Sea-ice forecasts in the Arctic: what conditions are likely at the beginning of the MOSAiC experiment?
During the MOSAiC expedition, in October 2019 the research icebreaker RV Polarstern will be moored to an ice floe in the northern Laptev Sea, a region where the ice conditions are largely unknown. Sea-ice forecasts based on numerical models represent an important tool for estimating the conditions in advance. The forecast for sea-ice conditions in October 2019 presented below is an essential component of planning the course for MOSAiC.
In the context of an international project on sea-ice forecasts, the Sea Ice Outlook (SIO), since 2008 the AWI has taken part in a scientific competition to provide the most accurate seasonal forecast for the minimum sea-ice extent in September. The SIO has been coordinated by the ‘Sea Ice Prediction Network project (SIPN)’ since 2014, and has been part of ‘Sea Ice Prediction Network–Phase 2 (SIPN2)’ since 2018. You can find descriptions of the two forecasting methods used at the AWI (dynamic and statistical) here.
Every year, the SIO prepares reports in June, July and August on the forecasts submitted for the September sea-ice minimum, which are supplemented by the latest analyses of the sea-ice conditions. Post-season reports provide detailed analyses of key factors influencing the sea ice, while also assessing the scientific methods that various international groups use for their seasonal forecasts, which range from heuristic to statistical and dynamic. The method described here pursues a dynamic approach, based on a numerical model of the sea ice and Arctic Ocean.
The system used for the pre-MOSAiC forecasts is similar to that used for predicting the September minimum, but there are several important differences. Both systems start with an analysis of the sea-ice conditions in March and April, which is produced by ‘fusing’ observational data and model simulations (‘data assimilation’). The data used includes satellite-based observations of daily ice concentrations and ocean surface temperature (both from the Ocean and Sea Ice Satellite Application Facility (OSI SAF)); of daily snow thickness (from the University of Bremen); and, most importantly, of the monthly mean ice thickness (CS2 / CS2SMOS from the AWI). The model is then integrated with atmospheric forcing data from a weather model (NCEP-CFSv2) up to the starting time for the forecast (for the MOSAiC expedition: 11 August 2019) and subsequently calculated through the end of December with forcing data from the past ten years (for the MOSAiC expedition: from 12 August to the end of the year for the years 2009 to 2018). This yields an ensemble of ten scenarios, which describe potential developments of the sea-ice conditions.
The system used for the MOSAiC forecasts differs from the September minimum system in that a) no bias correction is used, and b) a higher resolution is used. The parameters in the numerical model for the previous system were not optimally calibrated, as a result of which, in comparison to actual observations, too much sea ice melted in the summer. When the actually observed ice thicknesses in March and April were assimilated, the model predicted too little ice cover in the summer. A bias correction of the observed ice thicknesses compensates for this effect. Calibrating numerical models is a lengthy and complicated process. However, the numerical model used at the AWI was substantially improved with the help of a genetic algorithm (Sumata et al. 2019 a and b), which made a bias correction superfluous. Consequently, the system can now not only be used for September forecasts, but also for other times. Since the optimised parameters are largely scale-independent, the model’s resolution was also improved, allowing it to depict the Arctic at a grid scale of 25 km. The assimilation system remained unaffected by these changes.
In addition, in collaboration with colleagues from the AWI’s Sea Ice Physics working group, who provided field observations, suitable metrics for the pre-MOSAiC forecast were established. These metrics define regions with a) more than 70 % ice cover and b) an ice thickness of more than 1 m in the ice-covered cells. Figure 1 shows the likelihood (0-1) of encountering an ice concentration of 70 % or higher on 1, 15 and 31 October 2019, while Figure 2 shows the likelihood of encountering an ice thickness of more than 1 m on the same dates.
Needless to say, a forecast isn’t particularly useful without being validated by observational data. The system can be validated by retroactively simulating previous years. Principally speaking, this can be done starting in 2011, since it was the first year with ice-thickness observations. To date, however, only 2018 has been retroactively simulated. Figure 3 shows the likelihood of encountering 70 % or higher ice concentration on the same dates as in Figure 1 for the 2018 forecast (though the forecast was actually started somewhat earlier at 1 August 2018), together with the actually observed 70 % ice concentration isoline. In a perfect forecast, the likelihood 1 would be within the isoline, and 0 would be outside it. Here we can see high accuracy for 1 October and somewhat poorer accuracy levels for 15 and 31 October, which is due to the fact that the ice cover increased between 1 and 31 October, which the model can’t perfectly simulate.
A validation of the 1 m ice-thickness metric is not possible for individual days, but only for monthly data, since only monthly ice-thickness data is available. Accordingly, Figure 4 shows the ice-thickness forecast for the monthly mean, together with the monthly forecast for 2018. Here, too, we can see an high level of accuracy.
In summary, it can be said that, with the help of the dynamic forecasting system at the AWI, the ice conditions (concentration and thickness) for October 2019 in the starting region for the MOSAiC expedition can be predicted with a high level of accuracy. This allows us to provide initial guidance on the expedition route, before radar satellite images and, later, helicopter flights are used to determine the final target position for the sea-ice drift station.
Sumata, H., Kauker, F., Karcher, M. und Gerdes, R., 2019a: Simultaneous Parameter Optimization of an Arctic Sea Ice–Ocean Model by a Genetic Algorithm, doi.org/10.1175/MWR-D-18-0360.1
Sumata, H., Kauker, F., Karcher, M. und Gerdes, R., 2019b: Covariance of Optimal Parameters of an Arctic Sea Ice–Ocean Model, doi.org/10.1175/MWR-D-18-0375.1
Responsible for numerical ice forecasts and contact person at the AWI:
Dr Frank Kauker (Frank.Kauker@awi.de)