Detection and attribution of changes in global wildfire activity to anthropogenic drivers using machine learning

Exceptional wildfire activity in recent years highlighted the catastrophic impact these extreme events have on communities, ecosystems, and economies. Despite the growing concerns over wildfire activity under continued climate warming, there is little scientific evidence causally linking observed wildfire changes to anthropogenic forcings, climatic or non-climatic. In addition, future projections of burned area are marked by large uncertainties. This project will combine machine learning with detection and attribution techniques to uncover the anthropogenic imprint on past and future global wildfire activity. As attribution studies on wildfire activity are hampered by the absence of a long-term burned area record, I will first develop a global burned area reconstruction by applying deep learning to satellite observations, climate reanalyses and socioeconomic datasets. Subsequently, trend detection and attribution will be performed via optimal fingerprinting on new simulations from global wildfire models. Third, pattern recognition and attribution methods will be fused to link spatial wildfire patterns to climate or management changes. Finally, I will deploy machine learning architectures on a multi-model ensemble of global wildfire simulations to constrain regional burned area projections across the globe. The results of this research will deepen scientific understanding of the human imprint on wildfire and will inform climate change mitigation and adaptation strategies.