Fireline Analyzer — System capabilities

What the system does, how, and why it fits

Overview

Turning drone thermal into containment decisions

Fireline Analyzer reads thermal imagery from drone overflights, finds the fire fronts most likely to break the line, puts them on a live map with wind and terrain, and forecasts where they could move next.

Follow the pipeline in the chart above — each step is a slide. Jump to any of them, or scroll through in order.

A six-hour spread forecast on satellite imagery: nested arrival-time bands running from yellow at the outside through orange to red at the ignition point, a dashed ember-spotting ellipse downwind of it, and wind barbs across the scene. The control panel on the left shows the horizon buttons, fuel greenness, and the crown fire and spotting overlay toggles.
A six-hour forecast — arrival-time bands, the indicative ember zone downwind, and the wind that drove it.

Detection

01

Breakout-risk hotspot scoring

It finds the active-fire hotspots that most threaten the containment line — not just the hottest pixels, but the fronts most likely to escape.

How it works

The DJI Thermal SDK converts each thermal frame to a per-pixel temperature map. A scene-adaptive segmentation, anchored to the frame's own ambient, splits pixels into unburned, burned, and active fire. Each hotspot is scored intensity × area × unburned-fuel adjacency × proximity to the line — deterministic, no black box.

Why it fitsThe real question is where fire might cross the line, not where the heat is. Weighting hotspots by how much of their edge touches unburned fuel near the line surfaces exactly the encroaching fronts crews act on — and a transparent score is defensible where a black-box number is not.

Location

02

Geolocating every detection

Each hotspot is placed at real ground coordinates — with the method, and therefore the confidence, made explicit.

How it works

Three tiers, best first: projected (per-pixel from the camera model, GPS, altitude, gimbal), center (oblique frames placed at the image-center ground point), and a drone-GPS frame fallback — each point tagged with how it was derived. A jurisdiction-agnostic guard rejects any frame whose GPS is a far outlier (>50 km) from the rest of its own mission.

Why it fitsA score is only actionable at a coordinate. Tiering and tagging keeps positional uncertainty honest, and the outlier guard tolerates mutual-aid deployments anywhere while catching bad GPS fixes that would scatter detections hundreds of miles away.

Common operating picture

03

The interactive fire map

One shared map: containment lines, risk markers, and high-resolution aerial imagery over a choice of basemaps — the same picture in the office and on the line.

How it works

Built on Mapbox GL. Analysts draw and edit as many containment lines as an incident needs — a primary, a contingency, a section already walked — each named and measured. Risk markers are colored by severity and filterable (all / mid+high / high only); orthomosaics and thermal overlays stream in as map layers; basemaps switch between satellite, topographic, and street. Live NIFC/WFIGS fire perimeters can be pulled straight onto the map, shapes imported from GeoJSON, KML or shapefile, and everything exported back out as vector or as a georeferenced GeoPDF that opens in Avenza, QGIS or ArcGIS.

Why it fitsContainment is spatial. Seeing hotspots relative to the line, over real terrain and imagery, is the core of line planning — and the severity filter cuts hundreds of detections down to the handful that matter.

Atmosphere

04

Live wind field

An animated wind overlay shows current speed and direction across the fire area at a glance.

How it works

Current 10 m wind from the National Weather Service is rendered as a flowing, speed-colored particle field — the same style of wind map crews already read.

Why it fitsAfter fuel, wind is the dominant driver of spread, direction, and spotting. A live wind picture over the same map as the hotspots is baseline situational awareness — and it's the same wind that drives the spread forecast, so what crews see matches what the model uses.

Imagery

05

Georeferenced ortho imagery

High-resolution orthomosaics stitched from the flight, draped on the map so crews see the burn area exactly as it looks right now.

How it works

Uploaded orthomosaics are processed into georeferenced cloud-optimized GeoTIFFs on S3 and streamed to the map as tiled overlays via windowed range reads, so only the on-screen area loads. Overlay opacity is adjustable, and the imagery lines up with the containment line, hotspots, and wind on the same map.

Why it fitsA current, high-resolution aerial — captured on the same flight as the thermal — is ground truth a generic satellite basemap can't give: real-time fuel, structures, access routes, and how the fire actually sits on the landscape.

