Sea Ice Prediction Research Articles

Statistical seasonal prediction of Arctic sea ice concentration based on spatiotemporal anomaly persistent method

Abstract Accurate prediction of Arctic sea ice is crucial for high-latitude and even mid-latitude climate prediction. It significantly affects atmospheric circulation, the environment, ecology, and maritime transport. This study developed a statistical prediction model to predict monthly Arctic sea ice concentration (SIC) for up to one year based on the season-reliant empirical orthogonal functions (SEOFs) technique. Its prediction skill was compared with that of a dynamical prediction model. The spatiotemporal pattern of sea ice anomalies, which exhibit strong seasonality and are maintained for a significant period above the seasonal time scale by atmosphere-ocean interactions, was extracted using SEOFs. A prediction model was constructed by extrapolating from the recent anomalous state of sea ice to predict the future. Experimental retrospective predictions with monthly time resolution for 1982–2021 were performed to validate the prediction skill of Arctic SIC and areal extent. Statistically significant prediction skills were achieved over several months, even up to six months, exceeding the skill of the dynamical model.

Physics-informed deep convolutional network for combined sea ice concentration and velocity prediction

Accurate prediction of short-term fluctuations in Arctic sea ice is important for safe use of Arctic shipping routes. While deep-learning models can improve the accuracy of sea-ice predictions, the accuracy and predictability of short-term sea-ice conditions are limited in most purely data-driven deep-learning models that consider a single aspect of sea ice, without considering the physical variation law between several sea ice factors. We propose dual-task prediction models for sea ice concentration (SIC) and sea ice velocity (SIV), and incorporate a loss function that simultaneously addresses dynamic constraints of SIC and SIV into each model, based on dynamics terms in the sea ice control equation. Comparative experiments are performed to determine which model structure best performs at predicting SIC and SIV. We report a dual-task branching structure to be more suitable for predicting SIC and SIV, and a post-decoder branch network structure to best predict SIC and SIV. Adding a sea ice dynamics equation to the loss function improves the model fit with dynamics law, improving the predictability and prediction accuracy of SIC and SIV.

High-resolution regional sea-ice model based on the discrete element method with boundary conditions from a large-scale model for ice drift

Abstract Understanding sea-ice dynamics at the floe scale is crucial to comprehend polar climate systems. While continuum models are commonly used to simulate large-scale sea-ice dynamics, they have limitations in accurately representing sea-ice behaviour at small scales. DEMs, on the other hand, are well-suited for modelling the behaviour of individual ice floes but face limitations due to computational constraints. To address the limitations of both approaches while combining their strengths, we explored the feasibility of using a DEM within a continuum model, where the latter provides boundary conditions for a rectangular high-resolution DEM domain. This paper presents a feasibility study where a discrete model of a 100 × 100 km2 icefield was created using high-resolution optical satellite imagery. Sea-ice dynamics were simulated in the DEM considering environmental forces and integrating large-scale ice-drift velocities as boundary conditions. Model predictions were compared with satellite observations for ice drift and deformation parameters. This numerical approach showed potential for offering accurate, high-resolution predictions of sea ice, particularly in coastal areas and near islands, and may find applications in ice navigation and climate studies. However, further development of the DEM, along with upgrades to the coupled ocean models providing data for the ice component, may be necessary. Additionally, challenges remain to develop a two-way coupling between the DEM and a continuum model, which may be needed to improve the accuracy of large-scale simulations.

Winter arctic sea ice volume decline: uncertainties reduced using passive microwave-based sea ice thickness

Arctic sea ice volume (SIV) is a key climate indicator and memory source in sea ice predictions and projections, yet suffering from large observational and model uncertainty. Here, we test whether passive microwave (PMW) data constrain the long-term evolution of Arctic SIV, as recently hypothesized. We find many commonalities in Arctic SIV changes from a PMW sea ice thickness (SIT) 1992-2020 time series reconstructed with a neural network algorithm trained on lidar altimetry, and the reference PIOMAS reanalysis: relatively low differences in SIV mean (4615 km3, 37%), SIV trends (46 km3, 17%), and phased variability (r2=0.55). Key to reduced differences is the consistent evolution of many SIV contributors: seasonal and perennial ice coverage, their SIT contrast, whereas perennial SIT provides the largest remaining uncertainty source. We argue that PMW includes useful SIT information, reducing SIV uncertainty. We foresee progress from sea ice reanalyses combining dynamical models and data assimilation of PMW SIT estimates, in addition to the already assimilated PWM sea ice concentration.

