Effects of the oceans benchmarking between the synthetics for the SPECFEM3D Globe, RegSEM, and the normal-mode synthetics

[earthinversion]

Hi,
I was benchmarking the SPECFEM globe (Version 6.0) synthetics with results of other solvers (normal-mode calculations, regional solver RegSEM) for the effects of ocean.

These are my input choices:

Model: Preliminary Reference Earth Model, with the ocean layer of thickness 3 km

• Spherical Earth: ON
• Anisotropy: ON
• Oceans: ON
• Topography: ON
• Gravity: OFF
• Attenuation: OFF
• Ellipticity: OFF

In addition to these, I have used the values of R_EARTH equals to 6368 in the setup/constants.h as suggested in the manual.

Findings

I found that the RegSEM and NMS results match fairly well (even though normal-modes are 1D solution) for the given input parameters while the SPECFEM results differ. The results differ significantly more than what we can expect following Komatitsch and Tromp, 2002.

Results

Please see the detailed test results at the Github repository: Compare-Synthetics-for-Earth-Models-Ocean-Effects

I am not sure what am I doing wrong. Please help. Thanks

[danielpeter]

hi Utpal, great idea to compare these methods and codes!

how does the comparsion look like for the simplest of cases, no topography, fully spherical, and PREM with and without the ocean layer?

in the past, one had to be careful with geographic vs. geocentric locations when benchmarking the globe version against other codes like normal modes. SPECFEM3D_GLOBE assumes that source/station/topography is given in geographic location and converts them internally to geocentric ones (for positioning etc.). to circumvent that, there is an option in the constants.h file:

logical, parameter :: ASSUME_PERFECT_SPHERE = .false.

which can be set to .true. to use the given source/station (& topography data) locations directly as geocentric ones.

note that for the geographic to geocentric conversion, SPECFEM3D_GLOBE uses a flattening factor:

double precision, parameter :: EARTH_FLATTENING_F = 1.d0 / 299.8d0

this is coherent with what spherical-symmetric models like PREM were using, but slightly different to other values, e.g., by the reference ellipsoid WGS 84 with f = 1/298.2572236.

[earthinversion]

Hi Daniel,
Thanks for the reply. I had compared the SPECFEM Globe and RegSEM for the case with:

• PREM Model
• no topography
• no ocean
• ASSUME_PERFECT_SPHERE = .true. in SPECFEM
• Others are same as above

The two cases seem to agree fairly well.

For the case with ocean, I have taken ASSUME_PERFECT_SPHERE = .true. as well. Could that be an issue?

[danielpeter]

looks better :)

and i like the apparent ScS reflection in the SPECFEM seismogram, which doesn't show up for RegSEM however, probably that mesh doesn't go as far deep.

i would guess the problem starts when using topography. topography data (ETOPOx) comes for geographic lat/lon positions. if you assume these are now geocentric positions, then that could be a difference between RegSEM. also, the ocean approximation is only applied when the ocean depth is more than 50m in SPECFEM3D_GLOBE. given your source/station setup includes some shallow ocean areas, that might be another discrepancy - although likely negligible for your long periods (> 40s).