possibly use spatial indexes (rtree) to speed up finding which adm to assign to points
may not be necessary since init is under 2 minutes for nepal dataset but other datasets may have more locations being aggregated to adm levels (vs points, buffers, country)