Predicting September Arctic Sea Ice: A Multimodel Seasonal Skill Comparison

Abstract This study quantifies the state of the art in the rapidly growing field of seasonal Arctic sea ice prediction. A novel multimodel dataset of retrospective seasonal predictions of September Arctic sea ice is created and analyzed, consisting of community contributions from 17 statistical models and 17 dynamical models. Prediction skill is compared over the period 2001–20 for predictions of pan-Arctic sea ice extent (SIE), regional SIE, and local sea ice concentration (SIC) initialized on 1 June, 1 July, 1 August, and 1 September. This diverse set of statistical and dynamical models can individually predict linearly detrended pan-Arctic SIE anomalies with skill, and a multimodel median prediction has correlation coefficients of 0.79, 0.86, 0.92, and 0.99 at these respective initialization times. Regional SIE predictions have similar skill to pan-Arctic predictions in the Alaskan and Siberian regions, whereas regional skill is lower in the Canadian, Atlantic, and central Arctic sectors. The skill of dynamical and statistical models is generally comparable for pan-Arctic SIE, whereas dynamical models outperform their statistical counterparts for regional and local predictions. The prediction systems are found to provide the most value added relative to basic reference forecasts in the extreme SIE years of 1996, 2007, and 2012. SIE prediction errors do not show clear trends over time, suggesting that there has been minimal change in inherent sea ice predictability over the satellite era. Overall, this study demonstrates that there are bright prospects for skillful operational predictions of September sea ice at least 3 months in advance.

Reconstruction of Arctic Sea Ice Thickness and Its Impact on Sea Ice Forecasting in the Melting Season

Abstract Generally, sea ice prediction skills can be improved by assimilating available observations of the sea ice concentration (SIC) and sea ice thickness (SIT) into a numerical forecast model to update the initial conditions. However, due to inadequate daily SIT satellite observations in the Arctic melting season, the SIC fields in forecast models are usually directly updated, which causes mismatch of SIC and SIT in dynamics and affects the model prediction accuracy. In this study, a statistically based bivariate regression model of SIT (BRMT) is tentatively established based on the grid reanalysis data of SIC and SIT to reconstruct daily Arctic SIT data. The results show that the BRMT can reproduce the spatial and temporal changes in the SIT in the melting season and capture the variation trend of SIT in some periods. Compared with the SIT observations from buoy and satellite, the reconstructed SIT shows better performance in the central Arctic than other datasets. Furthermore, when the reconstructed SIT is added to the forecast model with only assimilated SIC, the forecast accuracy of SIC, sea ice extent, and SIT in the Arctic melting season is improved and does not weaken with the increase in the forecast time. Especially in the central Arctic, the average absolute deviation between 24-h SIT forecast results and observations is only 0.16 m. The results indicate great potential for applying the reconstructed SIT to the operational forecast of Arctic sea ice during the melting season in the future. Significance Statement To improve the prediction skills of Arctic sea ice, it is necessary to assimilate the sea ice observation into the dynamic model to generate a more realistic initial prediction field. At present, the observation data of daily sea ice thickness (SIT) during the Arctic melting season are few, which cannot well meet the demand of operational SIT forecast. In this study, a bivariate regression model is put forward to construct SIT using the sea ice concentration (SIC) observation. Benefitting from the joint assimilation of the observed SIC and the reconstructed SIT, the forecast accuracy of sea ice variables is greatly improved. The reconstructed SIT is expected to provide an available dataset for further research on the Arctic sea ice forecast.

