Twitter could provide a quick and affordable way of detecting earthquakes and warning people of coming disaster, according to a study.

Japanese scientists have developed an earthquake-reporting system, Toretter, which, they say, works faster than their country’s official earthquake warnings and could be adapted to warn of other disasters.

Based on data from more than 1,000 earthquakes between 2009 and 2011, they showed that their system detected 93 per cent of Japan’s strong earthquakes by monitoring tweets — although this included a large number of false alarms.

Fine tuning the system allowed it to detect 80 per cent of the strong earthquakes with three-quarters of the alarms being accurate.

The system uses computer-based semantic analysis to classify tweets based on keywords, such as ‘shaking’ or ‘earthquake’, and their context.

It also uses information about when and where each tweet is sent to pin down an earthquake’s time and location, and then emails an alert within two minutes of an earthquake.

Official broadcasts by the Japan Meteorological Agency typically six minutes, according to the paper, published in the April issue of IEEE Transactions on Knowledge and Data Engineering.

“We proved that a social media user could be treated as a sensor and could be applied to earthquake observation,” lead researcher Takeshi Sakaki, from the University of Tokyo tells SciDev.Net.

He adds that tweets could be used in the developing world as the basis of low-cost observation systems for typhoons, tornados, tsunamis and even heavy traffic.

Some earthquake-prone countries in the developing world, such as Indonesia, also have a high density of Twitter users — necessary for the system to work accurately.

Adam Acar, professor of communication and social media at the Kobe City University of Foreign Studies, Japan, says that a Twitter-based system would be an affordable tool in developing countries that lack comprehensive early-warning systems.

“Having an algorithm that constantly checks for sudden increases in the use of certain words — such as quake, shaking, flood, fire — on Twitter is certainly cheaper than hiring many people or building a number of facilities,” he says. “It may not be 100 per cent reliable, but it’s definitely better than nothing.”

However, Acar says Twitter “should be used to supplement existing official information rather than replace it”.

Nicholas Sitar, professor of civil and environmental engineering at the University of California, Berkeley, in the United States, says he doubts the system would benefit developing nations because  a very sophisticated network of sensors and high-speed data processing are needed to ensure the delivery of useful warnings.

“Twitter is just one in a suite of information-delivery systems. The real issue is whether the information can be analysed, interpreted and forwarded sufficiently fast,” he says.

“This is not some kind of a magic bullet that makes up for the lack of safe construction practices in seismically active parts of the developing world,” Sitar concludes.

Twitter could provide a quick and affordable way of detecting earthquakes and warning people of coming disaster, according to a study.

Japanese scientists have developed an earthquake-reporting system, Toretter, which, they say, works faster than their country’s official earthquake warnings and could be adapted to warn of other disasters.

Based on data from more than 1,000 earthquakes between 2009 and 2011, they showed that their system detected 93 per cent of Japan’s strong earthquakes by monitoring tweets — although this included a large number of false alarms.

Fine tuning the system allowed it to detect 80 per cent of the strong earthquakes with three-quarters of the alarms being accurate.

The system uses computer-based semantic analysis to classify tweets based on keywords, such as ‘shaking’ or ‘earthquake’, and their context.

It also uses information about when and where each tweet is sent to pin down an earthquake’s time and location, and then emails an alert within two minutes of an earthquake.

Official broadcasts by the Japan Meteorological Agency typically six minutes, according to the paper, published in the April issue of IEEE Transactions on Knowledge and Data Engineering.

“We proved that a social media user could be treated as a sensor and could be applied to earthquake observation,” lead researcher Takeshi Sakaki, from the University of Tokyo tells SciDev.Net.

He adds that tweets could be used in the developing world as the basis of low-cost observation systems for typhoons, tornados, tsunamis and even heavy traffic.

Some earthquake-prone countries in the developing world, such as Indonesia, also have a high density of Twitter users — necessary for the system to work accurately.

Adam Acar, professor of communication and social media at the Kobe City University of Foreign Studies, Japan, says that a Twitter-based system would be an affordable tool in developing countries that lack comprehensive early-warning systems.

“Having an algorithm that constantly checks for sudden increases in the use of certain words — such as quake, shaking, flood, fire — on Twitter is certainly cheaper than hiring many people or building a number of facilities,” he says. “It may not be 100 per cent reliable, but it’s definitely better than nothing.”

However, Acar says Twitter “should be used to supplement existing official information rather than replace it”.

Nicholas Sitar, professor of civil and environmental engineering at the University of California, Berkeley, in the United States, says he doubts the system would benefit developing nations because  a very sophisticated network of sensors and high-speed data processing are needed to ensure the delivery of useful warnings.

“Twitter is just one in a suite of information-delivery systems. The real issue is whether the information can be analysed, interpreted and forwarded sufficiently fast,” he says.

“This is not some kind of a magic bullet that makes up for the lack of safe construction practices in seismically active parts of the developing world,” Sitar concludes.