Article body copy
To track a pandemic’s spread, and to calculate where a disease might pop up next, researchers create models. They scrape and parse all sorts of information: news articles, animal disease reports, flight data, and more. The algorithms that power these models are becoming increasingly sophisticated—to wit, the Canadian company BlueDot made headlines earlier this year when it spotted the emergence of the virus that causes COVID-19 before the World Health Organization declared its spread.
Epidemiological models, however, seem to share one common blind spot: they don’t account for ships.
Ships have been integral to the spread of every pandemic at least since the Black Death ravaged Europe in the mid-14th century. For COVID-19, Spain’s first case came in on the Canary Islands ferry. Ships, from the cruise liners Coral Princess, Grand Princess, and Ruby Princess, to the aircraft carrier USS Theodore Roosevelt, have been crucibles in the spread of this particular disease.
And yet, says Katherine Hoffmann Pham, an information systems doctoral candidate at New York University Stern School of Business who is collaborating with the United Nations Global Pulse initiative, the movements of ships are not being accounted for by those modeling the disease’s spread.
That includes the ships that left Wuhan, China, on the Yangtze and Han rivers—which includes more than 1,100 ships in just one two-week period in January.
In a new paper, recently posted to the preprint server arXiv, Hoffmann Pham and her colleagues show that automatic identification system (AIS) data from ships could be harnessed to improve epidemiological modeling.
AIS is a global tracking program that all passenger ships, international ships over 270 tonnes, and cargo ships over 450 tonnes are legally required to take part in. Over a half million vessels carry onboard transceivers that broadcast messages on the ship’s location, speed, course, destination, and estimated time of arrival, as well as static information like the ship’s name, type, and size.
With so many messages coming at any given time from the hundreds of thousands of ships at sea, scientists could better understand the risk of a disease crisscrossing the planet.
Despite ships’ close association with historical pandemics, Hoffmann Pham says they have been overlooked. “People are less naturally inclined to think that this is a way something like coronavirus could spread,” she says.
That’s largely down to the field’s reliance on aviation data, which dwarfs maritime traffic with nearly 40 million flights in 2019. The stories of cruise ships being floating infection hubs, however, might make using ship data seem less far-fetched, she says.
Another reason ships have been overlooked, says Enmei Tu, a machine-learning researcher at Shanghai Jiao Tong University in China, is that AIS data is not as easy to understand as some other transportation data.
“For flights, we know the destination and the start,” Tu says. “It’s intuitive. AIS is just a sequence of numbers.”
Flight data, he says, has been streamlined after years of research use. The companies providing AIS data, however, all share it slightly differently: one might include all the messages, or just a sampling. How much historical data they include also varies. This makes it hard to compare data from different providers. Tu says that creating an open-source, standardized data set with clear instructions on how to use it would go a long way toward making AIS more accessible.
Elaine Nsoesie, a computational epidemiologist and assistant professor at Boston University School of Public Health in Massachusetts, sees the value of considering ships as a potential way to improve models. “Having this data and incorporating it can give us new insights and observations.”
This is especially true, says Hoffmann Pham, for island nations or busy port cities that have an outsized risk for ship-borne diseases.
Hoffmann Pham says that AIS data alone is unlikely to be enough to predict a disease’s spread. But using it, along with flight data and other information, could help countries be better prepared. The data’s power, she says, would become more apparent if those who create pandemic models actually start considering it.
“Hopefully, it’s at the top of the mind after this,” Hoffmann Pham says. “We are hoping we can enhance these models with data people haven’t looked at before.”