Abstract

The mechanisms for and predictability of a drastic reduction in the Arctic sea ice extent (SIE) are investigated using the Model for Interdisciplinary Research on Climate (MIROC) version 5.2. Here, a control (CTRL) with forcing fixed at year 2000 levels and perfect-model ensemble prediction (PRED) experiments are conducted. In CTRL, three (model years 51, 56, and 57) drastic SIE reductions occur during a 200-yr-long integration. In year 56, the sea ice moves offshore in association with a positive phase of the summer Arctic dipole anomaly (ADA) index and melts due to heat input through the increased open water area, and the SIE drastically decreases. This provides the preconditioning for the lowest SIE in year 57 when the Arctic Ocean interior is in a warm state and the spring sea ice volume has a large negative anomaly due to drastic ice reduction in the previous year. Although the ADA is one of the key mechanisms behind sea ice reduction, it does not always cause a drastic reduction. Our analysis suggests that wind direction favoring offshore ice motion is a more important factor for drastic ice reduction events. In years experiencing drastic ice reduction events, the September SIE can be skillfully predicted in PRED started from July, but not from April. This is because the forecast errors for the July sea level pressure and those for the sea ice concentration and sea ice thickness along the ice edge are large in PRED started from April.

1. Introduction

The Arctic summer sea ice extent (SIE) has markedly decreased since satellite observations began in the late 1970s. For September Arctic sea ice, the 10 lowest minimum extents have occurred since 2000 (NSIDC 2017). In particular, in the summers of 2007 and 2012, extreme sea ice loss was observed. The September average SIE in 2007 and 2012 was 4.27 and 3.57 million km2 (−2.14 and −2.84 million km2 anomaly from the 1981–2010 average), which have been ranked as the second lowest and lowest recorded values, respectively (NSIDC 2018).

Various physical mechanisms have been proposed for explaining such extreme sea ice reduction events. The 2007 reduction was mainly triggered by a dipole anomaly pattern in the sea level pressure (Wang et al. 2009; Zhang et al. 2008; Serreze and Stroeve 2015) and by the Pacific water inflow (Woodgate et al. 2010). In addition, the ice thickness in March, the wind stresses in May to June, and the 2-m temperature in September have been identified as key contributors (Kauker et al. 2009). For the 2012 reduction, the recent thinning of sea ice thickness (or decreasing of multiyear ice) due to the Arctic Ocean warming (Kwok and Rothrock 2009; Comiso 2012; Polyakov et al. 2012) and a great cyclone over the Arctic Ocean (Simmonds and Rudeva 2012; Zhang et al. 2013) have been suggested as drivers. Moreover, the sea ice memory and the positive feedback under warm atmospheric conditions contributed to the exceptional sea ice loss in 2012 (Guemas et al. 2013).

These extreme sea ice losses have occurred under the ongoing warming of the Arctic. Both natural and anthropogenic forcings have influenced the observed sea ice decline (Kay et al. 2011; Notz and Marotzke 2012; Day et al. 2012). However, their contributions are not well understood, partly because the sea ice records from satellites are short. The role of the internal variability in modulating the downward trend in the SIE has been suggested to be significant (Swart et al. 2015) and internal variability, which controls most of the Arctic atmospheric circulation changes, accounts for about 30%–50% of the decline in the September SIE (Ding et al. 2017). Although the observed SIE minimum record in 2012 is likely caused by a combination of anthropogenic and natural forcings (Zhang and Knutson 2013), further investigations are needed to assess the role of internal variability in drastic reduction events and to improve Arctic sea ice forecasting systems.

Regarding the predictability of the Arctic sea ice, many studies have used climate models to consider forecasts on seasonal to interannual time scales (Blanchard-Wrigglesworth et al. 2011; Chevallier et al. 2013; Wang et al. 2013; Sigmond et al. 2013; Day et al. 2014; Msadek et al. 2014; Guemas et al. 2016a,b; Bushuk et al. 2017; Ono et al. 2018). The predictability of extreme sea ice anomalies has also been investigated with perfect model ensemble predictions. For example, Holland et al. (2011) assessed the inherent predictability of Arctic sea ice focusing on the role of the preconditioning of ice and natural variation in extreme SIE reduction and showed that models can capture the occurrence of sea ice loss events but not their amplitude. In addition, Tietsche et al. (2013) investigated the predictability of two extreme anomalies for the present day in a historical run and for the middle of the twenty-first century in the RCP4.5 experiment and showed that although the onset and amplitude of extreme anomalies are not predictable, the sea ice reduction after the onset can be predicted 1 year ahead. However, the previous studies include the influence of the warming trend, and thus mechanisms (or sources) for the drastic sea ice reduction associated with the internal variability are not fully understood.

Motivated by these previous studies, the present study examines the spatial pattern of extreme sea ice reduction under radiative forcing fixed at the present-day level (i.e., without the warming trend), based on the Arctic Prediction and Predictability on Seasonal-to-Interannual Time Scales (APPOSITE) project, the main goal of which is to quantify at what time scales the Arctic climate is predictable (Day et al. 2016). Furthermore, by comparing a control simulation with perfect model ensemble predictions, the predictability of a drastic reduction in SIE and possible mechanisms for this reduction will be discussed. In Day et al. (2016), a detailed description of the APPOSITE project is provided; however, the predictability of extreme events is not investigated. Thus, this study contributes to a better understanding of the predictability of drastic sea ice reduction events.

2. Methods

a. Climate model

The climate model used in this study is the Model for Interdisciplinary Research on Climate (MIROC) version 5.2 (Tatebe et al. 2018), which is a low-resolution and minor-update version of MIROC5 (Watanabe et al. 2010) that contributed to phase 5 of the Coupled Model Intercomparison Project and the Intergovernmental Panel on Climate Change Fifth Assessment Report (IPCC 2013). The horizontal resolution of the atmospheric component is a T42 spectral truncation (about 300 km), and there are 40 vertical levels up to 3 hPa. The warped bipolar horizontal coordinate system of the MIROC5 oceanic component has been replaced by a tripolar coordinate system. The oceanic component had 1° longitudinal grid spacing in the spherical coordinate portion south of 63°N. The meridional grid spacing varies from about 0.5° near the equator to 1° in the midlatitudes. There are 63 vertical levels, the lowermost level of which is located at 6300-m depth. The sea ice component implements one-layer thermodynamics (Bitz and Lipscomb 1999), elastic–viscous–plastic rheology (Hunke and Dukowicz 1997), and a subgrid ice thickness distribution (Bitz et al. 2001) with five categories. Snow cover on sea ice in the model affects the thermodynamic processes through changes in the surface albedo and vertical heat fluxes. The detailed framework and parameters have been described in Komuro et al. (2012). In the land surface model, parameterizations of a subgrid-scale snow cover distribution (Liston 2004; Nitta et al. 2014) and a simple wetland scheme (Nitta et al. 2017) have been newly implemented into MIROC5.2. An improved treatment of the turbulent kinetic energy input from the atmosphere (Komuro 2014) is also adopted in the Arctic Ocean sea ice area.

b. Experimental designs

Using MIROC5.2, we performed a simulation with radiative forcing fixed at present-day levels (year 2000) within the APPOSITE project (Day et al. 2016). After a spinup of 1000 years, the model is run for a further 200 years, which are arbitrarily labeled as 1 to 200. This experiment will be referred to hereinafter as CTRL. For all variables, the monthly mean values during the period of 1 to 200 are used for the climatology and their anomalies are defined as the deviation from the climatology.

To diagnose the predictability of a drastic reduction in the Arctic sea ice, we conducted a series of perfect model ensemble prediction experiments (PRED). The target years are 51, 56, and 57 as discussed in section 3. The start dates are 1 April and 1 July in the same year. The PREDs with start dates of 1 April are referred to as PRED.APR, and those with 1 July as PRED.JUL. An ensemble of eight members is generated for each start date. The initial conditions are taken from the CTRL and each member differs only by a perturbation of the sea surface temperature, which is generated by spatially uncorrelated Gaussian noise with a standard deviation of 10−4 K under the APPOSITE protocol. Each ensemble is run for 3 years and 9 months from 1 April and 3 years and 6 months from 1 July.

