New Online Tool ‘FireInSite’ Offers Local Wildfire Forecasts Tailored to UK Vegetation

Quick summary

• FireInSite is a free online tool that predicts wildfire spread, flame length and intensity for UK vegetation.

• It combines weather forecasts, terrain and four years of UK fuel measurements, including 5,845 moisture samples.

• Early tests against 22 observed flame lengths showed a correlation of 0.74, but more validation is needed.

• Predictions need checking against local fuel and site conditions, and the paper is not yet peer reviewed.

Firefighters and moorland managers can use FireInSite to explore how fires might behave in different vegetation under forecast weather conditions.

A new research paper describes the science behind the free online tool, which combines UK fuel measurements, weather and terrain to predict fire spread, flame length and intensity.

An initial test against 22 records of observed wildfire flame lengths found encouraging agreement. However, uncertainty in those records and the small number of observations mean that further testing is needed to establish how reliably the system performs across different fuels and conditions.

For land managers, its practical value is the ability to compare vegetation types and fuel loads under the same weather forecast. That can inform wildfire planning, although the paper does not test the effectiveness of particular management treatments. It is a preprint, posted on 22 September 2026, rather than a peer-reviewed journal article.

Forecasting what a fire could do

Fire weather information describes conditions that influence fire danger. FireInSite goes further by estimating how a fire could behave in particular vegetation.

Users select a location on a map. The system retrieves weather forecasts and terrain information, estimates vegetation moisture, and produces predictions for a range of fuel types. Users can then identify which descriptions best match their site.

The outputs include spread rate, flame length and fireline intensity: the rate of heat release along the advancing fire front. Hourly forecasts cover five days. A historical mode allows users to reconstruct modelled conditions for past dates, extending back to 1970.

These are predictions for specified vegetation and conditions, not observations of the ground. The system does not automatically establish which fuel type is present at the selected location.

FireInSite uses established fire behaviour equations, adapted with UK measurements and models that account for seasonal changes in vegetation. Its main focus is surface fires in fuels such as shrubs, grasses and litter.

Built from field samples and laboratory burns

The researchers drew on four years of data collection, including a monitoring network of 43 heathland, bog, acid grassland and coniferous forest locations.

Sampling covered live and dead heather, gorse, bracken and purple moor grass. Regular collections captured seasonal variation, supplemented by measurements during heatwave conditions and the 2025 drought. A citizen science campaign contributed another 713 samples during a dry week in April 2023.

Across the sampling campaigns, 5,845 samples were processed for moisture content. Researchers weighed vegetation before and after drying to determine how much water it held. Laboratory burning tests measured ignition and energy release, helping to describe how flammability changes through the year.

Around 3,600 moisture measurements were used to develop the prediction models. To test performance beyond the locations used for training, the researchers withheld measurements from one location at a time and compared predictions against those withheld observations.

Distinguishing short heather, tall heather and brash

FireInSite separates heather into four height bands: 0–20 cm, 20–35 cm, 35–50 cm and over 50 cm. Each has corresponding estimates of live and dead fuel loads.

This gives managers a way to compare different heather structures under the same conditions. The models apply to vegetation dominated by heather; separate models represent mixed heath, grass, gorse, bracken and bilberry.

There is also a model for heavy heather brash left after cutting, based on depth and density measurements from 47 plots on Broomhead Moor in the Peak District. Its inclusion allows this material to be represented in forecasts. It does not establish the overall effect of cutting on wildfire risk, which this paper did not test.

For grass, the system estimates the proportion that has died back and become available to burn. This is important because green and dead grass behave differently, and their proportions change with the season and drought.

The grass model was developed using satellite vegetation greenness from 47 grass-dominated locations. Its performance was checked against a satellite-derived estimate of dead vegetation, rather than direct field measurements of that proportion.

How well did the predictions perform?

The main fire behaviour check compared predicted flame lengths with 22 observations assembled from photographs, videos and incident records. The correlation was 0.74: larger observed flames generally corresponded with larger predicted flames.

That figure is not a measure of percentage accuracy. Flame lengths were often estimated visually, while the exact vegetation, location and observation time were sometimes uncertain. This comparison also does not establish the accuracy of other outputs, such as fire spread rate.

The live vegetation moisture model performed better overall than a simpler approach using average monthly moisture values for each fuel type. It outperformed that baseline at 14 of the 17 sites with enough observations for comparison.

Performance was uneven. The largest improvements were for gorse canopy and purple moor grass, particularly in July. Heather stem moisture was predicted about as well by monthly averages, while the model performed slightly worse for bracken leaves.

Those differences matter when interpreting the forecasts: a single overall measure of performance cannot describe reliability for every vegetation type.

Using forecasts alongside local knowledge

The paper reports positive feedback from an exercise involving 20 specialist wildfire advisers. Reported uses included planning for increased fire activity, supporting decisions during incidents and training. This indicates practical usefulness, but practitioner feedback is separate from testing forecast accuracy.

Some limits are particularly relevant on moorland. Terrain information represents the surrounding landscape at roughly 180 metres, so small slopes and hollows may be missed.

Moss and the organic soil layer were excluded from the dead fuel moisture model; the forecasts should not be read as predictions of peat moisture or peat combustion.

FireInSite can help managers explore how forecast weather interacts with the vegetation on their ground. Its predictions need to be interpreted alongside checks of fuel type, structure and actual site conditions. Observations from further fires will help establish where the tool performs reliably and where its models need improvement.

Key Takeaway

FireInSite makes vegetation-specific wildfire forecasts accessible to UK land managers, but early validation is limited and predictions need to be checked against local conditions.