Project description

In case of natural disasters, modern sensor networks provide high quality data. These measurements, however, are only mapping disjoint values from their respective locations for a limited amount of parameters. Using observations of witnesses represents one approach to enhance measured values from sensors ("humans as sensors"). These observations are increasingly disseminated via social media platforms such as Facebook, Twitter, Youtube and Flickr. Every user of these social networks can be regarded as a mobile, virtual sensor ("social sensor"). These "social sensors" offer several advantages over common sensors, e.g. high mobility, high versatility of captured parameters as well as rapid distribution of information. Moreover, the amount of data offered by social media platforms is quite extensive. On the other hand these data are often subjectively influenced by the observer and of varying quality and quantity.
Methods and techniques, which identify appropriate information, need to be developed to enable the usage of social media data for applications in disaster management. The project's goal is to gather data from social media platforms that can be applied for rapid damage estimation in combination with common sensors.

Natural disasters can have different strength and according effects, depending on geographical circumstances. Thus, the intensity and consequences of disastrous events can be inferred from the spatial distribution of social sensor observations. The accurate estimation of the geographic location of these observations is therefore an important requirement for rapid damage assessment. It is necessary to analyze and evaluate the often vague information concerning its quality in order to be able to locate a high amount of observations. Therefore, a sub-goal of the project is to put the partially implicit spatial references into a local spatial context and thus to locate them automatically. References of this type are given by vague, spatial descriptions like “next to the subway station“ or by spatial concepts like “Big Apple”.

The realization of quality estimations and analyses of social sensor observations for practical problems, represents another sub-goal of the project. The observations are therefore classified via machine learning techniques according to their content. The great amount of observations captured through social sensors can thus be evaluated, distinguished by content, and be used as data resource for rapid damage assessment and forensic disaster analysis.
The corresponding methods and prototypes are being developed in cooperation with the CEDIM projects “Rapid Flood Risk Analysis“ and “Forensic Disaster Loss Analysis“.