Assessing the Impact of Initialization on Decadal Prediction Skill

Abstract

Abstract An alternative approach to evaluating the correlation skill of near‐term climate predictions is proposed. The approach separately quantifies the skill of the initialized and uninitialized components of the forecast and their contributions to overall correlation skill. The initialized component consists of the predictable part of the internally generated natural variability and that part of the externally forced component affected by initialization. The methodology is applied to results from the latest Canadian Centre for Climate Modelling and Analysis' decadal prediction system. The initialized component of annual mean temperature forecasts exhibits regional skill over ocean and land. The contribution to overall skill is modest at longer ranges and for multiyear temperature averages, partly because of the strong temperature response to external forcing. Larger ensembles increase the percentage of global area with predictable variance due to initialization and the contribution to overall skill from the initialized component of the forecast.

Keywords

• decadal climate predictions
• impact of initialization on prediction skill
• CanESM5 decadal hindcasts
• decadal forecast decomposition
• correlation skill decomposition
• large ensemble predictions