In 2012, 18,000 volunteers classified 1 million cyclone images; their crowdsourced assessments helped scientists study past storms

Somewhere around 2012, thousands of people spent some of their spare time gazing at grainy black-and-white photos taken from orbit of storm clouds, trying to determine if the swirl fell more in line with a comma or a spiral. Unpaid, and without leaderboard clout, somehow the combination of that odd little hobby and the work of thousands of volunteers shaped our understanding of hurricanes in ways that still inform forecasting today.

The initiative was named Cyclone Center, and it wasn't some kind of Silicon Valley moonlighting project; rather, it emerged as a response to an actual problem within science. According to a 2015 article titled “Cyclone Center: Can Citizen Scientists Improve Tropical Cyclone Intensity Records?,” published in the Bulletin of the American Meteorological Society, meteorologists traditionally used the so-called Dvorak technique, which is a decades-old approach that attempts to measure wind speed based on cloud structure in the satellite image of a hurricane. The problem, however, is that the technique is subjective, inconsistent over time, and has been used in various forecasting centers across the world in their own ways.

The problem nobody had the staff to solve


Re-analyzing all the past storms since the advent of satellite monitoring in the late 70s would have required a team of expert meteorologists more than a decade. And so, in 2012, a team led by NOAA's National Climatic Data Center (the agency's name at the time; it wasn't renamed the National Centers for Environmental Information until 2015), in collaboration with the University of North Carolina Asheville and the Citizen Science Alliance (which runs the popular Zooniverse online crowdsourcing platform), took the task on as a way to both utilize the internet's free time and to complete the boring part that would have otherwise kept us busy for years.

Online, volunteers on cyclonecenter.org were asked to help out by looking at an infrared satellite photo and running through a few basic guided questions about whether or not the storm has an identifiable eye, if the shape was more curved or comma-like, and how tight the swirl was, to create a basic set of classifications similar to what professional meteorologists do. Each image was reviewed by many different volunteers, accounts from the project describe up to several dozen per image, so no single person's judgment could skew the result.

Did it actually work?

Somehow, yes. By the time the researchers went back to check in on the results, there were over 440,000 classifications entered by volunteers, and a follow-up 2016 study titled “Identification of Tropical Cyclone Storm Types Using Crowdsourcing,” by Knapp and colleagues, published in the Monthly Weather Review, found that when compared to algorithms that automatically classify hurricanes, the crowdsourced estimates were often as good, or better at certain points, such as when a tropical storm was shifting into hurricane mode. Not bad at all for a bunch of amateurs

Why should this matter

Being able to accurately classify past hurricanes is important in its own right, but knowing what impact these storms have on climate warming trends, and how they're a part of climate change, is crucial to our understanding of whether the global warming problem is really as serious as we've always been told. NOAA's own summary of their citizen science program notes that the project ultimately lasted for 7 years on Zooniverse, and by the time it wrapped up, roughly 18,000 volunteers had logged over a million individual image classifications, giving researchers a far more consistent baseline for studying long-term shifts.

Also an interesting thing to note is that an entire field of research depended on unpaid volunteers doing the busywork that the budget-conscious government didn't want to pay scientists to do in the first place. Citizen science is a genuinely valuable model for tackling large-scale, labor-intensive datasets that agencies can't fully staff, though it also raises fair questions about how research work gets funded more broadly. So, the next time you see a hurricane statistic from 1985, there's a good chance that somewhere in that number is a contribution from your random neighbor, checking her email during lunch shift.