Examining wildfire dynamics using ECOSTRESS data with machine learning approaches: the case of South‐Eastern Australia's black summer

Abstract

Wildfires are increasing in risk and prevalence. The most destructive wildfires in decades in Australia occurred in 2019–2020. However, there is still a challenge in developing effective models to understand the likelihood of wildfire spread (susceptibility) and pre‐fire vegetation conditions. The recent launch of NASA's ECOSTRESS presents an opportunity to monitor fire dynamics with a high resolution of 70 m by measuring ecosystem stress and drought conditions preceding wildfires. We incorporated ECOSTRESS data, vegetation indices, rainfall, and topographic data as independent variables and fire events as dependent variables into machine learning algorithms applied to the historic Australian wildfires of 2019–2020. With these data, we predicted over 90% of all wildfire occurrences 1 week ahead of these wildfire events. Our models identified vegetation conditions with a 3‐week time lag before wildfire events in the fourth week and predicted the probability of wildfire occurrences in the subsequent week (fifth week). ECOSTRESS water use efficiency (WUE) consistently emerged as the leading factor in all models predicting wildfires. Results suggest that the pre‐fire vegetation was affected by wildfires in areas with WUE above 2 g C kg−1 H₂O at 95% probability level. Additionally, the ECOSTRESS evaporative stress index and topographic slope were identified as significant contributors in predicting wildfire susceptibility. These results indicate a significant potential for ECOSTRESS data to predict and analyze wildfires and emphasize the crucial role of drought conditions in wildfire events, as evident from ECOSTRESS data. Our approaches developed in this study and outcome can help policymakers, fire managers, and city planners assess, manage, prepare, and mitigate wildfires in the future.