Data Sources

Surface PM2.5 from the CAMS global atmospheric MODEL, not ground measurements. Shown as a colour field across Southeast Asia with 10 m wind particles (NOAA GFS) moving through it, because haze does not stop at borders: Riau's smoke reaches Kuala Lumpur and Singapore. The field is built on a schedule twice a day and served as a static file from the CDN, so the browser never calls a weather API. District points are NOT blended into the colour: they are the same CAMS model sampled at centroids, so mixing them in only produced a blob around each district centre without adding information. They remain the source of the per-region numbers in the popup. Opacity follows severity, so clean air is deliberately left transparent. Used because inland Riau, Jambi, and Central Kalimantan have almost no monitors — exactly where the smoke is thickest — so every station-based source is blank where it matters most. The trade-off: a 45 km cell smooths local peaks, so these values run LOWER than the air beside a burning block. The AQI shown alongside is computed by us from this PM2.5 using the official US EPA breakpoints (6 May 2024 revision); we do not take anyone's ready-made AQI. Values above AQI 500 are our own extension of the Hazardous segment, outside the range EPA defines, and are always flagged. The defensible number is µg/m³; the AQI is only its translation.

True-colour satellite mosaic for ONE chosen day, streamed straight from NASA GIBS tiles (keyless, not processed by us). Used to check land & forest fires: smoke and burn scars are visible to the naked eye, so a hotspot can be matched against what the sensor actually recorded. The product name shown on the map is the exact Worldview layer name, so the same day can be opened and compared in Worldview.

FIRMS thermal-anomaly detections for one chosen day, so a fire season can be replayed day by day — unlike the Hotspots (48h) layer, which is near-real-time only. Because GIBS publishes these vector tiles in EPSG:4326 only, this layer is drawn via the GIBS WMS as an image, and is therefore not clickable. A hotspot is an unverified heat detection, not proof of a fire nor of any party's involvement.

Weather-satellite tiles served straight from BMKG's own API (keyless, not processed by us) — five products: Himawari-9 cloud-top temperature (IR) and full disk, Himawari-9 high-resolution daytime imagery (VS), and GK-2A cloud-top temperature and water vapour. Played as an 18-frame loop, 10 minutes apart, counting back from the current time and further offset by each satellite's own processing delay (Himawari-9 ~40 min, GK-2A much slower) so the requested frame actually exists on BMKG's server.

Surface PM10 from the same CAMS model run as PM2.5, one extra field per step file, not a separate fetch. The AQI shown in the popup is computed from this figure using the official US EPA breakpoints.

10 m wind speed and direction from the GFS model, shown as moving particles. Can be toggled independently of PM2.5.

2 m air temperature from the GFS model, one of four weather fields riding the same request as the wind field.

2 m relative humidity from the GFS model.

Surface air pressure from the GFS model.

Hourly UV Index from Open-Meteo.

BMKG's severe-weather early warnings (strong wind, heavy rain) in CAP (Common Alerting Protocol) format, fetched live with no intermediate database.

Significant aviation hazards from AWC's International SIGMET feed, filtered to Indonesia and neighbours server-side because AWC's own bbox parameter does not work on this endpoint (confirmed directly).

Airport weather observations (METAR), already decoded: temperature, dewpoint, wind, visibility, QNH, flight category (VFR/MVFR/IFR/LIFR), plus the raw coded text.

Live aircraft positions from OpenSky's public state-vector feed, icons rotated to true heading and coloured by altitude band.

Vessel positions from AISStream, a WebSocket-only source with no REST snapshot endpoint. The backend opens a short subscription per request cycle then closes it — a recent snapshot, not truly continuous tracking, a consequence of the backend running as a serverless function with no always-on process.

BMKG's own official port-location catalogue: name, code, province. Reference data, not a live feed.

Wave, weather, wind, and current forecast per BMKG sea zone, joined onto BMKG's own zone-boundary polygons. BMKG's own encoding is undocumented; the field mapping was verified directly against BMKG's own frontend source code, not guessed.

OpenStreetMap-derived place names, boundaries, roads, and coastlines served by NASA GIBS. Shown automatically with the NASA satellite imagery (not a separate layer): that imagery covers the basemap and its labels, so this overlay restores locational context. Orientation only — not an environmental data source, and not used in any calculation.