Prediction

06

Fire-spread forecasting

A short-horizon forecast of where fire could travel — isochrones from 15 minutes to 24 hours, seeded from the detected hotspots, a point, or the fire's own perimeter.

How it works

It pulls LANDFIRE surface fuel plus slope and aspect for the fire's neighborhood, combines them with live wind and a fuel-moisture scenario, and computes a Rothermel rate of spread. Fire is advanced with elliptical propagation; arrival times contour into nested isochrones. Past a few hours the weather is stepped over the run rather than held constant, so a forecast follows the wind shift instead of ignoring it. Beyond the surface fire it flags crown fire initiation from LANDFIRE canopy bands, an indicative ember-spotting zone downwind, and a burn-probability mode that runs the forecast many times over a spread of wind directions.

Why it fitsRothermel is the surface-fire model behind BehavePlus and FARSITE — the lineage the fire community trusts. Grounding it in real fuel, terrain, and wind gives a defensible, directional read on near-term movement. It is short-horizon decision support, not a validated perimeter.

In the field

07

The same fire, with no signal

A native iPad app that carries the mission to the line — hot spots, containment lines, imagery and the spread model — and works with the radio off.

How it works

A crew downloads the mission while they still have coverage: the agency's own records, stored on the device and read back through the same code that reads them online. The Rothermel engine is ported to Swift and runs on the iPad, checked against the server's Python to within a rounding error, so a forecast needs no connection at all. Work made in the field — a line walked, a shape drawn, a photo taken, a forecast run — is written to the device and reaches the agency by itself when the signal comes back.

Why it fitsCoverage on a fireground is the exception, not the rule. A tool that needs a bar of signal is a tool that works at the truck and not at the line — so the device holds everything it needs, and syncing is something that happens later, without anyone remembering to do it.

Backbone

08

Data & compute that hold up in the field

The infrastructure that keeps all of the above fast, reliable, and cheap to run.

How it works

Fuel and terrain come from LANDFIRE, auto-falling-back through data vintages so any covered region resolves. Each agency's region is clipped once and read window-by-window instead of loaded whole; the heavy spread compute runs in a separate process so the app never stalls; results are cached.

Why it fitsFire programs run on lean budgets and modest hardware over field connections. Reading only the window a fire needs, computing off the request path, and caching aggressively keeps the tool responsive — and a new agency comes online by clipping its area once, with no per-incident setup.

Governance

09

Multi-agency isolation, security & audit

Multiple agencies share the platform without sharing data, under role-based access and an audit trail.

How it works

Every mission, image, detection, and forecast is scoped to its agency; roles gate who can edit versus view; sessions time out on inactivity with a clean re-login; and privileged changes are recorded to an audit log.

Why it fitsWildfire is inter-agency by nature — mutual aid, cooperators, and overlapping jurisdictions. Strict data separation with a defensible record of who changed what is exactly what shared-but-accountable operations require.

Honest scope

What it is — and what it is not

Credibility with fire professionals depends on being clear about limits.

  • The spread forecast is short-horizon and directional. It is not a validated perimeter forecast and does not model multi-day growth.
  • Crown fire is modelled only where LANDFIRE carries canopy data; elsewhere it is not assessed, and the map says so rather than implying no crowning.
  • Spotting is indicative, not a loft model — a flame-length heuristic pointing downwind, a cue about where to look rather than where embers will land.
  • Burn probability varies the wind only, holding fuel, moisture and terrain fixed. It is not a full uncertainty treatment.
  • Fuel and weather are US-based (LANDFIRE + National Weather Service); detections carry a geolocation-confidence tag.
  • The risk score is deterministic, explainable, and relative to its own mission — “high” means high on this fire, not on an absolute scale.
  • A downloaded mission is a timestamped snapshot of what the agency held, and the iPad says when it was taken.
In shortUsed within that scope — locating encroaching heat, watching wind, and getting a defensible directional read on near-term spread — it compresses a slow, manual thermal-review-and-mapping task into a live operating picture.
01 / 10

Use the chart above to jump to any step — or scroll sideways, use ←/→, or the arrows.