Package for GIS / Earth science, e.g. coord transform, satellite utils

[PeterDSteinberg]

Build a prototype showing how to coalesce the following sets of spatial utilities, such as satellite data transforms like reflectance/radiance calculations or NDVI, coordinate transformations, and regridding, broadcasting operations among rasters with grids of dif't resolutions. While prototyping / discussing, keep in mind the goal of making most or all the prototype part of Xarray, e.g. transitioning the .canvas attribute that is part of what is now Earthio.ElmStore (and the related canvas idea from datashader) to Xarray. This new prototype should avoid file loading logic, with goal of having Xarray do that (see #12 ).

• bokeh-geo
• elm.sample_util - See also this elm issue 149
• Earth data download utilities, such as those shown here for the ulmo package
• datashader's Canvas

Plan is for me to research some ideas on #12 over the next 2 days, then meet with @philippjfr and @jbednar later this week regarding how to address this issue in more detail, focusing on the .canvas ideas first.

[PeterDSteinberg]

The package created as part of this issue should have a lean conda recipe requirements, and rely on an environment.yml file here and/or elsewhere (e.g. datashader/examples, elm/examples) to tie together the GIS tools with ML, viz, and other tools (i.e. the conda.recipe for the new package is minimal but the environment.yml file may take an install-everything-viz-GIS-ML approach).

[jbednar]

Note that HoloViews "core" is essentially this Canvas functionality already, and it has no dependencies other than Numpy and Param (which themselves have no required dependencies).
See http://holoviews.org/Tutorials/Continuous_Coordinates.html for more details.

It's possible to simply use this portion of HoloViews as-is, which (a) has the benefit of then providing visualizations for free, if other libraries are installed, and (b) has the benefit of working across many different data backends (xarray, Numpy, Pandas, etc.). The disadvantage is that it's a fairly substantial collection of code, and may not all be relevant here. A middle ground is to provide a subset of what HoloViews does using xarray, and then use those new lower-level features to implement what HoloViews does now but more simply. This is essentially how HoloViews relates to datashader; HoloViews translates its own canvas-based properties into the arguments for ds.Canvas, such that the HoloViews user does not need to care much about how datashader works but datashader does not depend on HoloViews. @philippjfr and @jlstevens can weigh in when we meet about this later this week.

[jbednar]

And we can't fail to add Hunt's suggestion for a name, corresponding to EarthIO and EarthSIm, which accurately expresses this new repo's functionality: EarthMunge. Sure, it's horrible, but it's the one to beat!

[jbednar]

Actually, we've now agreed that most of this functionality can go into GeoViews, preferably in a file that doesn't depend on the rest of GeoViews. If it's very specific (e.g. just for one certain satellite), it can go in EarthIO, but hopefully there won't be much of that. So, no need for EarthMunge.

[gbrener]

@jbednar @PeterDSteinberg Are there any action items that should come out of this issue, or can we close it?

[PeterDSteinberg]

@gbrener Everything has been addressed via other issues/PRs except for the question of what to do with the bokeh-geo assets. Can you figure out what to do with bokeh-geo? I think it is mainly viz stuff but may have a couple satellite utilities like radiance or reflectance. Also not sure if it is a general tool we can use or if it is more of a one-off demo of Bokeh + LANDSAT. It may be that we just move the idea of bokeh-geo to an example in datashader or elm, both of which already have LANDSAT examples, and we either add the ideas to the existing notebooks or make new notebook(s). Feel free to close this issue once another issue has been made in the right repo regarding bokeh-geo.

[gbrener]

Ok, noted. I'll incorporate this into issue ContinuumIO/elm#175 as well (since it seems related to the examples), and return here with my findings once I get the chance.