GISS Lunch Seminar
Speaker: Jatan Buch (Columbia/LDEO)
Title: Towards multiscale fire frequency and burned area prediction in the western United States using machine learning

Abstract:
The annual area burned due to wildfires in the western United States (WUS) increased by more than 300% between 1984 and 2020, resulting in severe environmental and social impacts. However, accounting for the nonlinear, spatially heterogeneous interactions between climate, vegetation, and human predictors driving the trends in fire frequency and sizes at different spatiotemporal scales remains a difficult problem for statistical fire models. Recent advances in machine learning present a promising avenue for addressing this challenge. In this talk, I will introduce a novel stochastic machine learning (SML) framework, SMLFire1.0, to model observed fire frequencies and sizes in 12km × 12km grid cells across the WUS on monthly timescales. This framework is implemented using two Mixture Density Networks trained on a wide suite of input predictors. Our ML model captures the interannual variability and the distinct multidecade increases in wildfire activity for both forested and non-forested ecoregions. I will also present preliminary seasonal and subseasonal-to-seasonal (S2S) forecasts of wildfire frequency and burned area for the western United States at different lead times using SMLFire1.0. In particular, I will discuss results obtained by forcing the SMLFire1.0 model with observed climate and vegetation predictors as well as fire month climate forecasts derived using a probabilistic autoregressive ML model. I will also comment upon the relative contribution of uncertainties, from climate forecasts and fire model simulations respectively, in projections of wildfire activity across several spatial scales and lead times. Altogether our findings serve as a promising use case of ML techniques for wildfire prediction in particular and extreme event modeling more broadly. They also highlight the power of ML driven parameterizations for potential implementation in the fire modules of Dynamic Global Vegetation Models (DGVMs) and Earth System Models (ESMs).