Tech Specs

Where the station list comes from

The registry of which stations exist is built from two sources. For the National Electricity Market (NEM — QLD, NSW, VIC, SA, TAS), we use AEMO's DUID Registry— AEMO's authoritative list of every generating unit it dispatches, keyed by DUID (Dispatchable Unit Identifier, AEMO's internal code for a generator). For Western Australia's WEM grid, the registry comes from OpenElectricity.

Where the coordinates come from

AEMO's registry doesn't include GPS coordinates. We match each station by name to the WRI Global Power Plant Database, an open dataset of ~35,000 power plants worldwide maintained by the World Resources Institute. For stations not in WRI, we fall back to OpenElectricity coordinates, then a small manually curated table.

What the live output numbers mean

NEM live MW figures come from AEMO NEMWeb Dispatch SCADA. SCADA (Supervisory Control and Data Acquisition) is the real-time telemetry feed that every NEM generator sends to AEMO every 5 minutes as part of dispatch. Each reading is at the DUID level; we aggregate to station level.

WEM (Western Australia) uses a different source: the AEMO WA infographic generation CSV, which publishes per-facility output at 30-minute intervals in near-real-time as each trading interval closes. WEM's SCADA endpoint exists but is a daily batch file delayed ~1 day, so the infographic CSV is used instead. This means WEM output figures are coarser and slightly less fresh than NEM's 5-minute dispatch data.

Emissions data

Annual Scope 1 greenhouse gas emissions come from NGER (National Greenhouse and Energy Reporting), Australia's mandatory annual emissions disclosure scheme for large facilities. The dataset used is for the 2023–24 financial year.

Assumptions & limitations

Station matching relies on fuzzy name normalization — generic industry words (power, station, farm, wind, solar, battery, renewable, storage, facility, energy, and 40+ others) are stripped from both sides before comparison, with fuel-type compatibility checks as a guard — so there's a small risk of mismatches for stations with very similar names. A handful of newer stations may be missing from WRI if not yet added to the database. NEM output is refreshed every 5 minutes; WEM every 30 minutes. Both may lag by up to one dispatch interval. NGER emissions are annual averages, not real-time.

What AIS is

AIS (Automatic Identification System) is a radio broadcast standard that every commercial vessel over 300 tonnes is legally required to operate. Ships transmit on two VHF frequencies (161.975 MHz and 162.025 MHz), broadcasting their GPS position, speed, course, heading, navigational status, and vessel identity roughly every 2–30 seconds depending on speed. Each vessel has a globally unique MMSI (Maritime Mobile Service Identity) — a 9-digit number similar to a phone number.

How we receive the data

We subscribe to AIS Stream, a real-time WebSocket service that aggregates AIS signals from both coastal shore stations and low-Earth-orbit satellites. Terrestrial AIS range is roughly 40–50 km from shore; satellite AIS covers the open ocean but has slightly higher message loss rates in congested shipping lanes (multiple vessels broadcasting in the same time slot can collide). We collect 7 seconds of live data per refresh cycle, then maintain a rolling 10-minute window so vessels don't flicker in and out.

How ships are classified

The AIS Type Code (a number between 70–89) classifies the vessel by category — cargo, tanker, LNG carrier, etc. We refine this further using the vessel's declared destination port and ship name. A bulk carrier heading to Port Hedland becomes "iron ore"; one heading to Hay Point becomes "coal". Ships without a useful type code are inferred by proximity and heading toward known mining export ports.

Assumptions & limitations

AIS can be deliberately disabled ("going dark") by vessels, legitimately or not. Type codes and destinations are self-declared and sometimes incorrect or blank. Our classification heuristics will occasionally miscategorise vessels. Maximum 300 ships are displayed; priority is given to energy vessels (LNG, oil) then speed. The Australian EEZ bounding box we monitor is approximately 10°S–44.5°S, 112°E–154.5°E.

The two-tier approach

Fire data combines official agency perimeter polygons (when available and fresh) with satellite-derived hotspot clusters as a fallback. Official polygons come from NSW RFS Major Incidents, VIC Emergency Management Victoria, and the WA DFES Warnings API. QLD, SA, NT, and TAS rely on hotspot-derived polygons as those agencies don't publish equivalent feeds.

