GIS, Remote Sensing and Machine Learning Applications for Forest Fire Mapping, Susceptibility Assessment and Risk Analysis: A Systematic Review
This systematic review of 51 studies highlights the growing dominance of machine learning models like Random Forest and XGBoost, combined with remote sensing data such as MODIS and Sentinel-2, in forest fire mapping and risk assessment, while identifying critical geographical gaps in African research and calling for future advancements in data resolution, model validation, and global coverage.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Forest fires are more than just a dramatic spectacle of smoke and flame; they are powerful forces that reshape landscapes, alter the air we breathe, and threaten communities. To manage these risks, scientists need to know exactly where fires have burned, how severe the damage was, and where the next fire might start. For decades, gathering this information required teams of people on the ground, a slow and dangerous process that could not cover vast, remote forests. Today, the view has shifted from the ground to the sky. Satellites orbiting Earth act as constant watchers, capturing images of the planet's surface in visible light and other invisible wavelengths. These images reveal changes in vegetation and heat that signal a fire's presence. By combining these satellite pictures with computer programs that can spot patterns—known as machine learning—researchers can now map fire damage and predict future risks with a speed and scale that was once impossible.
A new systematic review by Lucius Malalu and Henry Chinthuli brings together the latest work in this field, examining how these technologies are being used to understand forest fires. The researchers looked at 51 studies published between 2021 and 2026, searching for the most effective ways to map burned areas and assess fire danger. Their goal was to see which tools scientists are using most often, where the research is happening, and what challenges remain. The review reveals a field that is rapidly evolving, moving away from simple manual checks toward complex, automated systems that can process huge amounts of data to protect forests and people.
The researchers found that the work is heavily concentrated in Asia, which accounted for nearly two-thirds of the studies reviewed. Countries like China, India, and Türkiye led the way, while Africa, despite having significant forest fire risks, was represented by only four studies. This gap suggests that while the technology is advancing, its application is not yet evenly distributed across the globe. In terms of the tools used, the review highlighted that the MODIS satellite sensor was the most popular choice, appearing in 27 of the studies. This sensor is valued for its ability to take frequent pictures of the Earth, allowing scientists to track fires as they happen. Sentinel-2, which provides sharper, more detailed images, was the second most common tool. The choice of sensor often depends on the specific need: MODIS for watching the big picture over time, and Sentinel-2 for seeing the fine details of a specific burned patch.
When it comes to the computer programs that analyze these images, machine learning has become the standard. The most frequently used model was Random Forest, a method that builds many decision trees to make a final prediction, appearing in 28 of the studies. It was followed by XGBoost and Support Vector Machines. These programs are trained to recognize the unique "signatures" of burned land. They look at a wide range of clues, such as the temperature of the ground, how much rain fell recently, the steepness of the terrain, and how close the forest is to roads or towns. One of the most common clues used was a measurement called the Normalized Difference Vegetation Index, which tells the computer how healthy the plants are. By feeding all these factors into the computer, the models can create maps that show not just where a fire burned, but also how likely an area is to burn in the future.
Despite these advances, the review points out that the technology is not perfect. A major hurdle is the resolution of the images. If the satellite picture is too blurry, small fires can be missed entirely, or the edges of a burned area might look fuzzy. Data quality and availability also pose problems, especially in regions where satellite coverage is inconsistent or where clouds hide the ground. The authors note that while the models are getting better, they still struggle to work equally well in every part of the world. A model trained on fires in one country might not work accurately in another because the forests, weather, and human activities are different.
The path forward, according to the review, involves filling the gaps in global coverage, particularly in Africa, and improving the quality of the data. Researchers are encouraged to combine different types of satellite images to get a clearer picture and to use more advanced deep learning techniques that can learn from the images themselves without needing as many manual instructions. The ultimate goal is to create systems that can monitor forests continuously and provide reliable, up-to-date information to help managers make better decisions. By refining these tools, the scientific community hopes to turn the chaotic nature of wildfires into a manageable risk, using the eyes in the sky to protect the forests below.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.