H3 — Hexagonal Hierarchical Geospatial Indexing System

Introduction

H3 is a geospatial indexing system that partitions the globe into hierarchical hexagonal cells at 16 resolution levels. Originally built at Uber for ride pricing, ETAs, and supply modeling, it provides a uniform grid that avoids the distortion artifacts of square grids near the poles.

What H3 Does

• Indexes any latitude/longitude coordinate into a hexagonal cell at a chosen resolution (0-15)
• Traverses the hierarchy: parent, children, and compact/uncompact operations across resolutions
• Computes neighborhoods: k-ring (grid disk), grid distance, and grid path between cells
• Converts between H3 indexes and boundary polygons for rendering on maps
• Supports polyfill to find all H3 cells covering a given polygon geometry

Architecture Overview

The core library is written in C for performance and portability. Bindings exist for Python (h3-py), JavaScript (h3-js), Java, Go, and Rust. H3 uses an icosahedral face-centered projection, subdividing each face into hexagons (with 12 pentagons at vertices). Each cell is identified by a 64-bit integer encoding its resolution and position, enabling fast bitwise operations for traversal.

Self-Hosting & Configuration

• Install language-specific bindings: pip install h3, npm install h3-js, or use the C library directly
• No server or API key needed; H3 is a pure computation library
• Choose resolution based on your use case: res 9 (~0.1 km²) for city-level, res 7 (~5 km²) for regional
• Use with PostGIS via the h3-pg extension for spatial SQL queries on hexagonal grids
• Integrate with deck.gl H3HexagonLayer or Kepler.gl for browser-based hex visualization

Key Features

• Uniform cell areas at each resolution for unbiased spatial aggregation
• 16 resolution levels from continental (res 0, ~4M km²) to sub-meter (res 15)
• Hierarchical containment: every child cell nests within its parent
• Edge and vertex indexing for modeling adjacency and connectivity
• Compact representation reduces storage by replacing groups of children with parents

Comparison with Similar Tools

• S2 Geometry (Google) — uses square cells on a sphere; H3 hexagons provide more uniform neighbor distances
• Geohash — rectangular cells with edge-sharing artifacts; H3 hexagons share edges uniformly
• PostGIS grid functions — arbitrary grids on projected data; H3 provides a standardized global hierarchy
• Uber's Placekey — built on H3 for POI identification; H3 is the lower-level spatial index underneath
• Open Location Code (Plus Codes) — encodes locations as short strings; H3 is optimized for spatial analytics

FAQ

Q: Why hexagons instead of squares? A: Hexagons have uniform adjacency (every neighbor shares an edge, not just a corner), which eliminates directional bias in distance calculations and spatial smoothing.

Q: How do I visualize H3 cells on a map? A: Convert cell indexes to GeoJSON boundary polygons with cell_to_boundary, then render with Leaflet, Mapbox GL, or deck.gl's H3HexagonLayer.

Q: Can H3 be used in SQL databases? A: Yes. The h3-pg extension brings H3 functions to PostgreSQL/PostGIS. DuckDB and ClickHouse also have H3 extensions.

Q: What is the performance like for large datasets? A: H3 operations are O(1) per cell using bitwise index arithmetic. Polyfilling a polygon scales with the number of cells it covers, which is efficient at appropriate resolutions.

Sources

• https://github.com/uber/h3
• https://h3geo.org/