ECMWF joins Sea-Ice Drift Forecast Experiment

Helge F. Goessling (AWI, Germany)
Steffen Tietsche (ECMWF)
Simon Reifenberg (AWI)
Axel Schweiger (University of Washington)


Since October 2018, ECMWF has contributed forecasts of the drift of Arctic sea-ice buoys derived from operational seasonal forecasts to the Sea-Ice Drift Forecast Experiment (SIDFEx). SIDFEx was launched in 2017 as part of the Year of Polar Prediction (YOPP). It is a community effort to collect and analyse Arctic sea-ice drift forecasts at lead times from days to a year, based on any method, for a number of sea-ice buoys and, ultimately, the research icebreaker Polarstern.

Aims of SIDFEx

SIDFEx is motivated in part by the need to determine an optimal start position for Polarstern’s year-long MOSAiC drift across the Arctic commencing in autumn 2019. Specifically, it will test whether forecasting systems that account for initial conditions and simulate the evolving atmosphere, sea ice and ocean system can provide additional skill over drift forecasts that rely on historical sea-ice velocity fields. The MOSAiC drift provides an exciting opportunity to assess current capabilities to forecast sea-ice drift for a number of applications, ranging from logistics support for field experiments to potential search and rescue operations. In addition, the examination of sea-ice drift forecasts will help us to understand many aspects of the coupled atmosphere–sea-ice–ocean system and to identify and resolve model shortcomings. SIDFEx specifies and enforces a common data format for submission, but otherwise contributors are free to choose their method and time range. As of November 2018, ten groups from the US, Canada, Germany, Norway, and the UK have submitted almost 20,000 forecasts.

ECMWF’s contribution

ECMWF contributes a fully coupled seasonal forecast every month in near real time. This has been made possible with the implementation of SEAS5 in November 2017, which predicts the movement of sea ice in response to surface winds, ocean currents, and internal forces within the ice. For the SIDFEx contribution, predicted trajectories of ice-tethered buoys are computed from SEAS5 daily-mean ice velocity fields archived on the native NEMO ocean grid. Although in principle these computations simply involve adding up daily displacements, the intricacies of the geometry of the tri-polar NEMO grid make this a non-trivial task. Therefore, we use the specialised software tool ARIANE, which has been developed by Bruno Blanke and Nicolas Grima at UBO/LOPS in Brest, France.

First results

ECMWF trajectory forecasts have been produced and submitted to cover initialisation times from 1 June 2018 onward. The figure shows forecasts valid on 15 September for one of the SIDFEx buoys and four consecutive forecast start dates. It illustrates the considerable forecast uncertainty on the seasonal time scale, which decreases as the initial date approaches the target date. Although the buoy position observed on 15 September is close to the predicted buoy positions for individual ensemble members in each case, for two of the initial dates the observed position is outside the likely (90%) range of forecast outcomes. Results for other buoys are broadly similar. More cases are needed to establish whether this is chiefly the result of forecast model biases or of the atypical drift pattern that occurred this summer: the eastward drift in the Russian and Alaskan sectors was rather unusual. In any event, it is likely that there are forecast model biases which call for calibration to improve reliability, as is common practice for these timescales. As more trajectory forecasts are added to the SIDFEx database, a clearer picture of the forecast quality will emerge.

Drift forecasts for an Arctic sea-ice buoy. The forecasts are based on the ECMWF seasonal system SEAS5. Initial positions on the first day of successive months from June to September 2018 are marked by triangles on the observed drift trajectory (black curve). The dots denote ensemble forecast positions, all valid on 15 September 2018. Ellipses enclose 90% probability areas of a bivariate normal distribution fitted to the points comprising the forecasts. The verifying observed position is shown by the cross in the box.


The SIDFEx community has only just started to exploit the constantly growing dataset, which is made publicly available in near real time ( ECMWF will benefit from the systematic forecast quality assessment of sea-ice drift carried out by the SIDFEx community, which will help to identify issues and ways to improve forecasts from days to seasons. The Centre is considering adding medium- and extended-range trajectory forecasts to its contribution. These could provide additional support for research and operations associated with the drifting icebreaker Polarstern.