In January 2014, heavy monsoon rains hit Jakarta, Indonesia, flooding the 14 million-strong city. During the deluge, tweets mentioning flooding peaked at almost 900 a minute.
Deltares, a research institute that works on flood risk, and Floodtags, a social enterprise that uses social media to monitor flooding, studied the event to test whether tweets could be used to accurately map flooded areas.
Dirk Eilander, a flood expert from Deltares, presented the work at the European Geosciences Union General Assembly in Vienna, Austria, this month (14 April). He said his research team collected tweets mentioning both the location and water depth, and, after filtering out spam and retweets, were left with about 1,000 usable observations.
Through statistical analysis, the researchers determined the likelihood of any one tweet being accurate. They used the principle that the more independent tweets in a district saying the water depth is at a certain level, the more likely it is that this is true.
From this, the researchers created a flood map showing the water level in different city districts. To validate it, they checked it against a digital elevation map of the city’s low-lying areas as well as photos that had been tweeted but which were excluded from their original sample. They found that two-thirds of the photos agreed with their map.
“This new method will eventually give crisis managers a better view of what is actually happening during a flood so they can make more effective decisions.”
Dirk Eilander, Deltares
Eilander tells SciDev.Net that typically little real-time information is available during floods. Instead, ground observations and satellite images are used to create a picture of the damage in the aftermath.
“This new method will eventually give crisis managers a better view of what is actually happening during a flood so they can make more effective decisions,” he says.
When implemented in an operational flood warning system, the technique could create real-time maps based on tweets sent just a minute earlier, he adds. The researchers hope it will eventually give authorities a better idea of where future floods will strike.
Eilander also says the information his team collected could help validate the computer models that scientists use to predict which areas will be worst hit by floods. Because there is little data on floods, “it’s hard to get an idea of how well your models are actually doing, so this data could be very useful”.
Leonardo Alfonso Segura, a civil engineer at the UNESCO-IHE Institute for Water Education in the Netherlands, says the experiment “provides a lot of new insights”. For instance, he says, because the tweets were written in Indonesian, they were easier to mine as place names were not repeated by tweeters from other regions. This would be more of a problem with tweets in a widely spoken language such as English, he says.
But Alfonso adds that “you cannot validate your model in real time”, indicating that the experiment’s use of tweeted photos to confirm the map may not be reliable.
Etienne Turpin is codirector of PetaJakarta.org, an open source platform that collects and disseminates flooding information. He says the predictive aspect of the project is a “large waste of time, money and energy” because flooding situations are so inherently chaotic. He favours the approach PetaJakarta uses, which is to create a community of users who can share flood information, ahead of tweet mining.
In the Sendai framework for disaster risk reduction 2015-2030 agreed last month, countries called for social media to be used to reduce disaster risks.
In light of this, says John Harding from UNISDR, the UN’s office for disaster risk reduction, all practical examples of social media being used for risk reduction are welcome. “The use of tweets to map the extent of floods in Jakarta is a useful case,” he says.
Dirk Eilander and others Real-time flood extent maps based on social media (European Geosciences Union, April 2015)
Sendai framework for disaster risk reduction 2015-2030 (UN World Conference on Disaster Risk Reduction, 18 March 2015)