Emergency flood detection using multiple information sources: Integrated analysis of natural hazard monitoring and social media data

Abstract

Extreme weather events are occurring more frequently as a result of climate change. In October 2019, eastern Japan was hit by Hagibis, a large and high-speed typhoon. This unprecedented typhoon caused the evacuation of over 4000 people, injured more than 300 people, and damaged more than 98,000 dwellings throughout the affected area. Because floods are one of the most devastating natural disasters in Asia, providing an effective early warning system (EWS) is critical to reducing disaster impacts. However, warnings based only on natural hazard monitoring do not offer sufficient protection. Integrating natural hazard monitoring and social media data could improve warning systems to enhance the awareness of disaster managers and citizens about emergency events. We analyzed time-series data including rainfall intensity, 90-min-effective rainfall, and river water level as well as Twitter data related to disaster events during the 5-day period from 11 to 15 October, focusing on the most affected areas in Japan. The analysis included more than 60,000 tweets. Our analysis confirmed the utility of the statistical approach of outbreak detection with social media data in the early detection and local identification of multiple-flood events, and the results from the municipality-level analyses show that tweet frequencies related to the flood disaster ontological categories were significantly correlated to temporal variations in the hazard monitoring data. Thus, flood detection at the administrative level using social media data combined with current hazard monitoring data can enable a decision-driven EWS design. Interactive approaches for decision-making and knowledge production should continue to be considered in the face of climate-change-induced disasters.