Hotspots
Raw thermal detections arriving every half hour
Context
Terrain, fuel load and population enriching every cluster
OzHazard
Powered by the Terrakio Agent
The Challenge
Australia receives a relentless stream of thermal detections, geostationary imagery, lightning strikes and weather updates that could signal an emerging bushfire. Turning that raw firehose into a live, trustworthy operational picture has traditionally demanded human analysts working around the clock.
A single thermal pixel is not a fire. It has to be correlated against the land it sits on — the fuel it might burn, the slope it will accelerate across, the communities downwind — and cross-checked against authoritative agency feeds before anyone can say what is actually happening.
The system needed to move from retrospective dashboards and manual triage to a continuous, autonomous loop — one that could reason over continental-scale evidence, remember what it had already decided, and publish a live, auditable picture of active hazards to operators and the public alike.
The Solution
OzHazard is a domain profile over the Terrakio Agent — Haizea's AI-driven geospatial intelligence service — running on top of the Terrakio analytics engine. The same framework that answers "what do the pixels say?" is used here to answer "what is happening right now, what does it mean, and what should we do about it?".
Autonomous Sense–Reason–Publish Loop
OzHazard runs the Terrakio Agent as an always-on service: it ingests FIRMS, Himawari, lightning and agency feeds, filters and clusters them, and reasons over each event with a purpose-built tool set — opening, updating, merging and closing incidents without human intervention.
Context-Aware Enrichment
Every candidate cluster is enriched through Terrakio DSL queries against the Copernicus DEM, crown cover, land use and population density, so the agent sees not hotspots but fronts — each framed by terrain, fuel and exposure.
Traceable, Correlated Intelligence
Every cycle appends to an immutable reasoning log and every incident carries a correlation record against NSW RFS and WA DFES, with measured detection lead time — turning autonomous judgement into quantifiable, auditable performance.
The result is a continental bushfire intelligence service delivered as both a live public dashboard and a standards-compliant API — the same agent framework Haizea re-profiles for flood watch, asset surveillance and environmental compliance elsewhere.
How It Works
A continuous sense, enrich, reason, publish and self-verify loop powered by the Terrakio Agent
Sense
FIRMS thermal detections every half hour, Himawari-8 imagery every ten minutes, a live lightning stream and authoritative feeds from NSW RFS and WA DFES are filtered, deduplicated and clustered before the agent ever reasons about them.
Enrich
Each fire cluster is enriched through Terrakio DSL queries against a 30 m DEM, crown cover, land-use classification, population density and live Himawari imagery — framing every event in terrain, fuel and exposure.
Reason
A large-language-model agent with a purpose-built tool set opens, updates, merges or closes incidents, declares risk zones, issues predictions with deadlines, records observations and persists memory across cycles.
Publish
Conclusions are written to state and broadcast on a real-time event bus, feeding a live dashboard, a versioned GeoJSON API and notification channels within seconds of the agent's latest cycle.
Self-Verify
Predictions are scored against observed reality when their deadlines pass and extents are back-tested against Geoscience Australia boundaries — producing continuously measurable lead time, IoU and F1 metrics.
OzHazard now:
Watches the continent without sleeping: an always-on agent loop that ingests thermal, geostationary, lightning and agency feeds on their native cadence and reasons over every meaningful cluster, replacing shift-based triage with a continuous operator.
Turns pixels into named incidents: each fire is a governed object with geometry, severity, an active front, a cumulative burnt area, and a narrative assessment — not a list of hotspots waiting to be interpreted.
Measures itself against ground truth: every agent-generated incident is correlated against NSW RFS and WA DFES records, with detection lead time, false-positive rate and missed-event rate continuously quantifiable.
Publishes live to the public and to agencies: the same output drives a public dashboard with a streaming reasoning timeline and a versioned GeoJSON API ready to drop into existing government GIS workflows.
Explore Our Case Studies
Evidence-based environmental policy in the ACT
ACT Government Sustainability
Annual environmental reporting across Australia
TERN & ANU Environmental Monitoring
Enabling Australia's Nature Repair Market
Australian Government
Continental bushfire intelligence with an AI agent
OzHazard — Terrakio Agent in production
A continental, EUDR-ready map of Australia's forests
VMAP — WWF, ANU & Haizea Analytics
Configure an always-on geospatial agent for your domain
OzHazard is one domain profile over the Terrakio Agent framework. The same always-on sense–reason–publish loop can be re-targeted to flood watch, asset surveillance, agricultural advisory or environmental compliance — change the profile, not the system.
