Expected and actual DOFS

[alvv1986]

Name and Institution (Required)

Name: Angel Vela
Institution: Aarhus University

New IMI feature or discussion

Hi IMI community,
Running the preview mode for a small domain over northern Europe provides a expected DOFS of ~12, but then when I run the inversion, it is calculated an actual DOFS of ~2. Can anyone please explain what might be causing this significant difference?
Thank you

[elenagjyli]

Here is another example of this issue as I have faced it a lo t during my work.
The preview gives expDOFs=41.82352 and from the notebook visualisation from the inversion the real DOFs=6.7951803
In most of my runs I notice that the difference between the expected and the real DOFs is more than 15.
But this is cancelling the value of the preview run, as it always gives a very big number.

imi_output.log
Neth_May_June_2019_g1_p1.pdf

[msulprizio]

Hi @alvv1986 and @elenagjyli. Thanks for writing. We were able to confirm this is happening for other regions as well. @laestrada tells me it is because of the updates in pull request #269. The calculation of DOFs in the IMI preview was updated there because the expected DOFs were previously too low. In particular, the mask around observations was expanded to account for transportation (see code below). Two concentric rings are currently used which was found to work well for the Permian case but now we see it might not work for other regions. It may be that we need to reconsider this code to better estimate DOFs across all regions in the IMI preview.

integrated_methane_inversion/src/inversion_scripts/imi_preview.py

Lines 593 to 603 in 0dc30a6

	# Following eqn. 11 of Nesser et al., 2021 we increase the mask
	# size by adding concentric rings to mimic transport/diffusion
	# when counting observations. We use 2 concentric rings based on
	# empirical evidence -- Nesser et al used 3.
	structure = np.ones((5, 5))
	buffered_mask = binary_dilation(mask, structure=structure)
	buffered_mask = xr.DataArray(
	buffered_mask,
	dims=state_vector_labels.dims,
	coords=state_vector_labels.coords,
	)