Geographic and Content-Based Prioritization of Tweets for Emergency Response: The Case of WeatherRelated Disaster Events

Content shared on Twitter usually represents timely, updated, and sometimes on-site reports. Because of this, and the speed and ease with which information is shared through this micro-blogging platform, forecasters and emergency managers have tried to leverage tweets as an on-the-ground reference for both confirming the occurrence of weather-related disaster events and for tracking their local impacts and damage. However, since Twitter data is massive, semantically varied, noisy and spatio-temporally uncertain (Steiger et al. 2015), little of this information is made useful and when it is, it usually requires exhaustive manual filtering - searching individually for tweets related to a specific event. This raises the questions of how can we both identify and prioritize tweets on weather-related disaster events based on their reliability in a timeframe that is useful for emergency response, and ultimately, which factors define their reliability. This study presents an approach for prioritizing tweets so that they can be used by emergency managers and decision makers to assess local effects and damage caused by weather-related disaster events. Using Hurricane Harvey as a case study, we demonstrate how authoritative data, such as on-the-ground reports provided by weather spotters trained by National Weather Service (NWS), as well as the National Flood Hazard1 information and damage maps created by the Federal Emergency Management Agency (FEMA) can be used to prioritize tweets based on their geographical location, and so allow them to assist in weather impact assessments.