Compare GDAL layer intersection vs. rtree/shapely

[phargogh]

For a lot of our projects, we often test intersections of a number of smaller polygons (buildings, let's say) against a much larger Area of Interest polygon. Our classic solution here is to use an rtree spatial index filled with shapely polygons to test the bounding boxes and then check the resulting shapely polygons against the prepared AOI geometry.

While this approach works, it also looks like the GDAL devs have done some work in the last few years on layer intersections (see OGR 2.1.0 notes), so maybe that's a superior solution?

I feel like we settled on the rtree/shapely approach because the GDAL one was slower and/or the GDAL approach might have silently skipped geometries that were invalid. We should test recent versions of GDAL to see if either of these methods is better than the other.

[dcdenu4]

I know @megannissel was maybe looking at this comparison a bit with DataHub vector clipping. Anything interesting to note here from that Megan?