Skip to main content
U.S. flag

An official website of the United States government

Publication Date
1 June 2023

Machine Learning-based Global Fire Modeling and Control Attributions

Global fire modelling and control attributions based on the ensemble machine learning and satellite observations.
Print / PDF
Powerpoint Slide

We used an ensemble of five machine-learning models and satellite observations to model global fires and identify the driving factors from 2003-2019. Our model accurately predicted burned area, total fire numbers, and fire size while capturing spatial patterns and trends. We found that anthropogenic activity played the most significant role in reducing burned area, followed by climate control, with vegetation having little effect. Climate had a larger impact on burned areas than human or vegetation control, but regional wetting and drying trends weakened this effect. Fire number was more affected by climate, while fire size was more influenced by human activities.


The research provides insights into the factors driving global wildfires, which can inform the development of strategies to manage and control wildfires. The use of machine learning techniques and satellite observations improves our ability to accurately predict the likelihood of wildfires occurring and to identify the most significant factors contributing to wildfire occurrence. This research has the potential to reduce the negative impacts of wildfires on human health, biodiversity, and ecosystems. Additionally, this work contributes to a better understanding of the complex relationship between climate, vegetation, and anthropogenic activities and their impact on global fire dynamics.


Contemporary fire dynamics are complex and poorly understood, as global fire controls related to climate, vegetation, and anthropogenic activity are often interdependent and challenging to disentangle. In this study, we leveraged five machine-learning models and multiple satellite-based observations to conduct global fire modeling for three fire metrics and determined the driving mechanisms underlying annual fire changes for the period 2003-2019. The optimized ensemble machine learning model was found to accurately reproduce the annual dynamics of global burned area, total fire numbers, and averaged fire size. Additionally, the model captured key spatial patterns for different fire metrics. The study highlighted the dominant role of enhanced anthropogenic activity in reducing global burned area, followed by climate control and insignificant positive vegetation control. The results showed that climate dominated a much larger burned area than human or vegetation control, but regional wetting and drying trends weakened the net climate impacts on the global burned area. The study's findings contribute to a better understanding of contemporary fire regimes and can assist in robust fire projections in a changing environment.

Point of Contact
Jiafu Mao
Oak Ridge National Laboratory
Funding Program Area(s)