When conducting sea ice forecasts with coupled global climate models, forecast errors arise from errors in the initial conditions and model biases. In the perfect model experiments, however, the model can predict itself with perfect initial conditions and no biases (Collins 2002; Day et al. 2014; Tietsche et al. 2014; Blanchard-Wrigglesworth et al. 2015). Therefore, an upper bound of the predictability can be estimated with climate models.

3. Control experiment

a. Model performances

The features of the SIE and sea ice volume (SIV) from CTRL are examined (Fig. 1). In this study, the SIE and SIV are defined as the cumulative area of grid cells with at least 15% sea ice concentration (SIC) and the sum of grid cell volumes obtained by multiplying the ice thickness by the ice concentration, respectively. Interannual variability in SIE is positively correlated with that in SIV for the 200 model years. The correlation coefficients are higher (r = 0.61–0.82) in June to October and lower (r = 0.47–0.51) in November to May. In addition, a period of drastic ice reduction can be seen around year 57. These events are described in detail in section 3b.

Fig. 1.

Time series of the monthly mean (a) SIE and (b) SIV in a 200-yr CTRL experiment.

Fig. 1.

Time series of the monthly mean (a) SIE and (b) SIV in a 200-yr CTRL experiment.

Here, the basic performance of the model used in this study is examined based on the mean state, variability, and diagnostic predictability (persistence) of sea ice. The model is also evaluated against observations [NSIDC, HadISST (Rayner et al. 2003), and ProjD (Ishii and Kimoto 2009)]. Note that one of the reasons for differences between the model and observations could be that the model was not run under the historical forcing, and instead the forcing was fixed at year 2000 levels. The seasonal cycle of the SIE monthly climatology in MIROC5.2 is smaller than the observations (NSIDC and HadISST) by 1–3 million km2 except for August to November (Fig. 2a). This is partly because the sea ice coverage is small in the Bering and Okhotsk Seas (not shown), and these regions have model biases in MIROC5.2 as documented by Komuro et al. (2012). The simulated SIE is smaller than the ProjD observations by 1–2 million km2 from January to April, and larger by ~1 million km2 from July to October due to the differences in the Barents and Kara Seas. The standard deviations of the SIE in MIROC5.2 are larger in all months than those in the observations by 0.1–0.2 million km2 (Fig. 2b). This implies that there is more variability in the model than in the observations. Following Day et al. (2014), the lagged correlation of the SIE in MIROC5.2 and three observational datasets for comparison is shown in Figs. 2c–f for each start month against different lead times. The linear trend was removed from the time series of SIE of MIROC5.2 and the observations before the correlations were calculated. The lagged correlations in MIROC5.2 are higher than those in the observations at all lead times except for 0 months. This could be due to sampling error, detrending, autocorrelation, or inadequate representation of processes in MIROC5.2, as suggested by Day et al. (2014). The reemergence in winter is stronger in MIROC5.2 than in the observations. In addition, the pattern of the lagged correlation in MIROC5.2 is similar to that in the Hadley Centre Global Environment Model version 1.2 (HadGEM1.2; Shaffrey et al. 2009) [see Fig. 1 of Day et al. (2014)].

Fig. 2.

Seasonal cycle of (a) the monthly mean SIE and (b) standard deviation of the SIE anomaly in a CTRL experiment. The NSIDC (red), HadISST (green; Rayner et al. 2003), and ProjD (blue; Ishii and Kimoto 2009) observations of the SIE from 1980 to 2009 are also included for comparison. Also shown are lagged correlations of the SIE anomaly from (c) a CTRL simulation, (d) NSIDC, (e) HadISST, and (f) ProjD observations, for each start month, against lead time as in Day et al. (2014). These data were linearly detrended before the calculations.

Fig. 2.

Seasonal cycle of (a) the monthly mean SIE and (b) standard deviation of the SIE anomaly in a CTRL experiment. The NSIDC (red), HadISST (green; Rayner et al. 2003), and ProjD (blue; Ishii and Kimoto 2009) observations of the SIE from 1980 to 2009 are also included for comparison. Also shown are lagged correlations of the SIE anomaly from (c) a CTRL simulation, (d) NSIDC, (e) HadISST, and (f) ProjD observations, for each start month, against lead time as in Day et al. (2014). These data were linearly detrended before the calculations.

Similarly, the mean state and standard deviation for the SIV are shown in Figs. 3a and 3b, and are almost the same as those indicated by magenta lines in Figs. 4b and 4d of Day et al. (2016) except for the length of the simulation. The SIV in MIROC5.2 is larger by 3000–5000 km3 than in PIOMAS (Pan-Arctic Ice-Ocean Modeling and Assimilation System; Zhang and Rothrock 2003; Schweiger et al. 2011) for all months (Fig. 3a). The standard deviation of the SIV is close to 1500 km3 throughout the year and is larger than in the PIOMAS data (Fig. 3b). The diagnostic predictability (persistence) of the SIV (Fig. 3c) is higher than that of SIE (Fig. 2c) and the correlation pattern with lead times seems to be similar to that in HadGEM1.2 and the Max Planck Institute Earth System Model (MPI-ESM; Jungclaus et al. 2013) [see Fig. 5 of Day et al. (2014)].

Fig. 3.

Seasonal cycle of (a) the monthly mean SIV and (b) standard deviation of the SIV anomaly in a CTRL experiment. The values for the PIOMAS data (magenta; Schweiger et al. 2011) from 1980 to 2009 are also included for comparison. (c) Lagged correlation of the SIV anomaly from a CTRL experiment, for each start month, against lead time as in Day et al. (2014). These data were linearly detrended before the calculations.

Fig. 3.

Seasonal cycle of (a) the monthly mean SIV and (b) standard deviation of the SIV anomaly in a CTRL experiment. The values for the PIOMAS data (magenta; Schweiger et al. 2011) from 1980 to 2009 are also included for comparison. (c) Lagged correlation of the SIV anomaly from a CTRL experiment, for each start month, against lead time as in Day et al. (2014). These data were linearly detrended before the calculations.

b. Drastic ice reduction events

To define an extreme sea ice loss event, years in which both the anomaly and year-to-year change in the September SIE were less than minus two standard deviations were extracted from the time series of CTRL. Figure 4 shows that three (model years 51, 56, and 57) drastic ice reductions occur even though the radiative forcing is fixed at the present-day (year 2000) level. Previously in a 600-yr control simulation of the Community Climate System Model version 3, Cullather and Tremblay (2008) reported that three extraordinary minima occurred. In the 200-yr CTRL of this study, however, three events occur within just 7 years, and therefore they are likely to be related to each other. These individual events and their composite are examined below in more detail.

Fig. 4.

Time series of the September (a) SIE anomaly (106 km2) relative to a 200-yr climatology and (b) year-to-year change (106 km2) in a 200-yr CTRL experiment. The three vertical lines in (a) and (b) denote years 51, 56, and 57. In (a) and (b), plus or minus two standard deviations are indicated by horizontal dashed lines. These data were linearly detrended before the calculations.

Fig. 4.

Time series of the September (a) SIE anomaly (106 km2) relative to a 200-yr climatology and (b) year-to-year change (106 km2) in a 200-yr CTRL experiment. The three vertical lines in (a) and (b) denote years 51, 56, and 57. In (a) and (b), plus or minus two standard deviations are indicated by horizontal dashed lines. These data were linearly detrended before the calculations.

Overall, the Arctic sea ice in September has a high concentration and thickness from the Chukchi and Beaufort Seas to the western central Arctic Ocean, and a low concentration and thickness from the Barents Sea through the Kara and Laptev Seas to the East Siberian Sea (Figs. 5a,c). Figures 5b and 5d show that in all years the sea ice edge retreats poleward in the Laptev and East Siberian Seas, and thus the SIC is lower than the climatology by 20%–60% along the ice edge and the corresponding sea ice thickness (SIT) is thinner by 0.5–1.0 m. Moreover, the SIC and SIT drastically decrease in the northern part of the Barents and Kara Seas in year 51, in the Chukchi Sea in year 56, and in the Chukchi and Beaufort Seas in year 57. As a result, a drastic sea ice reduction was identified in these three years. In particular, the year 57 SIC and SIT anomalies are the largest for all regions of the Arctic Ocean.

