Spatial pattern in Nonlinloc results

[gerocampo]

Good afternoon, Im trying to use NonLinLoc to relocate earthquakes in La Palma (Canarias), where there are hundreds of seismic events in 2020, for the search method I using Octtree and I´m got some kind of spatial patterns in the earthquake relocations (please see the attached image), I tried different parameters in the LOCSEARCH sentence from the control file, without luck.

The velocity model is 1D, the grid size is dx=500m dy=500m

Can you please advice, which parameter is necessary to change to avoid this pattern?

Best regards

german

[gerocampo]

Good morning

I followed the recommendations from the issue number 17 about the LOCSEARCH parameters, using a 1D velocity model with a very small dy, dz model (50m, 100m, 200m) and different parameters for LOCSEARCH sentence, including maxNumNodes and stopOnMinNodeSize and still I´m getting the same result a nice square pattern with 183m between locations.

Any advice on what parameter can I change to avoid this effect?

[alomax]

Hello @gerocampo

Another cause of the gridded pattern is that either the pick uncertainties (GAU) or travel-time uncertainties (LOCGAU, LOCGAU2) are too large. Then the pdf for each location will be broad without a localized minima, and the search will not converge, instead it maintains the ~183m step and wanders across the pdf.

If you post your control file I can take a look if I see this or other problems.

And here is a control file I set up for La Palma recently, this may help you: https://www.dropbox.com/s/z5qsaehxfo20iza/LaPalma_2021_analytic.in?dl=0 Here are some of the results I obtained: https://twitter.com/ALomaxNet/status/1491457888057171968?s=20&t=cipiYFdjU-OOlgHCSLkfmg

Anthony

[gerocampo]

Anthony

Many thanks for the answer, your control.in file helps a lot! I compared the file to your file. and most of the parameters are quite similar. I have only three differences:

• My velocity model is simpler than your model, I think is not the big issue.
• the value in LOCGAU2. was slightly higher 0.05 instead of 0.02 .
• the parameters for LOCPHSTAT were different. I setup with your values.

Now, making the changes 2 and 3, in the control file, the output locations are not following the 183m grid. I will make some runs this week, to compare with your results.

Regarding to your results in twitter are quite impressive! very good!

Many thanks for your help,

German

[alomax]

Hi German

OK - good. LOCPHSTAT should not make a difference, as this only affects the various stat outputs. So it must be LOCGAU2. Changing this from 0.05 to 0.02 would reduce the pdf extent and should improve convergence, but you may still have some smaller scale pattern in the hypocenters. Here is what I get:

There is always some pattern with oct-tree is because the oct-tree search is deterministic with regards to the cell positions.

Anthony

[gerocampo]

Good afternon Anthony

In the last week I has been testing different values for LOCGAU2, at the end with Oct tree I got a small pattern, not significant.. many thanks for your help.

I have a question after processing all the seismic events I can see for some seismic events, changing just 0.01 in the value of LOCGAU or LOCGAU2, there is a change of 3 or 4 kms in the final position. As I newie in the matter, I just wonder if for a particular seismic event I make multiple runs varying LOCAU and LOCGAU2 parameters in certain range for example 0.01 to 0.05, Can I choose the best solution based in the lower RMS or minimum volume of the ellipsoid error of all?

Best regards

german

[alomax]

Such a sensitivity to a small change in the values seems odd. However, it depends on the scale of your study and extent of uncertainty (e.g. ellipsoid len3 or PDF extent). If this extent is >> 3-4km, and the ellipsoid or PDF is very elongated, then there the hypocenter may be only constrained along a line. In this case, small changes in the data or parameters may shift the maximum likelihood hypocenter strongly along this line. The expectation hypocenter should be more stable.

Does the above perhaps seem to explain your results?

Anthony