Wildfire Dynamics with ECOSTRESS, Sentinel, and MODIS: The Case of Australia’s Black Summer

Abstract:

In the 2019–2020 Australian summer season, also widely known as the Black Summer, the country was devastated by the worst wildfires observed in decades. New South Wales experienced the longest continuously burning in the history of Australia’s bushfire. It consumed more than 4 million hectares. New South Wales also has the highest number of homes lost followed by Victoria. The overarching goal of this study is to understand wildfire dynamics in southeast Australia using machine learning algorithms on satellite data. To achieve the goal, we (1) establish a geospatial database, including MODIS MCD64A1 fire product, digital elevation model (DEM), slope, aspect, ECOSTRESS data (i.e., evapotranspiration (ET), evaporative stress index (ESI), land surface temperature (LST), water use efficiency (WUE), NDVI generated from Sentinel-2 data, and rainfall data, (2) quantify wildfire probability using geographically weighted regression, logistic regression, and random forest algorithms, (3) determine the importance of explanatory variables, and (4) evaluate susceptibility of cities using wildfire probability values within a 5-km buffer around them. NDVI, ESI, WUE, and rainfall variables were found to be the high impact factors in predicting and understanding wildfire dynamics with a lead time of two weeks. The logistic regression model using ECOSTRESS data can be employed to effectively predict wildfire probability anywhere in the world at any time step. This can be achieved without a prior knowledge of machine learning algorithms and unavailability of other explanatory variables. The analytical methods, data processing procedures, regression models, and susceptibility mapping approach for cities can be used to help policy makers, fire managers, forest rangers, and city planners to assess, manage, prepare, and mitigate wildfires in the future.