Spatial aggregation

[ecomodeller]

In case one has a lot of measurements, either from many stations or from satellites, it is useful to aggregate the model skill in defined subdomains.

[Hendrik1987]

For satellite data, this is exactly what we have done for the GWM, building on existing MATLAB tools. One way of aggregation were regular lon/lat bins with results like this example. Having an improved support for altimetry data would be great and something I could happily contribute to.

[jsmariegaard]

I guess this is "solved" now by the area argument to the different skill and scatter methods (filtering) and by spatial_skill() from #38.

[Hendrik1987]

I think yes and no.

Yes, a skill dataframe per subdomain can already be achieved in 2 ways:

1. Loop over a list of areas and stack the resulting skill dataframes.
2. Extact the data with .all_df(), add a subdomain category column and add the column to the by argument of .skill() (similar to the low and high sea state example).

However, I think a more user friendly solution would be nice; And maybe a method to create a plot similar to above example.
The xarray based spatial_skill() is only suitable for a spatial aggregation on a regular grid.

One option could be to allow a list of areas as area argument. skill() would return a dataframe with subdomain as additional multi-index, scatter() could produce one scatter plot per domain (optionally as subplots), hist() could have stacked histograms and spatial_skill() could return a list of datasets.

I like the simplicity of this solution. I don't like that it would mix the selection and grouping part, which could be confusing. It is also very specific to the subdomain use case. Grouping by other variables or attributes would still need other solutions.

So, maybe it's time to think about a smart, scalable way to add categories of various types.