01
Tells you
Email and SMS when a detection arrives in or near a project. One message per batch, a cooldown on running events, de-duplication so the same fire never alerts twice — and a stand-down message when it is over.
Overwatch · Live monitoring
Know the moment fire starts. And where it's heading.
Overwatch watches your land around the clock, tells the right people when something is detected nearby, and has the first round of analysis done — simulated, assessed and written up — before anyone is awake.
Detections in
- Satellite hotspots
- Lightning strikes
- Agency incidents
- Your own feeds
Assessed by
Overwatch agent
Weather and fire danger
Your assets
Local fire history
Every step logged
Modelled with
Ensemble simulation
Hectares and assets reached
Confidence range
Spread animation
Sent to you
Email and SMS
An operational brief
— or a reasoned decision
not to disturb you
Detections are assessed against the landscape you have already built in Mitigate — fuel loads, treatments, assets and fire history.
Why it matters
The first hour decides most fires. It is usually spent finding out.
Detection data already exists. What is missing is someone watching it at 2am, filtering out the noise, and turning a pixel into a decision before the fire has made it for you.
Without Overwatch
With Overwatch
Fires start when nobody is looking — overnight, on a weekend, on the far side of an estate no one has driven this week.
Every project is watched around the clock, and satellites re-scan every ten minutes. You hear about it while there is still time to act.
Raw hotspot feeds are noisy. Burn-offs, industrial heat and the same fire re-detected on every pass soon teach people to ignore their phone.
Detections are grouped into events, judged against your instructions and de-duplicated. When your phone goes off, it is worth reading.
A hotspot is a dot on a map. The first hour goes on pulling the forecast, finding the fuel layers and working out what it could reach.
The alert arrives with that work done: weather, fire danger, an ensemble spread simulation and the assets in its path.
Watching a large estate properly means someone on screens, or a roster of people checking feeds after hours.
Cover a whole portfolio without adding a shift. People are brought in when judgement is needed, not to sort noise.
After an event it is hard to show what was known, when, and why someone did — or did not — act.
Every assessment is logged with its reasoning, including the decision not to alert. A record for your team, a regulator or an insurer.
Built for
- Plantation forestry
- Power utilities
- Land management
- Fire and rescue
- Insurance
- Carbon and environmental projects
What it watches
Every way a fire announces itself.
Each detection type has its own map layer, a 1h / 6h / 24h / 72h time window and a live count. Click any detection to inspect it, or turn it straight into an ignition.
- Satellite hotspots
- Thermal detections from geostationary and polar-orbiting satellites, grouped into distinct fires by proximity.
- Lightning strikes
- Ground strikes near your land, with strike polarity — the ignitions nobody saw happen.
- Agency incidents
- Incidents and warnings published by state emergency services.
- Your own feeds
- Camera networks, sensors or crew-reported ignitions, watched exactly the same way.
What it does
02
Simulates it
Any detection becomes a fire spread simulation. An ensemble of runs under the current forecast gives the likely area burnt and the assets reached, with a range around both.
03
Works the problem
An AI agent assesses each detection against weather, fire danger, your assets and local fire history, decides whether it warrants a simulation, publishes an operational brief and sends you the link — recording every step it took.
The alert
What lands on your phone.
The SMS is the short version for 2am: what was seen, where, and the headline simulation figures. The email carries the full picture — distance to your boundary, how long it has been burning, new detections since, and a link to the brief with the spread animation and exposed assets.
You will hear again if a separate event appears, the area flares up after going quiet, or the fire grows into your alert radius. Otherwise, you won't.
12:32
Mitigate: single satellite hotspot detected ~1km SW of Conjola National Park, NSW, 12:21pm 18 Sep, one pass, unconfirmed. IF an uncontrolled fire, simulation under forecast conditions suggests ~55ha/24h median spread, 1–5 structures potentially affected. Brief: mitigate.fi-sci.ai/s/conjola-national-park-hotspot…[AI-generated – verify in Mitigate]
FiSci Mitigate
Every decision, written down
Quiet by design. Accountable either way.
Most detections are routine — a small, low-intensity cluster in country with plenty of burn history, under benign weather. Overwatch is built to stay quiet about those, and to write that decision up as fully as a decision to alert.
Below is a real activity log. The agent woke on a hotspot, pulled the forecast and the day's fire danger, checked the project and its assets, calculated the fire danger index — then closed it without waking anyone.
- Agent monitoring auto-started (project auto-monitor is on)08:39
- list_realtime_observations08:39
- get_weather_forecast08:39
- get_daily_fdi08:39
- get_project08:39
- calculate_ffdi08:40
- list_assets08:41
- Single hotspot detection, benign weather, no confirming detections — closed, nobody notified08:41
You set the rules
Tell it what matters. It listens.
AI incident monitoring is switched on per project, so you can run it on the ground that matters most. Instructions can raise or lower the threshold for waking people — but never stop the agent reporting a life-safety threat it can evidence.
Standing orders
Tell the agent, in your own words, what matters most and what local practice explains a detection. Set once for the organisation, refined per project.
“Do not notify anyone for a single detection more than 25 km from an asset unless the forecast fire danger is severe or above.”
Alert radius
Choose how far beyond your boundary a detection can be and still raise an alert. The fire just outside the fence is usually the one that matters.
Automatic simulations
Have a fire simulated the moment a given kind of detection opens an incident, and optionally published straight into your own GIS as time-enabled layers.
Who hears what
Per person, per project, per detection type, per channel. Administrators can cover a whole district team without waiting for anyone to log in.
Run your own
A detection is a pixel, not a point.
Want your own spread prediction, or a different weather assumption? Click any detection and create an ignition from it. For satellite hotspots, fill the pixel's area of uncertainty with a grid of ignitions and run each as its own fire — so the result reflects everywhere the fire could really have started.
What sets it apart
5 m
Resolution at scale
Five-metre detail even on the largest estates, so a simulated run follows the roads and trails that are really there.
P10–P90
Confidence, not a guess
Ensembles vary weather, fuel and ignition together. You get a range around the answer, not a single fire line.
24/7
Always on
Satellite re-scans every ten minutes, assessed as they land — including at 2am, when nobody is looking.
Logged
Every decision on record
Why it alerted, and why it did not. Reviewable by your team, a regulator or an insurer.
Overwatch is available as an add-on to a Mitigate licence; detection type availability varies by organisation. Everything Overwatch produces is AI-generated and advisory. It supports your judgement and does not replace official warnings or instructions from emergency services.
See it watching ground you know.
We'll build a project over your own landscape — your assets, your fuel loads, your fire history — and show you what Overwatch would have told you.