| dc.description.abstract | Many cities around the world are exposed to extreme flooding events. As a result of rapid population growth and urbanization, cities are also likely to become more vulnerable in the future and subsequently, more disruptions would occur in the face of flooding. Resilience, an ability of strong resistance to and quick recovery from emergencies, has been an emerging and important goal of cities. Uncovering mechanisms of flooding emergencies and developing effective tools to sense, communicate, predict and respond to emergencies is critical to enhancing the resilience of cities. To overcome this challenge, existing studies have attempted to conduct post-disaster surveys, adopt remote sensing technologies, and process news articles in the aftermath of disasters. Despite valuable insights obtained in previous literature, technologies for real-time and predictive situational awareness are still missing. This limitation is mainly due to two barriers. First, existing studies only use conventional data sources, which often suppress the temporal resolution of situational information. Second, models and theories that can capture the real-time situation is limited.
To bridge these gaps, I employ human digital trace data from multiple data sources such as Twitter, Nextdoor, and INTRIX. My study focuses on developing models and theories to expand the capacity of cities in real-time and predictive situational awareness using digital trace data. In the first study, I developed a graph-based method to create networks of information, extract critical messages, and map the evolution of infrastructure disruptions in flooding events from Twitter. My second study proposed and tested an online network reticulation theory to understand how humans communicate and spread situational information on social media in response to service disruptions. The third study proposed and tested a network percolation-based contagion model to understand how floodwaters spread over urban road networks and the extent to which we can predict the flooding in the next few hours. In the last study, I developed an adaptable reinforcement learning model to leverage human trace data from normal situations and simulate traffic conditions during the flooding. All proposed methods and theories have significant implications and applications in improving the real-time and predictive situational awareness in flooding emergencies. | en |
