Session: Using Machine Learning to Quantify and Improve Earth System Predictions
Quantifying the drivers and predictability of seasonal changes in African fire
Wednesday, August 4, 2021
Link To Share This Presentation: https://cdmcd.co/QM8kGY
Jiafu Mao, Peter E. Thornton, Stan D. Wullschleger and Xiaoying Shi, Environmental Sciences Division and Climate Change Science Institute, Oak Ridge National Laboratory, Oak Ridge, TN, Yan Yu, Princeton University, Michael Notaro, Center for Climatic Research, University of Wisconsin-Madison, Madison, WI, Forrest Hoffman, Computational Earth Sciences Group, Oak Ridge National Laboratory, Oak Ridge, TN, Yaoping Wang, University of Tennessee
Environmental Sciences Division and Climate Change Science Institute, Oak Ridge National Laboratory Oak Ridge, TN, USA
Background/Question/Methods Africa contains some of the most vulnerable ecosystems to fires. Successful seasonal prediction of fire activity over these fire-prone regions remains a challenge and relies heavily on in-depth understanding of various driving mechanisms underlying fire evolution. Here, we assess the seasonal environmental drivers and predictability of African fire using the analytical framework of Stepwise Generalized Equilibrium Feedback Assessment (SGEFA) and machine learning techniques (MLTs). Results/Conclusions The impacts of sea-surface temperature, soil moisture, and leaf area index are quantified and found to dominate the fire seasonal variability by regulating regional burning condition and fuel supply. Compared with previously-identified atmospheric and socioeconomic predictors, these slowly evolving oceanic and terrestrial predictors are further identified to determine the seasonal predictability of fire activity in Africa. Our combined SGEFA-MLT approach achieves skillful prediction of African fire one month in advance and can be generalized to provide seasonal estimates of regional and global fire risk.