The weather forecast is notoriously unreliable. But with the aid of better data sources and analytics, they have become much more accurate in the short term than they used to be.
Epidemiologists who model the spread of flu face some of the same challenges. And although it may sound silly, disease modelers can look to weather forecasts for ideas — including literally incorporating weather data into their models.
Fariss Samarrai filed this report for UVA Today:
“One reason it is difficult to predict a flu season is because we are using data that spans both space – such as regions – and time periods,” said Srinivasan Venkatramanan, a research scientist at the Biocomplexity Institute. “And what we know, or think we know, is affected by the behavioral changes of people who are responding to an outbreak, such as closing down schools, staying away from work and getting vaccinated. There are so many variables in real time to factor in, so many nuances to interpret, all of them affecting the future, and therefore the outcome. We’re working to sort out how these variables come into play, and learning from our mistakes.”
Flu forecasters are always working behind the curve. Data, which is gathered from a variety of sources – state health departments, the Centers for Disease Control, sales of medications – is usually about two weeks old. So, the very material with which modelers are making their forecasts is based on what’s already happened, not on what is happening at the very moment. If the data shows that people are rushing out to buy cold medicines, or are Googling the symptoms of the flu, that possibly means they already are sick. They might even be recovering.
Venkatramanan said that as modelers take into consideration a multitude of influencing factors that affect their results, they become better at identifying problems in data collection and analysis that have resulted in forecasting failures. But much of it is a matter of human interpretation – part art, part science – just as weather forecasters must interpret the different projections of a hurricane’s path when several computer models point to different routes.
“We’re looking to tie in weather data with search trends to see if we can get better at predicting where the flu will show up next, how many people may be infected, and how serious the infection could be,” Venkatramanan said. “It’s challenging work, but we are getting better at it, the goal being to ever improve our ability to forecast the spread of infectious disease.”
