鈥淚t鈥檚 an absolutely unbelievable scientific achievement,鈥� says Andrew Charlton-Perez, talking to me by video from his office at the University of Reading, UK. His colleague, Simon Driscoll at the University of Cambridge, nods enthusiastically. 鈥淭here are so many different applications and so many different uses for it.鈥�
No, they aren鈥檛 referring to quantum computing or nuclear fusion. They are talking about weather prediction. 鈥淧eople just moan about the weather forecast and how bad it is,鈥� says Charlton-Perez. As a meteorology professor, he hears this a lot. But that is because most people don鈥檛 realise that our ability to predict the weather, given the complexity of the atmosphere, is practically a superpower. 鈥淭his is an incredibly complicated system that we don鈥檛 observe very well. And we can put it onto your phone and it鈥檚 pretty accurate most of the time,鈥� he says.
Driscoll, a maths and physics researcher, has spent a lot of time working with Charlton-Perez on the miracle of 鈥減retty accurate鈥� forecasting. They have sliced and diced the many petabytes of weather data accumulated since the 1990s by satellites, weather balloons, ships and ground sensors. Now, they are testing new AI models that could change the way we predict the weather. No, you moaners, it isn鈥檛 going to become perfectly accurate. But it is about to change how you learn if tomorrow will be sunny.
Advertisement
Some big scientific insights of our time came from attempts to predict the weather. Edward Lorenz discovered chaos theory while modelling atmospheric circulation. He knew the way a storm develops is both chaotic and highly dependent on initial conditions. Lorenz fed those initial conditions into an early digital computer, using variables like temperature and wind speed. He found that a tiny shift in one of those variables led to a wildly different prediction of the storm鈥檚 path. He called it 鈥渄eterministic chaos鈥�. In popular parlance, it is known as the butterfly effect.
Every time you get a weather alert on your phone, it is partly thanks to Lorenz and partly thanks to a daily analysis produced by weather centres. For their starting variables, they use meteorological data gathered by thousands of sensors, on Earth and in orbit, and then feed it into a large computer, which spits out pretty accurate forecasts of the sort that tell you there is a 鈥�30 per cent chance of rain鈥�. This is known as numerical weather prediction and it has ruled the roost for decades.
The problem is that it requires expensive supercomputers to ingest huge amounts of current weather data, compare it with past events and subject all of it to the rules of physics to get an idea of what will happen. Global teams have cooperated to produce your rain forecast. Driscoll, for example, has contributed expertise on how ocean ice is affecting the climate. Ultimately what this means is that only a few countries can afford to generate weather reports, leaving most of the world dependent on the generosity of a small number of government agencies.
We could be about to democratise access to weather prediction, which would help smaller countries
All of that could change with new AI models. In a , Charlton-Perez and Driscoll stress-tested four popular AI models to see how well they could predict an unusual stormy event known as a bomb cyclone. They did decently, but 鈥渢he big difference is that it鈥檚 thousands of times faster鈥�, says Charlton-Perez. Plus, 鈥渢he forecasts we used鈥� I ran them on my laptop鈥�.
So AI could potentially allow forecasters to predict weather with fewer resources and smaller teams, meaning less dependence on, say, the US or the European Union for information about the temperature in Barbados. We could be about to democratise access to weather prediction. This would help smaller countries, but would also allow anyone to track niche weather phenomena. If you love rainbows, you could ask an AI model to predict where the next one might appear.
Still, Charlton-Perez warns there may be new roadblocks. The input data required to make a forecast has traditionally been shared freely. But as the cost of analysing it comes down, 鈥渢he data becomes even more king than it was鈥�, he says. He worries that firms behind AI weather models, such as Google, Microsoft and Nvidia, might enter into exclusivity relationships with meteorological services for such data. In other words, much of the globe would be dependent on tech companies for weather reports instead of government bodies.
Worse still, it could cut public access to free forecasts at a time when we need it most. Heat waves are getting deadlier. Storms that were once inconvenient now cause killer floods. This worries Charlton-Perez, who believes meteorological prediction is humanity鈥檚 鈥減rimary climate change adaptation tool鈥�. In an era when extreme weather is on the rise, we need to know what is coming. Having that information may increasingly be the difference between life and death.
Annalee鈥檚 week
What I鈥檓 reading
Historian Josephine Quinn鈥檚 In Search of the Phoenicians, because I want to understand the Punic world.
What I鈥檓 watching
Murderbot !!! Need I say more?
What I鈥檓 working on
Getting ready to visit Knossos 鈥� one of the great Bronze Age city-states 鈥� on the island of Crete in Greece.
Annalee Newitz is a science journalist and author. Their latest book is Stories Are Weapons: Psychological warfare and the American mind. They are the co-host of the Hugo-winning podcast Our Opinions Are Correct. You can follow them @annaleen and their website is techsploitation.com
