FireSim: Canadian FBP wildfire spread simulator

Fire growth, fire behaviour and building exposure from the Canadian Forest Fire Behavior Prediction (FBP) System, built for municipal Emergency Operations Centre (EOC) planning, preparedness and training in the wildland-urban interface.

• Live app: https://wildfire-simulator.vercel.app (API: https://firesim-api.fly.dev/api/v1/health)
• Intended use (now): preparedness, training, exercises and what-if planning. FireSim's FBP equations are verified against the CFS reference implementation and its spread against WISE on uniform fuel. A first validation against observed Alberta fires (docs/validation.md) shows modest one-day skill with over-prediction, like WISE with default settings. Do not use its output as an operational prediction or as the basis for evacuation decisions. See the model card and verification and limits.

Evidence status

Component	Status	Where
FBP System (18 fuel types, ST-X-3 + GLC-X-10)	Verified against CFS cffdrs to floating-point precision	docs/fbp-reference.md
FWI System, hourly FFMC	Verified against cffdrs	docs/verification.md
Spread models (Huygens, level set)	Burned area 0.90-1.07 x the FBP ellipse; vs WISE on uniform fuel and slope, area ratio 0.955-1.09 and head 0.98-1.05	docs/verification.md
Head / flank / back, head summary	Fastest head cell spreads at FBP ROS and rearmost back cell at BROS on uniform fuel; head direction within 2° of RAZ	engine/tests/spread/test_deployment_data.py
Spotting	Maximum distance (Albini/Chase/Morris) reproduces published examples; emission and landing are heuristic (illustrative only)	docs/verification.md
Building exposure	Reproduces Cohen's (2004) SIAM worked values; exposure, not ignition probability	docs/building-exposure.md
Observed fires	First validation on 143 Alberta fire-days (CFSDS, Bennett et al. 2026 protocol): one-day F1 0.15-0.24 at the default 17 h window, over-predicts growth on most days; in the same low band as WISE, not an operational forecast	docs/validation.md

What it does

• FBP fire behaviour for all 18 fuel types: rate of spread (head, flank, back), fuel consumption, crown fraction burned, head fire intensity, fire type, length-to-breadth, slope through net effective wind, point-ignition acceleration, foliar moisture from date.
• Fire growth on spatial fuel grids with a deterministic Eulerian level-set model (the framework of ELMFIRE, Lautenberger 2013, driven by the Huygens velocity of the FBP ellipse; wraps around water, roads and buildings), or Huygens wavelets on uniform fuel. Hourly weather streams with hourly FFMC; Open-Meteo forecast aligned to a scenario start time.
• Deployment data: per-cell front speed and head / flank / back, the head's speed, direction, intensity and Albini maximum spotting distance per frame, and an arrival-time grid (GET /simulations/{id}/arrival).
• Starting from an observed fire: an RPAS (drone) perimeter or the previous day of a multi-day run.
• Building exposure: distance bands, worst-case radiant flux and a flux-time dose index per building (exposure, not ignition).
• Burn probability (Monte Carlo over ignition point, wind speed and RH).
• EOC console: incidents and operational periods, ICS forms and an ICS Canada 209-WF situation report built from the run (docs/ics-canada-209.md).
• Neighbourhoods: the modelled time the fire is first within 500 m of each neighbourhood. The evacuation status (Order / Alert / Watch) is set by Planning and drawn as blue outlines with line styles and labels. The model never suggests a tier.
• Classes from published sources: head fire intensity classes 1-6 (Cole & Alexander 1995; CWFIS map limits) and FWI classes (CWFIS national FWI map intervals).
• Uncluttered screen: panels show labels, values and short badges ("Low one-day skill", "Range too narrow", "Model output", "Status set by Planning", "Illustrative", "Unsourced"...); explanations open as accessible tooltips (hover, keyboard focus or tap), and intended use, limits, methods, references and data sources are in the About & sources tab (frontend/README.md). A progress bar follows the main run; "New ignition" re-runs at another spot without a reload.

Stack

Layer	Technology
Fire engine	Python 3.10+ (NumPy, SciPy, Shapely), no web dependencies
API	FastAPI + WebSocket (Fly.io)
Frontend	React 19 + Vite + TypeScript + MapLibre GL (Vercel)
Tests	pytest (engine, API), Vitest and Playwright + axe (frontend), GitHub Actions

Quick start

# API
PYTHONPATH=engine/src:api/src uvicorn firesim_api.main:app --port 8000
# Frontend (second terminal)
cd frontend && npm install && npm run dev

Open http://localhost:3000, click the map to set an ignition point, set the weather and run. Edmonton data paths and other settings: docs/deployment.md.

