AI initiative to revolutionise extreme weather forecasting systems

Extreme weather events causing floods, drought, and wildfires are becoming more frequent and more severe with climate change, affecting countries around the globe.

The Horn of Africa is particularly prone to heavy rains which cause devastating localised flooding, loss of livelihoods, and even life.

Now, scientists are using artificial intelligence (AI) to predict extreme weather in the region which could help save lives as climate change becomes more severe.

The United Nations World Food Programme is working with researchers from the University of Oxford University physics department to develop a reliable AI weather forecasting system through machine learning.

"So it's a hybrid approach where we have artificial intelligence to then fill in the gaps where the physical models lack understanding or representation because it's too complex,” says Oxford climate scientist, Shruti Nath.

“They're then able to represent these complicated processes in a data-driven method so that we can best be able to forecast what the true observed reality is."

By using satellite data of clouds banks and the temperatures at the tops of the cloud, the AI makes predictions and tries to identify upcoming extreme weather.

"So for the AI, there is the part where it predicts. But to predict, it needs to first train. And we train on historic observational data. So the input towards our AI device is the weather forecasts,” she says.

It is taking these weather forecasts and training to be able to match them to the observed reality based on the state-of-the-art satellite observations and station data.

Like all AI models, it is constantly improving, the code rewards accurate predictions and penalises inaccurate ones.

"So as the training time progresses, the model will learn more and more, and it's rewarded for giving accurate predictions that match true observed reality. And if it doesn't do that, it's penalized."

In countries like the United Kingdom, supercomputers are used to forecast the weather. But the cost of the computers, the data gathering stations, and radar banks that feed them are high, making them inaccessible for developing nations.

However, the AI forecasting code produced by the University of Oxford can be operated from a laptop.

"So, once it's trained, you can just run it on your laptop to produce 50 predictions of all the possibilities of what the future weather is going to look like," says Nath.

Predicting the weather in the Horn of Africa is notoriously hard because of its changeable nature and a lack of weather stations observing and recording data.

Currently the pilot programme is being operated in Kenya and Ethiopia, but there are plans to roll it out across the wider area.

If it proves successful, it could be used in other parts of the world where extreme weather, driven by climate change, is devastating lives.

With a 48 hour warning from the AI weather forecast, those in danger can be forewarned through text messages, emails, and even radio and television broadcasts.