Aardvark Weather, an AI weather forecasting system that uses AI and runs on a standard desktop PC, is much faster with thousands of times less computing power than conventional systems



A team from Cambridge University has developed a new system called ' Aardvark Weather ' that can forecast weather conditions tens of times faster and with thousands of times less computing power than conventional weather forecasting systems that combine AI and physical simulations.

Fully AI driven weather prediction system could start revolution in forecasting | University of Cambridge

https://www.cam.ac.uk/research/news/fully-ai-driven-weather-prediction-system-could-start-revolution-in-forecasting

Project Aardvark: reimagining AI weather prediction | The Alan Turing Institute
https://www.turing.ac.uk/blog/project-aardvark-reimagining-ai-weather-prediction

AI can forecast the weather in seconds without needing supercomputers | New Scientist
https://www.newscientist.com/article/2472659-ai-can-forecast-the-weather-in-seconds-without-needing-supercomputers/




Weather forecasts bring great benefits to many parts of society, such as important decisions for agriculture, transportation, and energy, as well as warnings of meteorological disasters such as floods and heat waves. Methods for accurately predicting the weather have been researched for many years, and as of March 2025, the following three-step approach has been established.

Step 1
Data is collected from satellites, weather stations, weather balloons, ships, buoys, aircraft, etc. to estimate the current state of the atmosphere.

Step 2
Computational models are used to predict atmospheric conditions over time to create weather forecasts.

Step 3
They are processed to correct biases, create more detailed data, and incorporate input from human forecasters to produce location-specific forecasts.

To forecast the weather like this, it takes huge supercomputers, complex software, and large teams of people to do the calculations.

Weather forecasting is important to society, so it is natural that AI weather forecasting, which can predict the weather with less computing power, is attracting a lot of attention. So far, Google, Huawei, Microsoft, and others have developed AI weather forecasting.

Google DeepMind announces 'GenCast', an AI model that provides faster and more accurate weather forecasts up to 15 days ahead - GIGAZINE



However, conventional AI weather forecasting only replaces 'step 2' of the three steps with AI, leaving step 1, which requires just as much computing power as step 2, untouched. Therefore, even when AI weather forecasting was introduced, supercomputers and large teams were still required.

This time, the research team at the University of Cambridge developed a system called 'Aardvark Weather' that can process all steps from step 1 to step 3 with AI. The Aardvark Weather model can immediately output global and local forecasts by incorporating data from sensors such as satellites and weather stations, and succeeded in replacing the entire weather forecast pipeline with a single simple machine learning model.

Aardvark Weather is already able to match the accuracy of the US Meteorological Agency's forecasts by using just 10% of the data from existing observation systems. In fact, in the video below, you can see that the Aardvark Weather wind speed forecast shown on the left matches the actual wind speed shown on the right.

Aardvark weather - YouTube


Aardvark Weather also has the advantage of being able to run forecasts on a desktop PC without using a supercomputer, and can quickly create custom forecasts for specific industries or locations, such as temperature forecasts for agriculture in Africa or wind speed forecasts for wind power generation in Europe.

However, while the European Centre for Medium-Range Weather Forecasts uses a 0.3 degree grid for its weather forecasts, Aardvark Weather uses a 1.5 degree grid at the time of writing, which is 'too coarse to capture complex and unexpected weather patterns,' David Schultz of the University of Manchester commented.

'These results are just the beginning of what Aardvark Weather can achieve,' said Anna Allen, lead author of the Aardvark Weather paper. 'Aardvark Weather's end-to-end learning approach can easily be applied to other weather forecasting problems, such as hurricanes and wildfires, as well as broader forecasts of the Earth system, such as atmospheric states and ocean dynamics.'

Related Posts:

in Software,   Science,   Video, Posted by log1d_ts