Appraising Situational Awareness in Social Media Data for Wildfire Response

Situational Awareness (SA) is the perception of the environment and events in time and space. It comprises three stages: perception, comprehension, and projection. In rapidly evolving situations with limited resources, disaster agencies need fast and reliable information to understand SA and support decisions. This paper proposes a social media-based approach to investigate SA using data from the 2020 western US wildfire season. The study uses the Twitter Search API with a key search term “wildfire” to scrape related tweets from September 2 to October 4, 2020. Subsequently, it applies Name Entity Recognition (NER) to identify geoinformation. Next, it optimizes and applies transformer-based topic modeling (BERTopic) to classify topics from the tweets. Evolution of the topics over time can help identify phases of wildfire propagation. Using the identified time and space information, the study investigates SA stages across geographic regions and validates the results against wildfire response acts by responders. The results suggest that the proposed approach can provide consistent and comparable real-time estimation of wildfire situations. The study suggests guidelines for enacting timely and appropriate decisions and prioritizing operational needs.