Each year, the expert councils of the Russian Science Foundation select top-10 of RSF-funded projects that produced amazing discoveries and brought global attention to crucial scientific issues in the past year. These research works change our understanding of the world and bring us closer to the creation of new technologies—from medicine and the materials of the future to next-generation communications and interfaces that enhance the capabilities of artificial intelligence. RSF’s top-10 is not an award or a ranking.
Neural networks have learned to accurately predict forest fires and take into account the characteristics of the area
Scientists from Skoltech have created a system based on artificial intelligence and satellite data to assess forest health and predict forest fires with an accuracy of up to 87%. Unlike similar systems, the system takes into account a variety of information—from weather to population activity in a specific region—improving the quality of predictions. This development will allow responsible organizations to take proactive measures to protect citizens and forests and reduce the risk of fires.
The researchers used machine learning technology, satellite imagery, and archival fire data over 10 years—a total of approximately 17,000 events from 2012 to 2022. The algorithm analyzed this data and automatically identified patterns. Furthermore, the characteristics of specific study areas, such as vegetation type, terrain characteristics, and meteorological data—temperature, precipitation, wind, and evaporation—were separately considered to assess the likelihood of fire outbreak and spread. The data was compiled into a single database for further processing and forecasting. Particular attention was paid to detailed forest cover characteristics, which can be obtained through satellite monitoring of the area.
The forecast was made five days in advance. In tests, the developers achieved a prediction accuracy of 70% to 87%, depending on the region. This is sufficient for the practical use of the system, allowing responsible authorities to take measures to prevent fires. For example, by watering the forest or closing it to unauthorized access. In addition to regional authorities, the system could be useful to nature reserves, research centers, and other organizations working with forest ecosystems.
.png)
Preprocessed remote sensing data for the study region in the Irkutsk region as of July 11, 2020, with detailed vegetation characteristics. Source: Illarionova, S., Shadrin, D., Gubanov, F. et al.
A