Jump to content Jump to footer
Install App

Install this app on your device for a better experience.

News detail view

New Evidence Regarding the Reliability of Sea-ice Predictions on Sub-seasonal Time Scales

Sea ice forecast Sea ice models Arctic

The prediction of Arctic sea ice extent and sea ice edge is by now a major challenge but is increasingly in demand in traditional forecasting systems.

Recent findings show that the ability to predict the location of the Arctic sea ice edge varies considerably among sub-seasonal forecasting systems. The best of these forecasting systems outperforms the lead times of climatological forecasts by more than 1.5 months, with the highest accuracy in late summer. This higher accuracy in late summer suggests that useful long‐range forecasts can be provided to stakeholders during a time of the year when marine operations peak. Further advances will be fostered by a more comprehensive representation of the initial states and reduction in model formulation errors.

“The demand for reliable forecasts that describe the evolution of the sea ice from days to months in advance has substantially grown in the last decades,” explains Lorenzo Zampieri –a PhD student at the Alfred Wegener Institute’s Young Investigator Group Seamless Sea Ice Prediction (SSIP). He is the first author of the paper ‘Bright Prospects for Arctic Sea Ice Prediction on Sub-seasonal Time Scales’, the results of which are presented here. “Sea ice forecasts are fundamental tools for managing the risks associated with these activities in the Arctic, which, despite the observed sea ice decline, remains an extreme environment,” Zampieri adds. Climate projections indicate that the retreat of the Arctic sea ice will lead to an ice-free Arctic Ocean in late summer by the second half of this century if anthropogenic emissions of greenhouse gases aren’t reduced. Yet this scenario could also serve the broader economic interests of both private and public actors in the region. Examples include the increasing investigations for new commercial navigation routes, the expansion of fossil fuel and mineral extraction, and the development of tourism.

Operational weather forecast centres around the globe have come to recognise the need for a more accurate representation of the Earth’s weather and climate in their forecast products and are therefore moving towards using fully coupled forecast models. This more advanced model configuration goes beyond those typically employed in classical weather forecasting and is characterized by a global ocean and sea ice model that actively interacts with the atmospheric model. The forecast models are combined with in-situ and satellite observations to obtain a realistic representation of the initial state. These observations are, however, particularly scarce in the polar regions, which correspondingly limits forecasting reliability. Nevertheless, these forecast products increasingly hold the potential to forecast events beyond the classical weather timescale, and to predict the sea ice evolution from weeks to months in advance.

Most of our knowledge regarding the reliability of current Arctic sea ice prediction systems comes from the Sea Ice Outlook — a joint initiative of the international scientific community that has been collecting and evaluating seasonal sea ice forecasts to explore their prediction capabilities since 2008. Until now, Sea Ice Outlook dynamical prediction systems have demonstrated only limited reliability, and their estimates are comparable with those produced by statistical forecasts approaches. However, perfect‐model studies indicate the potential for significant improvements at seasonal time scales, suggesting that dynamical forecast systems can be meaningfully optimised.

The Subseasonal to Seasonal (S2S) Prediction Project is a joint initiative of the World Climate Research Program (WCRP) and the World Weather Research Program (WWRP), the goal of which is to improve both forecasting accuracy and our understanding of forecasts on S2S timescales (i.e., forecasts targeting events from weeks to months in the future; see Figure 1). A database of forecasts is being collected in the context of the S2S Prediction Project, including several forecast systems with a sea ice model coupled to an atmospheric and ocean model and, thereby, producing dynamical sea ice forecasts. “This database represents a unique opportunity for a thorough assessment of state‐of‐the‐art operational predictions of the Arctic sea ice at the sub-seasonal time scale,” says Zampieri. “In addition to quasi-real-time forecasts, reforecasts are available for the past two decades and cover the complete seasonal cycle. This provides the perfect basis for a robust assessment of forecasting accuracy.”

The comparison of the S2S forecasts with satellite observations was carried out by focusing on the position of the ice edge, a variable that provides useful information to potential final users. Specifically, we employed the recently introduced Spatial Probability Score (SPS), which can be considered an extension of the Integrated Ice Edge Error (IIEE) to probabilistic ice edge forecasts. These metrics have been specifically designed to overcome the limitations of the more commonly used difference in either pan‐Arctic sea ice extent or area. The latter only evaluate the total extent of the ice cover, but can’t provide useful information on its spatial distribution. In contrast, the SPS and the IIEE take into account not only differences in the total sea ice extent, but also ice that is forecast at the wrong location.

Results

Our study represents the first overview of state‐of‐the‐art coupled forecast systems’ sub-seasonal ability to predict the sea-ice edge in the Arctic. The results of six different forecast systems were compared (for model identification see Figure 2). The forecast systems contributing to the S2S database show a surprisingly large range of ability. Some of the systems aren’t reliable at all (CMA and Météo France), even at short time scales, while the best system (ECMWF) produces reliable forecasts up to 45 days in advance. The remaining systems (KMA, UKMO, NCEP) deliver reasonable performance, but only up to 10/20 days (see Figure 2). For longer lead times, a sea-ice description based on an observed climatological state provides more accurate results. The fact that accuracy is highest in the late summer suggests that useful long‐range forecasts can be provided to stakeholders during a time of the year when marine operations peak. These results are also reinforced by the 30-day sea ice forecasts for the September sea-ice minimum in 2007 (the first anomalous event that generated substantial attention from the general public). The reader should note the differences in reliability among different forecast systems, with ECMWF being arguably the most reliable (see Fig. 3).

Furthermore, our study reveals that significant sources of error are introduced in the data assimilation stage, which is the procedure that adjusts the initial state used by the forecast system to available observations, before each forecast is made. On the one hand, this suggests that the operational centres could improve their data assimilation approach for sea ice, but also for atmospheric and ocean initial conditions. On the other hand, the results point to the need for a more comprehensive characterisation of sea ice’s initial state that goes beyond the sea-ice concentration alone. New thickness measurements from satellite remote sensing will be fundamental to improving the predictive capabilities of sub-seasonal and seasonal forecast systems in the near future. “Seasonally dependent model biases play a critical role in compromising the accuracy of the forecasts. This calls for dedicated efforts to improve the realism of coupled models in the Arctic, with the ultimate aim of reducing systematic model errors,” says Zampieri, summarising the results of their current analysis.

Literature

Zampieri, L., Goessling, H. F., & Jung, T. (2018). Bright prospects for Arctic sea ice prediction on subseasonal time scales. Geophysical Research Letters, 45, 9731–9738. doi.org/10.1029/2018GL079394

Contact

  • Lorenzo Zampieri (AWI)

Questions?

Contact us via E-Mail or our contact form.