Abstract
Recent years have seen a remarkable proliferation of studies attempting to establish relationships between observable online human behaviour and various types of crisis (social, political, economic and natural). Methods utilizing user generated content (UGC) have been already applied to various environmental hazards, such as floods, wildfires, earthquakes, tsunamis and other kinds of emergencies. However, what is currently lacking are more detailed insights into differences between the ways people use social media to report various natural hazard events. In this study we make use of the YFCC100M dataset in order to verify whether statistically robust relationships exist between the volumes of uploaded content during different natural hazards and estimated human and economic losses in the affected countries. Our findings demonstrate that Flickr reflect impacts of events with the highest frequency of occurrence (such as floods or storms) and/or with the recurring spatial structure (such as landslides or earthquakes).