PERFORMANCE ASSESSMENT OF DATA-DRIVEN MODELS IN ARCTIC SEA ICE PREDICTION

Data-driven models are emerging as a brand-new approach in Arctic sea ice prediction. In this study, we perform the effectiveness comparison of different machine learning models in predicting the sea ice extent in the whole Arctic Sea and its subregions: the Bering Sea, the East Siberian Sea, the Chukchi Sea, the Beaufort Sea, and the Central Arctic Sea. The machine learning models used in this study include Ridge Regression, Lasso Regression, Decision Tree, k-Nearest Neighbors, Random Forests, XGBoost, as well as two stacked learning models. Our research reveals that due to the different features of sea ice evolution in different sea areas, these data-driven models generate significantly different prediction performances for sea ice evolution in different seas. In detail, two stacked learning models performed the best in the whole Arctic Sea, the stacked learning model LR_Random Forests performed the best in the Beaufort Sea, Bering Sea, and Chukchi Sea, the Random Forests model performed the best in the Central Arctic Sea, and the Decision Tree model performed the best in the East Siberian Sea.

Effectiveness of wind-constrained sea-ice momentum on formation of sea-ice distribution and upper halocline of Arctic Ocean in climate model

Initialization of sea ice and the upper halocline in the Arctic Ocean is crucial for sea-ice prediction, but their representation in climate models still remains biased. Here, using historical and four different simulations by a single climate model, we find that constraining the sea-ice momentum by surface wind stress contributes to a better representation of the sea-ice velocity, area, and concentration. Moreover, the wind-constrained sea-ice drift modifies the underlying ocean structure via ice-ocean stress, leading to an improved climatological halocline’s vertical structure in the Canada Basin. This is because the excessively represented negative wind and ice-ocean stress curl in the climate model is weakened when constraining the sea-ice momentum and consequently the downward vertical speed, including the Ekman pumping, is also weakened at depths of 0–500 m, alleviating the deepening of isohalines. From these results, the improvement of sea-ice and ocean states by constraining sea-ice momentum is expected to make sea-ice prediction more accurate.

The role of upper-ocean heat content in the regional variability of Arctic sea ice at sub-seasonal timescales

Abstract. In recent decades, the Arctic Ocean has undergone changes associated with enhanced poleward inflow of Atlantic and Pacific waters and increased heat flux exchange with the atmosphere in seasonally ice-free regions. The associated changes in upper-ocean heat content can alter the exchange of energy at the ocean–ice interface. Yet, the role of ocean heat content in modulating Arctic sea ice variability at sub-seasonal timescales is still poorly documented. We analyze ocean heat transports and surface heat fluxes between 1980–2021 using two eddy-permitting global ocean reanalyses, C-GLORSv5 and ORAS5, to assess the surface energy budget of the Arctic Ocean and its regional seas. We then assess the role of upper-ocean heat content, computed in the surface mixed layer (Qml) and in the 0–300 m layer (Q300), as a sub-seasonal precursor of sea ice variability by means of lag correlations. Our results reveal that in the Pacific Arctic regions, sea ice variability in autumn is linked with Qml anomalies leading by 1 to 3 months, and this relationship has strengthened in the Laptev and East Siberian seas during 2001–2021 relative to 1980–2000, primarily due to reduced surface heat loss since the mid-2000s. Q300 anomalies act as a precursor for wintertime sea ice variability in the Barents and Kara seas, with considerable strengthening and expansion of this link from 1980–2000 and 2001–2021 in both reanalyses. Our results highlight the role played by upper-ocean heat content in modulating the interannual variability of Arctic sea ice at sub-seasonal timescales. Heat stored in the ocean has important implications for the predictability of sea ice, calling for improvements in forecast initialization and a focus upon regional predictions in the Arctic region.

