Social media for disaster response during floods

Abstract:

During floods it is difficult to obtain real-time accurate information about the extent and severity of the hazard. This information is very important for disaster risk reduction management and crisis relief organizations. Currently, real-time information is derived from few sources such as field reports, traffic camera’s, satellite images and areal images. However, getting a real-time and accurate picture of the situation on the ground remains difficult.

At the same time, people affected by natural hazards increasingly share their observations and their needs through digital media. Unlike conventional monitoring systems, Twitter data contains a relatively large number of real-time ground truth observations representing both physical hazard characteristics and hazard impacts. In the city of Jakarta, Indonesia, the intensity of unique flood related tweets during a flood event, peaked at almost 900 tweets per minute during floods in early 2015. Flood events around the world in 2014/2015 yielded large numbers of flood related tweets: from Philippines (85.000) to Pakistan (82.000) to South-Korea (50.000) to Detroit (20.000). The challenge here is to filter out useful content from this cloud of data, validate these observations and convert them to readily usable information.

In Jakarta, flood related tweets often contain information about the flood depth. In a pilot we showed that this type of information can be used for real-time mapping of the flood extent by plotting these observations on a Digital Elevation Model. Uncertainties in the observations were taken into account by assigning a probability to each observation indicating its likelihood to be correct based on statistical analysis of the total population of tweets. The resulting flood maps proved to be correct for about 75% of the neighborhoods in Jakarta. Further cross-validation of flood related tweets against (hydro-) meteorological data is to likely improve the skill of the method.