Quiet Map

Global environmental noise atlas: computed sound levels (Lden) from roads, railways, aircraft, industry, buildings and settlements, for the whole world at about 12 m base resolution (with an optional ~6 m detail tier), served as an interactive web map with per-point source breakdowns. The acoustic model follows CNOSSOS-EU for surface sources with ISO 9613-2 propagation and a Doc 29-inspired aircraft NPD; formulas, constants and intentional simplifications are specified in engine/noise-compute/SPEC.md.

Live instance: https://quietmap.org

Repository map

engine/       Rust crates — noise-compute (the shared acoustic kernel),
              tile-painter (batch tile builder + tile store), osm-extract and
              aircraft-extract (OSM PBF / ADS-B → H3-res-4 arrow),
              raster-reader, source-reader (NAPI addon for point queries),
              noise-gpu (CUDA ports of the scatter kernels)
frontend/     The map — React 19 + MapLibre + deck.gl (Vite build)
server/       Fastify — tile serving, point noise queries, search
pipeline/     TypeScript enrichment — patches OSM defaults with
              measured/registry data (traffic, industry, buildings) per country
scripts/      Product entry points: run-extraction.sh (rasters + OSM extract),
              build-heatmap.sh (whole tile build), osm-to-h3r4.sh,
              check-fast.sh, dataset-year.json
docs/         about/ (public About + methodology pages) · standards/
benchmarks/   Validation fixtures (popup/world point sets)

Build & run

Requirements: Node.js 22 and the Rust toolchain pinned by rust-toolchain.toml.

./start.sh

start.sh is standalone: it builds the engine native addon, the frontend and an immutable compiled server release, activates the release, and serves on $PORT (default 8520) — no external orchestration needed. Callers that own the process lifecycle themselves (e.g. a systemd unit) build and activate without serving:

QM_BUILD_ONLY=1 ./start.sh

Quality gate

./scripts/check-fast.sh [node|rust]

The required, data-free gate: server smoke/tests, frontend lint/build, offline pipeline tests, and rustfmt + Clippy + tests for the engine crates. The optional first argument runs only one side (CI split jobs). Keep it green before every commit.

Dataset year

The default dataset year is pinned in scripts/dataset-year.json (current_year); the DATA_YEAR environment variable overrides it. The pin controls which data/prepared/<year>/ and data/tiles/<year>/ trees a build produces and the server reads.

Data layout (not in git)

data/ is gitignored and large. Expected shape:

data/prepared/dem/             1°×1° DEM rasters, 30 m (shared across years)
data/prepared/rasters/         forest/IMD rasters (shared)
data/prepared/<year>/h3r4/     H3-res-4 source + vector-obstacle Arrow extracts
data/tiles/<year>/pmtiles/     built tile pyramids + manifest

H3R4_DIR / PMTILES_DIR env vars point a server at data living outside the checkout. scripts/run-extraction.sh fetches and prepares the inputs, then scripts/build-heatmap.sh computes the tiles.

Scope

This repo is the whole product: model, pipeline, tiles, web app. Deployment and orchestration of the author's own cluster (multi-checkout hosting, distributed world-build fleet) is intentionally out of scope here and lives in a separate private repository.

License & contact

See docs/about/credits.md for data-source attribution (OSM, GLO-30, SRTM, Overture, WorldCover, IMD, ADS-B) and terms. [email protected].