Global, Arctic, and Antarctic sea ice volume predictions using score-driven threshold climate models

The literature on sea ice predictions uses a variety of general circulation models (GCMs), which suggest diverse predictions of the date of ice-free or almost ice-free oceans, and focus mainly on the Arctic. GCMs are not sensitive enough to tipping points in the Atlantic meridional overturning circulation (AMOC), and they underestimate the sensitivity of Arctic sea ice to carbon emissions. We contribute to the literature in two ways: First, we use a novel time series model, named the score-driven threshold climate (SDTC) model, and we report global, Arctic, and Antarctic sea ice predictions. For the SDTC model, the estimations are computationally less demanding than those of the GCMs. Second, we combine long-run 1,000-year frequency climate data from 798,000 to 1,000 years ago, and short-run annual data from year 850 to year 2014. We present the evolution of long-run and short-run climate data with descriptive statistics. We estimate the SDTC model using annual data from 850 to 2014 for Arctic and Antarctic sea ice volume Icet and Antarctic land surface temperature Tempt. We use the atmospheric CO2,t concentration as a clustering variable to define periods of climate change. Hence, CO2,t concentration drives the regime-switching dynamics of parameters in the SDTC model. We report in-sample interval forecasts of global, Arctic, and Antarctic sea ice from 1980 to 2014. Observed global and Arctic sea ice volumes are below the forecasted interval from 2003. The observed Antarctic sea ice volume is below the forecasted interval from 2011. We report out-of-sample interval forecasts of sea ice from 2015 to 2314. The out-of-sample forecasts, μ[μ±2σ], indicate that if the current trend of climate change continues, then Arctic sea ice will disappear around 2058[2049,2068], and global and Antarctic sea ice will disappear around 2174[2123,2270].

Incorporating physical constraints in a deep learning framework for short-term daily prediction of sea ice concentration

Grasping the rapidly changing sea ice patterns has become the key to Arctic navigation. Starting from meteorology and oceanography, this study combines the empirical orthogonal function (EOF) analysis with convolutional long short-term memory (ConvLSTM) model to realize short-term sea ice prediction. To further improve the effectiveness and predictability of the ConvLSTM model, we adopt the EOF analysis to incorporate the spatial and temporal patterns of sea ice changes into the ConvLSTM model in three ways, namely, model input factors, model internal network structure, and model loss function. Through the mean absolute error (MAE), anomaly correlation coefficient (ACC), and anomaly metrics, we find that embedding the EOF physical constraints into the model's internal structure fully employs the spatial pattern and time series information, affording an optimal prediction effect. Additionally, the impact of the weight proportion between the physical constraint and the base loss and the effect of combining the three ways are discussed. The experimental results reveal that tuning the weight proportion in the loss function can improve the model's training effect. However, multiple combinations of the same physical law make it difficult to introduce new effective information.

Ross–Weddell Dipole Critical for Antarctic Sea Ice Predictability in MPI–ESM–HR

We use hindcasts from a state-of-the-art decadal climate prediction system initialized between 1979 and 2017 to explore the predictability of the Antarctic dipole—that is, the seesaw between sea ice cover in the Weddell and Ross Seas, and discuss its implications for Antarctic sea ice predictability. Our results indicate low forecast skills for the Antarctic dipole in the first hindcast year, with a strong relaxation of March values toward the climatology contrasting with an overestimation of anomalies in September, which we interpret as being linked to a predominance of local drift processes over initialized large-scale dynamics. Forecast skills for the Antarctic dipole and total Antarctic sea ice extent are uncorrelated. Limited predictability of the Antarctic dipole is also found under preconditioning around strong warm and strong cold events of the El Niño-Southern Oscillation. Initialization timing and model drift are reported as potential explanations for the poor predictive skills identified.

