Evolution and Optimization of Numeric Indexing for Geolocation in Elasticsearch

Business Background

LBS services require fast "search nearby POI" queries; Elasticsearch provides millisecond‑level geo_distance queries to satisfy this need.

Background Knowledge

It explains how to precisely locate an address using latitude/longitude, compute distances with the Haversine formula, and share coordinates via Geohash.

GET /my_locations/_search { "query": { "bool": { "must": { "match_all": {} }, "filter": { "geo_distance": { "distance": "1km", "pin.location": { "lat": 40, "lon": 116 } } } } } }

Solution Evolution

Pre‑2.0 (String Simulation) – Elasticsearch relied on Lucene's inverted index and simulated numeric ranges with term prefixes.

Elasticsearch 2.0 – Introduced geo_distance using numeric range queries on separate lat and lon fields, calculating a bounding rectangle and then applying a Haversine filter.

public static DistanceBoundingCheck distanceBoundingCheck(double sourceLatitude, double sourceLongitude, double distance, DistanceUnit unit) { ... }

Elasticsearch 2.2 – Added Quadtree‑based indexing, storing lat/lon as a single Morton‑encoded numeric field, enabling more efficient coarse filtering before precise distance checks.

double centerLon = 116.433322;
double centerLat = 39.900255;
double radiusMeters = 1000.0;
GeoRect geoRect = GeoUtils.circleToBBox(centerLon, centerLat, radiusMeters);
System.out.println(geoRect);

Elasticsearch 5.0+ – Switched to BKD‑tree (a multidimensional B‑tree) for numeric and geo indexing, offering superior memory usage and query speed. Queries intersect the query rectangle with BKD‑tree cells to quickly include or exclude large groups of points.

// Core query class
public class GeoPointDistanceQuery extends Query { ... }

The article concludes that these indexing advances have transformed Elasticsearch from a pure full‑text engine into a versatile analytics platform capable of handling high‑performance geospatial queries, and hints at future directions such as R‑Tree support for shape indexing.

References

https://www.elastic.co/cn/blog/lucene-points-6.0

https://www.cs.cmu.edu/~ckingsf/bioinfo-lectures/kdtrees.pdf

https://www.csee.usf.edu/~tuy/Literature/KDtree-CACM75.pdf