12.07.23 - Forecasting Extreme Heatwaves: A Deep Learning Approach for Early Prediction

Although rare, extreme heatwaves can have significant consequences for society and the environment. For instance, the 2003 heatwave in Western Europe conducted to a death toll of 70’000. As a result of climate change, heat waves are becoming increasingly frequent and severe. Therefore, better understanding and anticipating their occurrence is key for risk assessment and population protection. However, as these events are rare, it is difficult to predict them using statistical analysis of observation data only and physical climate forecast models.
 
A recent study published by George Miloshevich et al. (2023) in Physical Review Fluids has unveiled a methodology using a deep learning approach to build a forecasting model to predict extreme heatwaves events weeks in advance. The model has been trained with 8’000 years of simulated climate data from the University of Hamburg's PlaSim climate model and takes into account several environmental conditions such as soil moisture and the state of the atmosphere. This probabilistic method complements climate forecasting models and has the advantage to provide prediction in a matter of seconds whereas climate models require intensive computation. 

The main challenge of this method is the lack of information, as these events are rare, but this weakness can be mitigated by the development of a new algorithm dedicated to the study of rare extreme events.

#deeplearning  #artificialintelligence  #heatwaves  #climatechange #data