[Qopen] Using Qopen for coda Q determination or coda normalization

[Tom Eulenfeld]

Hello Petr!

you are welcome.

On 23.06.20 09:05, Petr Predein wrote:
> Hello Tom!
> I’m very grateful to you for your response and for further developing 
> Qopen, it’s very helpful.
> And I noticed that you updated Qopen package, as I can see there are 
> some changes in the imaging process, and also coda Q determination and 
> coda normalization added, isn’t it?
> And I would like to test this features. So, I tried to update Qopen 
> package, and for coda normalization I see, the global change is that now 
> total spectral energy density is normalized by mean value of coda-window 
> 20 length (from 180 to 200 seconds in your example, or am I misunderstood > But I get almost the same values for `g0` and `b` parameter but site
> amplifications differs, compared to results with original options, for 
> the Tutorial dataset
> So, could you please explain to me how to use Qopen to compare with coda 
> normalization.

Yes, coda normalization is a new option that is turned off by default. 
It was added to allow a better comparison to MLTWA (multiple lapse time 
window analysis) which employs coda normalization. That you get the same 
values for g0 and b is actually a good sign. If it produced a different 
result you would need to have a closer look.

As a side effect, site amplifications and source spectra are useless 
when using coda normalization. The only advantage of coda normalization 
is that the number of unknowns in the optimization is reduced, but if 
the original method gives similar results I would prefer this one.

> Also, unfortunately coda Q determination (these 4 options) qopen fails 
> with error:
> AttributeError: module 'qopen.rt' has no attribute 'G_diffapprox3d'

These new options are only available in the development version. How did 
you install? I suggest to install qopen-dev with

pip install git+https://github.com/trichter/qopen.git

Please run "qopen-runtests" or all tests with "qopen-runtests -a".

Does the error persist?

Note that this coda Q in Qopen is also only for comparison purposes. The 
result is an average coda Q. Often coda Q is calculated separately for 
each station and event. This is not possible with Qopen.

> ps. I must say, also it’s my first experience to make CodaNorm package 
> using a similar method. That script was a bit trivial and only 
> calculates RMS amplitudes in fixed windows for bulk wave and for 
> coda-wave (the further calculations was not my task, however, but they 
> showed a large scatter of results depending on the geometric spreading 
> value, for example) > Thanks in advance, cheers!

I have seen this code on github, but never used it.
But it makes sense that the result depends on the geometric spreading 
factor. You have to make some assumptions (single scattering vs 
diffusion) for calculation and interpretation of coda Q. The 
"G_diffapprox3d" from Qopen uses the diffusion approximation. Other 
Green's functions are possible and have its impacts on the results.

Cheers!
Tom