@klaundal I've wrote some tests that compare ppigrf results against a precomputed grid I got from the NCEI Calculator. I'm seeing some discrepancies between them. Do you have any idea why it might be happening? It's due to floating point precision? Maybe a different set of coefficients? Would love to hear your opinion on this.

@santisoler I did a quick comparison for one point with the NOAA calculator now. The deviations are very small, but too large to be floating point precision I think. I doubt that the coefficient files are different. One possibility is that it is an issue with how the time interpolation between model coefficients is implemented. And/or the conversion between geodetic and geocentric coordinates and components (I get a very good match in the eastward component, which is not affected by this).

Still, the deviations are tiny, typically less than 1 nT. This rarely has any practical consequence, but it would be good to know why it happens. Comparison with an open code would be preferable though.

I suspect you're right and this is coming from the interpolations on the dates. I also checked ppigrf against outputs of igrf (which I suspect the NCEI calculator is running on the background) and I get the same differences. I couldn't generate a grid with igrf because their code was failing due to some Xarray error (probably they were using a different version of Xarray). But I checked with some individual points and the differences are still there.

I've just run a quick check comparing on 2020-01-01 (so no interpolation should be carried out) and the tests pass. So apparently the difference is coming from the date interpolation. I will push new tests with no interpolation on date.

Hi @klaundal! After a few weeks (time flies!) I managed to fix the tests so they pass. I've increased the atol in the second case (in which we have to interpolate along the dates). I think this is sufficient for now and it should be ready to be merged. Let me know what you think!