Subseasonal to seasonal Arctic sea-ice prediction: A grand challenge of climate science

K Wi , Jiping Liu , Qing Bao , Bian H , Jiao Ma , Ming Li , Mirong Song , Zhu Zhu , d

a Center for Monsoon System Research, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing, China

b Department of Atmospheric and Environmental Sciences, University at Albany, Albany, NY, USA

c State Key Laboratory of Numerical Modeling for Atmospheric Sciences and Geophysical Fluid Dynamics, Institute of Atmospheric Physics, Chinese Academy of Sciences,Beijing, China

d University of Chinese Academy of Sciences, Beijing, China

e Polar Research & Forecasting Division, National Marine Environmental Forecasting Center, Beijing, China

On 15September2020,theArctic sea-ice extent (SIE) reached its annual minimum, which, based on data from the National Snow and Ice Data Center ( NSIDC, 2020a ), was about 3.74 million km(1.44 million square miles). This value was about 40% less than the climate average (～6.27 million km) during 1980—2010. It was second only to the record low (3.34 million km) set on 16 September 2012, but significantly smaller than the previous second-lowest (4.145 million km, set on 7 September 2016) and third-lowest (4.147 million km, set on 14 September 2007) values, making 2020 the second-lowest SIE year of the satellite era (42 years of data).

Although ranked second lowest, the year 2020 carries special meaning. In August 2012, the Arctic experienced the Great Arctic Cyclone,which was the most extreme cyclone in summer on record since the beginning of the satellite era, reaching the level of strong cyclones in winter ( Simmonds and Rudeva, 2012 ). It led to a dramatic reduction in Arctic sea ice and thus set the historical record still standing today.In 2015/16, the occurrence of a super El Ni?o was linked to extremely warm anomalies in the Arctic, which affected sea-ice loss ( Petty et al.,2018 ; Stroeve and Notz, 2018 ). By contrast, during the melting season of 2020, there was no powerful cyclone, and the sea surface temperature of the equatorial central and eastern Pacific was neutral (La Ni?a had been developing since autumn).

The Arctic has experienced amplified warming and extensive sea-ice retreat in recent decades. Decreasing Arctic SIE has been observed for each month since the beginning of the satellite era ( IPCC, 2019 ), with the largest reduction in September —specifically, a decline of 12.8% per decade. The Arctic has lost about 40% of its sea ice compared with the late 1970s. This amount of change is likely the largest for at least the past 1000 years ( IPCC, 2019 ).

The low Arctic SIE in 2020 may have been related to the extremely warm air and sea surface temperatures in the Arctic ( NSIDC, 2020b ).The temperature in northern Siberia and the Canadian Archipelago had been on the high side since spring of that year; indeed, Verkhoyansk in northern Siberia set a daily maximum temperature record of greater than 38°C at the end of June ( WMO, 2020 ), and the sea surface temperature anomaly in July and August reached more than 5°C in the Beaufort Sea. A large amount of warm seawater entered the polar region, creating favorable conditions for lower Arctic sea ice ( NSIDC, 2020b ). Therefore,the low SIE in 2020 was contributed by both the long-term trend and the particularity of summer 2020.

Changes in sea-ice properties also contribute to abnormally low sea ice. Usually, the older the sea-ice age, the more difficult it is to melt.Before the 1980s, multi-year ice that was more than four years old covered more than 1/3 of the Arctic Ocean. Since then, in recent years,it has decreased by more than 95% ( Perovich et al., 2019 ). Continued loss of multi-year sea ice has prevented the recovery of the September SIE. Meanwhile, as snow and ice melt, meltwater accumulates at the ice surface, creating melt ponds ( Zhang et al., 2018 ). Extensive melt-pond coverage was observed during the melting season of 2020. This would have lowered the surface albedo and increased the absorption of solar radiation, leading to a strong positive albedo feedback contributing to the observed sea-ice loss. Therefore, predictions based on dynamical models, statistical relationships, or heuristic methods should have the ability to properly reflect the change in Arctic sea-ice properties.

Since 2008, the Sea Ice Prediction Network (SIPN) has called for a Sea Ice Outlook (SIO) from the international research community every year. The participants have been requested to submit their predicted September Arctic SIE in early June, July, and August. SIPN represents the current prediction level and community understanding of the state and evolution of Arctic sea ice on the sub-seasonal-to-seasonal (S2S)time scale.

