Lat / Lon coordinate building

[DavidMcCann-NOC]

There is a problem with the way the latitude and longitude coordinates are built. Currently they are created using the LatImage, LonImage or OrbLatImage, OrbLonImage variables in the OSCAR datasets.

Like all of the other data arrays there are NaNs present in these variables (e.g., those sloping edges to the aquisitions) which mean that building the coordinates like:

level2.coords['longitude'] = level1.sel(Antenna='Fore').LonImage level2.coords['latitude'] = level1.sel(Antenna='Fore').LatImage

results in NaNs present in longitude and latitude. This causes issues when trying to plot with xarray, for example, as imshow doesn't like having non-numeric or masked values in the coordinate axes.

Additionally this has until now prevented the use of joining the beam datasets by 'outer', which would be preferable.

Either we need to:

• Build the lat/lon coordinates by interpolating out from what we have (undesirable)
• Find a way of having this handled when plotting (i.e. ignoring the NaNs, masking)
• Re-calculate the latitude and longitude images properly so there are no NaN values - i.e. like CrossRange and GroundRange images have.
• Ask Metasensing to provide the Lat/Lon coordinates as part of the data files

[DavidMcCann-NOC]

A visual example:

LatImage from the Mid antenna Fore antenna Aft antenna

So we have CrossRange, GroundRange coordinates in each dataset that go further than the data present in LatImage and LonImage. It might be as simple as adding lat/lon as coordinates to each dataset before xr.concat in merge_beams. I'll try this next