LOCTOPO_SURFACE causing NLLoc to crash for some shallow earthquakes

[CedricDM]

Hi Anthony,

I have been trying to use LOCTOPO_SURFACE to help locate my earthquakes below the topography/bathymetry. I am using quite a large DEM (the whole South Island of New Zealand with a large extension offshore).

I have been testing different NLLoc setups on a dataset containing 1000 events scattered across the region. Since using LOCTOPO_SURFACE, some events (39 exactly) are failing to locate. The reason for this is that NLLoc crashes when when generating the scatter file after completing the octtree grid search.

Without LOCTOPO_SURFACE, those events get located in the upper 5km, so just below the topography/bathymetry. I do not assume NLLoc is running out of memory, since it works fine for the other 900 or so events, which also contain shallow earthquakes.

Could it be an issue with the size of the scatter cloud extending above the LOCTOPO_SURFACE DEM?

Do you have any suggestions to solve this?

Many thanks,
Cedric

[alomax]

Hello Cedric,

Could it be an issue with the size of the scatter cloud extending above the LOCTOPO_SURFACE DEM?

I think that part of the scatter cloud extending above the topo surface should not be a problem, this probably happens often when the locations are shallow. But I wonder if maybe all or the majority of the scatter cloud being stuck up at the LOCTOPO_SURFACE might cause the crash. Can you take a look at the depth extent of the scatter cloud relative to the topo surface for the problem events when not using LOCTOPO_SURFACE? Are these events well constrained, or do they have extensive pdfs, ellipsoids, large residuals, etc ?

Is there any informative message when the crash occurs?

Thanks,
Anthony

[CedricDM]

Hi Anthony,

I looked at a number of these events. Generally, at least a quarter of the scatter cloud is above the topography, often by several 100m. The highest frequency of scatter points is also well above the topography. In addition, these events are generally not well constrained, they have rms values > 0.5s, depth uncertainties are quite variable but are always larger than the returned depth of the event, and most events have quite large ellipsoids (often >5km and >10km for the min. and max horizontal uncertainty, respectively). An example of the depth distribution of the scatter is shown below, where the red line indicates the elevation of the earthquake's epicentre.

I have tested removing some phases from the input file, namely those that are far away and those that have really high residuals. This caused NLLoc to not crash when running it with LOCTOPO_SURFACE for some of the affected events. After locating with those phases removed, the resulting locations were still shallow with no real change in epicentral location. However, depth generally increased and depth uncertainty decreased, resulting in, what I would guess, an omission of the LOCTOPO_SURFACE algorithm.

NonLinLoc does not give any error messages when reading in the LOCTOPO_SURFACE grid. However, after completing the octtree search for the affected events it prints the following:

Generating scatter file
Segmentation fault (core dumped)

Thanks,
Cedric

[alomax]

It sounds like not well constrained events with maximum likelihood hypocenter above the topo may be causing the problem. But there must be additional characteristics involved, otherwise the crash would probably have occurred for me and others.

Perhaps I can see see a setting in your NLL control file that is related to the problem.

But otherwise, I would need more information to debug. Can you reduce your study area, topo grid, vel grid (if 3D) and other file sizes to reproduce the problem with one or a few events? Then I could isolate where the code is failing.

An alternative might be to put very low velocity above the topo in your model (if 3D) - I suspect this will not give the crash.
But if you are using a layered model, this is not possible.

Anthony

[CedricDM]

Hey Anthony,

Below are the NLLoc settings of my control file:

LOCHYPOUT SAVE_NLLOC_ALL SAVE_NLLOC_EXPECTATION

LOCSEARCH OCT 36 96 21 0.001 295000 5000 1 0

LOCGRID 600 1600 350 -300.0 -200.0 -1.0 1.0 1.0 1.0 PROB_DENSITY SAVE

LOCMETH EDT_OT_WT 1000.0 4 -1 -1 -1 0 -1.0 0

LOCGAU 0.05 0.0

LOCGAU2 0.02 0.04 1.6

LOCPHASEID P P p G PN PG Pb Pn Pg PB
LOCPHASEID S S s G SN SG Sb Sn Sg SB

LOCPHSTAT 9999.0 -1 9999.0 1.0 1.0 9999.9 -9999.9 9999.9

LOCANGLES ANGLES_YES 5

LOCMAG ML_HB 1.0 1.0 0.0029

LOCQUAL2ERR 0.05 0.1 0.2 0.6 99999.9 99999.9 99999.9 99999.9 99999.9 99999.9

LOCFILES IN_1/*.nll NLLOC_OBS TIME_frd/frd29 OUT_1/located

LOCTOPO_SURFACE ./topo.grd.asc 0

I am currently using a model with that is corrected for bathymetry, but not for topography I think. For offshore, a low velocity and extremely high vp/vs layer are set in between -1 and 2-3km depth. This model, together with the expected location pulls most of the earthquakes down, especially offshore, except those really shallow and uncertain ones, which are all located onshore.

I tried reducing area size. However, the network I am dealing with is quite sparse (the national network with station spacing up to 60-80km in some places so I could not really go down to small grid sizes as I would not get a sensible location out of it. The smallest grid size I tested was 400km by 250km and down to 30km depth to include only 10 stations. For this grid, the problem still persisted it seems.

Cedric

[alomax]

Hi Cedric,

The problem might be SAVE_NLLOC_EXPECTATION. This option was added fairly recently, and as much NLLoc processing and post-processing makes use of the maximum likelihood hypocenter, there may (sometimes) be a conflict with the topo mask and enabling SAVE_NLLOC_EXPECTATION.

If this solves the problem, I will investigate further and see if I can find a fix...

Thanks,

Anthony

[CedricDM]

Hey Anthony,

Thanks for the suggestion. Disabling SAVE_NLLOC_EXPECTATION indeed solved the problem for the reduced grid size.
For the original grid it did not work initially. However, I noticed the bathymetry grid I obtained did not extend far enough south, so it was slightly smaller than my velocity grid. After solving this, the LOCTOPO_SURFACE working for the larger grid as well.

Cedric