A 3D Convolutional Neural Network for Predicting Wildfire Profiles

Wildfire has become an unavoidable natural disaster that continues to threaten fire-prone communities and the frequency is expected to increase due to climate change. Therefore, predicting a wildfire spread profile is an essential tool for firefighters when planning an evacuation strategy. The current traditional, physics, and empirically based fire spread models require extensive inputs, which are often difficult to obtain. Thus, we propose a 3D Convolutional Neural Network (CNN), named WildfireNet, that can predict the profile of wildfire of the next day when given historical wildfire profiles and accessible remote-sensing data. WildfireNet utilizes 3-dimensional spaces to extract features from both the temporal and spatial dimensions to better understand the relationship between historical fires and upcoming fires. The motivation behind WildfireNet is to locate fires in a precise manner and be able to accurately predict fire profiles. Pixels that were labeled as fire but not on the previous days were extracted to calculate Intersection over Union (IoU) and recall. WildfireNet outperformed 2D CNN and logistic regression model in both IoU and recall.