The way to do this in a distributed way is to do chunk aware regridding. So for each destination chunk, you get a list of intersecting source chunks. That becomes one dual-DFS call. Then, for each destination chunk, you do another dual-DFS call on the implicit quadtrees formed by each source chunk, and assemble the regridder that way.
The weights would somehow have to be saved to file or so, because if you're doing distributed regridding you probably can't fit the destination grid in memory.
An interacting-chunks method also gets us closer to SphericalSpatialTrees.jl in some sense, which is good.
The way to do this in a distributed way is to do chunk aware regridding. So for each destination chunk, you get a list of intersecting source chunks. That becomes one dual-DFS call. Then, for each destination chunk, you do another dual-DFS call on the implicit quadtrees formed by each source chunk, and assemble the regridder that way.
The weights would somehow have to be saved to file or so, because if you're doing distributed regridding you probably can't fit the destination grid in memory.
An interacting-chunks method also gets us closer to SphericalSpatialTrees.jl in some sense, which is good.