Comparison of Canada model disaggregations for versions 3.11 vs 3.23

[Alberto91]

Hello,

I’m running a PSHA for a site in North British Columbia, using both OpenQuake 3.11 and 3.23. I have observed that in the disaggregation:

- in v3.11 all percent‐contribution bars appear at short distances (< 200 km).

- in v3.23 I see additional small bars at ~600–700 km and even farther.

After this I forced v3.23 to run a single realization (num_rlzs_disagg=1) and confirmed that its output matches v3.11, verifying that v3.11 was simply disaggregating one “closest-to-mean” logic-tree leaf. From this, I deduce:

• v3.11’s disaggregation picks the single realization whose hazard curve is closest to the mean hazard curve. Because crustal sources at < 200 km dominate the mean curve, that realization j is always crustal, so subduction/craton sources at > 600 km never appear.

• v3.23’s default disaggregation sums Λi​(m,r) over all realizations i, weighted by logic-tree probability. Hence, even ∼5 %-weight of the Queen Charlotte and Haida Gwaii Thrust or other distal sources could show up as small bars.

My questions are:

1. Is the behavior I see in  v3.23 (small ∼3–5 % bars at ∼600 km) fully expected for a Canadian logic‐tree or if there's something that I might be missing causing an error in my results?

2. When one realization is chosen (v3.11 or v3.23 with num_rlzs_disagg=1), that realization j is “closest to mean.” Is there ever a case where a subduction branch could win that “closest to mean” test? In other words, under what conditions (logic-tree weights or site location) might a subduction branch actually end up being the “closest‐to‐mean” realization?

Thank you in advance for any further insights!

[Sujan Raj Adhikari]

Just to adding things in your question  Have you done this for the Alberta, say, Peace River? The result in the Seismic Hazard Calculator Canada and the value from OpenQuake are also very different.  

[Yves Robert]

After running several tests with OpenQuake, I found that version 3.19 matches the Seismic Hazard Calculator Canada fairly well. Starting with version 3.20, the results provided by OpenQuake do not match the Seismic Hazard Calculator.

[Prajakta Jadhav]

Hello,

We are experiencing similar issue, for sites in North of British Columbia, for eg. Houston. The UHS from OQ 3.23 does not match with those using Seismic hazard calculator (NBCC 2020 tool); with % deviation up to 83% as given in the table below.

@OQ team: Please advise. Also, find attached the input files used for this analyses.

Thanks,

Prajakta

[Yves Robert]

I did an additional research to try to explain the difference in the seismic hazard provided OpenQuake 3.20+ and the seismic hazard calculator for the NBCC 2020.

What I found is that OpenQuake 3.19 and 3.20+ give the similar hazard curves. The problem is with the mean curve. It seems that OpenQuake 3.20+ no longer considers generic weights when an IMT-specific entry is absent.

To correct this problem for the Canada SHM6, you can change the .xml file where the weights are defined. The CanadaSHM6-GMM-LT_AA13_explicit_IMT_weights.xml is a modification of the official .xml file. The weights have not been changed, but explicit values have been added to get the results OpenQuake 3.19 gives.

I tested the file for the city of Montreal (eastern Canada), since published seismic hazard and disaggregation values are available, and OpenQuake 3.26.2 produced the correct results. I didn’t do any testing for western Canada.