Wildfire spread simulator

Runs in your browser. Keys stay in this browser.

One-time settings · Accounts

NASA FIRMS map key Google OAuth client ID Earth Engine project ID CARTO basemap API key (optional) Remember in this browser

Not signed in.

Mode

Area of interest

Draw a polygon or rectangle with the tools at top left of the map. Edit or delete with the same toolbar.

Load GeoJSON Load road/river lines GeoJSON (optional)

If online geometry is unavailable, supply road/river LineString GeoJSON here.

Fire data

Choose one or more sources. MODIS via Earth Engine is the fallback if FIRMS fails.

Sources VIIRS S-NPP VIIRS NOAA-20 MODIS MODIS via Earth Engine

Ignore low-confidence detections Load FIRMS CSV (optional, replaces API)

Fire history fromto

Gap between event ranges (days)

Choose one or more ranges; start and duration follow your selection.

Environmental data

Satellite (sets cell size) Cell size (m)

Fuel curing = green-season NDVI minus dry-season NDVI.

Green NDVI fromto

Dry NDVI fromto

NDMI fromto

NDWI fromto

Weather fromto

Pre-fire LST fromto

DEM source

Minimum ΔNDVI: no fuel below this 0.10

Choose an AOI and both NDVI date ranges, then press Preview. Other data need not be prepared.

NDMI, NDWI, WorldCover and Copernicus GLO-30 load during Prepare. Weather layers show the mean over the selected range.

Weather point spacing (km) Skip Earth Engine (uniform test terrain)

Optional Field Data

A. Weather Calibration

Only supplied measurements are corrected. Enter latitude, longitude and UTC date/time before Prepare.

LatitudeLongitude

Date (UTC)Time (UTC)

2 m temperature (°C)Wind speed (m/s)

Wind direction (degrees)Relative humidity (%)

Precipitation (mm/h) Upload weather CSV

CSV Format / Example

UTC date and time are required; leave unused measurements blank. Example:

latitude,longitude,date,time,temperature_2m,wind_speed_10m,wind_direction_10m,relative_humidity_2m,precipitation
29.450,79.140,2026-04-10,12:00,27.5,3.2,235,37,0

Weather calibration method

B. Vegetation & Fuel Observation Dominant species (one per line)

Dry foliage / litter present Foliage flammability class

Only species with explicit published evidence make a uniform AOI layer. Unresolved species have no effect.

Simulation parameters

Duration (hours) Step length (minutes) Base spread rate R₀ (m/min) Burn duration (minutes) Moisture multiplier Rain multiplier Ignition size Ignition radius (m) Fire lookback (hours)

Conditioning influence (0 = off; 1 = normal) Fuel curing NDMI moisture Soil moisture Slope (DEM) Elevation effect on weather Wind speed Wind direction Air temperature Relative humidity Precipitation Land surface temperature (LST)

Water, non-burnable land cover, and the hard ΔNDVI cutoff always apply.

Base and calibrated runs are recorded automatically. Their download buttons appear in Results when each recording is ready.

1 · Prepare & baseline run

Run the first simulation with your current parameters.

After Prepare, you can also click the map to add an ignition point.

Map layers

Choose layer Layer opacity (%)

2 · Observed burn & baseline validation

Compare satellite dNBR or RBR with the baseline. FIRMS footprints are a provisional reference.

Load satellite reference during Prepare Burn index Burnt if index >

Before fire (days)After fire (days)

Tolerance (cells)

3 · Optimization & calibration

Calibration requires satellite burn rasters. A held-out event is reserved for independent evaluation.

Optimization depth Final repeats

Search effort / optimization runs 24

The slider goes to 5,000; Custom also accepts larger numbers. Simulation runs in your browser. Keep this tab open for long runs; sleep or closing the tab may interrupt the search. More runs do not guarantee a better fit.

Parameters to tune

4 · Optimized run, validation & decision

The complete model reruns automatically with the best combination. Baseline parameters remain saved until the final decision.

5 · Results and downloads

Ensemble runs (10–30)

Run videos

After a run finishes, its video will be available here.

About the simulator · README

Developer: Zainab Khan, PhD, former postdoctoral fellow, IIT Roorkee, India.

khanzainub@gmail.com

Enter a FIRMS key in Accounts. Sign in with Google for Earth Engine rasters.
Draw an area on the map or load a GeoJSON.
Choose fire sources, dates and event ranges in Fire data.
Review Environmental data. Optionally add UTC weather measurements or CSV, look up species in research, and select dry litter class in Optional Field Data. Then review Simulation parameters and press Prepare.
Review the green checks and red crosses. Run baseline, load observed satellite burn and validate the baseline.
Choose ranges for training and, if possible, hold out one event for independent testing. Select parameters and bounds, then run automatic calibration.
Watch benchmark and real optimization runs, compare Base and optimized results, then accept or restore Base after independent validation. Download GeoTIFFs, videos, plots and CSVs.

Nearby detection dates may merge into one range. All selected points in each range ignite at its t=0. Multiple ranges are simulated separately and their burnt areas are combined. The animation does not independently establish the historical route.

The first three actual runs benchmark this area and device. The search explores broadly, then refines near promising candidates. Each candidate runs the full CA model.

Improvement on training events measures fit. Independent validation needs a held-out event. A worse holdout score rejects the proposal and restores baseline; without a holdout the proposal remains unverified.

Weather correction uses only supplied variables; quantile mapping needs at least two matched observations. A species without explicit research evidence has no effect. High/Medium/Low field flammability multiplies the existing spread by 1.15/1/0.85 for each supplied uniform AOI layer; water and the fuel threshold still block fire.

This is a research prototype. Validate results against independent observations.