Data Science problem: Predict the probability of a new forest fire event escaping the initial attack
Portugal is a country that has been severely affected by fires in recent decades, which generate high social, economic and environmental impacts. There are several factors that contribute to this reality: (1) territory with high rural abandonment, which combined with the low profitability of rural spaces, creates a flammable landscape, with high continuity and quantity of fuel; (2) favorable weather conditions, especially during summer, aggravated by climate change; (3) a high number of ignitions promoted by a culture of negligent use of fire.
The fires that generate the largest burned area and negative impacts are usually concentrated in a period of 10-15 days. This fact introduces complexity into fire management, since large fires are often accompanied by simultaneity (i.e. several occurrences occurring at the same time) and high geographic dispersion across the national territory.
In this context, fire suppression management is a complex task that requires knowledge, tools and concerted efforts from all involved. This must take into account the environmental conditions that influence the development of fires, but also the spatial, temporal and capacity heterogeneity of resources available for fighting.
In order to improve the forest fire fighting system, it is important to increase the effectiveness of its response. This improvement may involve the promotion of anticipatory measures, which promote more effective management of resources, maintaining the balance of the combat system (i.e. avoiding its collapse).
The possibility of predicting the type of event, based on the analysis of the history of occurrences, is an important element in supporting resource management decisions.
In this Hackathon, the challenge is to predict the probability of a new forest fire event escaping the initial attack that allows it to be extinguished in 90 minutes, given the terrain and meteorological conditions associated with the occurrence. The aim is also to identify which factors determine a forest fire to escape this initial attack. For this project, historical data on occurrences in Portugal since 2018 is used, including operational data, meteorological data and geographic data.
Hackathon details
- Partner: Autoridade Nacional de Emergência e Proteção Civil - FEPC
- Domain: Wildfire management
- Year: 2023
- Team Members: António Lacerda, Benjamin Hilliger, Christoph Fischer, Inês Silveira, Joana Esteves, Johanna Rauberger, Vasco Florentino
- Faculty advisor: Akli Benali, Rui Figueira
- Partner advisor: Fábio Silva, Alexandre Penha, Carlos Mota
- Github: https://github.com/isa-ulisboa/greends-hack2023-wildfire (available soon)