资源环境科技发展态势分析平台

A significant barrier to understanding and quantifying current skill of Arctic sea ice forecasts is a lack of a central database to enable model evaluation and intercomparison. This study addresses this issue by introducing a central server and web portal housing multimodel ensemble forecasts. We present an overview of the portal and provide an analysis of 2018 forecast skill. Among the 16 participating models, forecasts of sea ice concentration varied widely; yet the multimodel mean generally offered skillful forecasts for up to 5 months. Models that assimilated observed concentrations with more advanced methods performed better on average than other models. Similarly, one model that incorporated satellite-based sea ice thickness thereafter compared most favorably with thickness measured along IceBridge flight tracks. These results highlight the benefits from multimodel predictions and assimilating sea ice variables and the insights gained from near-real-time evaluation of operational forecasts.

Plain Language Summary Prediction of regional Arctic sea ice on subseasonal-to-seasonal time scales represents a significant challenge to dynamical and statistical methods. Yet accurate forecasts of regional sea ice conditions are critical for local community supply logistics, shipping, fishing, and feedbacks to weather and ocean. A significant barrier to understanding current model skill and targeting improvements is a lack of a central database to enable intermodel comparisons and evaluations and drive model innovations. This study develops such a database and provides analysis of model skill during the 2018 melt season. We find large intermodel differences in sea ice presence and thickness forecasts. Forecast skill was improved by averaging across multimodel ensembles and through assimilation of sea ice observations. As this data set continues to grow, we also envision it being used to evaluate changes in model forecast configurations, providing immediate feedback to model developers and end users.