In June 2020, 33 institutions and organizations worldwide submitted their SIO of the pan-Arctic September SIE, of which 13 institutions also submitted a predicted spatial distribution. The submissions increased to 38 in July and 39 in August. From the June to August SIO, the median of all predictions remained quite stable (4.33 million kmin June, 4.36 in July, and 4.3 in August), which were much higher than the observed value of 3.92 million km( NSIDC, 2020c ). This indicates that most forecasting systems overestimated the coverage of sea ice in September 2020.This year’s observation is beyond the interquartile range of the 110 submissions, leading to the situation that only “radical ”predictions are close to the observation, i.e., for the June SIO, the prediction of the University of Washington/APL was the smallest (3.2 million km),followed by GFDL/NOAA with 3.5, and ANSO IAP-LASG with 3.8, deviating wildly from the median value of 4.33 and suggesting a possible second record low.

It is still a challenge to accurately predict the Arctic SIE on S2S timescales, especially in extreme years. From 2009 to 2020, most years’observed values (8 out of 12) fell outside the predicted interquartile range of dynamical models ( Fig. 1 ). In 2012 (record low) and 2020(second-lowest extent), the median of the predicted values was significantly higher than the observation; while in 2013, 2014, and 2017, the prediction results were considerably lower than the observed values. In addition, prediction of the spatial distribution of the Arctic sea ice is even more challenging than the total SIE. As it turned out, only a few participating institutions submitted a predicted spatial distribution.

Fig. 1. The median and interquartile range (IQR) of the September Arctic Sea Ice Outlook (SIO) based on the July predictions of dynamical models that participated in the Sea Ice Prediction Network.

Accurate simulation of the Arctic SIE requires that models can reasonably consider the following processes:

(1) Assimilation system. The assimilation of the sea-ice data (i.e., concentration, thickness), the snow depth, and sea surface temperature,is still lacking in many models. Recent studies suggest that different observations and initialization methods used in the initialization significantly affect the predictability of Arctic sea ice. Thus, dynamical models need to better assimilate the observed sea-ice data, as well as important atmospheric and oceanic observations, to generate a skillful initialization to increase the predictive skill.

(2) Better atmospheric forcing. For regional models and ocean—sea ice coupled models, the atmospheric forcing is essential. This requires that the S2S system produces a high-quality prediction of the atmosphere, especially in the extreme conditions of summer.

(3) Sea-ice property changes due to global warming. Most models do not have processes that can capture the changes in sea-ice properties due to the changes in sea-ice age, ice thickness, and the increased occurrence of melt ponds. Thus, improved descriptions of sea-ice processes in the sea-ice model component of the prediction system are needed, i.e., realistic evolution of melt ponds over sea ice.

Associated with global warming, the summertime Arctic is on track to be functionally ice-free, which is defined by it possessing an SIE of less than 1 million km. Sea ice remains mainly in the Canadian Arctic Archipelago and northern Greenland Sea. As sea ice melts, the exposed seawater absorbs more solar radiation, leading to more melting and warming —a positive feedback mechanism for Arctic warming and sea-ice loss. This makes the Arctic approach a dangerous tipping point( Steffen et al., 2018 ; Lenton et al., 2019 ) in which domino-like irreversible processes might be triggered.

As an indicator and “amplifier ”of global climate change, the Arctic’s health and stability is the cornerstone of the stability of our climate system. It has far-reaching impacts on ecosystems, coastal resilience, and human settlements in the middle and high latitudes. It can also influence the frequency and intensity of midlatitude blocking and extreme events( Overland et al., 2016 ; Vavrus et al., 2017 ).

In addition to the need to improve the ability to predict the total SIE,prediction of the spatial distribution of the Arctic sea ice in September is necessary but more challenging ( Zheng et al., 2021 ). The number of participating institutions is obviously lower than that involved in prediction of the total sea ice. In conclusion, studies need to improve their ability to make more accurate predictions and achieve a better understanding of the physics of sea-ice processes.

Funding

This work was supported by the National Key R&D Program of China[grant number 2018YFA0605901] and the National Natural Science Foundation of China [grant numbers 41861144016 and 42011530082].