Getting caught out in the rain might soon be a thing of the past thanks to a powerful new AI weather forecaster.
Google DeepMind has unveiled an AI-powered weather model called GenCast which it claims is faster and more accurate than traditional forecasts.
Compared to the top-performing supercomputer Google’s GenCast model was more accurate across 99.8 per cent of predictions up to 15 days in advance.
According to Google, this will not only help commuters decide whether to bring an umbrella but also spot natural disasters like Typhoons before it is too late.
Normally, weather agencies like the Met Office predict the weather by using huge supercomputers to crunch the complex maths which simulates the climate.
GenCast, on the other hand, uses AI to spot patterns in historical weather data and create 50 possible outcomes which form the basis of an ‘ensemble forecast’.
When the majority of these possibilities show the same weather events occurring, scientists can predict the weather with a high degree of confidence.
Google DeepMind engineer Ilan Price says: ‘Such ensemble forecasts are more useful than relying on a single forecast, as they provide decision makers with a fuller picture of possible weather conditions in the coming days and weeks and how likely each scenario is.’
Google DeepMind has unveiled an AI-powered weather forecast which can predict the weather better than the best supercomputers up to 15 days in advance (stock image)
Across 1,320 tests, the new AI forecast was more accurate than 98 per cent of predictions created by a traditional supercomputer forecast (stock image)
Since weather patterns are so complex, the best weather forecasts are ‘probabilistic’ – meaning that each outcome is assigned a likelihood of occurring.
Although this provides a fuller picture of the weather, it is also extremely demanding in terms of time and computer power.
In a new paper published in Nature, Google DeepMind shows that its new AI model is more accurate than the top-performing ENS model from the European Centre for Medium-Range Weather Forecasts (ECMRWF).
Both GenCast and ENS are probabilistic, but the way they create their set of predictions is totally different.
Rather than trying to simulate the complex physics of the atmosphere, GenCast uses a type of AI called a diffusion model which is typically found in video, image and music generators.
When provided with the most recent state of the weather, the AI generates 50 predictions for the future state of the weather just like some AIs might make images from a text prompt.
The difference is that GenCast has been specifically adapted to work on the Earth’s spherical surface and has been trained on 40 years of weather data.
Google claims that this method is not only faster, but also provides a better forecast for both day-to-day weather and extreme events than ENS.
Traditional weather forecasts rely on massive supercomputers crunching the numbers to simulate how the weather will evolve over time. Instead, the new AI model looks for patterns in past weather data to make a series of predictions about what the weather might look like in the future. Pictured Laura Tobin presents the weather forecast on Good Morning Britain
In a new study, Google DeepMind shows that its new AI model is more accurate than the top-performing ENS model from the European Centre for Medium-Range Weather Forecasts (ECMRWF)
The AI was trained on data leading up to 2018 and then evaluated against the real weather data from 2019 and ENS’s forecasts for that year.
GenCast was more accurate than ENS on 97.2 per cent of predictions and on 99.8 per cent when making predictions more than 36 hours in advance.
Most notably, when both systems were tasked with predicting the arrival of Typhoon Hagibis, GenCast was able to produce a warning 12 hours earlier.
When Typhoon Hagibis hit Japan in 2019, it was the worst storm in 60 years and led to widespread devastation.
The engineers behind GenCast hope that, by giving authorities an earlier warning, AI-powered weather forecasts could help save lives.
Mr Price says: ‘As climate change drives more extreme weather events, accurate and trustworthy forecasts are more essential than ever, yet weather cannot be predicted perfectly and forecasts are especially uncertain beyond a few days.
‘Getting better and more advanced warnings of where they’ll strike land is invaluable.’
Poor predictions based on traditional methods have led to deadly consequences in the past when reports downplayed the dangers of incoming storms.
The AI generates a set of possible outcomes based on the latest weather data which become closer and more accurate as they get closer to the time. This image shows the predicted paths for Typhoon Hagibis (purple) compared to the real path (red)
Typhoon Hagibis was Japan’s deadliest storm in 60 years and caused widespread flooding (pictured). GenCast was able to warn of its arrival with 12 hours of extra lead time, which could have helped coordinate emergency responses in advance
For example, in 1987 BBC weatherman Michael Fish assured viewers that there was no hurricane making its way to the UK.
The next day, devastating hurricane-force winds hit Britain killing 18 people and causing £1 billion ($1.3 billion) of damage.
However, the power of computational weather forecasts has come a long way since the 1980s.
When Michael Fish made his fateful prediction, the Met Office’s supercomputer had the processing power equivalent of an average smartphone today.
Currently, the Met Office has upgraded to the Cray XC40 supercomputing system which is capable of more than 14,000 trillion arithmetic operations per second.
Even the ENS model against which GenCast was measured has improved significantly in the last few years.
In their paper, Google DeepMind used ENS 2019 predictions but ECMRWF has made some major improvements since this time.
In particular, ENS is now capable of producing significantly higher-resolution forecasts than GenCast can produce.
Google says that the world will still need traditional forecasts like those created by the Met Office’s Cray XC40 supercomputer (pictured) but says that AI forecasts will become more useful over time
Dividing the world into a grid, GenCastlooksat squares that are 0.25 degrees across by latitude and longitude.
By contrast, the ENS weather forecast now operates at a resolution of just 0.1 degrees, which means finer predictions.
Google DeepMind admits that traditional models are likely to be irreplaceable for the foreseeable future – not least because they provide the data to train AI.
However, AI forecasts have a major advantage in terms of speed and computer power.
Traditional forecasts such as the ENS take hours on a supercomputer with tens of thousands of processors.
GenCast, on the other hand, takes just eight minutes to produce a 15-day forecast using a single processing unit.
In the future, this means AI models could become far more common for applications like extreme weather prediction or planning around renewable energy sources like solar and wind.
