VOXI magnetics: TMI or RTP and database vs. grid

PeterHedin

I'm a bit new to aeromagnetic data and to VOXI inversions.
My first question is if reduction to pole should be done on the data input to VOXI? I tested 1) using the TMI without the reduction to pole and then specifying the IGRF for the survey date in the VOXI data import dialog, and then 2) using data that were already reduced to the pole beforehand and setting the IGRF to I=90 and D=0 in the VOXI data import dialog. I end up with quite a large difference in terms of the location of subsurface bodies. So, I'm thinking if the IGRF in the VOXI data import is then not used for RTP or similar and whether RTP should or shouldn't be done when doing VOXI inversions...? The learning paths mention that it is good to remove a regional beforehand and not in the data import, but I see no mention of RTP.
Secondly I'm wondering if there is a general idea if it's better to invert on the database ("optimized" subsampling of real data) or grid (interpolated/averaged data)? Are there any rules of thumbs for when one or the other is more likely to give a better/stable result?

TaronishPithawala

Hi @PeterHedin,

Thanks for your question and your interest in VOXI.

RTP can be useful at times, such as when you're compiling different vintage surveys together with different IGRF values. Since VOXI can only deal with one IGRF setting at a time, RTP can be used to bring different surveys to one common IGRF prior to inversion.

RTP can also be seen as additional processing which may introduce additional artefacts into the data prior to inversion. Artefacts in usually equal artefacts out. Therefore, I don't personally recommend RTP as a common practice for inversion.

Regarding the difference you saw in the two results, one with RTP and the other... I can't speak to your exact instance because I'm not aware of the suitability of RTP to your particular dataset (e.g. was it near the equator to begin with?). However, I can show you an example where the inversion result of the RTP data was not appreciably different from the inversion result without RTP filtering (see below).

If all else is equal, including consistent trend removal between each project, the results should be the same assuming the RTP filter didn't introduce harmful artefacts into the grid. Trend removal can be done within the VOXI Add Data process - the options are simple however: no background removal, constant background removal, and linear background removal. While these are great places to start, you can of course experiment with more sophisticated regional-residual separation of your data prior to adding data to VOXI. I personally, usually begin with a simple linear trend removal. In the example below, I stuck with constant background removal to ensure that trends introduced or removed by the RTP filter don't skew the point I am trying to demonstrate.

Regarding the use of databases or grids, this is debatable. If the mesh design, the survey geometry, and the frequency content of interest in the data are all suitable for each other, then using a database as input is fine. Generally speaking most surveys, especially those of earlier vintage, were not designed with 3D inversion in mind. So you may find that your line spacing is too wide for the cell size of your mesh. In that case, I'd recommend gridding the data if you're fixated on a particular mesh cell size. Of course, you'll reach a point where the interpolation in the gridding is so vast, that you'll want to question whether you really should be inverting with such a small cell size given the spatial sampling of your survey.

PeterHedin

Thank you Taronish, I believe you've answered my questions. These tips on best practice and pros and cons of different options is valuable when you have limited experience to compare with.

FYI, and in case you have any spontaneous comments regarding the above questions on the use of RTP and input from grids vs. gdb: my AOI is centered at latitude ~50 degrees N and was surveyed in one go over about two months time. Line spacing is 100 m, average in-line point spacing is ca 3.5 m, and sensor was at 30 m above the terrain. I generated grids and also inverted with 25 m cell size in X and Y.
In some places the TMI input data (with IGRF specified in VOXI) and the RTP input data agree, but in some places the subsurface bodies resulting from inverting the RTP input data appear "shifted" to the North as compared with the other inversion results.

Thank you once more! Have a nice weekend!

TomPopowski

Hi @PeterHedin,
An additional factor - RTP assumes that the magnitude of any remanent magnetism is small compared to the induced magnetism. The differences in your results may reflect the distribution of remanently-magnetized rock in your survey area.

PeterHedin

Thank you Tom, That's a good point.

SteveLabranche1

Some articles discuss how to use RTP to locate remanent magnetization volumes. Over the last year, we have worked in proximity of large banded iron formation. In such cases, RTP is a bad idea prior to inversion. You can look at, paper 102 in "Proceedings of Exploration 07: Fifth Decennial International Conference on Mineral Exploration" p. 1095-1098

TaronishPithawala

@PeterHedin , just wanted to provide thoughts in addition to @TomPopowski 's comment regarding RTP suitability and shortcomings.

@AdamKroll had a great LinkedIn post a few years ago in which I commented on the possibility to create an RTP grid that does not suffer from the same numerical shortcomings of conventional RTP.

The method I described can be thought of as an "equivalent-source RTP", whereby a magnetization vector distribution is sought that fits the TMI (RMI) response at the usual IGRF, and then is forward modelled at the pole's IGRF to calculate an equivalent "RTP" response.

VOXI facilitates this approach with MVI...

One would simply use VOXI to run an unconstrained MVI of the TMI data at the local IGRF. Once a good data fit is reached, you can use the vector voxel result to forward calculate the TMI at the Pole. The result will not be affected by the low magnetic latitude, and will also allow for non-induced source magnetization.

I have included some slides that compare conventional RTP methods to the MVI RTP. The slides were prepared by @IanMacLeod for the ASEG, 2013.

PeterHedin

It's nice to see how the thread grows with more great stuff.

@SteveLabranche1 , thanks for the reference, I will check it out. There seems to be several nice papers in those proceedings.

@TaronishPithawala , it's a nice post, and together with your method gives me a few ideas that I think I have to test and see how they work in practice.

NickWilliams

There are many advantages to using VOXI MVI to generate an RTP image the way Taronish suggests:

1. As noted above RTP is not accurate if the magnetization direction changes within the area, which is true wherever there is remanent magnetization
   
2. An FFT RTP becomes unstable as the magnetic inclination gets low. There are many methods that address this, but all involved sacrifices.
   
3. FFT is not strictly accurate where there is drape, and more-so where there are different data altitudes within a survey with significant topography
   
4. FFT is not strictly accurate if the survey covers a finite area, or has holes or gaps. We manage this with padding and filling, but they are approximations.
   

Using a VOXI MVI to get an estimate of the magnetization in the earth allows us to manage all 4 of those limitations in a single solution.