AI could force us to find a new go-to topic for small talk by bringing about a future where we already know everything about the weather.
Driving the news: A new study found that Google Deepmind’s AI weather model, called GraphCast, was more accurate than the leading conventional three-to-ten-day forecast system run by the European Centre for Medium-range Weather Forecasts (ECMWF).
GraphCast did better at predicting 90% of the 1,380 observable metrics used (i.e., temperature, humidity) in forecasts.
- In one instance, GraphCast predicted Hurricane Lee would make landfall in Nova Scotia nine days before it happened, three days earlier than ECMWF’s prediction.
How it works: GraphCast uses a neural network trained on over 40 years of data about how weather systems develop and move. Scientists then input current weather data, which the model compares to past weather events, to generate a forecast.
It’s very different from current high-end weather forecasting systems, which use an army of supercomputers to crunch complex equations.
- Meanwhile, GraphCast can make its predictions using a single cloud computer, meaning it could be up to “1,000 times cheaper in terms of energy consumption.”
Why it matters: Studies have found that earlier and more accurate weather predictions reduce climate-induced damages and economic costs. In an era of increased extreme weather events, forecasts that are faster, more precise, and cheaper would be a godsend.
Almost 60% of Canadian small- and medium-sized businesses were directly impacted by extreme weather this year, per a KMPG survey.
- Meanwhile, analysis from last year projected extreme weather like floods and major storms will cost Canada’s economy $139 billion by 2050.
Yes, but: Some experts are wary about AI weather models since they don’t actually understand meteorology; they simply identify patterns. As climate change brings with it more unusual and extreme weather, that could make AI-driven forecasting less reliable.—QH