Observations below the physical model bottom

[denispauline]

[Comment from Kenny]

Bathymetric/Grid smoothing means that some observation depths, particularly around the shelf break, are deeper than the depth represented by the model .

The original use of numpy.interp silently assigns the deepest model-layer value to observations below the model bottom.

This all ties into how the depth is calculated as well because GETM reports layer thicknesses rather than physical layer depths, I added code that calculates depth from the sum(h), but that has cascading impacts on decisions like how we want to handle these values that are outside of the model domain. I currently retain observations below that bottom in the output table, but assign mod = NaN and exclude them from the statistics.

How do you want this to be handled, scaling depths to observations? Ie deepest observation is equivalent to the bottom layer, everything is scaled relatively to that?

[acapet]

My vote is that using the bottom model value by default for observations referenced deeper than the local model bathymetry is OK.

Potentially, we could consider a threshold of depth difference that would disregard the data point (e.g. if the obs is more than 10 m lower than the model deepest point). But at some point we would need to start opting for simple solutions in CodeBlue :).

Eventually we could tag those points in some way but I 'don't think it's relevant.
Also consider that many "layers" product are for bottom waters (typically AVG 0-10 m above bottom). Here bottom would mean "bottom", so we want to be able to validate the bottom value, even if the model's bottom is technically 2 meters above the true bottom.