Jiafu Mao (12/2/2022): Machine-Learning Applications in Understanding and Prediction of Wildfire

Machine-Learning Applications in Process-understanding and Prediction of Wildfire
Jiafu Mao. Senior Staff Scientist. Oak Ridge National Laboratory.
Wildfires are a major land disturbance and aerosol emission source, affecting the global carbon budget, climate, and socioeconomic development. However, the driving mechanisms underlying fire evolution and reliable prediction of fire activity remain to be explored, especially in the fire prone regions. Here, I will present our recent studies aimed at investigating the wildfire drivers and predictability using machine learning techniques (MLTs), satellite observations and Earth system model (ESM) simulations. We quantified the natural and anthropogenic controlling factors underlying global fire changes for the period 2003-2019 and highlighted the dominant role of enhanced anthropogenic activity in reducing global burned area. We assessed the seasonal environmental drivers and predictability of African fire and achieved skillful prediction of African fire one month in advance. Moreover, we constrained fire carbon emissions simulated by the latest ESMs during the twenty-first century and refined the regional wildfire exposure in different socioeconomic factors. Overall, our research confirmed the feasibility and efficiency of ensemble MLTs in wildfire attribution, modeling and prediction.