H3 geohashes and Uxarray ... is there a match in heaven?

[dev360]

Hello,

First off, let me preface my question by saying I have working knowledge of xarray but data science and GIS is not my specialty.

We are prototyping different ways of solving a geographical intersection problem, and our team has been looking at a few different solutions. Its essentially a lat/long grid space, so we looked at geohashing, and H3 in particular appeals to us.

So we essentially have potentially thousands of variables across a x/y grid space, and each variable is a sparse matrix (filled 2-3%, e.g. country sized at most). When we walk down this path, the obvious thing you realize with H3 is that a state like Texas would be like 300k cells at resolution 10 (5km), but would only be 5300 something cells when you collapse it and express it as a combination of resulting resolution 1 through 10 cells.

My question is this: would uxarray be the right tool for the job for this? We want to essentially be able to query for the intersection (values of all matrices) at a given x/y level, expressed at the highest resolution (10 in example above).

Here is a drawing of it for the visually inclined:

[philipc2]

Hi @dev360

This is some very interesting work! I don't believe that uxarray would be the best fit for this task since we are focused around extending xarray to support unstructured grids and this problem appears to be dependent on high-performance GIS algorithms and data structures.

Since you are already familiar with xarray, I would take a look at xarray-spatial. Another package worth looking into (especially if this problem can be GPU accelerated) would be cuSpatial.

Hope this helps!

[dev360]

Interesting. Yeah the first thing to came to my mind was actually GPU/Cuda, but there is no appetite for very low-level solution (Rust). The tooling is painful on Mac too since AWS only supports Nvidia. I'll def check out the libraries you suggested! Thanks

[dcherian]

I would ask over at https://github.com/martinfleis/xvec