reusable interpolation weights?

[ctessum]

Is your feature request related to a problem? Please describe.

I think there are two main ways one might want to use an interpolation package:

1. to interpolate a single field or dataset at many new locations; or
2. to interpolate many fields or datasets at the same locations.

For the most part, existing julia interpolation packages focus on the former use case above. However, the latter case is needed in situations where data is available in one spatial basis and needs to be regridded to a different spatial basis before use.

Describe the solution you’d like

For the latter use case, one would want to precalculate the weights that would be used to interpolate a dataset at a given set of locations, and then those weights could be reused to interpolate multiple datasets.

Describe alternatives you’ve considered

The only existing package I can find in julia with this capability is CimaCoreTempestRemap.jl, but this package is closely integrated with the Clima API and requires writing all the inputs to temporary files and then calling an external executable.

Additional context

Within the SciML ecosystem, something like this would be needed for example to create an interface between two MethodOfLines.jl simulation domains which use different grids.

[rafaqz]

This should be almost trivial now that weight calculation is abstracted in #33. We would just need to run the prepare and weight methods over the vector of t_eval for each dimension in their constructors. Then we can store it in a new weights field.

[rafaqz]

My profiles have weight being a reasonable fraction of compute time in a bunch of cases.

I guess we should make it optional to precompute? I can see it being annoying in 1 dimension, as the weights may be larger than the arrays.

[ctessum]

I think that makes sense