Fig. 5.

(a) SIC, (b) SIC anomaly, (c) SIT, and (d) SIT anomaly in September (top to bottom) for years 51, 56, and 57 and for the composite of the three years. The black lines in all panels indicate the 15% SIC contours from a CTRL experiment. The red line in the composite of the SIC indicates the 15% SIC contours in September 2007 from the ProjD observations. The SIC and SIT anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

Fig. 5.

(a) SIC, (b) SIC anomaly, (c) SIT, and (d) SIT anomaly in September (top to bottom) for years 51, 56, and 57 and for the composite of the three years. The black lines in all panels indicate the 15% SIC contours from a CTRL experiment. The red line in the composite of the SIC indicates the 15% SIC contours in September 2007 from the ProjD observations. The SIC and SIT anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

The composite of the three events also shows that the Arctic sea ice retreats noticeably in the Pacific sector of the Arctic Ocean, while the large negative anomalies in the SIC and SIT are overall lessened due to the compensation of the opposite sign anomaly. Compared with the observations (red contours in Fig. 5a), the ice edge in the Pacific side of the Arctic Ocean is located farther south and sea ice remains in the Kara Sea partly because of the model topography, while the spatial distribution of sea ice is similar to that observed in September 2007. Previous studies have revealed various causes for the record minimum sea ice in September 2007. For example, anomalous atmospheric circulations, such as the Arctic dipole anomaly (ADA) (Wang et al. 2009), the thinning of ice cover in spring (Stroeve et al. 2008), and the enhanced inflow of warm Pacific water (Woodgate et al. 2006, 2010), are thought to be contributing factors. Possible mechanisms for the extreme ice reduction in CTRL are next investigated based on the results for years 51, 56, and 57 and their composite.

c. Possible mechanisms

The sea level pressure (SLP) in the summer [June–August (JJA)] for years 56 and 57 is characterized by a high pressure anomaly over Greenland and the Canada Basin and a low pressure anomaly over the Kara and Laptev Seas (Fig. 6a). This ADA-like pattern causes winds to blow from the Siberian and Alaskan coasts and sea ice is moved offshore by anomalous winds (see vectors in Fig. 6b), and thereby the ice edge retreats greatly from July (blue) to August (red). In August, the heat flux from the atmosphere to the ocean has a large positive anomaly in the increased open water areas (Fig. 6c). Correspondingly the sea surface salinity (SSS) decreases by 0.5 to 2.0 psu near the sea ice edge between July and August (Fig. 6d), indicating a freshening due to increased sea ice melting. It is therefore suggested that the sea ice retreat from July to August is triggered by anomalous winds associated with the ADA-like pattern and then further ice melt occurs due to the increased heat input through the open water (namely, ice–albedo feedback), leading to the drastic ice reductions in September of years 56 and 57. In year 51, the driver for sea ice retreat is different from that described above because the SLP anomaly does not show an ADA-like pattern. However, the sea ice in July moves away from the East Siberian Sea and the Laptev Sea to the north, and therefore the ice edge retreats poleward from July to August. This is partly because of a small dipole anomaly that is characterized by a weak high pressure anomaly over the Canada Basin and a low pressure anomaly across northern Eurasia.

Fig. 6.

(a) JJA SLP anomaly, (b) sea ice vectors in July, (c) heat flux anomaly in August, and (d) sea surface salinity changes from July to August (top to bottom) for years 51, 56, and 57 and for the composite of the three years. In (a) and (b), 15% SIC contours in July (blue) and August (red) are superimposed. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

Fig. 6.

(a) JJA SLP anomaly, (b) sea ice vectors in July, (c) heat flux anomaly in August, and (d) sea surface salinity changes from July to August (top to bottom) for years 51, 56, and 57 and for the composite of the three years. In (a) and (b), 15% SIC contours in July (blue) and August (red) are superimposed. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

As shown in Fig. 6, the ADA-like pattern is suggested to be a key driver of the extreme ice reduction, except in year 51. To investigate the relationship between the September SIE and the ADA, an ADA index is defined using the SLP in CTRL (Fig. 7a). Following Wang et al. (2009), an empirical orthogonal function (EOF) analysis of the summer (JJA) SLP north of 70°N was conducted and the second mode, which accounts for 17.7% of the variance, is shown in Fig. 7b. In this study, the ADA index is defined as the difference in the area-averaged SLP values over the center of action (Greenland and the western central Arctic, indicated by dots in Fig. 7b), instead of the EOF time series, and these values are normalized by their standard deviation. A positive value of the ADA index means favorable conditions for sea ice retreat, especially, in the western part of the Arctic Ocean. There is no significant correlation between the ADA index and the September SIE anomaly (Fig. 7a; r = 0.03). In this study, statistical significance of a correlation coefficient at a 95% confidence level is based on a two-sided Student’s t test. While the ADA index has significant correlation with the ice volume export from the Fram Strait (r = 0.53), the correlation between the September SIE anomaly and the ice volume export from the Fram Strait is only 0.02. The regressed SIC anomalies onto the ADA index are negative in the western Kara Sea and the northern parts of the East Siberian and Chukchi Seas and are positive in the eastern Kara Sea, the Fram Strait, and the Greenland Sea (Fig. 7c). These spatial patterns in the SIC are similar to those in the SIC for year 56 (see Fig. 5). This indicates that the SIC in the Pacific Sector of the Arctic Ocean decreases by 2%–8% per one standard deviation of the ADA index. For year 56, the SIC decreases by 5%–18% in this region, which accounts for about half of the negative SIC anomaly in year 56.

Fig. 7.

(a) Time series of the normalized September SIE (black), Arctic dipole anomaly (ADA) index (red), and ice volume export from the Fram Strait (blue), obtained from a 200-yr CTRL experiment. Four vertical lines denote the model years 28, 51, 56, and 57. (b) Spatial pattern of the second EOF mode for the JJA SLP anomaly north of 70°N. (c) Correlation (colors) and regression (contours; %) coefficients between the SIC anomaly in September (%) and the normalized ADA index. Note that the sign of the ADA index is reversed in (a). White dots in (b) denote grid points used for calculating the ADA index. Stippling in (c) indicates areas where the correlation coefficient is not statistically significant at the 95% confidence level. These data were linearly detrended before the calculations.

Fig. 7.

(a) Time series of the normalized September SIE (black), Arctic dipole anomaly (ADA) index (red), and ice volume export from the Fram Strait (blue), obtained from a 200-yr CTRL experiment. Four vertical lines denote the model years 28, 51, 56, and 57. (b) Spatial pattern of the second EOF mode for the JJA SLP anomaly north of 70°N. (c) Correlation (colors) and regression (contours; %) coefficients between the SIC anomaly in September (%) and the normalized ADA index. Note that the sign of the ADA index is reversed in (a). White dots in (b) denote grid points used for calculating the ADA index. Stippling in (c) indicates areas where the correlation coefficient is not statistically significant at the 95% confidence level. These data were linearly detrended before the calculations.

However, the September SIE is not always extreme negative in years with a high positive ADA index (Table 1). If a threshold of 0.6 is used following Wang et al. (2009), there are 37 cases that have an ADA index higher than 0.6 standard deviations. Among them, only two cases (years 56 and 57) lead to extreme sea ice loss. To consider the reason why sea ice does not necessarily decrease drastically when there is a high ADA index, results for year 28, which had the highest ADA index (2.50), are shown in Fig. 8. The ice edge retreats from July to August in the Pacific sector (Fig. 8f). However, the SIC and SIT have positive anomalies in the northern Barents Sea and Laptev Sea (Figs. 8a–d), which is different from years 51, 56, and 57. As shown in Fig. 8e, the ADA structure is clearly formed, but sea ice is advected from the Barents Sea through the Kara Sea to the Laptev Sea under the influence of another dipole anomaly that consists of a negative anomaly over the western central Arctic and positive (but weak) anomaly over the Siberian continent. Furthermore, the ice edge extends toward the East Siberian Sea from August to September due to the ice advection (not shown). This is one of the reasons why the sea ice does not decrease drastically in year 28. This suggests that atmospheric patterns with wind directions that cause offshore ice motion are a key driver of extreme ice reduction, in addition to the ADA.

