Anomalies and Hovmoeller plots break with multiple countries

[luigibrancati]

I selected multiple countries (Italy, Spain, France and Portugal) on the Streamlit app and tried generating the Anomalies and Hovmoeller plots, but they broke. The issue is with the clipping of the xarray dataset most likely.

[elisaliv]

@GiuliaFiantanese and I started investigating this bug tonight. We tried to replicate your error selecting the same countries listed in the issue:

• Anomalies is not working, and the error looks related to clipping as you speculated:
• Latitude Hovmoeller is working, but it does not involve clipping anyway. We should make it clearer that it's not related to the selected countries.
• Pressure level Hovmoeller is working: However, it only works when selecting multiple levels. We should add a note that you cannot create a pressure level Hovmoeller plot with only one level.
• Model level Hovmoeller is not implemented.

So, these are the next steps to solve this issue:

• [x] Debug the error for the anomalies plot, by calling the function on a Jupyter notebook
• [ ] Add a note or re-think the UI for the latitude Hovmoeller plot, to make it clear that it's a global plot and unrelated to selected countries
• [ ] Add a note for the pressure level Hovmoeller plot, and/or force in some way selecting multiple pressure levels. This kind of plot does not make sense for a single level anyway
• [ ] [Out of scope for this issue] Implement model level Hovmoeller or consider removing it if it's not needed

[elisaliv]

We further investigated the issue. The problem seems to be with Portugal data, as we had already noticed at the beginning. If we try to create the anomalies plot with multiple countries that don't include Portugal, we do not get any error.

In the clipping function (clip_and_concat_countries, line 24 with rio.clip), if we have drop=True as argument we get this error:

NoDataInBounds: No data found in bounds. Data variable: tcno2

If we use drop=False, the function works but there seem to be no data in the anomalies plot (for Portugal only):

We also tried to create a geospatial plot out of the same data, which supports this idea:

Strangely enough, we don't have such issues if we select Vatican as a country. So it's probably not a matter of small shapefile, but something else.

Our suggested solution would be to leave drop=False as default. This way, plots wouldn't break anyway. We should also discuss with mentors about this issue.

[luigibrancati]

I agree with everything except with the Latitude Plot: we should provide a clipping based on the bounding box of the country, i.e. the latitudes included in the selection