Cryospheric Sciences

polar prediction

Image of the Week – Stuck in the ice: could it have been predicted?

Image of the Week –  Stuck in the ice: could it have been predicted?

Expeditions in the Southern Ocean are invaluable opportunities to learn more about this fascinating but remote region of the world. However, sending vessels to navigate the hostile Antarctic waters is an expensive endeavor, not only financially but also from a human perspective. When vessels are forced to turn back due to hazardous conditions or, even worse, become stuck in the ice (as shown in our Image of the Week), a mission full of expectations can quickly turn into a nightmare. Hence there is an increasing demand for reliable information on the navigability of the Southern Ocean a few weeks to a few months in advance. This information could support the final decision whether to start the journey or not, and would allow minimizing the associated risks.

What’s the problem?

In late February 2018, the British vessel RRS James Clark Ross was heading to the Eastern Antarctic Peninsula to investigate the consequences of the calving of a massive iceberg from the Larsen C ice shelf. Unfortunately the vessel had to turn back before reaching its goal due to the unexpected presence of thick sea ice in the region. This story is not unusual. During Christmas 2013, a Russian ship named the Akademik Shokalskiy also got stuck in several meters of Antarctic sea ice. Ironically, one of the rescuing vessels itself (the Chinese Xuě Lóng) got trapped in the ice as well. To prevent such events from happening again, we need to be able to predict the upcoming sea-ice conditions. Can sea-ice conditions be forecast at seasonal time scales? If so, how?


Antarctic sea ice, the Year of Polar Prediction and SIPN South

To prevent accidents and unforeseen problems, one goal of the Year Of Polar Prediction is to enhance environmental forecasting capabilities from operational (hours to days) to tactical (weeks to months) time scales in high latitude regions. Several studies support the notion that Antarctic sea ice may be predictable a few months ahead, at least in certain regions (Holland et al. 2017, Chen and Yuan 2004, Holland et al. 2013, Marchi et al. 2018).

To investigate further the predictability of Antarctic sea ice, the Sea Ice Prediction Network South (SIPN South) was launched in 2017. It is a two-year international project endorsed by the YOPP. SIPN South pursues three strategic objectives:

  • Hosting seasonal outlooks of Antarctic sea ice to better understand the sources of sea-ice predictability and the origins of systematic forecast errors in different types of models.
  • Providing news and information on the current state of Antarctic sea ice, disseminating research to a wider audience and reporting ongoing field campaigns.
  • Coordinating realistic seasonal prediction exercises to investigate the potential use of this information for users and customers, primarily ships navigating in the region.


February 2018 seasonal sea-ice forecasts

As a first major milestone, SIPN South provided coordinated forecasts of sea ice for February 2018. February is the month with the smallest sea-ice area in the Antarctic, and therefore most of the shipping traffic in the region happens around that time. Participants were asked to provide an estimation of sea-ice coverage (area, concentration) for each day of February 2018, and were asked to issue their predictions by mid-December 2017. 13 research groups participated in this first forecasting experiment, following different approaches: several groups used fully coupled climate dynamical models, while others applied statistical regression methods to predict future ice conditions.

As we all know, the weather is unpredictable beyond a few days. However, previous research has suggested that the statistics of weather (its mean, its variability) can potentially be predicted from months to decades, due to the coupling of the atmosphere with “slower” components of the climate system like the ocean. To reflect this and to accurately estimate the statistics of weather, groups tend to provide not just one forecast, but several of them. These “ensembles” of forecasts provided by each group therefore represent all possible states of the atmosphere, ocean and ice that may prevail in February 2018 – given the known initial conditions of December.

