Bushuk, Mitchell, Sahara Ali, David A Bailey, Qing Bao, Lauriane Batté, Uma S Bhatt, Edward Blanchard-Wrigglesworth, Ed Blockley, Gavin Cawley, Junhwa Chi, François Counillon, Philippe Goulet Coulombe, Richard I Cullather, Francis X Diebold, Arlan Dirkson, Eleftheria Exarchou, Maximilian Göbel, William Gregory, Virginie Guemas, Lawrence C Hamilton, Bian He, Sean Horvath, Monica Ionita, Jennifer E Kay, Eliot Kim, Noriaki Kimura, Dmitri Kondrashov, Zachary M Labe, Woo-Sung Lee, Younjoo J Lee, Cuihua Li, Xuewei Li, Yongcheng Lin, Yanyun Liu, Wieslaw Maslowski, François Massonnet, Walter N Meier, William J Merryfield, Hannah Myint, Juan C Acosta Navarro, Alek Petty, Fangli Qiao, David Schröder, Axel Schweiger, Qi Shu, Michael Sigmond, Michael Steele, Julienne Stroeve, Nico Sun, Steffen Tietsche, Michel Tsamados, Keguang Wang, Jianwu Wang, Wanqui Wang, Yiguo Wang, Yun Wang, James Williams, Qinghua Yang, Xiaojun Yuan, Jinlun Zhang, and Yongfei Zhang, July 2024: Predicting September Arctic sea ice: A multi-model seasonal skill comparison. Bulletin of the American Meteorological Society, 105(7), DOI:10.1175/BAMS-D-23-0163.1. Abstract
This study quantifies the state-of-the-art in the rapidly growing field of seasonal Arctic sea ice prediction. A novel multi-model 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 2001–2020 for predictions of Pan-Arctic sea ice extent (SIE), regional SIE, and local sea ice concentration (SIC) initialized on June 1, July 1, August 1, and September 1. This diverse set of statistical and dynamical models can individually predict linearly detrended Pan-Arctic SIE anomalies with skill, and a multi-model 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 three months in advance.
In this study, we perform online sea ice bias correction within a Geophysical Fluid Dynamics Laboratory global ice-ocean model. For this, we use a convolutional neural network (CNN) which was developed in a previous study (Gregory et al., 2023, https://doi.org/10.1029/2023ms003757) for the purpose of predicting sea ice concentration (SIC) data assimilation (DA) increments. An initial implementation of the CNN shows systematic improvements in SIC biases relative to the free-running model, however large summertime errors remain. We show that these residual errors can be significantly improved with a novel sea ice data augmentation approach. This approach applies sequential CNN and DA corrections to a new simulation over the training period, which then provides a new training data set to refine the weights of the initial network. We propose that this machine-learned correction scheme could be utilized for generating improved initial conditions, and also for real-time sea ice bias correction within seasonal-to-subseasonal sea ice forecasts.
Data assimilation is often viewed as a framework for correcting short-term error growth in dynamical climate model forecasts. When viewed on the time scales of climate however, these short-term corrections, or analysis increments, can closely mirror the systematic bias patterns of the dynamical model. In this study, we use convolutional neural networks (CNNs) to learn a mapping from model state variables to analysis increments, in order to showcase the feasibility of a data-driven model parameterization which can predict state-dependent model errors. We undertake this problem using an ice-ocean data assimilation system within the Seamless system for Prediction and EArth system Research (SPEAR) model, developed at the Geophysical Fluid Dynamics Laboratory, which assimilates satellite observations of sea ice concentration every 5 days between 1982 and 2017. The CNN then takes inputs of data assimilation forecast states and tendencies, and makes predictions of the corresponding sea ice concentration increments. Specifically, the inputs are states and tendencies of sea ice concentration, sea-surface temperature, ice velocities, ice thickness, net shortwave radiation, ice-surface skin temperature, sea-surface salinity, as well as a land-sea mask. We find the CNN is able to make skillful predictions of the increments in both the Arctic and Antarctic and across all seasons, with skill that consistently exceeds that of a climatological increment prediction. This suggests that the CNN could be used to reduce sea ice biases in free-running SPEAR simulations, either as a sea ice parameterization or an online bias correction tool for numerical sea ice forecasts.