Table 1.

September SIE anomaly, normalized ADA index, Fram Strait ice volume export, spring SIV anomaly, Arctic Ocean state anomaly, and horizontal ocean heat transport anomaly from the Atlantic and Pacific Oceans, for three extreme years. The OHT anomaly from the Atlantic Ocean is shown for 4 years before (years 47, 52, and 53 for the 51, 56, and 57 ice reduction events).

September SIE anomaly, normalized ADA index, Fram Strait ice volume export, spring SIV anomaly, Arctic Ocean state anomaly, and horizontal ocean heat transport anomaly from the Atlantic and Pacific Oceans, for three extreme years. The OHT anomaly from the Atlantic Ocean is shown for 4 years before (years 47, 52, and 53 for the 51, 56, and 57 ice reduction events).
September SIE anomaly, normalized ADA index, Fram Strait ice volume export, spring SIV anomaly, Arctic Ocean state anomaly, and horizontal ocean heat transport anomaly from the Atlantic and Pacific Oceans, for three extreme years. The OHT anomaly from the Atlantic Ocean is shown for 4 years before (years 47, 52, and 53 for the 51, 56, and 57 ice reduction events).
Fig. 8.

(a) SIC, (b) SIC anomaly, (c) SIT, and (d) SIT anomaly in September, (e) JJA SLP anomaly, (f) sea ice vectors in July, (g) heat flux anomaly in August, and (h) sea surface salinity changes from July to August, for year 28. In (a)–(d), the 15% SIC contour in September is superimposed as a black line. In (f) and (h), the 15% SIC contours in July and August are superimposed as blue and red lines, respectively. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

Fig. 8.

(a) SIC, (b) SIC anomaly, (c) SIT, and (d) SIT anomaly in September, (e) JJA SLP anomaly, (f) sea ice vectors in July, (g) heat flux anomaly in August, and (h) sea surface salinity changes from July to August, for year 28. In (a)–(d), the 15% SIC contour in September is superimposed as a black line. In (f) and (h), the 15% SIC contours in July and August are superimposed as blue and red lines, respectively. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

d. Effects of preconditioning

Other factors have been identified that contribute to drastic ice reduction. As suggested by Hutchings and Perovich (2015), thinner (thicker) sea ice and warmer (colder) ocean temperatures are favorable conditions for sea ice retreat (advance). In the real world, the summer upper ocean has warmed over the last decade and enhanced the sea ice melt (Steele et al. 2008, 2010). In addition, the thinning of sea ice in spring and the heat flux from the Pacific water have been thought to be key factors for the drastic ice reduction observed in 2007 and 2012 (Stroeve et al. 2008; Woodgate et al. 2010; Serreze and Stroeve 2015). Hence, the effects of preconditioning on extreme sea ice loss are examined here, based on the oceanic and sea ice states (Fig. 9). In this study, the spring [March–May (MAM)] mean SIV anomaly is used as an indicator for the sea ice state and the summer potential temperature averaged over the region north of 65°N for the oceanic state. The spring SIV anomaly has significant correlation with the subsequent September SIE anomaly (Figs. 9a,b). The lagged correlation between the spring SIV and the September SIE is greatest when the September SIE leads by 1 year (black line in Fig. 10). Focusing on the years of drastic ice reduction (51, 56, and 57), the spring SIV greatly decreases after the drastic reduction in the September SIE. This means that a delay in the fall to winter ice growth due to the increased open water areas causes the reduction of the thicker ice in spring, as inferred from a previous study (Cullather and Tremblay 2008). Although the spring SIV anomalies for the years 51 and 56 are positive, 1110 and 370 km3, respectively (Table 1), that for the year 57 has a large negative value of −2110 km3 due to the drastic ice reduction in year 56. This indicates that the large negative anomaly in the spring SIV played a role in the preconditioning for the drastic ice reduction in year 57.

Fig. 9.

Time series of (a) the September SIE anomaly and (b) the spring (MAM) SIV anomaly. (c) Depth–time plots of the summer (JJA) potential temperature anomaly averaged over the region north of 65°N. Time series of (d) depth-averaged JJA potential temperature anomaly [Arctic Ocean state (AOS)] and the annual integrated horizontal ocean heat transport (OHT) anomaly from (e) the Atlantic Ocean across the circle of 60°N and (f) the Pacific Ocean across the Bering Strait. The positive (negative) value is the northward (southward) OHT. In (d)–(f), thick black lines denote the 11-yr running average. Black vertical lines indicate the model years 51, 56, and 57. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

Fig. 9.

Time series of (a) the September SIE anomaly and (b) the spring (MAM) SIV anomaly. (c) Depth–time plots of the summer (JJA) potential temperature anomaly averaged over the region north of 65°N. Time series of (d) depth-averaged JJA potential temperature anomaly [Arctic Ocean state (AOS)] and the annual integrated horizontal ocean heat transport (OHT) anomaly from (e) the Atlantic Ocean across the circle of 60°N and (f) the Pacific Ocean across the Bering Strait. The positive (negative) value is the northward (southward) OHT. In (d)–(f), thick black lines denote the 11-yr running average. Black vertical lines indicate the model years 51, 56, and 57. Anomalies were relative to a 200-yr climatology and were linearly detrended before the calculations.

Fig. 10.

Lagged correlation coefficients between the September SIE anomaly and the spring SIV anomaly (black line with closed circles), those between the September SIE anomaly and OHT anomaly from the Atlantic (red line with closed circles) and Pacific (blue line with closed circles), and those between the September SIE anomaly and the depth-averaged summer potential temperature anomaly (green line with closed circles). Horizontal dashed lines denote statistical significance at the 95% confidence level with 66 degrees of freedom based on a two-sided Student’s t test. These data were linearly detrended before computing the correlations.

Fig. 10.

Lagged correlation coefficients between the September SIE anomaly and the spring SIV anomaly (black line with closed circles), those between the September SIE anomaly and OHT anomaly from the Atlantic (red line with closed circles) and Pacific (blue line with closed circles), and those between the September SIE anomaly and the depth-averaged summer potential temperature anomaly (green line with closed circles). Horizontal dashed lines denote statistical significance at the 95% confidence level with 66 degrees of freedom based on a two-sided Student’s t test. These data were linearly detrended before computing the correlations.

The potential temperature in the Arctic Ocean interior is characterized by the strong warm (cold) anomaly from the surface to a 20-m depth, due to the heat gain (loss) associated with the negative (positive) SIE anomaly, and the internal variability on longer time scales (Figs. 9a,c). For statistical analyses, the Arctic Ocean state (AOS) (Fig. 9d) is defined as the vertically averaged potential temperature anomaly from 0 to 154 m based on Fig. 9c. The AOS has a warm anomaly in the three extreme reduction years (Table 1). On the interannual time scale, the September SIE anomaly is significantly correlated with the AOS at lags of −1 to 4 years (green line in Fig. 10). This suggests that the ocean state in the Arctic Ocean is influenced by the September SIE anomaly a year before, and then plays a role in preconditioning the sea ice variability until 4 years after. In addition, large negative anomalies in the SIE during years 56 to 70 are consistent with the warm ocean state (Figs. 9a,c), indicating that the internal low-frequency variability of the ocean state in the Arctic Ocean interior contributed to delay the recovery from the negative anomaly in the SIE.

