Seeking guidance on parameter optimization for volcanic setting

[10go-L]

Hello Anthony,

First and foremost, I want to express my gratitude for your continued support and invaluable advice on using NonLinLoc. Your guidance has been instrumental in my journey of learning and applying this powerful tool. However, I've encountered a few challenges that I'm hoping you might help me navigate.

I'm currently working on a volcanic setting with a seismic network comprising 4 stations within a 40km radius of the area of interest and 2 additional stations located approximately 80 and 100km away. My primary goal is to fine-tune the NonLinLoc parameters to improve the accuracy of event locations.

Here is the control file setup for the P picks:

VGGRID 121 271 85 0.0 0.0 -5 0.5 0.5 0.5 SLOW_LEN   

LOCFILES ./NLL/picks/pickfile.hpf NLLOC_OBS ./NLL/time/80km0.5 ./NLL/loc/ALL
LOCHYPOUT SAVE_NLLOC_ALL SAVE_HYPOINVERSE_Y2000_ARC SAVE_HYPOINV_SUM SAVE_HYPOELL_ALL SAVE_HYPO71_ALL SAVE_HYPOELL_SUM
LOCSEARCH OCT 10 10 10 0.05 250000 10000 1 1
LOCSTAWT 1 80
LOCGRID 121 271 65 0.0 0.0 -2.5 0.5 0.5 0.5 PROB_DENSITY SAVE
LOCMETH EDT_OT_WT_ML 9999 3 -1 -1 1.75 -1 -1 1
LOCGAU 0.1 20.0
LOCGAU2 0.1 0.01 5.0
LOCPHASEID P P p G PN PG
LOCPHASEID S S s G SN SG
LOCQUAL2ERR 0.05 0.1 0.2 0.5 99999.9
LOCPHSTAT 9999.0 -1 9999.0 9999.9 9999.9 9999.9 -9999.9 9999.9
LOCANGLES ANGLES_YES 5

While the error range (10km > len3 > 30km) seems decent, I believe there is room for improvement. I've experimented with various parameters, including LOCSEARCH, LOCGRID, LOCSTAWT, LOCGAU, and LOCGAU2, but I'm seeking your expertise for further optimization.

Additionally, I'm interested in using multiple grids, as LOCGRID is a repeatable parameter. However, I'm not sure if the algorithm is utilizing the second or third grids I've entered. My approach has been as follows:

# Coarse grid
LOCGRID 121 271 65 0.0 0.0 -2.5 0.5 0.5 0.5 PROB_DENSITY SAVE
# Refined grid
LOCGRID 50 50 40 -1.0e30 -1.0e30 -1.0e30 2 2 1 MISFIT NO_SAVE
LOCGRID 85 85 40 -1.0e30 -1.0e30 -1.0e30 0.5 0.5 1 MISFIT NO_SAVE
LOCGRID 50 50 50 -1.0e30 -1.0e30 -1.0e30 0.25 0.25 0.25 PROB_DENSITY SAVE

Could you provide any insights or suggestions on how to better configure these parameters for more precise location results? Your advice would be greatly appreciated.

Thank you once again for your time and support.

Best regards,
D

[alomax]

Hello D,

Thanks for your compliments. I attempt to answer your basic questions below, but without running the locations myself I do not have some of the basic understanding of your locations setup and behavior. Here are a few quick comments and suggestions:

While the error range (10km > len3 > 30km) seems decent, I believe there is room for improvement.

That seems a bit large, though with few stations nearby these errors may be reasonable. With good station coverage, many of the events might have ellipsoid len3 < 2-5 km.

Additionally, I'm interested in using multiple grids, as LOCGRID is a repeatable parameter. However, I'm not sure if the algorithm is utilizing the second or third grids I've entered.

There is only one LOCGRID needed and used for LOCSEARCH OCT. I highly recommend LOCSEARCH OCT, it performs extremely well, I now never use the nested grid search GRID or Metropilis MET.

Also, I recommend EDT_OT_WT and not EDT_OT_WT_ML, which does an extra, final search over origin time that does not seem to improve results much but is much less efficient.

VGGRID 121 271 85 0.0 0.0 -5 0.5 0.5 0.5 SLOW_LEN
This specifies a 3D grid. If you have a 1D, layered model, you can use a 2D grid, e.g. something like:
VGGRID 2 300 85 0.0 0.0 -5 0.5 0.5 0.5 SLOW_LEN
The grid can be made quite fine with a 2D grid, which will give a better representation of the model (if boundaries not exactly at 0.5 depth intervals, and of the travel-time field. For example, the above grid with 0.2 cell size:
VGGRID 2 750 213 0.0 0.0 -5 0.2 0.2 0.2 SLOW_LEN

LOCSEARCH OCT 10 10 10 0.05 250000 10000 1 1
The initial grid size should usually be proportional the the LOCGRID, and I generally use 50000 samples (25000 is often sufficient). I usually do not use useStationsDensity So I might modify the above to:
LOCSEARCH OCT 18 41 14 0.05 50000 1000 0 1

LOCGAU 0.1 20.0
The CorrLen I usually have as 0.0, as it is difficult to set and, in theory , depends on a scale length un-modelled velocity heterogeneity. I would have to dig into the code, but I think that with EDT CorrLen is not used.

Best regards,

Anthony