Aardvark Weather Blows Away All Weather Forecasts (pun intended)
It applies Occam's Razor - the famous KISS Principle
Where do usual weather predictions come from?
Meteorological stations use a long-established approach called numerical weather prediction. It requires a pipeline like this:
One. Meteorologists gather information from satellites, weather stations, weather balloons, boats, buoys, ships and aircraft. They combine this data with a recent forecast to estimate the current state of the atmosphere.
Two. Then, they produce numerical weather prediction (or NWP) models. They simulate the atmosphere, ocean, land surface and sea ice. Mathematically, they use Navier Stokes and some other differential equations to predict future weather. "Time" is used as a variable. It does not have the same characteristics of *real* time - where it flows in one direction only (Time flies like an arrow, fruit flies like a banana). [These models are not statistical. They have no concept of “error in the model”. Estimated parameters come as god-given numbers.]
Three. The forecasts are processed to ensure that they are ready for use at particular locations. It requires correcting biases and increasing spatial granularity with input from human forecasters.
These three steps require supercomputers, complex software and large support teams.
They can be done in countries with lots of resources, lots of data. They are infeasible for areas with sparse data and even sparser human resources.
Enter Aardvark
Aardvark replaces all of the steps in the weather forecasting pipeline with a Artificial Neural Network (of half a dozen hidden layers) model trained and run on a laptop.
Aardvark’s forecasts are thousands of times faster than all existing traditional forecasting systems. Yes, you read that right. It is cheaper, faster and requires simpler machines.
Aardvark takes in multimodal data from satellites, weather stations and weather balloons, and produces a ten-day global forecast.
In theory it is simple. But, not all data comes with the same time frame - some come every hour, others come every few hours or even once a day.
A new deep learning architecture is necessary to process it. This architecture is trained to forecast weather up to ten days into the future using a large historical dataset.
How good is Aardvark?
Comparing an Aardvark wind speed forecast with the actual weather (ground truth) shows that Aardvark accurately predicted wind speeds of the formation of 2018 tropical cyclone Berguitta outside Madagascar. Four day ahead forecast of Berguitta for Aardvark was better than *any* existing model.
Aardvark is already as accurate as America’s Global Forecast System (GFS), but only using 10 percent of the available data to make its forecasts. Nothing like this has ever been accomplished before.
This new paradigm could easily replace the traditional numerical approach in developing countries. It will empower developing or data-sparse countries to build capacity and create specific weather forecasting systems that previously would have required large teams to operate, deploy and maintain.
What’s next, Aardvark?
As yet, Aardvark has been implemented on short term forecasts. In the current form, it seems to do worse than the GFS for longer horizons.
The system, Aardvark Weather, has been produced by researchers from the Alan Turing Institute, Microsoft Research and the European Centre for Medium Range Weather Forecasts.
Aardvark is flexible. Applications can easily extend to broader Earth system forecasting, including air quality, ocean dynamics, and sea ice prediction.
The results are reported in the journal Nature on 20 March 2025. If you go to the website of Nature, it will ask you to pay 25 pounds.
https://www.nature.com/articles/s41586-025-08897-0
Or you can get it for free here.
Your choice.
Thanks, I'll certainly take a look. I know some unquestionably clever data experts (you know them too) who remain very sceptical of AI-based predictions
I hope , Aardvark helps us to get better weather forecasts than what we get today. But even today countries like Norway produce good forecasts. I wonder if they use Aardvark.