The interannual and longer variabilities in the AOS are likely caused by the horizontal ocean heat transport (OHT) from the North Atlantic across the circle of 60°N and the North Pacific across the Bering Strait (Figs. 9d–f). Focusing on the interannual time scale, the September SIE anomaly is significantly correlated with the OHT from the Pacific Ocean at a lag of 0 years and from the Atlantic Ocean at a lag of 4 years (blue and red lines in Fig. 10). The variability in the OHT in the Pacific and Atlantic Oceans leads that in the September SIE by 0–4 years and played a role in preconditioning the sea ice loss through the ocean temperature anomalies in the Arctic Ocean. The September SIE is therefore influenced by the ocean temperature anomalies in the Arctic Ocean with 0- to 4-yr time scales through the OHT variability. Compared to a previous study (Holland et al. 2006), the time lag for the OHT to the Arctic Ocean is longer by 2–3 years. For years 56 and 57, the positive OHT anomalies from the Atlantic Ocean contribute to the warmer state in the Arctic Ocean (Table 1), which is a favorable condition for sea ice loss. For year 51, although the Arctic Ocean state has a small positive anomaly, the spring SIV and ADA index are unfavorable for sea ice loss. Compared to years 56 and 57, the retreat of the ice edge in the region north of Svalbard partly contributes to the drastic ice reduction for year 51 (Figs. 5 and 6).

4. Perfect model ensemble prediction experiments

Here, whether and to what extent extreme sea ice loss events can be predicted with climate models is examined, based on the results of PRED. Following previous studies (Collins 2002; Day et al. 2014; Tietsche et al. 2014), predictability in this study is assessed by two metrics: the root-mean-square error (RMSE) and the anomaly correlation coefficient (ACC). The ensemble RMSE is defined as

 
formula

where is the monthly quantity of interest, for instance the sea ice extent, at lead time for the ith member of the jth ensemble and , in which (= 3) is the number of start dates and (= 8) is the number of ensemble members. Here, the normalized RMSE (NRMSE) is defined as the RMSE divided by , where is the standard deviation of CTRL. There is significant predictability when the NRMSE is significantly lower than 1 using an F test. As in Day et al. (2014), to compare the predictability of the initialized perfect model predictions with the lagged correlation of CTRL, the ACC is defined as

 
formula

where indicates the expected value, which is calculated by summing over the specified index, and is the monthly mean climatology of at lead time . There is significant predictability when the ACC is statistically significant (0.413) at the 95% confidence level with 23 degrees of freedom based on a two-sided Student’s t test. The current study will judge whether the drastic sea ice reduction is predictable based on the above two metrics. As mentioned in section 3c, mechanisms for the drastic ice reduction differ between year 51 and years 56 and 57. However, these three events are here grouped in the same category as extreme ice reduction to investigate the predictability of drastic ice reduction events.

Figure 11 shows the ACC and NRMSE of the SIE and SIV in PRED started from April (lead months 1 to 45) and July (lead months 1 to 42). The ACC values are statistically significant until October (lead months 7 and 4) in predictions started from April and July. The NRMSE values are significant during the first three lead months for both predictions. In PRED.APR, although the ACC values reemerge in August and September, the NRMSE values are not significant. This indicates that the predicted sea ice extents tend to decrease drastically but vary widely among the ensemble members. Focusing on the second prediction year, the ACC is significant in both PRED, while the NRMSE is large. From both the ACC and RMSE metrics, therefore, the drastic sea ice reduction in September can be predicted from July but not from April. In contrast, as suggested by previous studies (Day et al. 2014; 2016), the SIV is much more predictable than the SIE in predictions started from both April and July (Figs. 11b,d) and is predictable up to a lead time of 4 years.

Fig. 11.

ACC for (a) SIE and (b) SIV and normalized RMSE for (c) SIE and (d) SIV in PRED experiments started from 1 Apr (blue) and 1 Jul (red). Vertical lines denote September. Blue and red closed circles indicate where metrics have significant predictability at the 95% confidence level, based on an F test with 23 degrees of freedom.

Fig. 11.

ACC for (a) SIE and (b) SIV and normalized RMSE for (c) SIE and (d) SIV in PRED experiments started from 1 Apr (blue) and 1 Jul (red). Vertical lines denote September. Blue and red closed circles indicate where metrics have significant predictability at the 95% confidence level, based on an F test with 23 degrees of freedom.

The regional predictability of sea ice is investigated by comparing the ice edge and thickness for PRED with that for CTRL (Fig. 12). In CTRL (Fig. 12a), negative anomalies in the SIT are found in the Barents and parts of the Kara, Laptev, and East Siberian Seas in April and July, and further extend to the Chukchi and Beaufort Seas in August in association with the retreat of the ice edge. In PRED.JUL (Fig. 12c), the sea ice edge location and the spatial patterns and magnitudes of the SIT anomalies are consistent with those of CTRL. In PRED.APR (Fig. 12b), the differences in the SIT between PRED and CTRL are small at lead months 1 and 4 (April and July) but become large at lead months 5 and 6 (August and September). Particularly, the predicted SIT increases by 0.5–1.0 m in the Pacific sector of the Arctic Ocean, delaying the ice edge retreat. This may be because the model cannot predict the sea ice motion driven by the SLP dipole anomalies that developed from July to August in CTRL (Fig. 13a). In CTRL (Fig. 13a), the composite of the three drastic reduction years shows the ADA-like SLP structures from July to August. Compared to PRED.APR (Fig. 13b), the differences in the SLP between PRED and CTRL of PRED.JUL (Fig. 13c) are small in July to August, especially in the Canada Basin. These results suggest that the negative SIE anomaly itself can be captured even in PRED.APR, partly because of the initial ocean states, but accurate prediction of the ADA-like SLP structure is one of the key factors for a skillful forecast of drastic ice reduction.

Fig. 12.

(a) SIT anomaly in (left to right) April, July, August, and September for the composite of years 51, 56, and 57 in a CTRL experiment. Also shown are differences in SIT between a CTRL experiment and PRED experiments started from (b) 1 Apr and (c) 1 Jul for years 51, 56, and 57. Black and green lines indicate the 15% SIC contours in each month for a CTRL experiment and PRED experiments, respectively.

Fig. 12.

(a) SIT anomaly in (left to right) April, July, August, and September for the composite of years 51, 56, and 57 in a CTRL experiment. Also shown are differences in SIT between a CTRL experiment and PRED experiments started from (b) 1 Apr and (c) 1 Jul for years 51, 56, and 57. Black and green lines indicate the 15% SIC contours in each month for a CTRL experiment and PRED experiments, respectively.

Fig. 13.

As in Fig. 12, but for the SLP anomaly.

Fig. 13.

As in Fig. 12, but for the SLP anomaly.

As shown in Tietsche et al. (2014), spatial patterns of potential forecast errors for the SIC and SIT are useful information for operational forecasts of Arctic sea ice. Using Eq. (1), the RMSE for the SIC and SIT is calculated at a specific grid point in PRED started from April and July. Figure 14a shows the SIC RMSE in July (lead months 4 and 1) and September (lead months 6 and 3) for PRED.APR and PRED.JUL. In PRED.APR, the large SIC RMSE is found in the marginal ice zone of the Arctic Ocean and in the central Arctic. In PRED.JUL, the SIC RMSE in July is considerably smaller over most of the Arctic Ocean and in September except for the central Arctic. The spatial distribution of the SIT RMSE in PRED.APR shows that the potential forecast errors are large in the ice edge, but small in the central Arctic basin. However, as in the SIC, the potential errors for the SIT in PRED.JUL are fairly small in the Arctic Ocean, while regions of large errors are found in the Pacific sector in September.

Fig. 14.

(a) SIC RMSE and (b) SIT RMSE in lead months 4 (July) and 6 (September) in PRED started from 1 Apr and those in lead months 1 (July) and 3 (September) in PRED started from 1 Jul. Stippling denotes areas where the potential predictive skill (not shown) is below 10%, following Tietsche et al. (2014).

Fig. 14.

(a) SIC RMSE and (b) SIT RMSE in lead months 4 (July) and 6 (September) in PRED started from 1 Apr and those in lead months 1 (July) and 3 (September) in PRED started from 1 Jul. Stippling denotes areas where the potential predictive skill (not shown) is below 10%, following Tietsche et al. (2014).

Similarly, the spatial forecast errors for the SLP and the atmospheric temperature at 2 m above the surface (T2) are shown in Fig. 15. The spatial distribution of the SLP RMSE in July (lead months 4 and 1) clearly shows that the potential forecast errors in the Arctic Ocean are higher in PRED.APR than PRED.JUL, whereas the differences between them are quite small in September (lead months 6 and 3) (Fig. 15a). Some consistent features are found in the spatial distribution of the T2 RMSE (Fig. 15b), except that the errors are small compared to the SLP.

