Scaling properties of wildfire frequency and intensity in the western United States

The total annual area burned due to wildfires in the western United States (US) has dramatically increased (~300%) over the past four decades. In the same period, anthropogenic climate change has increased the likelihood of warm temperature extremes, high aridity, and prolonged drought conditions, all of which have been shown to play a key role in increasing fire risk and intensity. While there has been some progress in the literature toward understanding equilibrium climate-fire relationships at large spatiotemporal scales, there is currently no unified framework to model the scale dependence of wildfire activity in a non-stationary climate with dynamic vegetation and human variables. In this talk, I will a) provide a brief overview of our novel machine learning (ML) approach based on hybrid Mixture Density Networks coupled by a Random Forest classifier, b) discuss our results for the scaling properties of wildfire frequency and sizes, c) and highlight ongoing work that seeks to disentangle the role of anthropogenic forcing from natural variability in future climate-fire relationships using a scale aware generative ML fire model and climate simulations.