Because 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.
Research over the past decade has demonstrated that dynamical forecast systems can skillfully predict pan-Arctic sea ice extent (SIE) on the seasonal time scale; however, there have been fewer assessments of prediction skill on user-relevant spatial scales. In this work, we evaluate regional Arctic SIE predictions made with the Forecast-Oriented Low Ocean Resolution (FLOR) and Seamless System for Prediction and Earth System Research (SPEAR_MED) dynamical seasonal forecast systems developed at the NOAA/Geophysical Fluid Dynamics Laboratory. Compared to FLOR, we find that the recently developed SPEAR_MED system displays improved skill in predicting regional detrended SIE anomalies, partially owing to improvements in sea ice concentration (SIC) and thickness (SIT) initial conditions. In both systems, winter SIE is skillfully predicted up to 11 months in advance, whereas summer minimum SIE predictions are limited by the Arctic spring predictability barrier, with typical skill horizons of roughly 4 months. We construct a parsimonious set of simple statistical prediction models to investigate the mechanisms of sea ice predictability in these systems. Three distinct predictability regimes are identified: a summer regime dominated by SIE and SIT anomaly persistence; a winter regime dominated by SIE and upper-ocean heat content (uOHC) anomaly persistence; and a combined regime in the Chukchi Sea, characterized by a trade-off between uOHC-based and SIT-based predictability that occurs as the sea ice edge position evolves seasonally. The combination of regional SIE, SIT, and uOHC predictors is able to reproduce the SIE skill of the dynamical models in nearly all regions, suggesting that these statistical predictors provide a stringent skill benchmark for assessing seasonal sea ice prediction systems.
The continuing decline of the summertime sea ice cover has reduced the sea ice path that must be traversed to Arctic destinations and through the Arctic between the Atlantic and Pacific Oceans, stimulating interest in trans–Arctic Ocean routes. Seasonal prediction of the sea ice cover along these routes could support the increasing summertime ship traffic taking advantage of recent low ice conditions. We introduce the minimum Arctic sea ice path (MIP) between Atlantic and Pacific Oceans as a shipping-relevant metric that is amenable to multidecadal hindcast evaluation. We show, using 1992–2017 retrospective predictions, that bias correction is necessary for the GFDL Seamless System for Prediction and Earth System Research (SPEAR) forecast system to improve upon damped persistence seasonal forecasts of summertime daily MIP between the Atlantic and Pacific Oceans both east and west of Greenland, corresponding roughly to the Northeast and Northwest Passages. Without bias correction, only the Northwest Passage MIP forecasts have lower error than a damped persistence forecast. Using the forecast ensemble spread to estimate a lower bound on forecast error, we find large opportunities for forecast error reduction, especially at lead times of less than 2 months. Most of the potential improvement remains after linear removal of climatological and trend biases, suggesting that significant error reduction might come from improved initialization and simulation of subannual variability. Using a different passive microwave sea ice dataset for calculating error than was used for data assimilation increases the raw forecast errors but not the trend anomaly forecast errors.
The current GFDL seasonal prediction system, the Seamless System for Prediction and Earth System Research (SPEAR), has shown skillful prediction of Arctic sea ice extent with atmosphere and ocean constrained by observations. In this study we present improvements in subseasonal and seasonal predictions of Arctic sea ice by directly assimilating sea ice observations. The sea ice initial conditions from a data assimilation (DA) system that assimilates satellite sea ice concentration (SIC) observations are used to produce a set of reforecast experiments (IceDA) starting from the first day of each month from 1992 to 2017. Our evaluation of daily sea ice extent prediction skill concludes that the SPEAR system generally outperforms the anomaly persistence forecast at lead times beyond 1 month. We primarily focus our analysis on daily gridcell-level sea ice fields. SIC DA improves prediction skill of SIC forecasts prominently in the June-, July-, August-, and September-initialized reforecasts. We evaluate two additional user-oriented metrics: the ice-free probability (IFP) and ice-free date (IFD). IFP is the probability of a grid cell experiencing ice-free conditions in a given year, and IFD is the first date on which a grid cell is ice free. A combined analysis of IFP and IFD demonstrates that the SPEAR model can make skillful predictions of local ice melt as early as May, with modest improvements from SIC DA.
