Sentimental wildfire: a social-physics machine learning model for wildfire nowcasting

Abstract The intensity of wildfires and wildfire season length is increasing due to climate change, causing a greater threat the local population. Much this population are increasingly adopting social media, sites like Twitter being used as real-time human-sensor network during natural disasters; detecting, tracking documenting events. concept currently largely omitted by models, representing potential loss information. By including data source in our we aim help disaster managers make more informed, socially driven decisions, detecting monitoring online media sentiment over course event. This paper implements machine learning prediction model, using geophysical sources with Sentiment Analysis predict characteristics high accuracy. We also use wildfire-specific attributes dynamics, has been shown be indicative localised severity. may useful for management teams identifying areas immediate danger. combine satellite from Global Fire Atlas provided Twitter. perform collection subsequent analysis & visualisation, compare regional differences expression. Following this, contrast different models predicting attributes. demonstrate predictor activity, present which accurately model work develops human sensor context wildfires, users’ Tweets noisy subjective sentimental accounts current conditions. contributes development conscious incorporating into modelling.