Fig. 15.

As in Fig. 14, but for (a) SLP and (b) atmospheric temperature at 2 m above the surface.

Fig. 15.

As in Fig. 14, but for (a) SLP and (b) atmospheric temperature at 2 m above the surface.

In this study, only eight ensemble members were used for the perfect model experiments based on the APPOSITE protocol. However, Jahn et al. (2016) have shown that 10–15 ensemble members are needed to capture well the internal variability. In addition, Hawkins et al. (2016) have concluded that probabilistic measures of predictability were not reliable even with 16 members, which is the maximum ensemble size in APPOSITE. Therefore, care may be needed when interpreting the results of this study.

5. Concluding remarks

We have conducted a 200-yr control simulation with radiative forcing fixed at the year 2000 level using MIROC5.2 as part of the APPOSITE project and perfect ensemble prediction simulations to evaluate the predictability of a drastic reduction in the summer sea ice and to clarify driving mechanisms. The key implications regarding the present study are as follows:

  1. Extreme Arctic sea ice reductions comparable to the years 2007 and 2012 in the real world occur even for radiative forcing fixed at present-day levels (year 2000).

  2. Wind directions favoring the offshore ice motion, including ADA-like patterns, are one of the key drivers of drastic ice reduction events.

  3. Warm ocean states associated with the OHT from the Atlantic and Pacific Oceans and the SIV in spring play a role in preconditioning for sea ice loss.

  4. An extreme reduction in the September SIE has predictability from 1 July but not for April, because the forecast errors for the July SLP and those for the SIC and SIT along the ice edge are large in PRED started from April.

In this study, three extreme events occurred within just 7 years. This is because the atmospheric circulations were favorable wind directions for sea ice loss and the warm states in the Arctic Ocean continued for 20–30 years from the year 51. However, only a single climate model, MIROC5.2, was used in this study. Therefore, the robustness of the results should be assessed using different climate models under APPOSITE.

Here the implications for the future are discussed by comparing the present study with previous studies. Holland et al. (2006) showed that abrupt changes in the summer Arctic sea ice could occur in the early twenty-first century and suggested that reductions in future greenhouse gas emissions reduce the likelihood of abrupt events. In this study, however, CTRL showed that drastic reductions in the September SIE occur under radiative forcing fixed at year 2000 levels. The sea ice loss from a 200-year climatology (7.10 million km2) is −1.10, −0.89, and −1.98 million km2 for years 51, 56, and 57, respectively. These are comparable to the 42%–93% and 31%–70% observed ice loss in 2007 and 2012 from the 1981–2010 average, while the comparison of a CTRL experiment in MIROC5.2 with observations may not be appropriate. This indicates that under the warm climate state an extreme ice loss event can occur even without a warming trend. Regarding the frequency of extreme events, Cullather and Tremblay (2008), who conducted a control simulation under 1990 constant forcing, suggested that extreme ice loss will increase under climate states warmer than the 1990 level. In a 3600-yr-long control simulation of the Geophysical Fluid Dynamics Laboratory Coupled Model version 2.1 (CM2.1; Delworth et al. 2006) by Zhang (2015), several drastic ice reductions occur even under 1860-level constant forcing. Thus, the frequency of extreme sea ice loss events is expected to increase if the length of a control simulation with year 2000–level constant forcing is extended to longer periods.

On the other hand, the formation mechanism for the ADA itself is highly unclear. Although the physical processes behind the significant signal were not examined in detail in the present study, preliminary analyses suggest that the SST variability in the tropical Indian Ocean is involved in the ADA formation. Recent studies have indicated that Arctic warming and sea ice loss are strongly linked to remote influences from the tropical and extratropical regions (e.g., Ding et al. 2017; Tokinaga et al. 2017). Their findings provide motivation to investigate these mechanisms more thoroughly. To improve the predictability of sea ice associated with internal variability, further analyses and numerical experiments focusing on the relationship between sea ice and other important modes of climate variability will be performed in future work.

Acknowledgments

This work was supported by the Arctic Challenge for Sustainability Project (ArCS Project) and the Program for Generation of Climate Change Risk Information (SOUSEI project), of the Japanese Ministry of Education, Culture, Sports, Science and Technology. J.O. was supported by the Japan Society for the Promotion of Science (JSPS) Grant-in-Aid for Young Scientists (B) Grant JP17K12830. Numerical experiments in this study were conducted using the Fujitsu PRIMEHPC FX10 System (Oakleaf-FX, Oakbridge-FX) in the Information Technology Center, The University of Tokyo. We thank Masayoshi Ishii and Masato Nodzu for their helpful discussions. We thank a reviewer, Dr. François Massonnet, two other anonymous reviewers, and editor Prof. James Screen for their useful comments and suggestions.

