search

seismology / mc_kernel

Calculate seismic sensitivity kernels using Axisem
GNU General Public License v3.0
24 stars 8 forks source link

Kernel values are not the same as Dahlen Kernels #23

Open sstaehler opened 9 years ago

sstaehler commented 9 years ago

comparison_dahlen_mckernel

I have taken this part from #22, since it is a new question.

I have calculated the Kernels with (presumably) the same physical parameters as @kasra-hosseini did with the Dahlen method, and they look very similar, alas, they differ by 7 orders of magnitude (left: MC Kernel, right: MC Tony Dahlen).

comparison_dahlen_mckernel_source Also, there are differences in the source region and with a second Fresnel zone above the kernel. On the one hand, this is something we expected, on the other hand, it might affect tomography quite significantly. Exciting times, people!

kasra-hosseini commented 9 years ago

Cool! Have you done this comparison for 60degrees? That should be interesting since Dahlen kernels should be quite reasonable for that epicentral distance. However for 90 degrees, the sensitivity kernel already starts to sense the CMB...deviating from ray theory approximation.

kasra-hosseini commented 9 years ago

Why the colors are flipped by the way?

sstaehler commented 9 years ago

What do you mean by flipped?

kasra-hosseini commented 9 years ago

That the blue part of the kernel is positive in MC and negative in Dahlen.

sstaehler commented 9 years ago

That is a very good question, I'll look into how that happened. Another question: Are the Dahlen Kernels absolute or relative?

kasra-hosseini commented 9 years ago

by relative you mean K.dV/V? Yes, and I am saying it because the sum over all elements will be the travel-time. Does it make sense?

sstaehler commented 9 years ago

I am not sure. Never got to think that through

tnissen commented 9 years ago

Did you honor the minus sign in the basic kernel definitions (eq. 5.13) ?

On 17/03/2015 19:09, Kasra Hosseini wrote:

That the blue part of the kernel is positive in MC and negative in Dahlen.

— Reply to this email directly or view it on GitHub https://github.com/sstaehler/kerner/issues/23#issuecomment-82539007.

Tarje

<>--<>--<>--<>--<>--<> Dept. of Earth Sciences Oxford University South Parks Road Oxford OX1 3AN; UK tel: +44 1865 282149 fax: +44 1865 272072 web: seis.earth.ox.ac.uk http://seis.earth.ox.ac.uk <>--<>--<>--<>--<>--<>

sstaehler commented 9 years ago

@tnissen Right, I forgot the minus sign, thanks. Fixed in 970a5328cbdfa622971ac1a8de83f613d6f44ae6

sstaehler commented 9 years ago

The different amplitudes might be related to an apparent error in normalization (see #24)

sstaehler commented 9 years ago

After fixing #24, the difference is 10^19 instead of 10^7.

comparison

sstaehler commented 9 years ago

It seems that the division by amplitude of the AxiSEM run is pretty inconsistent. Fans of the code do remember this discussion.

Might be related. Stay tuned for more

sstaehler commented 9 years ago

Almost solved by 6afd224. Now only a factor of roughly 5 is remaining. Also interesting is how the relative amplitudes in the middle and at the ends is different for MC Kerner (left) and MC Tony (right)

comparison

kasra-hosseini commented 9 years ago

Awesome!

About the relative amplitude in the middle and at the ends, and in general detailed comparison in the shapes of the kernels, I would suggest to do it in a very fine inversion grid, something like the following picture. I can prepare the required inversion grids as needed.

Kernel_fine

sstaehler commented 9 years ago

Yes, that would be great!

auerl commented 9 years ago

Cool! In terms of how relative kernels in [s/m^3] compare to published plots of kernels, they are now also relatively similar 5x10^-17 (our kernels) vs 5x0^-15 (published kernels).

screenshot

(for comparison, from Zhao & Chevron, 2012, in s/km^3)

zhao

kasra-hosseini commented 9 years ago

@sstaehler I thought that it is better to put the information for fine inversion grid here for future references.

Here is the address where you can find the fine inversion grid (ETH machine): /home/khosseini/inversion_mesh/mesh_example/fine_20_96

The most interesting part is: -45<=lat<=45 and lon=0 ---> edge-length = 20km the rest: 600km

WARNING: I changed the source code of the Dahlen kernels on ETH machine to have exact latitudes (in the original version, they will be translated into geocentric..., so please do not use the version in the ETH machine for tomography :) )

An example for 60degrees epicentral distance (source latitude = 30, receiver latitude = -30, longitudes = 0, depths = 0):

The sum over all elements were:-602.57562715 Theoretical arrival time: 608.28 Error: 0.9% (which should be acceptable with the criteria defined in Tian et al, 2007)

60degree_dahlen