Seasonal Prediction of Regional Arctic Sea Ice Using the High‐Resolution Climate Prediction System CMA‐CPSv3

AbstractSea ice is a central part of the Arctic climate system, and its changes have a significant impact on the Earth's climate. Yet, its state, especially in summer, is not fully understood and correctly predicted in dynamical forecast systems. In this study, the seasonal prediction skill of Arctic sea ice is investigated in a high‐resolution dynamical forecast system, the China Meteorological Administration Climate Prediction System (CMA‐CPSv3). A 7‐month‐long retrospective forecast is made every other month from 2001 to 2021. Employing the anomaly correlation coefficient as the metric of the prediction skill, we show that CMA‐CPSv3 can predict regional Arctic sea ice extent and sea ice thickness up to 7 lead months. The Bering Sea exhibits the highest prediction skill among the 14 Arctic subregions. CMA‐CPSv3 outperforms the anomaly persistence forecast in the Bering Sea, Sea of Okhotsk, Laptev Sea, and East Siberian Sea. The sources of the sea ice prediction skill partly come from the good performance of upper ocean temperature in CMA‐CPSv3. This holds true not only for winter sea ice in the Arctic marginal seas but also for summer sea ice in several Arctic central seas. Furthermore, CMA‐CPSv3 exhibits a strong relationship between the variability of sea ice and surface heat fluxes. This underscores the importance of enhancing the representation of air‐sea heat exchanges in dynamical forecast systems to improve the prediction skill of sea ice.

Deep Learning Shows Promise for Seasonal Prediction of Antarctic Sea Ice in a Rapid Decline Scenario

Deep Learning Shows Promise for Seasonal Prediction of Antarctic Sea Ice in a Rapid Decline Scenario

Important Influence of North Atlantic Oscillation on Interannual Variability of the Boreal Autumn East Siberian–Beaufort Sea Ice

Abstract The East Siberia–Beaufort Sea (EsCB) sea ice during boreal autumn has recently been reported to play important roles in the climate over Eurasia and North America. The EsCB sea ice exhibits remarkable year-to-year fluctuations in autumn (August–October), the season in which its minimum extent usually occurs. However, the physical driver of the autumn EsCB sea ice interannual variability remains unclear, impeding the seasonal prediction of local sea ice. Here we find that the autumn EsCB sea ice variability is largely driven by the preceding summer (May–July) dipolar atmospheric anomalies over the North Atlantic, resembling the North Atlantic Oscillation (NAO) pattern. During the negative NAO-like phase, the circumpolar anticyclonic anomalies tend to transport the warm air from Greenland toward the EsCB region, which triggers rapid sea ice melt there. The associated EsCB sea ice anomalies can be maintained or even intensified by the local sea ice–albedo positive feedback until autumn. Therefore, the abnormal signals of Arctic sea ice tend to show significant persistence in summer and autumn. The influence of the summer NAO-like atmospheric circulation on the ensuing autumn EsCB sea ice can be realistically reproduced in the historical simulation of the E3SM-1-0 model, supporting our findings based on the observation. This lagged relationship provides a promising pathway for skillful seasonal prediction of the EsCB sea ice and its related climatic impacts.

Interrogating Sea Ice Predictability With Gradients

Interrogating Sea Ice Predictability With Gradients

Improvements in September Arctic Sea Ice Predictions Via Assimilation of Summer CryoSat‐2 Sea Ice Thickness Observations

AbstractBecause of a spring predictability barrier, the seasonal forecast skill of Arctic summer sea ice is limited by the availability of melt‐season sea ice thickness (SIT) observations. The first year‐round SIT observations, retrieved from CryoSat‐2 from 2011 to 2020, are assimilated into the GFDL ocean–sea ice model. The model's SIT anomaly field is brought into significantly better agreement with the observations, particularly in the Central Arctic. Although the short observational period makes forecast assessment challenging, we find that the addition of May–August SIT assimilation improves September local sea ice concentration (SIC) and extent forecasts similarly to SIC‐only assimilation. Although most regional forecasts are improved by SIT assimilation, the Chukchi Sea forecasts are degraded. This degradation is likely due to the introduction of negative correlations between September SIC and earlier SIT introduced by SIT assimilation, contrary to the increased correlations found in other regions.

