“A peatland is a type of wetland whose soils consist of organic matter from decaying plants, forming layers of peat.”
Peatlands cover a tiny share of the land, but hold more carbon than all the world’s forests.
600+ billion tonnes of carbon in peat soils, more than in all other plants and trees
When peat burns, carbon stored for thousands of years goes back into the air as CO₂.
Scotland holds two-thirds of the UK’s peatland.
- 2 million ha nearly, across the country
- ¾ damaged nearly three-quarters is degraded
- 16% of Scotland’s emissions in 2023 came from peatland
Scottish Government, 2025 · Climate Change Plan, 2026 · NatureScot, 2024
The Scottish Government is restoring peat and controlling fire.
-
Restore
£250m to restore 250,000 ha by 2030
90,000 ha restored since 1990
-
Regulate
Burning on peat over 40 cm deep needs a licence
Muirburn Act, 2024
-
Warn
Wildfire danger assessments flag the areas most at risk
Scottish Fire and Rescue Service
Scottish Government, 2025 · IUCN UK Peatland Programme, 2024 · SFRS
Where do we come in?
PeatPulse ranks peat squares every day, highest fire risk first.
Highest riskLowest risk
Each square is 1 km. Its score uses 16 weather and fire-danger readings.
PeatPulse beat the Fire Weather Index baseline by 20 points.
- 52% PeatPulse: 13 of 25 burn areas found
- 32% Fire Weather Index: 8 of 25
Historical test, 2018–2025. Each method picked five squares a day from the same candidates.
How did we do it?
- Join open data
- Train four algorithms
- Test against the baseline
We joined public datasets by place and date.
- NatureScot peat mapWhere the peat is
- ERA5 weatherRain, heat, humidity, wind
- Copernicus fire dangerFWI and drought codes
- Sentinel-1 radarGround wetness (tested)
- MODIS burned areaWhat burned: the answer
- NASA FIRMS, NatureScotFire records to check it
One row is one peat square on one day.
16 readings weather and fire danger
→Burned within 7 days? yes or no
- 346 peat squares
- 2018–2025
- 535,473 rows
We tested four algorithms. Random Forest was the clear winner.
Burn areas found out of 25, checking one square a day
- Random Forest6
- Gradient boosting3
- Extra Trees2
- Logistic regression2
- Fire Weather Index baseline2
Adding satellite radar never helped.
We tested on fires the model had never seen.
- Squares from the same fire stay together
- FWI ranks the same squares on the same days
- Both get the same number of checks
PeatPulse found more burn areas at every budget.
Accuracy can’t tell them apart
- PeatPulse
- 98.53%
- Fire Weather Index
- 98.46%
- Always “no fire”
- 99.96%, finds none
1 square a day
3 squares a day
5 squares a day
Burn areas found, out of 25
Let us walk you through one day: 14 April 2019.
- 10 days without rain
- 48% humidity, with a 24 km/h wind
- 103 peat squares to rank that day
PeatPulse put this square 1st. The Fire Weather Index put it 59th.
About five days later, satellites mapped burning there.
Everything else is in the report.
- The data
- The method
- How we tested it
- The results
- What this shows, and what it doesn’t
Thank you. Our code is open on GitHub.
- Public data
- No large models
- Random Forest, 4.6 MB
- Runs on a laptop CPU
