Allow non-regularly gridded and lat-lon coordinates

[joeloskarsson]

This issue is to document and discuss some enhancements I am looking into w.r.t. the coordinates used as inputs.

The current situation

Currently the grid node coordinates fed to wmg are a numpy array of shape [2, N_y, N_x]. This assumes that the grid nodes are on a regular grid. It is also the case that these are assumed to be cartesian coordinates, in the projection used for the LAM data.

Generalizing to non-regular gridded data

My idea to generalize this is to change so that the coordinate array xy instead has shape [N_grid_nodes, 2] with e.g. xy[:,0] containing the x-coordinate for all grid nodes. Existing mesh creation algorithms would still have to work with a rectangular area, for which they can use the axis-aligned bounding box of the grid nodes. Other, new, mesh creation methods could use the convex hull of the set of grid nodes.

Generalizing to allow lat-lon coordinates

For lat-lon coordinates the same [N_grid_nodes, 2] shape could be used for the coordinates. It would also be good to feed a projection specification (I propose a Cartopy projection instance, https://scitools.org.uk/cartopy/docs/latest/reference/generated/cartopy.crs.Projection.html). Using the projection the lat-lon coordinates can then be mapped to a euclidean space and all our current methods can be applied as is. Having the option of lat-lon coordinates however also opens up the opportunity for mesh creation methods that work directly with these.

A question is how to tell wmg if in-projection or lat-lon coordinates are given. Some options are:

• Have separate arguments xy and lat_lon where only one has to be given.
• Have only one argument coords, and then a separate boolean flag e.g. uses_lat_lon.
• One could also enforce that a projection has to be given when using lat-lon coordinates, and use that implicitly as the uses_lat_lon flag above.
• Another option is really to only allow lat-lon coordinates from here onwards.

Not really sure what is the best option of the ones above. Any input is appreciated.

[joeloskarsson]

I also realize that the concept of grid_refinement_factor is defined in terms of the number of grid nodes in the x- and y-directions. This will not make sense for irregularly positioned grid nodes. So likely this will have to be generalized to a refinement in terms of actual distances, rather than number of nodes. Which probably makes more sense anyhow (and you can always find a mapping in-between).

[joeloskarsson]

Currently being worked on in #32

[joeloskarsson]

Now the option

One could also enforce that a projection has to be given when using lat-lon coordinates, and use that implicitly as the uses_lat_lon flag above.

is what is implemented in #32 .