Compared to the Arctic, seasonal predictions of Antarctic sea ice have received relatively little attention. In this work, we utilize three coupled dynamical prediction systems developed at the Geophysical Fluid Dynamics Laboratory to assess the seasonal prediction skill and predictability of Antarctic sea ice. These systems, based on the FLOR, SPEAR_LO, and SPEAR_MED dynamical models, differ in their coupled model components, initialization techniques, atmospheric resolution, and model biases. Using suites of retrospective initialized seasonal predictions spanning 1992–2018, we investigate the role of these factors in determining Antarctic sea ice prediction skill and examine the mechanisms of regional sea ice predictability. We find that each system is capable of skillfully predicting regional Antarctic sea ice extent (SIE) with skill that exceeds a persistence forecast. Winter SIE is skillfully predicted 11 months in advance in the Weddell, Amundsen/Bellingshausen, Indian, and west Pacific sectors, whereas winter skill is notably lower in the Ross sector. Zonally advected upper-ocean heat content anomalies are found to provide the crucial source of prediction skill for the winter sea ice edge position. The recently developed SPEAR systems are more skillful than FLOR for summer sea ice predictions, owing to improvements in sea ice concentration and sea ice thickness initialization. Summer Weddell SIE is skillfully predicted up to 9 months in advance in SPEAR_MED, due to the persistence and drift of initialized sea ice thickness anomalies from the previous winter. Overall, these results suggest a promising potential for providing operational Antarctic sea ice predictions on seasonal time scales.
The current GFDL seasonal prediction system achieved retrospective sea ice extent (SIE) skill without direct sea ice data assimilation. Here we develop sea ice data assimilation, shown to be a key source of skill for seasonal sea ice predictions, in GFDL’s next-generation prediction system, the Seamless System for Prediction and Earth System Research (SPEAR). Satellite sea ice concentration (SIC) observations are assimilated into the GFDL Sea Ice Simulator version 2 (SIS2) using the ensemble adjustment Kalman filter (EAKF). Sea ice physics is perturbed to form an ensemble of ice–ocean members with atmospheric forcing from the JRA-55 reanalysis. Assimilation is performed every 5 days from 1982 to 2017 and the evaluation is conducted at pan-Arctic and regional scales over the same period. To mitigate an assimilation overshoot problem and improve the analysis, sea surface temperatures (SSTs) are restored to the daily Optimum Interpolation Sea Surface Temperature version 2 (OISSTv2). The combination of SIC assimilation and SST restoring reduces analysis errors to the observational error level (~10%) from up to 3 times larger than this (~30%) in the free-running model. Sensitivity experiments show that the choice of assimilation localization half-width (190 km) is near optimal and that SIC analysis errors can be further reduced slightly either by reducing the observational error or by increasing the assimilation frequency from every 5 days to daily. A lagged-correlation analysis suggests substantial prediction skill improvements from SIC initialization at lead times of less than 2 months.
The next‐generation seasonal prediction system is built as part of the Seamless System for Prediction and EArth System Research (SPEAR) at the Geophysical Fluid Dynamics Laboratory (GFDL) of the National Oceanic and Atmospheric Administration (NOAA). SPEAR is an effort to develop a seamless system for prediction and research across time scales. The ensemble‐based ocean data assimilation (ODA) system is updated for Modular Ocean Model Version 6 (MOM6), the ocean component of SPEAR. Ocean initial conditions for seasonal predictions, as well as an ocean state estimation, are produced by the MOM6 ODA system in coupled SPEAR models. Initial conditions of the atmosphere, land, and sea ice components for seasonal predictions are constructed through additional nudging experiments in the same coupled SPEAR models. A bias correction scheme called ocean tendency adjustment (OTA) is applied to coupled model seasonal predictions to reduce model drift. OTA applies the climatological temperature and salinity increments obtained from ODA as three‐dimensional tendency terms to the MOM6 ocean component of the coupled SPEAR models. Based on preliminary retrospective seasonal forecasts, we demonstrate that OTA reduces model drift—especially sea surface temperature (SST) forecast drift—in coupled model predictions and improves seasonal prediction skill for applications such as El Niño–Southern Oscillation (ENSO).
