Flashsourcing or Real-Time Mapping of Earthquake Effects from Instantaneous Analysis of the EMSC Website Traffic

Abstract

Earthquake response efforts are often hampered by the lack of timely and reliable information on the earthquake impact. Rapid detection of damaging events and production of actionable information for emergency response personnel within minutes of their occurrence are essential to mitigate the human impacts from earthquakes. Economically developed countries deploy dense real-time accelerometric networks in regions of high seismic hazard to constrain scenarios from in-situ data. A cheaper alternative, named flashsourcing, is based on implicit data derived from the analysis of the visits by eyewitnesses, the first informed persons, to websites offering real time earthquake information. We demonstrated in 2004 that widely felt earthquakes generate a surge of traffic, known as a flashcrowd, caused by people rushing websites such as the EMSC's to find information about the shaking they have just felt. With detailed traffic analysis and metrics, widely felt earthquakes can be detected within one minute of the earthquake's occurrence. In addition, the geographical area where the earthquake has been felt is automatically mapped within 5 minutes by statistically analysing the IP locations of the eyewitnesses, without using any seismological data. These results have been validated on more than 150 earthquakes by comparing the automatic felt maps with the felt area derived from macroseismic questionnaires. In practice, the felt maps are available before the first location is published by the EMSC. We have also demonstrated the capacity to rapidly detect and map areas of widespread damage by detecting when visitors suddenly end their sessions on the website en masse. This has been successfully applied to time and map the massive power failure which plunged a large part of Chile into darkness in March, 2010. If damage to power and communication lines cannot be discriminated from damage to buildings, the absence of sudden session closures precludes the possibility of heavy damage and provides further constraints to seismic scenarios. To better discriminate in case of large-scale session closure, we propose to analyze cell phone connections, whose networks may remain viable when land lines have gone down. This presentation aims to give an overview of the potential of this technique which is today the fastest way to collect in-situ information on an earthquake's effects. It also provides insightful information about the actual reaction of the population after experiencing an earthquake. There are indications that the relative level of shaking for a given event could be estimated through the ratio of actual visitors to the number of inhabitants. By focusing on the reaction of the population rather than the parameters of the seismic source, it identifies what is perceived as an important event by the public regardless its magnitude. Typical examples are low magnitude earthquakes located close to large cities, like the M2.8 located beneath Sofia, Bulgaria, on Aug. 27, 2010. Based on implicit data, the results cannot be easily altered, which differs from other user-generated content where reliability assessment remains essential. This mechanism is an extension of complementary crowdsourcing techniques such as online questionnaires, applied on flash crowds, leading to its name flashsourcing.