Improve treatment of missing data

[DaniJonesOcean]

At present, in the example notebook 1.0, missing data is either interpolated or replaced with a mean value. This is not likely to be the most appropriate method. We should decide how to treat missing data from the entire gyre dataset. For example, the data may be "missing" from certain depths because the profiles are taken from shallow bathymetry. It's not really sensible to replace with mean values below this bathymetry.

• Do we scale the data by the full depth of the bathymetry?
• Do we discard profiles that don't have the full depth, thereby missing the shelf entirely?
• Do we focus on just the shelf instead (not likely, there isn't much data there)

I plan to discuss this with collaborators at a meeting later today.

[DaniJonesOcean]

At present, we're using the approach of selecting the depth range and only keeping profiles that cover this entire depth range.

[DaniJonesOcean]

This approach seems fine for now. I'm not sure that scaling the data by the full depth of the bathymetry is a good idea. I'm not sure that makes much sense.