Skip to main content
European Commission logo
English English
CORDIS - EU research results
CORDIS

Improving Subseasonal and Seasonal sUmmer forecast over southern Europe through machine Learning

Article Category

Article available in the following languages:

Leveraging machine learning to better predict heatwaves

Researchers used an advanced sub-seasonal to seasonal forecast model based on machine learning algorithms to predict regional monthly mean temperatures over southern Europe with a 1-month lead time.

Climate Change and Environment icon Climate Change and Environment

With much of Europe seeing a record number of days with extreme heat stress, saying that the summer of 2023 was hot would be an understatement. In fact, southern Spain recorded at least 60 days where real-feel temperatures landed between 38 and 46 °C, with some areas seeing the mercury exceed 46 °C on numerous occasions. Because predicting such extreme heat events is particularly challenging, it is difficult for authorities to take mitigating measures to protect people from their potentially deadly effects. “Traditional weather forecasting systems have a predictability limit of about 10 days,” says Marie Drouard, a researcher at the Institute of Geosciences (website in Spanish) in Spain. “But to prepare and better protect society from climate change and the extreme weather events that come with it, we need reliable predictions on longer timescales.” Helping to answer this need is the ISSUL project, which was funded by the Marie Skłodowska-Curie Actions programme.

Improving the use of sub-seasonal to seasonal forecast models

ISSUL focused on developing advanced sub-seasonal to seasonal (S2S) forecast models. “Falling in-between weather and climate forecasting, S2S models are essentially extended weather forecasts that allow us to make predictions beyond two weeks but less than a season,” explains Drouard, who served as the project coordinator. Specifically, the project looked to improve the use of S2S for predicting heatwaves in southern Europe. To do this, the project utilised a fairly new approach in the field of extreme weather prediction: machine learning. “Our S2S forecast model is based on both an optimisation algorithm, which aims to select optimal predictors from a pool of drivers, and a regression algorithm to predict heatwave occurrence,” adds Drouard.

Successfully predicting regional monthly mean temperatures

After some fine-tuning and additional testing, researchers succeeded at using the machine learning model to skilfully predict regional monthly mean temperatures over the Iberian Peninsula with a 1-month lead time. This achievement confirms recent studies on the prediction of extreme heatwaves using shallow machine learning models. Using the same forecast model, researchers also demonstrated that it is more difficult to predict a complex target such as regional monthly heatwave intensity. “Our model was not able to achieve a skilful prediction for the regional heatwave intensity index,” notes Drouard. “Although this could be seen as a negative result, it actually illustrates that extreme events are difficult to predict and confirms that more work needs to be done.” That being said, researchers were able to show that there are recurrent sets of optimal predictors, suggesting that there is an opportunity to further improve the predictability of summer high temperatures and extremes.

Demonstrating the benefits of machine learning forecast models

The ISSUL project is part of a larger effort by the scientific community to improve our ability to predict extreme weather events on long timescales. “The ISSUL project confirmed not only that machine learning forecast models can produce skilful predictions of regional monthly mean summer temperatures, but that such predictions are easier to produce and thus ready for use,” concludes Drouard.

Keywords

ISSUL, machine learning, heatwaves, sub-seasonal to seasonal forecast, extreme weather, weather forecasting, climate change

Discover other articles in the same domain of application