Probabilistic forecasts of extreme heatwaves using convolutional neural networks in a regime of lack of data

Abstract

Forecasting extreme climate events, for instance extreme heat waves, is key for society and a scientific challenge. In this paper we propose a novel machine learning approach that successfully forecasts extreme heat waves up to 45 days before the end of the event. The approach allows for dynamical process studies. A key message is that optimal machine learning forecasts require a large amount of data. The image shows temperature (colors) and geopotential height (lines) anomalies for a typical atmospheric situation.