REFERENCES

REFERENCES
Bitz
,
C. M.
, and
W. H.
Lipscomb
,
1999
:
An energy-conserving thermodynamic model of sea ice
.
J. Geophys. Res.
,
104
,
15 669
15 677
, https://doi.org/10.1029/1999JC900100.
Bitz
,
C. M.
,
M. M.
Holland
,
A. J.
Weaver
, and
M.
Eby
,
2001
:
Simulating the ice-thickness distribution in a coupled climate model
.
J. Geophys. Res.
,
106
,
2441
2463
, https://doi.org/10.1029/1999JC000113.
Blanchard-Wrigglesworth
,
E.
,
K. C.
Armour
,
C. M.
Bitz
, and
E.
DeWeaver
,
2011
:
Persistence and inherent predictability of Arctic sea ice in a GCM ensemble and observations
.
J. Climate
,
24
,
231
250
, https://doi.org/10.1175/2010JCLI3775.1.
Blanchard-Wrigglesworth
,
E.
,
R. I.
Cullather
,
W.
Wang
,
J.
Zhang
, and
M.
Bitz
,
2015
:
Model forecast skill and sensitivity to initial conditions in the seasonal Sea Ice Outlook
.
Geophys. Res. Lett.
,
42
,
8042
8048
, https://doi.org/10.1002/2015GL065860.
Bushuk
,
M.
,
R.
Msadek
,
M.
Winton
,
G. A.
Vecchi
,
R.
Gudgel
,
A.
Rosati
, and
X.
Yang
,
2017
:
Skillful regional prediction of Arctic sea ice on seasonal timescales
.
Geophys. Res. Lett.
,
44
,
4953
4964
, https://doi.org/10.1002/2017GL073155.
Chevallier
,
M.
,
D.
Salas-Mélia
,
A.
Voldoire
,
M.
Déqué
, and
G.
Garric
,
2013
:
Seasonal forecasts of the pan-Arctic sea ice extent using a GCM-based seasonal prediction system
.
J. Climate
,
26
,
6092
6104
, https://doi.org/10.1175/JCLI-D-12-00612.1.
Collins
,
M.
,
2002
:
Climate predictability on interannual to decadal time scales: The initial value problem
.
Climate Dyn.
,
19
,
671
692
, https://doi.org/10.1007/s00382-002-0254-8.
Comiso
,
J. C.
,
2012
:
Large decadal decline of the Arctic multilayer ice cover
.
J. Climate
,
25
,
1176
1193
, https://doi.org/10.1175/JCLI-D-11-00113.1.
Cullather
,
R. G.
, and
L.-B.
Tremblay
,
2008
: Analysis of Arctic sea ice anomalies in a coupled model control simulation. Arctic Sea Ice Decline: Observations, Projections, Mechanisms, and Implications, Geophys. Monogr., Vol. 180, Amer. Geophys. Union, 187–211.
Day
,
J. J.
,
J. C.
Hargreaves
,
J. D.
Annan
, and
A.
Abe-Ouchi
,
2012
:
Sources of multi-decadal variability in Arctic sea ice extent
.
Environ. Res. Lett.
,
7
,
034011
, https://doi.org/10.1088/1748-9326/7/3/034011.
Day
,
J. J.
,
S.
Tietsche
, and
E.
Hawkins
,
2014
:
Pan-Arctic and regional sea ice predictability: Initialization month dependence
.
J. Climate
,
27
,
4371
4390
, https://doi.org/10.1175/JCLI-D-13-00614.1.
Day
,
J. J.
, and Coauthors
,
2016
:
The Arctic Predictability and Prediction on Seasonal-to-Interannual Timescales (APPOSITE) data set version 1
.
Geosci. Model Dev.
,
9
,
2255
2270
, https://doi.org/10.5194/gmd-9-2255-2016.
Delworth
,
T. L.
, and Coauthors
,
2006
:
GFDL’s CM2 global coupled climate models. Part I: Formulation and simulation characteristics
.
J. Climate
,
19
,
633
674
, https://doi.org/10.1175/JCLI3629.1.
Ding
,
Q.
, and Coauthors
,
2017
:
Influence of high-latitude atmosphere circulation changes on summertime Arctic sea ice
.
Nat. Climate Change
,
7
,
289
295
, https://doi.org/10.1038/nclimate3241.
Guemas
,
V.
,
F.
Doblas-Reyes
,
A.
Germe
,
M.
Chevallier
, and
D. S.
Mélia
,
2013
:
September 2012 Arctic sea ice minimum: Discriminating between sea ice memory, the August 2012 extreme storm, and prevailing warm conditions [in “Explaining Extreme Events of 2012 from a Climate Perspective”]
.
Bull. Amer. Meteor. Soc.
,
94
(
9
),
S20
S22
, https://doi.org/10.1175/BAMS-D-13-00085.1.
Guemas
,
V.
,
M.
Chevallier
,
M.
Déqué
,
O.
Bellprat
, and
F.
Doblas-Reyes
,
2016a
:
Impact of sea ice initialization on sea ice and atmosphere prediction skill on seasonal timescales
.
Geophys. Res. Lett.
,
43
,
3889
3896
, https://doi.org/10.1002/2015GL066626.
Guemas
,
V.
, and Coauthors
,
2016b
:
A review on Arctic sea-ice predictability and prediction on seasonal to decadal time-scales
.
Quart. J. Roy. Meteor. Soc.
,
142
,
546
561
, https://doi.org/10.1002/qj.2401.
Hawkins
,
E.
,
S.
Tietsche
,
J. J.
Day
,
N.
Melia
,
K.
Haines
, and
S.
Keeley
,
2016
:
Aspects of designing and evaluating seasonal-to-interannual Arctic sea-ice prediction systems
.
Quart. J. Roy. Meteor. Soc.
,
142
,
672
683
, https://doi.org/10.1002/qj.2643.
Holland
,
M. M.
,
C. M.
Bitz
, and
B.
Tremblay
,
2006
:
Future abrupt reductions in the summer Arctic sea ice
.
Geophys. Res. Lett.
,
33
,
L23503
, https://doi.org/10.1029/2006GL028024.
Holland
,
M. M.
,
D. A.
Bailey
, and
S.
Vavrus
,
2011
:
Inherent sea ice predictability in the rapidly changing Arctic environment of the Community Climate System Model, version 3
.
Climate Dyn.
,
36
,
1239
1253
, https://doi.org/10.1007/s00382-010-0792-4.
Hunke
,
E. C.
, and
J. K.
Dukowicz
,
1997
:
An elastic–viscous–plastic model for sea ice dynamics
.
J. Phys. Oceanogr.
,
27
,
1849
1867
, https://doi.org/10.1175/1520-0485(1997)027<1849:AEVPMF>2.0.CO;2.
Hutchings
,
J. K.
, and
D. K.
Perovich
,
2015
:
Preconditioning of the 2007 sea-ice melt in the eastern Beaufort Sea, Arctic Ocean
.
Ann. Glaciol.
,
56
,
94
98
, https://doi.org/10.3189/2015AoG69A006.
IPCC
,
2013
: Climate Change 2013: The Physical Science Basis. T. F. Stocker et al., Eds., Cambridge University Press, 1535 pp., https://doi.org/10.1017/CBO9781107415324.
Ishii
,
M.
, and
M.
Kimoto
,
2009
:
Reevaluation of historical ocean heat content variations with time-varying XBT and MBT depth bias corrections
.
J. Oceanogr.
,
65
,
287
299
, https://doi.org/10.1007/s10872-009-0027-7.
Jahn
,
A.
,
J. E.
Kay
,
M. M.
Holland
, and
D. M.
Hall
,
2016
:
How predictable is the timing of a summer ice-free Arctic?
Geophys. Res. Lett.
,
43
,
9113
9120
, https://doi.org/10.1002/2016GL070067.
Jungclaus
,
J. H.
, and Coauthors
,
2013
:
Characteristics of the ocean simulations in the Max Planck Institute Ocean Model (MPIOM), the ocean component of the MPI-Earth system model
.
J. Adv. Model. Earth Syst.
,
5
,
422
446
, https://doi.org/10.1002/jame.20023.
Kauker
,
F.
,
T.
Kaminski
,
M.
Karcher
,
R.
Giering
,
R.
Gerdes
, and
M.
Voßbeck
,
2009
:
Adjoint analysis of the 2007 all time Arctic sea-ice minimum
.
Geophys. Res. Lett.
,
36
,
L03707
, https://doi.org/10.1029/2008GL036323.
Kay
,
J. E.
,
M. M.
Holland
, and
A.
Jahn
,
2011
:
Inter-annual to multi-decadal Arctic sea ice extent trends in a warming world
.
Geophys. Res. Lett.
,
38
,
L15708
, https://doi.org/10.1029/2011GL048008.
Komuro
,
Y.
,
2014
:
The impact of surface mixing on the Arctic river water distribution and stratification in a global ice–ocean model
.
J. Climate
,
27
,
4359
4370
, https://doi.org/10.1175/JCLI-D-13-00090.1.
Komuro
,
Y.
, and Coauthors
,
2012
:
Sea-ice in twentieth-century simulations by new MIROC coupled models: A comparison between models with high resolution and with ice thickness distribution
.
J. Meteor. Soc. Japan
,
90A
,
213
232
, https://doi.org/10.2151/jmsj.2012-A11.
Kwok
,
R.
, and
D. A.
Rothrock
,
2009
:
Decline in Arctic sea ice thickness from submarine and ICESat records: 1958–2008
.
Geophys. Res. Lett.
,
36
,
L15501
, https://doi.org/10.1029/2009GL039035.
Liston
,
G. E.
,
2004
:
Representing subgrid snow cover heterogeneities in regional and global models
