren’t you using the OQ disaggregation output directly?

Regarding the comparison you are doing, we should discuss with Chris DiCaprio to understand how the original disaggregation results were computed.

Marco

--
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.
To view this discussion visit https://groups.google.com/d/msgid/openquake-users/1469d2bb-bd52-47dc-8005-1365e6c5486an%40googlegroups.com.

[Claudio Schill]

Hi Marco,

Thanks for the quick reply. The reason for not using the mean disagg from OQ directly is that I wanted it in traditional form (i.e. sums to 1), and from my understanding OQ only supports that for realisations, not stats. 
So that's why I decided to get the realisations with traditional, and then compute the mean disagg. However, I must be doing something wrong. 

I heard from Chris DiCaprio and he has successfully run disagg through OQ (with LT), and his results match the official results. The difference is that he extracted the mean disagg from OQ (in terms of PoE) and then converted that to contribution %. Will try and replicate his approach and see if I can get it to work.

Thanks,

Claudio

[Claudio Schill]

Hi Marco,

Thanks for the reply. The reason for not using the mean disagg from OQ directly is that I wanted it in traditional form (i.e. sums to 1), and from my understanding OQ only supports that for realisations, not stats. 

So that's why I decided to get the realisations with traditional, and then compute the mean disagg. However, I must be doing something wrong. 

I heard from Chris DiCaprio and he has successfully run disagg using OQ v3.20.1 (with sub-sampled LT), and his results match the official results. The difference is that he extracted the mean disagg from OQ (in terms of PoE) and then converted that to contribution %. Using this approach (and fixing a input difference), I have managed to replicate (with only very minor differences) the official NSHM results using OQ v3.20.1. 
However, when I do the same with OQ version 3.23.2, there are some differences compared to v3.20.1 and the official NSHM results. I have attached a notebook that compares the 3 different results, would be interested in your thoughts in regards to the differences between 3.20.1 and 3.23.2.
Note: I used the full LT to compute those results, so its not sub-sampling issue.

Turns out the html notebook is too large to upload here, have uploaded it on dropbox instead, 
https://www.dropbox.com/scl/fi/ggl5grbea2zxe7mbqfjms/disagg_v3p20p1_v3p23p2_comp.html?rlkey=4lyippakgdk95xez48rmbrj3d&st=zcpz5wea&dl=0

Thanks,

Claudio

On Wednesday, July 9, 2025 at 1:33:49 PM UTC+12 marco....@globalquakemodel.org wrote:

[Claudio Schill]

I have posted 2 replies, but both were deleted the next time I returned to this question. Not sure whats going on there.

If this stays up for more than a day then I'll try again...

On Wednesday, July 9, 2025 at 1:33:49 PM UTC+12 marco....@globalquakemodel.org wrote: