Both natural and man-made disastrous events have frequently shocked the world in recent years. In addition to the current sensors that are used for disaster relief such as buoys and street cameras, researchers are seeking the value of using social media users as sensors because of social media's popularity as a platform to share and receive information. Volunteered Geographic Information (VGI) in social media reports following disasters has been proven to be a prompt information source in various studies. However, future research and disaster management could benefit from further understanding of the nature of the data.

In this study, social media feeds in two different types of disasters (both man-made and natural) were analyzed to compare and contrast the patterns in spatio-temporal distribution and user contribution of the responses on social media. Twitter was chosen as the data source because of its real-time nature and the accessibility of its data. Data used in this research was shared by the DOLLY (Digital OnLine Life and You) project at the University of Kentucky, which collects billions of geolocated tweets. The first event considered in this study is the Boston Marathon bombing in April 2013, which serves as an example of a man-made disaster occurring at a definite location at a discrete point in time. The study also includes the Colorado Floods in September 2013, which was a natural disaster with a large geographic footprint that occurred over a longer period of time.

Geolocated tweets centering around the U.S. were extracted using combinations of keywords and processed using MySQL. Visual analytic methods such as sequenced word cloud graphs were used to explore the evolution of themes in disaster-related tweets. In order to explore the spatial and temporal distribution of tweets, spatial analyses including central tendency shift, directional distribution, and kernel density maps were performed.

The results of this study show that the attention from Twitter users about disastrous events does not necessarily correspond to the magnitude or impact area of the disasters. However, the user contribution patterns are similar in both events, and these patterns also resemble the user contribution patterns seen in other user-contributed web content. Additionally, the sequenced word cloud graph can be a useful tool for visualizing thematic changes in event-related tweets and for better understanding how the development of an event is reflected in social media responses. The results provide insights about the nature of VGI in social media following disasters.