My new Physical Review Fluids article delves into estimating the probability of extreme heatwaves and how we can forecast them using convolutional neural networks. Our methodology demonstrates positive predictive skills for the occurrence of long-lasting heatwaves and relies on the combination of fast and slow drivers. We discuss the challenges of training neural networks in a regime of lack of data and potential solutions, such as rare event simulations and transfer learning

Tweet

Freddy Bouchet presents our work on learning how to predict heatwaves from data at APS DFD 2021: