Understanding enhanced sub-seasonal predictability of cold spells with nudged experiments.

The weather-dependent planning and decision-making benefit greatly from subseasonal to seasonal (S2S) weather predictions made for up to six weeks ahead. At this timescale anomalies in the extratropical stratospheric circulation, which can last for several weeks in the Northern Hemisphere during winter, are known to affect climate at the surface and can extend the predictability of tropospheric weather conditions. Although state-of-the-art models can to some extent capture the surface impacts of such stratospheric variability and have demonstrated enhanced skill at the subseasonal timescales after major stratospheric events, the causal link between the events and the sources of predicted signals in real-time forecasts is difficult to establish.

We performed relaxation (nudging) experiments to uncover sources of predicted signals in the forecasts, specifically focusing on forecasts of high impact weather events, such as cold spells in the winter season. Nudging of winds and temperatures towards a reference state, usually towards observations is a technique commonly used to constrain model behavior. In addition to this traditional approach where nudging is done towards observations and climatology, we nudge towards control forecast ensemble members predicting anomalous behavior in the stratosphere. By comparing these nudged experiments with the control forecast ensemble, we identify the cause of predicted signal in real-time situations when observations may not yet be available. Therefore, the main benefit is that the methodology can be applied in operational practice as an additional tool to interpret forecast behavior. This approach also provides more accurate quantification of the signal-to-noise ratio because, when nudging is done towards observations, the predicted signal is artificially enhanced by the elimination of model biases. Understanding the conditions associated with enhanced sub-seasonal predictability allows to distinguish between signals associated with remote sources and cases when the signal has no clear origin.