Decadal oscillation provides skillful multiyear predictions of Antarctic sea ice

Over the satellite era, Antarctic sea ice exhibited an overall long-term increasing trend, contrary to the Arctic reduction under global warming. However, the drastic decline of Antarctic sea ice in 2014–2018 raises questions about its interannual and decadal-scale variabilities, which are poorly understood and predicted. Here, we identify an Antarctic sea ice decadal oscillation, exhibiting a quasi-period of 8–16 years, that is anticorrelated with the Pacific Quasi-Decadal Oscillation (r = −0.90). By combining observations, Coupled Model Intercomparison Project historical simulations, and pacemaker climate model experiments, we find evidence that the synchrony between the sea ice decadal oscillation and Pacific Quasi-Decadal Oscillation is linked to atmospheric poleward-propagating Rossby wave trains excited by heating in the central tropical Pacific. These waves weaken the Amundsen Sea Low, melting sea ice due to enhanced shortwave radiation and warm advection. A Pacific Quasi-Decadal Oscillation-based regression model shows that this tropical-polar teleconnection carries multi-year predictability.

Multidecadal variability and predictability of Antarctic sea ice in the GFDL SPEAR_LO model

Abstract. Using a state-of-the-art coupled general circulation model, physical processes underlying Antarctic sea ice multidecadal variability and predictability are investigated. Our model simulations constrained by atmospheric reanalysis and observed sea surface temperature broadly capture a multidecadal variability in the observed sea ice extent (SIE) with a low sea ice state (late 1970s–1990s) and a high sea ice state (2000s–early 2010s), although the model overestimates the SIE decrease in the Weddell Sea around the 1980s. The low sea ice state is largely due to the deepening of the mixed layer and the associated deep convection that brings subsurface warm water to the surface. During the high sea ice period (post-2000s), the deep convection substantially weakens, so surface wind variability plays a greater role in the SIE variability. Decadal retrospective forecasts started from the above model simulations demonstrate that the Antarctic sea ice multidecadal variability can be skillfully predicted 6–10 years in advance, showing a moderate correlation with the observation. Ensemble members with a deeper mixed layer and stronger deep convection tend to predict a larger sea ice decrease in the 1980s, whereas members with a larger surface wind variability tend to predict a larger sea ice increase after the 2000s. Therefore, skillful simulation and prediction of the Antarctic sea ice multidecadal variability require accurate simulation and prediction of the mixed layer, deep convection, and surface wind variability in the model.

Daily-Scale Prediction of Arctic Sea Ice Concentration Based on Recurrent Neural Network Models

Arctic sea ice prediction is of great practical significance in facilitating Arctic route planning, optimizing fisheries management, and advancing the field of sea ice dynamics research. While various deep learning models have been developed for sea ice prediction, they predominantly operate at the seasonal or sub-seasonal scale, often focusing on localized areas, and few cater to full-region daily-scale prediction. This study introduces the use of spatiotemporal sequence data prediction models, namely, the convolutional LSTM (ConvLSTM) and predictive recurrent neural network (PredRNN), for the prediction of sea ice concentration (SIC). Our analysis reveals that, when solely utilizing SIC historical data as the input, the ConvLSTM model outperforms the PredRNN model in SIC prediction. To enhance the models’ capacity to capture spatiotemporal relationships between multiple variables, we expanded the range of input data types to form the ConvLSTM-multi and PredRNN-multi models. Experimental findings demonstrate that the prediction accuracy of the four models significantly surpasses the CMIP6 model in three prospective climate scenarios (SSP126, SSP245, and SSP585). Of the four models, the ConvLSTM-multi model excels in assimilating the influence of reanalysis data on sea ice within the sea ice edge region, thus exhibiting superior performance than the PredRNN-multi model in predicting daily Arctic SIC over the subsequent 10 days. Furthermore, sensitivity tests on various model parameters highlight the substantial impact of sea surface temperature and prediction date on the accuracy of daily sea ice prediction, and meteorological and oceanographic parameters primarily affect the prediction accuracy of the thin-ice region at the edge of the sea ice.