Skill assessment of a set of retrospective decadal climate predictions with EC-Earth

Abstract

The climate system is changing with unprecedented consequences for the environment and many socioeconomic sectors. Hence the importance of predicting these changes. This study aims to produce an evaluation of the predictive skill in a decadal prediction system performed with EC-Earth. It specifically targets three variables of high relevance for human activities, such as sea surface temperature, the sea surface height anomaly (which quantifies sea level rise) and the total cloud cover (which is critical for storm development). The evaluation has mostly focused on two major ocean basins (Pacific and Atlantic), where important modes of variability like the El Ni˜no-Southern Oscillation and the Atlantic Multidecadal Variability take place, and also on the Equatorial stratosphere, where the Quasi-Biennial Oscillation, a highly predictable mode, occurs. Concerning the results, we have shown high prediction skill for all variables in the first forecast year. In the following years, we note a general reduction of the predictive skill, particularly in the southeastern Tropical Pacific, which might point to deficiencies in the model to simulate ENSO periodicity and/or regionality. Furthermore, a general lack of skill in the North Atlantic, may imply that the Atlantic Multidecadal Variability, at least in EC-Earth, is not a source of sea level predictability. Regarding the QBO, results have shown a high prediction skill, especially in the first 29 months. However, the QBO cycle periodicity is not well represented by EC-Earth, which degrades the credibility of the predictions in the subsequent forecast years