The results of the coordinated experiment are shown in Figure 2. The February mean sea-ice area is shown for each group (colors), along with two actual observational references (black). Bear in mind that the forecast data were issued two months before the actual target date! Here, the forecasts are expressed as anomalies with respect to a reference climatology. All forecasts tend to overestimate the February sea ice area in the Ross Sea. A reason for this wrong estimation might be a very unusual cyclone, which passed over the Ross Sea around the 20th of January 2018 (i.e., between the time the forecasts were issued and the period for verification). This cyclone brought relatively warm air into the region. Furthermore it fractured the ice, opening more areas of open water and possibly increasing the effect of the ice-albedo feedback. Events like this one are not individually predictable several weeks in advance, but a well-designed forecasting system should at least account for this possibility. Despite running ensembles of forecasts, the sea-ice reduction in the Ross Sea was not captured by most forecasts. This may point towards a common and systematic deficiency in these prediction systems.

Figure 2: February 2018 mean regional sea-ice area anomaly (compared to 1979-2014 observed climatology) by longitude, for the 13 submissions, with observed estimates given in black. Solid lines show the ensemble mean for each contribution, with transparent shading indicating the ensemble range (min-max) [Credit: F. Massonnet].

Communicating climate information

Sea-ice area, as shown in Fig. 2, is a primary parameter used by scientists to quantify ice presence in a given region. It is also a useful number to diagnose model-data mismatch. However, sea-ice area is of little use for those who actually need climate information. For someone operating a vessel, the important information is how likely that vessel is to encounter sea ice in a given region for a given day in February. Information from Fig. 2, while certainly useful to scientists, is meaningless to those willing to extract practical information for navigation.

Alongside the work to understand fundamental drivers of sea-ice predictability in order to eventually improve the predictions, it is necessary to consider how potential users will interact with the forecasts. As explained above, climate forecasts are probabilistic in nature. Communicating probabilistic information to a non-trained audience is always a challenging task: for example, how would you interpret a forecast saying that there is a 50% chance of rain for tomorrow?

To reflect the irreducible uncertainty of climate forecasts (see previous section), sea-ice forecasts are generally expressed in terms of sea-ice probability, i.e. the probability that a given region of the Southern Ocean has sea-ice concentration larger than 15%. This probability is derived for each day and each grid cell from the ensemble forecasts contributed by each group (Fig. 3). If well calibrated, this type of information can be useful to those planning operations weeks in advance. For example, all but one model had forecast a high (>80%) probability of ice presence in the Larsen C area (eastern tip of the Antarctic Peninsula) where the RRS James Clark Ross got stuck five months ago. That is, there was a high risk, according to those forecasts, that ice would be present in that area in February. Of course, this does not mean that navigation would have been impossible (ice breakers can still operate in icy waters, provided the ice is thin), but these forecasts provided a first-order warning that there was a significant risk of encountering hazardous ice conditions there.

Figure 3: Probability of sea-ice presence for 15th February 2018, as forecasted by the five groups that submitted daily sea-ice concentration information. The sea-ice edge as observed by two products is shown in white. The probability of presence for a given day corresponds to the fraction of ensemble members that simulate sea-ice concentration larger than 15% in a given grid cell for that day. A dynamic animation of the figure showing all 28 days of February is available on the SIPN South website. [Credit: F. Massonnet]

Forecasting February 2019

The core phase of the Year of Polar Prediction entails “Special Observing Periods”, that is, intensive efforts to monitor the Arctic and Antarctic regions but also to enhance modeling activities (see this previous post). The (unique) Special Observing Period in the Southern Ocean will take place between mid-November 2018 and mid-February 2019. A new call for contributions will be launched by SIPN South to collect sea-ice forecasts for austral summer 2019, hoping that the first exercise in 2018 will raise the interest of even more research groups. A key question will be to assess whether the systems will be able to forecast better the sea-ice conditions in the challenging Ross Sea area, where most forecasts failed. Better insights will hopefully be gained in tracing the origin of systematic model error and lead to an improvement of Antarctic sea ice predictions within the next decade. As reliable climate information is crucially needed in this remote but important region of the world, future efforts to predict Antarctic sea ice will be very welcome!


