Probabilistic Framework for Integrating Multiple Data Sources to Estimate Disaster and Failure Events and Increase Situational Awareness

Abstract

As data for monitoring the natural and built environments become increasingly prevalent, integrating information from varied data sources offers a fuller understanding of the impacts damaging events have on surrounding communities. In this paper, the authors present a probabilistic framework to integrate data from multiple sources to estimate disaster and failure events. The authors show how utilizing data from disparate sources, including physical sensors measuring environmental quantities and big data from social sensors reporting personal accounts and public perceptions within a community, contributes to increasing situational awareness during an event. The approach uses Bayesian updating to infer updated probabilities of event occurrence based on collected data and focuses on data fusion within first individual sensor networks and next across unrelated sensor types. The framework is flexible and applicable to estimate events in a variety of systems and environments using multiple, heterogeneous data sources. The authors apply the approach to estimate flood risks in Louisiana during a 4-d period in August 216 by integrating physical sensor data from the United States Geological Survey and social media data from Twitter. The results show the change in estimated flood risks across the state as additional data is introduced and how multiple data sources increase the amount of updating possible in real-time event estimation.