05.10.23 - Towards Improved Wildfire Management: Insights from Deep Learning Applications

Wildfires are a global problem, with considerable impact on the environment, wildlife and human health.
 
Thanks to the use of satellites, drones and other technologies, wildfires can be monitored. These monitoring systems collect data such as temperature, smoke development and spread patterns to enable a better understanding of fires. See our previous Swiss Territorial Data Lab (STDL) post on this topic.
 
However, monitoring wildfires requires the ability to detect them quickly, even in real time, on satellite images as soon as they become available. This is of crucial importance to draw attention to fires at an early stage and enable rapid reactions.
 
In a recent study of the Université de Moncton, deep learning models were found to be more effective than traditional methods in detecting, mapping and predicting forest fires using satellite data, taking into account factors such as vegetation, topography, weather and historical fire data. Various models, including FireCNN, FCN, Burnt-Net, 2D/3D CNN and FU-NetCastV2, have shown strong performance in tasks such as fire detection and susceptibility assessment.
 
Université de Berne and Berner Fachhochschule BFH have also studied how the severity of wildfires can be classified based on satellite images. Link and image source
 
Much work remains to be done, but it's a safe bet that new data science technologies will improve our detection of forest fires and even their prediction, in order to better protect our natural resources.
 
Switzerland, particularly in Valais and Ticino, also suffers the consequences of these fires. It would be interesting to put these new techniques into practice through a Swiss Territorial Data Lab (STDL) project !