Simplifying workflow to get from ESS Xarray datasets with a time dimension to a Pandas DataFrame for machine learning

[ThomasMGeo]

What should we add?

Many of our datasets in earth systems science have dimensions in x, y, z, and t. While this makes for easy multi-dimensional analysis, it does require a bit of work to get into the scikit-learn ecosystem for time series machine learning projects.

The workflow below might be out of scope for the MetPy project, but was curious if using some of the internal MetPy tools if this could be simplified and made more generalizable for all ESS xarray datasets with a time dimension.

Reference

Here is a short notebook with how the workflow could look using the xarray tutorial dataset.

[rpmanser]

If I understand correctly, you want to convert an xarray dataset to a pandas dataframe while keeping the dimension values as both indexes and column variables, right? If so, I think this may be a more streamlined approach:

import xarray as xr

ds_ar = xr.tutorial.load_dataset('air_temperature')
ds_ar['air_plus5'] = ds_ar['air'] + 5

coord_names = ['time', 'lat', 'lon']
df = ds_ar.to_dataframe().reset_index()
df = df.set_index(coord_names, drop=False)

[ThomasMGeo]

Hi @rpmanser , I need to close this. There is a new issue with more of a workflow sketched out. Thanks for the comment!

#3579