@fbriol, FWIW, I updated the benchmarks by including pyinterp.RTree (BTW, pyinterp looks very nice!):
I should mention that pys2index.S2PointIndex is a direct wrapper of s2geometry’s S2PointIndex, which I think uses a B-Tree in memory, still not very suited for indexing massive amounts of data in distributed environments.
It’s not relying on key-value embedded database of any kind, even though I think this would an idea very exciting to explore (among others? Dask’s Actors?), using either GeoHash, H3 or s2geometry.
That said, it’s not yet very clear to me when exactly (which use case? which data?) those alternatives would start showing real benefits over common, in-memory (tree-based) solutions…