Advancing Wildfire Predictive Models: A Novel Dataset for Next-Day Wildfire Spread

Wildfire spread prediction is crucial as wildfires destroy millions of acres of land and produce billions of dollars in economic and insured losses annually. Recently, there has been a marked increase in wildfire severity, duration, and frequency due to a multitude of environmental factors. Currently, there are two central issues in the field of wildfire modeling: the lack of high-quality meteorological datasets and the production of inaccurate spread prediction models. In this research effort, we address these concerns by granularizing and aggregating several years' worth of data into a comprehensive dataset for the purpose of wildfire spread prediction. Available data for wildfire spread prediction is standardized over a base interval of one day and an area of one km<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>. We selected feature data scientifically proven to influence wildfire spread from public datasets known for their high quality and extensive coverage, integrating and aligning them to create a robust and comprehensive dataset and pipeline. The created dataset is a tool that can be utilized with a convolutional long-short-term-memory model (CNN-LSTM). This model can thus be used by environmental protection agencies and insurance compa-nies to evaluate and minimize the environmental and economic impacts of wildfires. By improving the accuracy of wildfire spread prediction, our work aims to enhance preparedness and response strategies in wildfire disasters.