.plot() method for Grid Object

UXARRAY / uxarray

Xarray-styled package for reading and directly operating on unstructured grid datasets following UGRID conventions

https://uxarray.readthedocs.io/

Apache License 2.0

141 stars 31 forks source link

.plot() method for Grid Object #118

Closed philipc2 closed 9 months ago

philipc2 commented 1 year ago

Similar to Xarray, directly allow for plotting from a Grid Object (compared to a DataArray).

Proposed Functionality

Calling .plot() on a Grid object with no data variables (only grid info) would render an unfilled mesh
Calling .plot() on a Grid object with an argument for data variables, indexing, etc, would render a filled mesh according to the user inputs

PolyMesh Specific

Goal is to incorporate my PolyMesh code to handle visualizations, however due to package dependencies starting off with SciPy's or MPL's triangulation might be better (having both options would be good too)
Any mesh that's constructed during plotting would be stored in the Grid object to reduce computation time on repeated plots (ex. plotting multiple data variables on the same mesh)

philipc2 commented 1 year ago

@dcherian @rajeeja @paullric @erogluorhan

Any comments or suggestions? I think it might be worth discussing this topic during one of our bi-weekly meetings since there is a lot of design decisions we need to make before we can start working on it.

rajeeja commented 1 year ago

@dcherian @rajeeja @paullric @erogluorhan

Any comments or suggestions? I think it might be worth discussing this topic during one of our bi-weekly meetings since there is a lot of design decisions we need to make before we can start working on it.

What are the dependencies of PolyMesh? Is setting this up with SciPy or MPL's triangulation easy? IMO MPL has been the most reliable over the years.

philipc2 commented 1 year ago

@rajeeja

PolyMesh is written around Datashader, with the main dependencies being Shapely (since PyGEOS merged with Shapely), Spatial Pandas, and GeoViews. Shapely2.0 is currently in an alpha release, and is still merging a lot of the PyGEOS work as we speak.

Starting off with triangulation support using one of the SciPY or MPL methods would be a good start. We wouldn't be able to visualize large grids very efficiently, however it would be very easy to implement once we settle on a design.

zarzycki commented 1 year ago

This sounds most excellent @philipc2.

I keep coming back to the comments in https://github.com/UXARRAY/uxarray/discussions/46#discussioncomment-2747480 ... I think it's important for any unstructured viz to reflect (or at least have a boolean switch) to reflect the underlying mesh and cell topology.

So for the latter (Calling .plot() on a Grid object with an argument for data variables), I would consider that a relatively desirable feature that such contour maps properly show the underlying mesh.

A potentially relevant example @cdeciampa shared with me this AM is: https://psyplot.github.io/examples/maps/example_ugrid.html

Note https://psyplot.github.io/examples/_images/maps_example_ugrid_18_0.png, where the cells with more than 3 edges are correctly filled as one scalar value (it's possible the quads are really just two triangles glued together, I haven't dug deep enough into the source).

One of the benefits of having a full grid description (i.e., having cell edges and vertices versus just lat/lon points) means viz tools using the description can be faithful to the computational mesh!