RAPID ESTIMATE OF GROUND SHAKING INTENSITY BY COMBINING SIMPLE EARTHQUAKE CHARACTERISTICS WITH TWEETS

Abstract

Here we demonstrate a model that combines Tweets following significant earthquakes with basic site and earthquake characteristics to produce a rapid estimate of the ground shaking intensity. We used all geo-tagged Tweets around the world containing the keyword “earthquake ” or “tsunami ” in several languages that occurred in the first 10 minutes following Japanese earthquakes of magnitude 6 or greater from 2011 to 2012. Using different regression models, we found relevant features and the best model by minimizing the mean squared error between model predictions and historical estimates of shaking intensity. We found that the model with lowest error was based on a combination of earthquake- and Tweet-based features, such as local site conditions, source-to-site distance, and number of Tweets within a certain radius. Ground shaking intensity estimates from our model are comparable with historical recordings and conventional estimates provided, for example, by the United States Geological Survey ShakeMaps. Therefore Tweets may be a useful additional data source following significant earthquakes, especially in regions without an extensive network of recording stations.