Further reading

Edited by Adam Bateson and Clara Burgard


François Massonnet is a F.R.S.-FNRS Post-Doctoral Researcher at the Université catholique de Louvain and scientific collaborator at the Barcelona Supercomputing Center (Spain). He is assessing climate models as tools to understand (retrospectively and prospectively) polar climate variability and beyond. He tweets as @FMassonnet. Contact Email:



Image of the Week – Polar Prediction School 2018

Image of the Week – Polar Prediction School 2018

Early career scientists studying polar climate are one lucky group! The 29 young scientists who took part in the 10 day Polar Prediction School this year were no exception. They travelled to Arctic Sweden to learn and discuss the challenges of polar prediction and to gain a better understanding of the physical aspects of polar research.

The Year of Polar Prediction

The Year of Polar Prediction (YOPP) was launched on May 15th 2017; a large 2 year project that ‘aims to close gaps in polar forecasting capacity’ and ’lead to better forecasts of weather and sea-ice conditions to improve future environmental safety at both poles’. With these aims in mind, and with the support of the related APPLICATE project and the Association for Polar Early Career Scientists (APECS), a ten day Polar Prediction School took place in Abisko, Sweden in mid-April.

Abisko is a little town of 85 inhabitants, located north of the Arctic Circle (68°N) next to a National Park and a large lake. Due to the interesting habitats found in the region it is an excellent place to undertake polar research. Consequently, a scientific research station is located in the town, where research mainly focuses on biology, ecology, and meteorology.

Heading back to the research station (seen at the back of the picture) after a long hike [Credit: C. Burgard].

The 29 school participants were made up of Master students, PhD students, and PostDocs, with some studying the Arctic and some the Antarctic. The participants had diverse research backgrounds, with research that focused on atmospheric sciences, oceanic sciences, glaciers, sea ice and hydrology of polar regions, and used a range of techniques, from weather or climate models to in-situ or satellite observations. However, in the end, we were all linked together by our interest in the polar regions. Both this diversity and this link in our research helped us to exchange ideas about the common issues and the differences in all our disciplines.

The school programme

The course aimed to broaden students’ knowledge around their very specific PhD area. Therefore, the school covered a huge range of topics including polar lows, polar ocean-sea ice forecasting, remote sensing of the cryosphere, boundary layers, clouds and much more! Each day was made up of a mixture of lectures and practical sessions, which included:

  • Computer modelling exercises, for example using a simple 1D sea ice model
  • Observations, which included measuring temperature and wind from a weather station on the frozen lake next to the station, and daily radiosonde launches at lunchtime, in sync with radiosonde launches worldwide. These results were compared to model predictions each day.
  • Data assimilation, which focused on understanding the shortcomings in reanalysis products that we all use, including sources of uncertainty and error in the products and how they may impact our own work.

After dinner each evening a different group gave an informal weather briefing for the next day, which was often condensed down to how cloudy it would be, the amount of snow predicted (very little), and temperature (which averaged 2-3°C). Not quite the harsh, sub-zero temperatures that most of us had packed for! Each day was broken up by two coffee breaks (always accompanied by an obligatory cinnamon roll!) and meals which were taken all together in the main research building. This dragged everyone out of the lecture room to chat and refresh before the next session.

As is usual for any worthwhile meteorological fieldwork, we installed a small weather mast on the lake [Credit: C. Burgard].

Living Arctic weather for real

The usual weather in Abisko during April is fairly dry with temperatures ranging from 2°C to -6°C. In preparation for the cold, most of us had brought an abundance of wooly jumpers, thick thermal layers and numerous pairs of socks. However, on arrival in Abisko, the sun was shining and it was a balmy 7°C for the first two days. Whilst erecting the meteorology mast many of us were wearing T-shirts and sunglasses, after abandoning our warmer gear. The warm weather was not to last! Cloudy, relatively mild (2°C to -2°C) conditions persisted throughout most of the week, and it remained dry, which made it easier to forecast the weather but we were all hoping for a little snow! Finally, on the final day of the summer school, large snowflakes fell, although sadly it all melted quite quickly.

