Predicting the restoration pattern from hurricane-induced power outages from facebook data

Extreme events such as tropical storm, tornado, hurricane cause significant disruptions to infrastructure systems including power, water, transportation, telecommunication services. Faster restoration from power outages is critical since power outages substantially impact various sectors including education, financial transactions, healthcare, and leisure. Thus, it is important to study outage restoration patterns. To develop data-driven models and test its performance on unseen hurricanes, high-resolution data from multiple hurricanes are required. However, such high-resolution power outage data from utility companies are proprietary and not easily accessible to all. Thus, the aim of this study is to demonstrate the use of macroscopic location data available from Facebook for analyzing power outage during hurricanes. First, it shows the association between population activity in Facebook and hurricane-induced power outage using the data for Hurricane Ida at a ZIP Code level. Second, it develops a data-driven model to predict power outage restoration pattern at a ZIP Code level utilizing Facebook data for Hurricanes Ida and Ian. We found that Facebook data can explain 59 % of variance in by power outages at daily level and it can explain 65 % of variance in restoration times from power outages at a ZIP code level. The data-driven model can reliably predict the restoration pattern from power outages (R 2 =0.816). This study can aid researchers to choose alternative data for power outage analysis and help emergency managers and utility companies gain data-driven insights enhancing their decision-making for an impending hurricane.