Forecast opportunities for European summer climate ensemble predictions using Self-Organising Maps

Abstract

Current state-of-the-art dynamical seasonal ensemble prediction systems (EPS) still show limited predictive skill, particularly over Europe in summer. We propose a neural network-based classification of individual ensemble members before performing a hindcast skill analysis. This classification targets high skill cases emerging from large-scale atmospheric regimes associated with the dominant modes of summertime low-frequency variability in the North Atlantic-European sector (NAE). This classification allows to then select those ensemble members that better predict the phase of the summer North Atlantic Oscillation (SNAO) and East Atlantic Pattern (EAP). Our baseline is a set of teleconnection patterns in NAE identified by Self-Organising Maps (SOM) using ERA-20C reanalysis data. We illustrate our methodology with an example with one set of hindcast ensemble simulations with 30-members generated by the MPI Earth System Model. We achieve better predictive skills at 3-4 months lead time for sea level pressure and geopotential height anomalies at 500 hPa.