Crowdsourcing for Natural Disaster Response: An Evaluation of Crisis Mapping The 2010 Haitian Earthquake

On January 12, 2010, a magnitude 7.0 earthquake struck Haiti, causing catastrophic damages that resulted in at least 300,000 dead, 300,000 serious injuries, and 1.8 million homeless. The destruction was so complete that roads were no longer visible. While buildings, roads, power, and other infrastructure have taken years to restore, mobile phone service was restored almost immediately. A communications network based on mobile phone text messages became an innovative and valuable tool for relief.

Within four hours of the earthquake, a crisis map was established, geocoding messages for inclusion in a freely accessible, online database. Over the next three months, over 3,600 messages would be translated, mapped, and coded with labels indicating the messages’ actionable topics. This undertaking involved over 2,000 online volunteers from around the world. Analyzing and evaluating what happened, what worked, and what went wrong from a programmatic perspective is critical for the future use of crisis maps in disasters and for the future integration of new technologies into large bureaucratic entities.

The purpose of this study was to investigate the diffusion of a novel innovation; analyze aspects of the maps’ deployment that limited success; and posit solutions for improving crisis mapping in natural disasters. The manuscript comprises three papers, beginning with a review of literature and emerging tools for social media and health promotion. The second paper developed an automated algorithm to code the need expressed in texts and compared its reliability to the actual human-derived codes. The findings suggest that automated algorithms can enhance speed of response and overcome human biases. The result is improved situational awareness. Algorithm codes revealed a pattern of message topics, which transitioned from emergency needs, including finding missing persons, to health infrastructure requests, primarily for food and water. The third paper employed a social capital framework to understand the system users’ intents. The findings revealed that individuals far outnumbered aid organizations in users of the system. Also whereas the traditional rapid analysis takes six weeks, the messages revealed real-time needs. These findings suggest that machine coding methodology could increase accuracy of situational analysis and speed response in future disasters.