H3 grid

Uber’s hexagonal hierarchical index. Hexagons give uniform adjacency (every cell has six neighbours), which suits movement and coverage analysis.

The interactive explorer below is rendered with deck.gl and behaves like the h3geo.org map: the H3 resolution follows the zoom level and cells are generated in the browser for whatever is in view — zoom in for finer cells, zoom out for coarser, no resolution picker. Two neighbouring resolutions are shown at once: the current level draws a darker, heavier border and the next finer level a lighter, thinner one, so the hexagonal nesting stays visible. Cell geometry comes from the shared Rust/WASM core, identical to the Python package. GIS-native (lon, lat) order is used throughout.

Why H3?

Hexagons make neighbourhoods honest: every cell touches exactly six others at the same centre distance, with no corner-only contacts, so k-rings, flows and smoothing behave the same in every direction. The ecosystem is the other big draw — H3 ships in DuckDB, BigQuery, Snowflake and kepler.gl, so cells flow through analytics stacks unchanged. The price of hexagons: children only approximately nest into parents, twelve cells per resolution are pentagons, and areas are only roughly equal. When exact containment or exact equal area matters, see S2 grid, EA-Quad grid or A5 grid.

Usage

Encode a point and tile a small bounding box around Paris — same result in Python and JavaScript (both call the shared core):

Python

import m3s

cell = m3s.H3.from_geometry((2.35, 48.86))            # (lon, lat)
cells = m3s.H3.from_geometry((2.2, 48.8, 2.4, 48.9))  # bbox
print(cell.id, len(cells))

JavaScript

import * as m3s from "m3s";
await m3s.ready();

const cell = m3s.H3.fromPoint(2.35, 48.86);            // (lon, lat)
const cells = m3s.H3.fromBbox([2.2, 48.8, 2.4, 48.9]); // bbox
console.log(cell.id, cells.length);

from _deckmap import DeckExplorer, read_grid_js


DeckExplorer(
    center=(2.35, 48.86),
    zoom=12,
    grid_js=read_grid_js("h3"),
    wasm=True,
    hover="#ffeb3b",
)