Global data-driven prediction of fire activity
Top Cited Papers
Open Access
- 1 April 2025
- journal article
- research article
- Published by Springer Nature in Nature Communications
- Vol. 16 (1) , 1-12
Abstract
Recent advancements in machine learning (ML) have expanded the potential use across scientific applications, including weather and hazard forecasting. The ability of these methods to extract information from diverse and novel data types enables the transition from forecasting fire weather, to predicting actual fire activity. In this study we demonstrate that this shift is feasible also within an operational context. Traditional methods of fire forecasts tend to over predict high fire danger, particularly in fuel limited biomes, often resulting in false alarms. By using data on fuel characteristics, ignitions and observed fire activity, data-driven predictions reduce the false-alarm rate of high-danger forecasts, enhancing their accuracy. This is made possible by high quality global datasets of fuel evolution and fire detection. We find that the quality of input data is more important when improving forecasts than the complexity of the ML architecture. While the focus on ML advancements is often justified, our findings highlight the importance of investing in high-quality data and, where necessary create it through physical models. Neglecting this aspect would undermine the potential gains from ML-based approaches, emphasizing that data quality is essential to achieve meaningful progress in fire activity forecasting.Keywords
Funding Information
- European Commission (service contract no. 942604, ervice contract no. 942604)
This publication has 74 references indexed in Scilit:
- World and Human Action Models towards gameplay ideationNature, 2025
- Probabilistic weather forecasting with machine learningNature, 2024
- Assessing the 2023 Canadian wildfire smoke impact in Northeastern US: Air quality, exposure and environmental justiceScience of The Total Environment, 2024
- Drought triggers and sustains overnight fires in North AmericaNature, 2024
- Human-like systematic generalization through a meta-learning neural networkNature, 2023
- Human-ignited fires result in more extreme fire behavior and ecosystem impactsNature Communications, 2022
- Overwintering fires in boreal forestsNature, 2021
- Human exposure and sensitivity to globally extreme wildfire eventsNature Ecology & Evolution, 2017
- The collection 6 MODIS active fire detection algorithm and fire productsRemote Sensing of Environment, 2016
- Wildfires, complexity, and highly optimized toleranceProceedings of the National Academy of Sciences, 2005