Multi-model seasonal forecast of Arctic sea-ice: forecast uncertainty at pan-Arctic and regional scales

Dynamical model forecasts in the Sea Ice Outlook (SIO) of September Arctic sea-ice extent over the last decade have shown lower skill than that found in both idealized model experiments and hindcasts of previous decades. Additionally, it is unclear how different model physics, initial conditions or forecast post-processing (bias correction) techniques contribute to SIO forecast uncertainty. In this work, we have produced a seasonal forecast of 2015 Arctic summer sea ice using SIO dynamical models initialized with identical sea-ice thickness in the central Arctic. Our goals are to calculate the relative contribution of model uncertainty and irreducible error growth to forecast uncertainty and assess the importance of post-processing, and to contrast pan-Arctic forecast uncertainty with regional forecast uncertainty. We find that prior to forecast post-processing, model uncertainty is the main contributor to forecast uncertainty, whereas after forecast post-processing forecast uncertainty is reduced overall, model uncertainty is reduced by an order of magnitude, and irreducible error growth becomes the main contributor to forecast uncertainty. While all models generally agree in their post-processed forecasts of September sea-ice volume and extent, this is not the case for sea-ice concentration. Additionally, forecast uncertainty of sea-ice thickness grows at a much higher rate along Arctic coastlines relative to the central Arctic ocean. Potential ways of offering spatial forecast information based on the timescale over which the forecast signal beats the noise are also explored.

过去十年中，海冰展望（SIO）对9月北极海冰范围的动力模型预测显示，其技巧低于理想化模型实验以及对前几十年的回测结果。此外，尚不清楚不同的模型物理过程、初始条件或预测后处理（偏差校正）技术对SIO预测不确定性的贡献如何。在这项工作中，我们使用在北极中部具有相同海冰厚度初始化的SIO动力模型，对2015年北极夏季海冰进行了季节性预测。我们的目标是计算模型不确定性和不可约误差增长对预测不确定性的相对贡献，并评估后处理的重要性，以及对比全北极预测不确定性和区域预测不确定性。我们发现，在预测后处理之前，模型不确定性是预测不确定性的主要贡献因素，而在预测后处理之后，预测不确定性总体上降低，模型不确定性降低了一个数量级，不可约误差增长成为预测不确定性的主要贡献因素。虽然所有模型在其对9月海冰体积和范围的后处理预测中总体上一致，但海冰浓度情况并非如此。此外，相对于北冰洋中部，沿北极海岸线海冰厚度的预测不确定性增长速度要快得多。还探讨了根据预测信号优于噪声的时间尺度提供空间预测信息的可能方法。