Reliability and Usability of Climate Predictions and Projections

Climate predictions and projections from seasons to centuries have the potential to inform societal adaptation and mitigation to climate change – as long as they are robust and reliable. Many studies assess the skill of such predictions using deterministic measures such as the ensemble mean correlation or probabilistic measures such as reliability. However, these commonly used skill scores do not necessarily quantify the potential usability of such predictions from an end-user perspective. This potential usability is strongly linked to the sharpness, which describes the variation in forecast probabilities. By taking both reliability and sharpness into account, the potential economic value of the system can be estimated, based on a cost-loss framework.

In this study the reliability, sharpness and potential economic value is evaluated for current decadal predictions and climate projections. The added skill of initialized decadal predictions is assessed based on the residuals (after removing the uninitialized ensemble mean). Results for surface air temperatures suggest that there is added information carried by decadal predictions with regards to reliability. However, these residuals partly lack sharpness, indicating that the potential utility might not be as high as the correlation/reliability suggests.