Scope of Verde

[mdtanker]

I'm wondering what the current scope of Verde is. I have seem some phrasing in the past of Verde's focus being on interpolating point data to make grids (fatiando/verde#108 (comment)). However now there are several issues and PR's for adding functionality to Verde focused on working with already gridded datasets, such as merging grids, filling nan's in grids, and computing derivatives. Similarly, we have functions in Harmonica for filtering.

I think Verde has a much nicer interface than some other geospatial packages, and personally I would love to be able to do more of my processing in Verde instead of having to move to other tools like PyGMT, rioxarray, pyresample etc.

Is there interest in building out Verde to have more of the grid-based geospatial processing functionality, like those listed above as well as things like resampling grids to different resolutions, and interpolating gaps in gridded data? Alternatively, would there be interest in a new Fatiando package specifically for these grid-based operations?

[leouieda]

Hey @mdtanker I've struggled with this definition a lot myself. I've tented to think of Verde as mostly dealing with the point data and then letting other things handle the grids but needs of messing with grid data keep coming up and Verde often feels like the place for them. At this point, I think it's OK to think of Verde as our "geospatial" package and so some grid operations wouldn't be out of place.

What kind of functionality do you find missing?

[mdtanker]

There's a long list of functionality that would be awesome to include, but I think the bigs ones for me would be 1) resampling a grid to fine or coarser resolutions or to match the coordinates of a different grid, 2) converting between pixel and gridline registration, and 3) interpolating gaps in grids, ideally with using SplineCV with the data surrounding the gaps.

Something to think about is if we start expanding a bit more grid based operations, we may want to include a more grid-friendly interface to the existing verde functions. For example, allow passing a grid, instead of coordinates, to functions like get_region and inside.

[leouieda]

Thanks, Matt! Please feel free to open issues whenever you think of features you’d like added.

Alright, let’s say we start with the resampling part, which includes the change in registration. That would require a different interpolation approach since pretty much all the ones we have here (with the exception of the KNeighbors) would be inefficient for going between regular grids. We’d probably something like a bi-cubic interpolation implemented. Would be worth having a look at what pyresample does under the hood.

Interpolating in the gaps would be easy enough once we can correctly identify the gaps and data in their vicinity. Any thoughts on how to do that?

[mdtanker]

Ok sounds great, I'll start opening up some issues next week.

Not too sure, but I'll have a think and put my thoughts in an issue for a resampling function!