Data

The Edmonton layers in frontend/public/edmonton/ load automatically with the Edmonton fuel grid: critical assets (assets.geojson, 623 features: City of Edmonton Open Data, Government of Alberta continuing care list, Statistics Canada ODHF, © OpenStreetMap contributors), major roads (roads.geojson, motorway to secondary with ramps, © OpenStreetMap contributors) and neighbourhoods (City Open Data). Sources, licences, dates and refresh: docs/data-sources.md. Rebuild with python3 scripts/build_edmonton_assets.py and python3 scripts/build_edmonton_roads.py. Other jurisdictions add their own GeoJSON under Setup, "Add your own layer".

Testing

make test           # engine and API (pytest): cffdrs reference, FBP ellipse, WISE comparison, exposure, ...
cd frontend && npm test            # Vitest unit tests
cd frontend && npx playwright test # end-to-end with a mocked API, accessibility (axe), performance budget

See frontend/README.md for the frontend fixture and test setup.

Project structure

engine/     Fire science: FBP, FWI, spread models, spotting, exposure, data loaders (no web deps)
api/        FastAPI service: runs, streaming, arrival grid, FWI and weather endpoints
frontend/   React app: map, setup, situation panel, timeline, EOC console
docs/       Model card, verification, FBP reference, architecture, API, deployment

Documentation

• docs/model-card.md: intended use, users, evidence, limits, out-of-scope uses
• docs/verification.md: what is verified, how, history of fixes, known limits
• docs/fbp-reference.md: FBP equations and inputs
• docs/building-exposure.md: exposure metrics and their sources
• docs/architecture.md: components and spread models
• docs/data-sources.md: bundled datasets, licences, dates, refresh
• docs/api-reference.md, docs/deployment.md

Name

"FireSim" is also the name of a module in Technosylva's Wildfire Analyst. The name of this project is under review before any release outside the City of Edmonton.

References

Fire behaviour

• Forestry Canada Fire Danger Group (1992). Development and Structure of the Canadian Forest Fire Behavior Prediction System. Information Report ST-X-3.
• Wotton, B.M., Alexander, M.E., Taylor, S.W. (2009). Updates and revisions to the 1992 Canadian Forest Fire Behavior Prediction System. Information Report GLC-X-10.
• Van Wagner, C.E. (1977). Conditions for the start and spread of crown fire. Can. J. For. Res. 7: 23-34.
• Van Wagner, C.E. (1977). A method of computing fine fuel moisture content throughout the diurnal cycle. Information Report PS-X-69 (hourly FFMC).
• Van Wagner, C.E., Pickett, T.L. (1985). Equations and FORTRAN program for the Canadian Forest Fire Weather Index System. Forestry Technical Report 33.
• Wang, X. et al. (2017). cffdrs: an R package for the Canadian Forest Fire Danger Rating System. Ecological Processes 6: 5.
• Alexander, M.E., Cruz, M.G. (2012). Interdependencies between flame length and fireline intensity. Int. J. Wildland Fire 21: 95-113.
• Cole, F.V., Alexander, M.E. (1995). Head fire intensity class graph for FBP System Fuel Type C-2 (Boreal Spruce). Alaska DNR Division of Forestry and Canadian Forest Service.

Fire growth

• Richards, G.D. (1990). An elliptical growth model of forest fire fronts and its numerical solution. Int. J. Numer. Meth. Eng. 30: 1163-1179.
• Tymstra, C., Bryce, R.W., Wotton, B.M., Taylor, S.W., Armitage, O.B. (2010). Development and structure of Prometheus: the Canadian Wildland Fire Growth Simulation Model. NOR-X-417.
• Lautenberger, C. (2013). Wildland fire modeling with an Eulerian level set method and automated calibration. Fire Safety Journal 62: 289-298. doi:10.1016/j.firesaf.2013.08.014

Spotting

• Albini, F.A. (1979). Spot fire distance from burning trees: a predictive model. GTR INT-56; Albini (1981) Res. Note INT-309; Albini (1983) Res. Pap. INT-309; Chase, C.H. (1981) Res. Note INT-310 and (1984) INT-346; Morris, G.A. (1987) Res. Note INT-374. USDA Forest Service.

Structure exposure

• Cohen, J.D. (2004). Relating flame radiation to home ignition using modeling and experimental crown fires. Can. J. For. Res. 34: 1616-1626.
• National Research Council Canada (2021). National Guide for Wildland-Urban Interface Fires.