Plan the component

[mdpiper]

I'd like to make the planning and development of this project as open and as public as possible. GitHub provides a good place for this.

Goal

Make a CSDMS data component that allows a user to read data (and metadata) from a GeoTIFF file. I estimate this project will take about a week to complete.

Overview

• Start with a GeoTIFF file on the local filesystem. This component won't fetch data from the internets. Update: No--an URL to a remote file can also be used.
• Use xarray.open_rasterio to load the data and metadata from the file into memory. This implies raster data only.
• No CLI, only API and BMI.

Outline

• Package name: bmi_geotiff
• Module names: io.py, bmi.py
• Write the library and tests iteratively
• Write the BMI and run bmi-tester iteratively
• Make an example, in two parts:
  1. a script that can be run at the shell prompt (just API, save BMI for pymt implementation)
  2. a Jupyter Notebook based on the script
• README
  • include install instructions
  • include the script example
• Documentation
  • include README
  • include Jupyter Notebook example converted to rst
  • deploy to Read the Docs
• CI with GitHub Actions
• Stamp release using zest.releaser
• TestPyPI
• PyPI
• Make a conda-forge package
• DOI
• Babelize
• Make a pymt package
• Add to CSDMS Model Repository (this will populate Data Components page)
• Twitter announcement

[mdpiper]

I want to try using [options.package_data] instead of a MANIFEST.in file. See the xarray setup.cfg for a working example.

Update: I was mistaken. What I really want to use is setuptools-scm.

[mdpiper]

Call the library module io.py instead of geotiff.py.

[mdpiper]

A GeoTIFF typically stores raster imagery. The data could be many things: elevations, reflectances, intensities, etc. What should the Standard Name for these data be? Candidates:

• image__raster_data (image__aspect_ratio is in the registry)
• gis_raster__data (could also have gis_vector__data)

It would also be good to have a Standard Name for the projection string, although this will be obsolete with csdms/bmi#80. Candidates:

• map_projection__string
• coordinate_reference_system__descriptor

[mdpiper]

The xarray.openrasterio method automatically returns rectilinear data, so the BMI grid type can be rectilinear.

[mdpiper]

A file could hold a time series of images, although this isn't typical. I need to think about how to handle this. Is there a standard?

This is another use case for an is_steady_state function in the BMI (see csdms/bmi#79) because this would change the way we interpret the rank of the data and the way we use get_value.

Idea: a boolean is_time_series entry in the BMI config file and a _time attribute in BmiGeoTiff.

For now, neglect time information in the BMI.

[mdpiper]

Reference links

xarray.open_rasterio
http://xarray.pydata.org/en/stable/generated/xarray.open_rasterio.html#xarray.open_rasterio

xarray.plot.imshow
http://xarray.pydata.org/en/stable/generated/xarray.plot.imshow.html

Matplotlib imshow
https://matplotlib.org/stable/api/_as_gen/matplotlib.pyplot.imshow.html

GDAL GeoTransform
https://gdal.org/user/raster_data_model.html#affine-geotransform

rasterio plotting
https://rasterio.readthedocs.io/en/latest/topics/plotting.html

rasterio.plot.show
https://rasterio.readthedocs.io/en/latest/api/rasterio.plot.html

EPSG:32618
https://epsg.io/32618

Cartopy projection list
https://scitools.org.uk/cartopy/docs/latest/crs/projections.html

Cartopy image example
https://scitools.org.uk/cartopy/docs/v0.5/matplotlib/introductory_examples/04.images.html

[mdpiper]

When setting up the conda-forge recipe, try using PyPI for the source tarball instead of GitHub.

[mdpiper]

Should I use rioxarray instead of xarray? rioxarray does reprojection. It wouldn't hard to make the switch. See http://xarray.pydata.org/en/stable/io.html#rasterio

Should I use rasterio instead of xarray?

[mdpiper]

Idea: The BMI should always address single bands. Never load multiband data (even RGB) all at once. Use time and the update and update_until methods.

On the other hand, xarray does a lazy load of data from disk, so it may do no harm in keeping all bands.

Can I update_until a negative time?