At a Glance
- Researchers in Brazil developed a wildfire detection system using a Convolutional Neural Network (CNN) trained on satellite images. The system achieved 93% accuracy in identifying fire-affected areas.
- Traditional wildfire detection struggles with smaller or remote fires; the CNN model overcomes these challenges by analyzing specific infrared wavelengths that reveal vegetation and temperature changes.
- The CNN was trained on a balanced dataset of wildfire and non-wildfire images and successfully identified 23 out of 24 fire-affected areas and all non-fire areas in testing.
- The model could enhance existing wildfire detection systems, such as MODIS and VIIRS, by providing more localized and detailed fire monitoring.
- With further refinement and expanded datasets, the CNN model could also be used for tracking deforestation and other environmental threats in the Amazon and beyond.
Researchers at the Universidade Federal do Amazonas in Brazil have developed an innovative method to detect wildfires in the Amazon rainforest using advanced artificial intelligence. This new system, which uses a machine learning model called a Convolutional Neural Network (CNN), successfully analyzes satellite images to identify areas affected by wildfires. The CNN model, which was trained on images from the Landsat 8 and 9 satellites, can distinguish between areas with and without fires, showing a remarkable 93% accuracy rate during training.
The Amazon rainforest has experienced a significant increase in wildfires in recent years, with over 98,000 fires reported in 2023 alone. Traditional wildfire monitoring methods struggle to detect smaller or remote fires. The CNN approach overcomes these limitations by using specific wavelengths from satellite imagery, such as near-infrared and shortwave infrared, which are critical for identifying changes in vegetation and surface temperatures caused by fires.
During the study, the CNN was trained on a balanced dataset of images, half showing wildfires and half showing areas without fires. The trained model was then tested using additional images not part of the training data. The CNN correctly identified 23 out of 24 wildfire images and all 16 of the images without fires, demonstrating its robustness and potential for future wildfire detection efforts. This success highlights the system’s ability to generalize findings and accurately detect wildfires in various conditions.
This new technology could complement existing wildfire detection systems, such as MODIS and VIIRS, by providing more detailed and localized data. The researchers suggest that with further improvements, including a larger dataset for training, the CNN model could play a crucial role in detecting wildfires and monitoring other environmental issues, like deforestation, in the Amazon and beyond.
The full study can be read in the International Journal of Remote Sensing.
References
- Taylor & Francis. (2025, March 6). AI has ‘great potential’ for detecting wildfires, study of the Amazon rainforest suggests. Phys.Org; Taylor & Francis. https://phys.org/news/2025-03-ai-great-potential-wildfires-amazon.html
- Eleutério, C. L., Filizola, N. P., De Brito, A. P., Galiceanu, M., & Mendes, C. F. O. (2024). Identifying wildfires with convolutional neural networks and remote sensing: Application to Amazon Rainforest. International Journal of Remote Sensing, 1–24. https://doi.org/10.1080/01431161.2024.2425119