How satellites detect fire

Hotspot data comes from NASA FIRMS (Fire Information for Resource Management System). Three VIIRS satellites (Suomi-NPP, NOAA-20, NOAA-21) orbit Earth at ~824 km altitude, each completing ~14 passes over Australia per day. Each satellite carries an infrared imaging sensor that scans the surface in a 3,040 km wide swath. When the mid-infrared brightness temperature of a 375 m × 375 m ground pixel is significantly hotter than the surrounding pixels, the algorithm flags it as an active fire. MODIS (aboard Terra and Aqua satellites) provides similar detection at 1 km resolution as a cross-check.

Fire Radiative Power (FRP)

FRP is the rate of radiant heat energy being emitted by a fire, measured in megawatts. A grass fire might register 5–20 MW; a large crown fire can reach thousands of MW. FRP is a proxy for fire intensity and biomass combustion rate — it correlates with smoke emission and is widely used in fire behaviour modelling.

Confidence levels

VIIRS uses categorical confidence: high (mapped to 90%), nominal (70%), low (40%). MODIS uses a numeric 0–100 scale. We filter out detections below 50% confidence to limit false alarms. Hot sunlit surfaces (e.g. bare metal roofing, desert outcrops in summer) can produce false detections at low confidence.

Industrial heat filtering

Steel mills, gas flares, and refineries appear as persistent hotspots on every satellite pass — at the exact same location, for months on end. The fire pipeline uses a 48-hour lookback window; industrial filtering extends this to 7 days. Any cluster meeting all three criteria is classified as an industrial source and displayed separately from fires:

• ≥5 individual hotspot detections
• All detections within 1.5 km of the cluster centroid
• Detections spanning ≥18 hours (temporally persistent)

The clustering uses a tight DBSCAN pass (ε = 1.0 km, minPts = 5) before applying the spatial/temporal tests, to handle large industrial complexes like steelworks that span ~2 km but are still a single fixed-point emitter.

Assumptions & limitations

Cloud cover blocks satellite detection entirely — Australia's monsoon season degrades northern fire coverage significantly. Fires under dense tree canopy may be partially missed. Hotspot-derived polygon areas are rough estimates based on point clustering (DBSCAN, ε = 2.5 km, minPts = 3), not actual burn boundaries. NSW, VIC, and WA official polygons update when the agency publishes, typically every few hours. NASA FIRMS files update every ~10 minutes.

Site database

A curated static dataset of Australia's major industrial facilities, covering steel mills, aluminium smelters, alumina refineries, copper/zinc/lead/nickel smelters, cement plants, chemical and fertiliser plants, phosphate facilities, and coal export terminals. Each site has operator, state, and operational status (active, care & maintenance, planned, closed). The data is manually compiled and is not a live registry — it reflects the state of the industry at the time of last update.

Emissions data — two sources

Where available, facility-level emissions come from the CER Safeguard Mechanism covered emissions data for FY 2022–23. The Safeguard Mechanism covers 219 large industrial facilities that each emit ≥100,000 tonnes CO₂e per year, with a total covered emissions of ~138.7 Mt CO₂e — mandatory government reporting, so these numbers are statutory figures rather than estimates. For facilities not covered by Safeguard (smaller sites, or those with missing entries), emissions are filled from the Climate Trace API, which provides satellite and modelling-based estimates as a static snapshot (retrieved March 2026 — not a live API call). The popup on each facility shows which source is being used.

Industrial heat signals (satellite)

When the Industry layer is fully active, persistent thermal hotspots are overlaid from the same NASA FIRMS VIIRS pipeline used by the Fires layer. These are the clusters identified as fixed-point industrial sources — steel mills, gas flares, refineries — based on 7 days of orbital passes showing spatially compact (within 1.5 km radius) and temporally persistent (≥18 hour span) detections at the same location. Shown as red triangles at each confirmed industrial heat source. FRP (Fire Radiative Power in MW) and detection count are shown in the popup.

Assumptions & limitations

Site list is static — new facilities or closures since last update will not appear. Safeguard emissions are for FY 2022–23; the most recent full year available. Emissions figures are for the whole facility, not split by process or product. Climate Trace estimates carry wider uncertainty than statutory Safeguard figures. Industrial heat signals depend on cloud-free VIIRS passes; heavily cloud-covered periods may produce incomplete coverage.

Basin polygons

Approximate outlines of Australia's major sedimentary petroleum basins, with status (producing, development, exploration), resource type (gas, oil, oil & gas, coal seam gas, gas condensate), major fields, and operators. Sources are Geoscience Australia, NOPTA (National Offshore Petroleum Titles Administrator), and the Australian Energy Resource Assessment 2024. These are drawn as approximate display polygons, not exact regulatory survey boundaries.