.
J. Climate
,
17
,
1381
1397
, https://doi.org/10.1175/1520-0442(2004)017<1381:RSSCHI>2.0.CO;2.
Msadek
,
R.
,
G. A.
Vecchi
,
M.
Winton
, and
R. G.
Gudgel
,
2014
:
Importance of initial conditions in seasonal predictions of Arctic sea ice extent
.
Geophys. Res. Lett.
,
41
,
5208
5215
, https://doi.org/10.1002/2014GL060799.
Nitta
,
T.
, and Coauthors
,
2014
:
Representing variability in subgrid scale snow cover and snow depth in a global land model: Offline validation
.
J. Climate
,
27
,
3318
3330
, https://doi.org/10.1175/JCLI-D-13-00310.1.
Nitta
,
T.
,
K.
Yoshimura
, and
A.
Abe-Ouchi
,
2017
:
Impact of Arctic wetlands on the climate system: Model sensitivity simulations with the MIROC5 AGCM and a snow-fed wetland scheme
.
J. Hydrometeor.
,
18
,
2923
2936
, https://doi.org/10.1175/JHM-D-16-0105.1.
Notz
,
D.
, and
J.
Marotzke
,
2012
:
Observations reveal external driver for Arctic sea-ice retreat
.
Geophys. Res. Lett.
,
39
,
L08502
, https://doi.org/10.1029/2012GL051094.
NSIDC
,
2017
: Arctic Sea Ice News & Analysis. National Snow and Ice Data Center. Accessed 31 March 2018, http://nsidc.org/arcticseaicenews/.
NSIDC
,
2018
: State of the cryosphere. National Snow and Ice Data Center. Accessed 7 December 2018, https://nsidc.org/cryosphere/sotc/sea_ice.html.
Ono
,
J.
,
H.
Tatebe
,
Y.
Komuro
,
M. I.
Nodzu
, and
M.
Ishii
,
2018
:
Mechanisms influencing seasonal to inter-annual prediction skill of sea ice extent in the Arctic Ocean in MIROC
.
Cryosphere
,
12
,
675
683
, https://doi.org/10.5194/tc-12-675-2018.
Polyakov
,
I. V.
,
J. E.
Walsh
, and
R.
Kwok
,
2012
:
Recent changes of Arctic multiyear sea ice coverage and the likely causes
.
Bull. Amer. Meteor. Soc.
,
93
,
145
151
, https://doi.org/10.1175/BAMS-D-11-00070.1.
Rayner
,
N. A.
,
D. E.
Parker
,
E. B.
Horton
,
C. K.
Folland
,
L. V.
Alexander
,
D. P.
Rowell
,
E. C.
Kent
, and
A.
Kaplan
,
2003
:
Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century
.
J. Geophys. Res.
,
108
,
4407
, https://doi.org/10.1029/2002JD002670.
Schweiger
,
A.
,
R.
Lindsay
,
J.
Zhang
,
M.
Steele
,
H.
Stern
, and
R.
Kwok
,
2011
:
Uncertainty in modeled Arctic sea ice volume
.
J. Geophys. Res.
,
C00D06
, https://doi.org/10.1029/2011JC007084.
Serreze
,
M. C.
, and
J. C.
Stroeve
,
2015
:
Arctic sea ice trends, variability and implications for seasonal ice forecasting
.
Philos. Trans. Royal Soc.
,
373A
,
20140159
, https://doi.org/10.1098/rsta.2014.0159.
Shaffrey
,
L. C.
, and Coauthors
,
2009
:
U.K. HiGEM: The new U.K. high-resolution global environment model—Model description and basic evaluation
.
J. Climate
,
22
,
1861
1896
, https://doi.org/10.1175/2008JCLI2508.1.
Sigmond
,
M.
,
J. C.
Fyfe
,
G. M.
Flato
,
V. V.
Kharin
, and
W. J.
Merryfield
,
2013
:
Seasonal forecast skill of Arctic sea ice area in a dynamical forecast system
.
Geophys. Res. Lett.
,
40
,
529
534
, https://doi.org/10.1002/grl.50129.
Simmonds
,
I.
, and
I.
Rudeva
,
2012
:
The great Arctic cyclone of August 2012
.
Geophys. Res. Lett.
,
39
,
L23709
, https://doi.org/10.1029/2012GL054259.
Steele
,
M.
,
W.
Ermold
, and
J.
Zhang
,
2008
:
Arctic Ocean surface warming trends over the past 100 years
.
Geophys. Res. Lett.
,
35
,
L02614
, https://doi.org/10.1029/2007GL031651.
Steele
,
M.
,
J.
Zhang
, and
W.
Ermold
,
2010
:
Mechanisms of summertime upper Arctic Ocean warming and the effect on sea ice melt
.
J. Geophys. Res.
,
115
,
C11004
, https://doi.org/10.1029/2009JC005849.
Stroeve
,
J.
,
M.
Serreze
,
S.
Drobot
,
S.
Gearheard
,
M.
Holland
,
J.
Maslanik
,
W.
Meier
, and
T.
Scambos
,
2008
:
Arctic sea ice extent plummets in 2007
.
Eos, Trans. Amer. Geophys. Union
,
89
,
13
14
, https://doi.org/10.1029/2008EO020001.
Swart
,
N. C.
,
J. C.
Fyfe
,
E.
Hawkins
,
J. E.
Kay
, and
A.
Jahn
,
2015
:
Influence of internal variability on Arctic sea-ice trends
.
Nat. Climate Change
,
5
,
86
89
, https://doi.org/10.1038/nclimate2483.
Tatebe
,
H.
,
Y.
Tanaka
,
Y.
Komuro
, and
H.
Hasumi
,
2018
:
Impact of deep ocean mixing on the climatic mean state in the Southern Ocean
.
Sci. Rep.
,
8
,
14479
, https://doi.org/10.1038/s41598-018-32768-6.
Tietsche
,
S.
,
D.
Notz
,
J. H.
Jungclaus
, and
J.
Marotzke
,
2013
:
Predictability of large interannual Arctic sea-ice anomalies
.
Climate Dyn.
,
41
,
2511
2526
, https://doi.org/10.1007/s00382-013-1698-8.
Tietsche
,
S.
, and Coauthors
,
2014
:
Seasonal to interannual Arctic sea ice predictability in current global climate models
.
Geophys. Res. Lett.
,
41
,
1035
1043
, https://doi.org/10.1002/2013GL058755.
Tokinaga
,
H.
,
S.-P.
Xie
, and
H.
Mukougawa
,
2017
:
Early 20th-century Arctic warming intensified by Pacific and Atlantic multidecadal variability
.
Proc. Natl. Acad. Sci. USA
,
114
,
6227
6232
, https://doi.org/10.1073/pnas.1615880114.
Wang
,
J.
,
J.
Zhang
,
E.
Watanabe
,
M.
Ikeda
,
K.
Mizobata
,
J. E.
Walsh
,
X.
Bai
, and
B.
Wu
,
2009
:
Is the dipole anomaly a major driver to record lows in Arctic summer sea ice extent?
Geophys. Res. Lett.
,
36
,
L05706
, https://doi.org/10.1029/2008GL036706.
Wang
,
W.
,
M.
Chen
, and
A.
Kumar
,
2013
:
Seasonal prediction of Arctic sea ice extent from a coupled dynamical forecast system
.
Mon. Wea. Rev.
,
141
,
1375
1394
, https://doi.org/10.1175/MWR-D-12-00057.1.
Watanabe
,
M.
, and Coauthors
,
2010
:
Improved climate simulation by MIROC5: Mean states, variability, and climate sensitivity
.
J. Climate
,
23
,
6312
6135
, https://doi.org/10.1175/2010JCLI3679.1.
Woodgate
,
R. A.
,
K.
Aagaard
, and
T. J.
Weingartner
,
2006
:
Interannual changes in the Bering Strait fluxes of volume, heat and freshwater between 1991 and 2004
.
Geophys. Res. Lett.
,
33
,
L15609
, https://doi.org/10.1029/2006GL026931.
Woodgate
,
R. A.
,
T.
Weingartner
, and
R.
Lindsay
,
2010
:
The 2007 Bering Strait oceanic heat flux and anomalous Arctic sea-ice retreat
.
Geophys. Res. Lett.
,
37
,
L01602
, https://doi.org/10.1029/2009GL041621.
Zhang
,
J.
, and
D. A.
Rothrock
,
2003
:
Modeling global sea ice with a thickness and enthalpy distribution model in generalized curvilinear coordinates
.
Mon. Wea. Rev.
,
131
,
845
861
, https://doi.org/10.1175/1520-0493(2003)131<0845:MGSIWA>2.0.CO;2.
Zhang
,
J.
,
R.
Lindsay
,
M.
Steele
, and
A.
Schweiger
,
2008
:
What drove the dramatic retreat of Arctic sea ice during summer 2007?
Geophys. Res. Lett.
,
35
,
L11505
, https://doi.org/10.1029/2008GL034005.
Zhang
,
J.
,
R.
Lindsay
,
A.
Schweiger
, and
M.
Steele
,
2013
:
The impact of an intense summer cyclone on 2012 Arctic sea ice retreat
.
Geophys. Res. Lett.
,
40
,
720
726
, https://doi.org/10.1002/grl.50190.
Zhang
,
R.
,
2015
:
Mechanisms for low-frequency variability of summer Arctic sea ice extent
.
Proc. Natl. Acad. Sci. USA
,
112
,
4570
4575
, https://doi.org/10.1073/pnas.1422296112.
Zhang
,
R.
, and
T. R.
Knutson
,
2013
:
The role of global climate change in the extreme low summer Arctic sea ice extent in 2012 [in “Explaining Extreme Events of 2012 from a Climate Perspective”]
.
Bull. Amer. Meteor. Soc.
,
94
(
9
),
S23
S26
, https://doi.org/10.1175/BAMS-D-13-00085.1.

Footnotes

© 2019 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).