carpentries-incubator
Stackstac episode
It'd be cool to show folks how to access cloud-hosted geospatial imagery as a dask-backed xarray with stackstac: https://github.com/gjoseph92/stackstac
This would probably necessitate explanation of Dask, chunks, and relevant methods and how to take advantage of chunking. We'd also need to make these caveats very clear:
- stackstac discards geospatial info like the crs.
- it's difficult to associate data accessed by stackstac with rioxarray DataArrays
This episode should probably focus on the value of
- accessing cloud hosted STAC catalogs and the easy APIs to do so and
- the speed improvements that come from Dask/parrallelizing
#86 and #90 have addressed the background of data access and parallelization, respectively. The stackstac example is going to be addressed by #102.
@fnattino great to see these merged! Either @srmsoumya or I will work on #102 during the first week of May when we have some blocked off time to work on open source , possibly sooner.
update: not much movement on this episode to date. but I think I can shoot to have this done after Maptime Davis. Realistically, sometime in November. Some things to cover:
- selecting multiple single band assets with the assets arg in stackstac.stack to get a multiband DataArray
- compute multispectral indices using xarray-spatial https://github.com/makepath/xarray-spatial
- plot change over time at a single pixel.
- make a gif of spectral indices and label them by date?
- other kinds of plots, simple time series analysis? Maybe something that brings in the crop region information? Open to suggestions!
noting that ODC-stac is more actively maintained than stackstac, so it might be the option to consider for a stacking episode. I don't have bandwidth to work on this right now but happy to review PRs if someone would like to give this lesson a shot.
https://github.com/opendatacube/odc-stac