Pipelines

A static linestring dataset of major gas transmission pipelines — including the Dampier–Bunbury Natural Gas Pipeline (DBNGP), Eastern Gas Pipeline, South West Queensland Pipeline, and others — with operator, length, and daily throughput capacity where known.

LNG terminals & oil refineries

LNG export facilities and oil refineries are drawn from the same curated industry site database as the Industry layer, filtered to those categories. Popup cards show capacity (Mtpa for LNG, barrels/day for oil refineries, TJ/day for onshore gas plants), number of processing trains, feedstock, equity partner breakdown, and emissions (Safeguard or Climate Trace, same methodology as the Industry layer).

Assumptions & limitations

All data is static. Basin outlines are for orientation — do not use for regulatory or title research. Pipeline routes are approximate centrelines; exact alignments are not published. Facility capacities reflect nameplate design capacity, not actual throughput.

Three layers in one

The Floods layer combines three independent data sources rendered on top of each other: a NASA satellite raster showing actual inundation extent, BOM warning zone polygons, and a river network overlay for spatial context.

NASA GIBS — VIIRS Combined Flood 3-Day (satellite raster)

The primary visual layer is a live satellite raster from NASA GIBS (Global Imagery Browse Services), serving the VIIRS Combined Flood 3-Day product. Multiple VIIRS satellites (Suomi-NPP, NOAA-20, NOAA-21) each make ~14 passes over Australia per day; NASA composites 3 days of passes into a single classification layer to minimise gaps from cloud cover. Each 375 m pixel is classified into one of four states:

Flood (shown red) — surface that is inundated and was not water prior to the event. Possible flood (yellow) — likely inundated but with lower confidence, often due to partial cloud obscurement or marginal spectral signal. Permanent / seasonal water (cyan) — rivers, lakes, reservoirs that are always or normally wet. Cloud / no data (faint grey) — pixel obscured; no determination possible.

NASA classifies pixels by comparing multi-spectral reflectance signatures — water strongly absorbs near-infrared and reflects little back, making flooded pixels spectrally distinct from dry land. The 3-day composite means the raster updates once per day with approximately a 1-day latency. No API key is required; tiles are served via open WMS.

BOM Warnings — catchment zone polygons

Overlaid on the raster are polygon outlines from BOM's Warnings API. BOM issues warnings when their forecasting system — which fuses river and rain gauges (10,000+ across Australia), weather radar, and Himawari geostationary satellite rainfall estimates — indicates a catchment will flood. Catchment boundary polygons come from BOM's ArcGIS FeatureServer. We only display Major and Emergency warnings — the two tiers where significant inundation is occurring or imminent. (Full scale: Watch → Minor → Moderate → Major → Emergency.)

River network

A river network overlay is drawn from a bundled Australian rivers GeoJSON at low zoom, switching to Mapbox's waterway vector tiles at higher zoom. Rivers and streams are highlighted when clicked, with an animated flow dash to show direction.

Assumptions & limitations

The satellite raster is optical/spectral — cloud cover blocks it. The 3-day composite reduces gaps but a sustained overcast period (common during east-coast flood events) can still leave large areas classified as "no data". The raster has ~1 day latency, so a rapidly developing flood may not yet be visible in the satellite layer even though BOM has issued a warning. The BOM polygon is the warning catchment area, not the actual flooded footprint — it covers the whole zone under warning, which is typically much larger than the inundated land. BOM severity is parsed from message text (not a metadata field) so text format changes can break severity detection. Warnings refreshed every 10 minutes; satellite tiles refreshed daily.

Sources

Three feeds are combined: BOM Warnings (primary source for systems in Australia's Area of Responsibility — roughly 80°E to 175°E, south to 50°S), GDACS (Global Disaster Alert and Coordination System, maintained by the European Commission — supplements BOM for storms outside the Australian region), and the BOM 7-Day TC Outlook (developing tropical lows and disturbances not yet at cyclone intensity, with probability-of-cyclone forecasts out to 7 days).