When we arrived, the whole area was coated in a thick layer of white snow and the frozen lake was covered. However, by the end of our visit, the bare earth was visible, and the top of the lake was slushy puddles of water. The changes in weather throughout the summer school made for interesting observation records. The albedo (reflectivity) of the lake surface went from approximately 0.8 for the fresh, white snow, but was reduced to 0.4 for the darker, water covered lake surface. It was great to see some theory in action!

Exploring the region

Luckily, we were also given a free day , in which we could explore the region, go skiing or just relax. One large group went off hiking, whilst a smaller group went cross country skiing and a few had a walk to the nearby frozen waterfall. But don’t worry, the science still continued! A group of 3 people stayed close by to release the lunchtime radiosonde.

Abisko children launching a radiosonde! [Credit: J. Turton]

Our visit to the area coincided with the exciting annual ice fishing contest! Whilst cars and small DIY tools are common place in many cities, in Abisko it is a snow mobile (or skidoo) and an ice drill, so they were well versed in the art of ice fishing! The majority of the town’s occupants arrived at the lake and started drilling small holes to catch some fish. After two hours, a number of prizes were awarded (e.g for the longest fish caught). Unfortunately, some of the holes were a little too close to our meteorology mast, and some cables were pulled out, but thankfully we still collected some good data!

An important aspect of any research is engaging with the local communities and communicating effectively with them. So all of the summer school attendees gathered by the lake to watch the ice fishing contest, and a large number of the children from Abisko gathered to watch us release the radiosonde, even helping launch one. They found our activities just as exciting as we found theirs!

And we did some science communication as well!

A crucial aspect of science is how you communicate it to a variety of audiences. The way you might discuss your thesis to your viva panel should be completely different to the way you describe your science to your Great Aunt Linda or to a group of 10-year olds who are attending your outreach event. As part of the summer school, we learnt a range of tips and tricks for communicating science, thanks to Jessica Rohde. Jess is the communications officer for IARPC (Interagency Arctic Research Policy Committee) Collaborations and has years of science communication experience under her belt. Each evening we had a short lecture by Jess, which focused on a specific area of communication including presentation slide design, knowing your audience, listening to the audience and finding the story behind your science. Once we had learnt the theory we then put what we had learnt into practice. We did a bit of  improv’, which included 1-minute elevator pitches and tailoring your science to taxi drivers, the Queen of England and models (no not computer models, the Kate Moss variety). An important take-home message was that there is no such thing as the ‘general public’. When designing your outreach event, the ‘general public’ could involve children of all ages (and therefore all learning levels), parents, teachers, professors and pensioners. Therefore, you should listen to the needs of your audience and understand what their motivation is.

You can check out the final results of these sessions here!

In summary…

In the end, although the school was quite intense, everyone was sad to part. We are sure we will all remember this exciting time, where we learnt about the many aspects of polar prediction, and what to consider when tackling science communication. We hope that this school will be organized again in the next years to provide this amazing and unforgettable experience to all those who could not join this year’s Polar Prediction School!

Further reading

Edited by Morgan Jones

Rebecca Frew is a PhD student at the University of Reading (UK). She investigates the importance of feedbacks between the sea ice, atmosphere and ocean for the Antarctic sea ice cover using a hierarchy of climate models. In particular, she is looking at the how the importance of different feedbacks may vary between different regions of the Southern Ocean.



Jenny Turton is a post doc working at the institute for Geography at the University of Erlangen-Nuernberg, in the climate system research group. Her current research focuses on the interactions between the atmosphere and surface ice of the 79N glacier in northeast Greenland, as part of the GROCE project. 




Clara Burgard is a PhD student at the Max Planck Institute for Meteorology in Hamburg. She investigates the evolution of sea ice in general circulation models (GCMs). There are still biases in the sea-ice representation in GCMs as they tend to underestimate the observed sea-ice retreat. She tries to understand the reasons for these biases.