Can artificial intelligence prevent wildfires?

AI is a fast-developing technology that business leaders are looking to leverage to help solve the world’s biggest challenges. It can process vast amounts of data accurately and efficiently, analyzing the information to identify patterns and make predictions that could be overlooked by human researchers. One of the greatest challenges AI can help to overcome is climate change, but could AI also be the key to preventing wildfires?

Wildfires are a critical challenge for forest management in various ecosystems. It is difficult to anticipate when and where they will appear, and the impact can be devastating — resulting in infrastructure failures, greenhouse gas emissions, the destruction of biodiversity, and loss of human life.

Worryingly, wildfires are becoming more common due to environmental changes and human activity. In 2022, numerous fires broke out in southwest France between July and September and the scale of the fires reached a record level, with more than 72,000 hectares burnt. More recently, we have seen hundreds of devastating wildfires in Greece, Canada, Algeria and Hawaii during 2023.

A major hazard influenced by climate change and human activities

What causes wildfires? In the Sixth Assessment Report on Climate Change, experts from the Intergovernmental Panel on Climate Change (IPCC) found a link between carbon emissions and fire hazards, predicting a 30% to 60% increase in the probability of "catastrophic wildfires" by the end of the century, depending on greenhouse gas emission scenarios. There are also some scenarios where human activity is at play — for example in circumstances pertaining to criminal damage — but even when poor maintenance and inadequate management plans are to blame, the situation is aggravated by environmental factors such as repeated heat waves and drought.

If we look back at France’s incendiary summer of 2022, the regions of Gironde and Landes were threatened by repeated fires of various origins. The risk of fire was very high, driven by drought and a heatwave that produced temperatures exceeding 40°C and broke all-time records in several French towns.

Overall, more than 20,000 hectares of forest were burnt by the wildfires, resulting in the evacuation of more than 8,000 people, the devastation of crops, and the destruction of biodiversity. Sadly, Greece has recently faced a similar disaster, with over 80 fires being recorded since July of 2023. The question is: How can we prevent this from occurring again in the future?

Using satellite data and remote sensing to track fires

Anticipating wildfires requires a detailed understanding of the origins of the fire, with an accurate assessment of the level of damage and its spatial distribution. Having accurate and complete data on fire sites and burnt areas is the key to quantifying trends and modeling occurrences. In turn, this helps improve sustainable forest management by anticipating future occurrences using data-driven predictions.

The data is collected through satellite platforms such as Atos’s data platform named Mundi Web Services, which generates and quantifies earth observation satellite images mainly for the Copernicus program. Satellite imagery is one of the most helpful tools available for monitoring the earth. The images can cover much larger areas of land than images captured by aircraft or drones, as satellites operate hundreds of kilometers higher. The images are captured frequently, providing continuous coverage, and integrating datasets from different sensors provides richer information than analyzing the same data sources individually. Perfect for observing changes to the environment over time, the images can be analyzed for risk indicators (such as stress on the ground caused by farming and vegetation) that, compounded by rising temperatures and greenhouse gas emissions, could potentially lead to the outbreak of a fire.

AI and remote sensing can also use historical data to locate and map the dynamics of a fire, quantify the extent of the damage to the landscape, and replay its evolution while considering the influence of meteorological events. Geospatial data can map the affected areas while remote sensing techniques detect changes caused by the fire, using supervised and unsupervised classification at the level of each image pixel. Sentinel 2 satellite images can enable the detection of burnt areas, while AI and remote sensing can be used to classify the impact as low, moderate, or high severity.