The Australian cyclone category scale

Australia uses a 5-category scale based on 10-minute sustained wind speed — this is different from the US Saffir-Simpson scale which uses 1-minute winds: Cat 1: 63–88 km/h (gales, minor damage), Cat 2: 89–117 km/h (destructive winds, significant damage), Cat 3: 118–159 km/h (very destructive, severe structural damage — "severe cyclone" threshold), Cat 4: 160–199 km/h (devastating damage), Cat 5: ≥200 km/h (extreme devastation, catastrophic). Note: BOM's Cat 3–5 maps roughly to Saffir-Simpson Cat 2–5, so Australian categories aren't directly comparable to Atlantic/Pacific classifications you may see in international coverage.

Track data

For active BOM warnings, the track (past positions and forecast path) is extracted from the GeoJSON embedded in BOM's warning detail. For the 7-day outlook, the "confidence area" polygons show where the disturbance centre is forecast to be at 12-hour intervals — these widen rapidly beyond 48 hours as forecast uncertainty grows.

Assumptions & limitations

Track forecasts are probabilistic guidance, not a guaranteed path. For BOM warnings, wind speed and category are extracted by regex from advisory text — if BOM's format changes, category parsing may break. For GDACS, wind speed is a structured data field (not text-parsed) and is generally more reliable, but GDACS track data may be incomplete for weaker systems. The 7-day outlook entries carry no wind speed or category data at all.

De-duplication is partial: BOM outlook entries are suppressed by name once a BOM warning exists for the same storm. GDACS storms are excluded from the Australian region by geographic bounding box, but are not de-duplicated by name — in rare cases, a storm crossing the boundary may briefly appear in both feeds.

Source

A curated static dataset of Australia's major commodity export ports — Port Hedland, Dampier, Hay Point, Gladstone, Newcastle, and others — with each port's location, state, primary commodities handled, and contextual notes. The dataset is also used internally by the Ships layer to classify vessels by destination port and to infer approach/departure direction.

Assumptions & limitations

Static data — port capacities and commodity handling change over time and may not reflect current operations. Port Hedland is the world's largest bulk export port by tonnage; the dataset includes all ports material to Australia's resource export chains.

Source

A static GeoJSON file of Australia's major private heavy-haul railways — primarily the Pilbara iron ore lines (BHP, Rio Tinto, FMG networks), the Bowen Basin coal rail corridors, and other resource-dedicated lines not shown on standard transport maps. These railways are private infrastructure, not part of the national public rail network, and are not published by government transport agencies.

Assumptions & limitations

Static data compiled from public sources including satellite imagery tracing and published company infrastructure disclosures. Routes are approximate centrelines. No live train movement data is available or shown.

What we're tracking

FIFO (Fly-In Fly-Out) operations are a dominant feature of Australian mining — workers commute by air between capital cities and regional hubs rather than living near mine sites. We track commercial and charter aircraft whose routes touch known mining hub airports (Pilbara, Goldfields, Bowen Basin, NW QLD, NT/Gulf, SA outback, NSW mining belt).

How ADS-B works

ADS-B (Automatic Dependent Surveillance–Broadcast) is the aviation equivalent of AIS. Aircraft transponders broadcast a 1090 MHz radio signal every ~0.5 seconds containing the aircraft's ICAO24 code (unique 24-bit identifier), GPS position, altitude, speed, and heading. Unlike radar, it requires no interrogation — it's a continuous broadcast. Ground stations and an increasing number of satellites pick up these signals and relay them to aggregators.

OpenSky Network is an open, academic non-profit that aggregates ADS-B from thousands of volunteer receiver stations globally. We use their API for authenticated access to the full Australian airspace snapshot. If OpenSky is unavailable, we fall back to the ADS-B LOL community feed, a crowd-sourced alternative.

How routes are inferred

Aircraft don't broadcast their flight plan via ADS-B — only their current position and heading. We filter to Australian-registered aircraft (originCountry = "Australia") that are airborne and travelling at least 55 m/s (~107 knots) to exclude taxiing, holding-pattern, or hovering signals.

For each qualifying aircraft we calculate the bearing from its position to every tracked mining hub airport, then compare that to the aircraft's actual heading. Confidence tiers:

• High — heading within 22° of airport bearing AND within 700 km
• Medium — within 40° AND within 1,200 km; or no heading data available but within 60 km of the airport

At least one endpoint (origin or destination) must be a mining hub for the flight to appear. Up to 120 flights are tracked at a time; once the cap is reached, lower-scoring candidates are dropped (score = distance + heading delta × 6, lower is better).

