Continue work on transition of load_array and load_meta

[PeterDSteinberg]

How do load_array and load_meta from earthio relate to dataset loading capabilities in xarray?

For examples:

• Is there anything in the GeoTiff related logic for earthio that should be in xarray? The earthio GeoTiff reader (with rasterio) was a Phase I thing before xarray added GeoTiff support.
• Do earthio's load_array and load_meta need to exist? Can everything they do now be effectively pushed into xarray? This may already be the case for the NetCDF reader.

[PeterDSteinberg]

#17 is the same issue but more specific to GeoTiff file support.

[PeterDSteinberg]

SHould it be called 'load_mldataset'?

[gbrener]

Hey @PeterDSteinberg, responding to your last comment - do you think it would make sense to call this load_layers()? Since we support passing layer_specs (via a kwarg) and we've already centralized around the "layer" as a core abstraction? Just an idea. It might help us keep earthio decoupled conceptually from elm and xarray_filters. Happy to keep load_array() for now though, if you're not in favor.

In terms of integrating with xarray_filters / MLDataset, I was thinking we could add a MLDataset.load() method (or something similar), which could then call earthio.load_array() under the hood if necessary. That way we wouldn't have to return a data structure from earthio.load_array() that strictly depends on xarray_filters; perhaps it could be just a xr.Dataset or xr.DataArray? What do you think?

Regarding what we may be able to feed back to xarray from the GeoTiff work in earthio - if you recall the table I created a few months ago, I think we've closed the gap (to some extent) in xarray by preserving more of the GeoTiff metadata than was previously being collected. Also, the geotransform features - and I could be wrong here - were added to MLDataset with the Step work we did a few weeks ago. It might be a good idea to transition to using the xarray rasterio backend at some point if we want to reduce our maintenance burden.

[PeterDSteinberg]

• I like load_layers (plural form makes more sense + agreed that layers is more consistent with other work).
• Let's go with your suggestion of having load_layers return xr.Dataset (they all return MLDataset now, I'm pretty sure, so it will be simple to change to the more general xr.Dataset
• MLDataset.load makes sense to me (importing earthio as an optional import wherever it is used in xarray_filters
• Thanks for reminder on the GeoTiff capabilities table - Let's make a notebook documenting that (I'll create a separate docs issue reminder - I think overall one of the potential confusions of new users is what returns/uses DataArray vs Dataset vs MLDataset)

[gbrener]

Cool, thanks for the feedback. I'll get started on the implementation shortly.

[gbrener]

Closing this issue now that #39 is merged. Opened ContinuumIO/xarray_filters#43 to identify the remaining work left to be done, which is in the xarray_filters library.