부산대학교 도서관

Wildfires pose a significant threat with an increased risk of loss of life and property damage in recent years. Traditionally catastrophe modeling has relied on physical models to understand and forecast the behavior of such catastrophic events. In large part this has been due to the lack of a concise dataset that can bring together all the features required for properly modeling such phenomena, and also the required computational strength did not exist. In this paper, we produce a large-scale multivariate dataset to develop deep learning models to understand and forecast the spread of wildfires. We examine features such as topography, climate conditions, and population density which affect the severity of a natural disaster. We discuss challenges in deep learning approaches to next-day wildfires prediction. We expect that this approach can be utilized to produce state of the art deep learning models for other natural catastrophes as well.