Assumptions & limitations

Route inference is probabilistic — some general aviation or charter flights will appear that aren't FIFO operations. Aircraft seating capacity is estimated from the ICAO type code (e.g. B737 ≈ 162 seats) — actual load factor is unknown. OpenSky coverage is thinner in remote areas; some aircraft may not appear if no receiver is in range. Refresh cadence is 20 seconds.

Site database

Mine site locations and metadata come from a curated static database keyed by Geoscience Australia mine ID, covering active and historical mine sites across Australia with operator, primary commodity, and production status. Processing and enrichment data (ore grades, processing method, production volumes) is sourced from the Mudd minerals dataset, a peer-reviewed academic compilation of Australian mine production history.

Emissions data

Where a mine appears in the CER Safeguard Mechanism 2022–23 dataset (mines emitting ≥100,000 t CO₂e/yr), those statutory Scope 1 emissions are shown on the mine popup. This covers large operations like Olympic Dam, BHP Mitsubishi Alliance coal mines, Bowen Basin operations, and major WA gold mines.

Assumptions & limitations

Static data — new mines, closures, or ownership changes since last update will not be reflected. Commodity prices shown alongside mines are spot prices from the commodity prices feed (see below), not mine-specific contract prices or realised revenues. Safeguard emissions are FY 2022–23; smaller mines below the 100,000 t threshold do not have reported emissions shown.

What is shown

125 major Australian data centre facilities across all eight states and territories — operational, under construction, and publicly announced planned sites. Coverage spans wholesale colocation campuses, cloud hyperscale deployments, government-classified secure facilities, retail colocation, and key enterprise sites. Small server rooms and retail edge nodes are not included.

Where the data comes from

This is a curated static dataset compiled from publicly available sources. Each entry links directly to the operator's own facility page (via the “Operator source” link in the popup). Key operators covered include:

• NEXTDC — ASX-listed operator; capacity and commissioning years from annual reports and investor presentations.
• AirTrunk (BlackRock) — campus locations and headline capacity from company website and press releases.
• Equinix — International Business Exchange (IBX) locations from the Equinix data centre finder.
• CDC Data Centres — Eastern states and ACT campus details from company disclosures and government contracts.
• DigiCo Infrastructure (formerly HMC Capital) — campus locations and capacity from company announcements.
• Digital Realty, STACK Infrastructure, Keppel, Vantage — wholesale colo campuses; capacity and timelines from company websites and press releases.
• Hyperscale cloud regions (AWS, Azure, Google, Oracle) — region presence confirmed via provider documentation; capacity not publicly disclosed.
• Regional and government operators (DXN, DCI, iseek, DC Two, Vocus, Telstra, state government data centres) — locations from company and government disclosures.

Coordinates are approximate. For campuses with multiple closely spaced buildings, a single representative point is used.

How size and colour work

Each circle is coloured by build status: cyan = operational, amber = under construction, violet = planned. Circle radius scales with rated IT load capacity in MW (sqrt scaling so the largest sites don't dominate the map). Sites where capacity is not publicly disclosed use a small default radius.

Power supply

Each popup shows the known power supply arrangement — the electricity grid connection and any publicly disclosed renewable energy contract or PPA (Power Purchase Agreement). Most large operators have committed to 100% renewable energy matching via LGC (Large-scale Generation Certificates) or direct PPAs with wind and solar generators. Specific counterparty names are rarely disclosed publicly. ACT-based facilities benefit from the ACT Government's 100% renewable electricity scheme, which contracts directly with wind and solar generators on behalf of territory consumers.

Assumptions & limitations

Capacity figures are rated IT load (the power drawn by servers and storage), not the facility's total power draw (which is higher by a factor of PUE, typically 1.2–1.5). Data is updated manually — new announcements or capacity expansions may not be reflected immediately. Planned facilities may be cancelled or delayed; under-construction timelines are estimates from public disclosures. Hyperscale cloud regions do not disclose capacity, so their circles reflect only presence, not scale.

Source

Commodity prices are static research estimates based on TradingEconomics data. 17 commodities are covered: Gold, Silver, Copper, Zinc, Iron Ore, Coal, Uranium, PGEs (platinum group), Cobalt, Diamonds, Lead, Lithium, Manganese, Nickel, Rare Earths, Tin, and Tungsten. Most prices are as of February 2026; Diamonds uses a Q1 2025 estimate and Tungsten a December 2025 estimate (both thinly traded with infrequent public price data).

This is not a live feed — prices are indicative only and will drift from market rates over time. A live commodity price feed is on the roadmap.