“Weather Rescue” sounds like it could be a TV show about the adventures of an emergency response team. But the Weather Rescue project led by University of Reading researcher Ed Hawkins is actually focused on that need rescuing.
The UK Met Office has an incredible trove of historical weather data in its archives that is trapped on paper.
After a storm sunk 200 ships off the coast of Wales (including the Royal Charter and its crew of 450), FitzRoy set about creating a network of UK weather stations that could telegraph daily observations to him in London. In February 1861, he put out the first forecast storm warning. After some of the fishermen who ignored this new-fangled sorcery sank in the storm, the forecasts encountered an increasingly attentive audience.
The Weather Rescue project uses volunteers—a group you could join by visiting the website—to read the scanned paper records of the daily measurements from the network Fitzroy created, which span a century. The data have scientific value beyond historical curiosity.
It’s a matter of pressure
“Up until recently [in climate science], our ‘exam questions’, if you like, have been: Is the climate changing and is it our fault?” Hawkins told Ars. “Those are comprehensively answered, and so our new exam questions are: How is the weather changing in response to our warming climate? Are we going to get more or fewer storms to hit the UK, is the storm track going to move north or south, are they going to become more intense or less intense?”
Longer records of stormy weather are enormously useful for answering these questions. That means accessing historical measurements of air pressure. Pressure data reveal the meteorology of storms passing through more even more usefully than temperature or rainfall. And if rare events are the most interesting, you really need as much data as possible in order to capture them.
That data isn’t just useful for scientists. “October 1903 is still the wettest month in UK history, but if that happened now, it would be even wetter, because we’ve got a warmer atmosphere with more moisture,” Hawkins said. The context of a weather event has to be understood in order to build to withstand future flooding.
The UK isn’t the only country with vast stores of inaccessible data. In the US, for example, NOAA is working on reconstructing weather across the 19th and 20th centuries based on air pressure data, but is similarly limited by the number of older records that have been digitized.
“To me it’s a no-brainer that this data is just gold that we’re currently missing out on, and I think it will help us answer some really interesting and important questions about how our weather is changing,” Hawkins said.
A little help, please?
There is so much data in so many places that ultimately the digitizing is going to have to be automated. Machine-reading handwritten data can’t be done very effectively yet, however. “We would love to get a really big tech company involved in this,” Hawkins hinted. Machine learning would probably be up to the task, but no one with the algorithmic firepower has stepped up to try.
“If someone like Amazon, or Google, or Microsoft, or someone like that—or Apple—wanted to come along and do something really beneficial for climate science, then I would suggest that a bit of time spent helping us read these typeset and handwritten pages would be of enormous benefit to climate science, in general,” he added.
About 5,000 volunteers have participated in several versions of this project over the last few years, putting in the total equivalent of at least 10 years of full-time work, digitizing more than a million weather observations. Not only is that a lot of “rescued” data with its own value, but it could provide the training sets needed to get future machine learning projects off the ground.
Currently, the project is working through the earliest observations after having been promoted as a citizen science project for the annual British Science Week. Previous runs digitized the first decade of the 20th Century and ship-based data from World War I.
The project website will hand volunteers one page at a time, asking them to read the temperature, rainfall, and air pressure for a given location. Each datapoint gets read by seven volunteers, ensuring that any mistakes can be filtered out. It’s a straight-forward task that anyone can quickly get rolling on.
Luckily, FitzRoy and his successors had pretty neat handwriting, so the data that once led to life-saving forecasts can easily contribute to science once again.