Early Freeze-Up over the Bering Sea Controlled by the Aleutian Low

Abstract

Early freeze-up affects the local marine environment and ecosystem throughout the entire Bering Sea. However, the process governing early freeze-up, which is responsible for the most significant interannual variation in the December sea ice area (SIA), is not well understood. Here, we show that the SIA in December is modulated by the Aleutian low in November by altering poleward heat transport (PHT). The stronger the November PHT is, the lower the December SIA. The rise in heat transport across the Bering Strait in November is consistent with the decrease in SIA in December, with a correlation of −0.71, further validating the regulatory role of PHT. The Aleutian low anomaly controls the local wind field, further altering the sea surface temperature and PHT. The center of the anomalous low-pressure in the east (west) generates the northerly (southeasterly) anomaly over the northern Bering Sea, leading to acceleration (suppression) of seawater cooling and weakening (enhancement) of the PHT. It is also found that a strong northerly surface current has a greater influence on the later SIA than warm water temperature. Hence, atmospheric forcing causing changes in ocean forcing is imperative to understand early freeze-up.

1. Introduction

Sea ice is one of the key driving factors impacting the climate and environment in the Bering Sea [1]. Seasonal sea ice fluctuations influence water stratification [2,3], the cold pool area [2,4], phytoplankton communities [3,5], benthos taxa [3,6], ecosystems [7,8], and even economic activity [9]. The Bering Sea is undergoing delayed freeze-up and earlier sea ice retreats [10,11] combined with higher ocean temperatures and freshwater content [5,12,13,14,15,16,17]. These cause the duration of open water to increase, which results in seawater absorbing more solar radiation, warming the atmosphere and weakening the stability of the atmospheric boundary layer. Over the past decade, the significant retreat of sea ice has resulted in water stratification weakening, the disappearance of cold pools, and shifts in marine species composition and carbon cycling [13]. Additionally, as open water rises, economic activities such as resource exploitation, shipping, and tourism become more accessible [18,19].

Sea ice in the Bering Sea is produced in the north, moved southward by the action of the wind, and melted in the south by the warmer seawater [20]. This process of changing the sea ice area in winter might be compared to a “conveyor belt” [21,22,23]. Up to now, researchers have essentially demonstrated the impact of large-scale atmospheric circulation [24,25,26], cyclone activity [27], wind anomalies [4], sea ice transport southward across the Bering Sea strait [28], sea surface temperature, and Unimak warm water inflow on sea ice [29,30,31]. However, previous studies have concentrated mostly on mid-late winter sea ice and its interannual variation in the Bering Sea, with less attention given to early winter sea ice [26].

This paper focuses on the early freeze-up in the Bering Sea. From 1979 to 2021, the early freeze-up onset (EFO) of the Bering Sea is November 11, as shown in Figure 1a. Additionally, EFO has increased by 5.7 days per decade. In November, sea ice is mostly produced in Norton Sound [20,32]. This has a limited impact on the hydrology and environment of the Bering Sea. However, in December, the northern Bering Sea observes the commencement of significant sea ice production. The correlation between December SIA (Figure 1b) and EFO reaches −0.69 (p value < 0.01) in the last 30 years, meaning that the SIA in December can properly reflect the initial status of early freeze-up. In December, sea ice in the Bering Sea is primarily produced as a result of local cooling. A downward linear trend in December of SIA over the 44-year SMMR/SSMI microwave radiometer record is 3.0 × 10³ km² per year (Figure 1b), or 5.3% per decade relative to the 1981 to 2010 average, which is the largest value of all months [20], suggesting an overall decrease in sea ice, and highlighting the need for further research into its cause. In addition, a substantial negative correlation between the SIA in December and the sea ice area increment (ΔSIA) in January indicates that the December sea ice may influence the formation of sea ice in January via atmospheric and oceanic feedback.

The suppression of early freeze-up may be due to the increased ocean temperature and warmer sea air temperature (SAT) [26]. The sea surface temperature (SST) is determined by air–sea heat flux, advection, and dissipation [20]. Although the air–sea heat flux in the Bering Sea shelf area is substantially higher than the advection term in November and December, the interannual variation in SST is primarily controlled by warm advection [20]. The SAT anomalies in the Bering Sea are often associated with changes in the position of the Aleutian low and its central pressure [33,34]. Despite the early freeze-up being of large importance for the climate in the Bering Sea, the processes controlling it and its large interannual variability are only partially understood. The most crucial factor is yet unclear. In addition, the surface wind has a considerable role in promoting the expansion of sea ice in the winter [4,23,29,35]. The role of the wind field performing in the early freeze-up period is also an essential issue in this paper. These questions are addressed utilizing 43 years of ERA5 data, independent sea ice areas, and 30 years of ocean heat fluxes derived from dynamic topography data and Ekman drift data.

2. Materials and Methods

2.1. Materials

Data on sea ice concentration are obtained from the Scanning Multichannel Microwave Radiometer (SMMR) on the Nimbus-7 satellite and the SSM/I sensors on the DMSP-F8, -F11, and -F13 satellites according to the NASA-team algorithm, whose resolution is 25 km [36]. The period chosen is from 1 January 1979 to 31 December 2021. In this paper, it is the entire Bering Sea (51–66°N, 165–205°E) that is of interest. As a result of the area-weighted SIC on a grid-cell level, SIA is computed and integrated across a region of interest. In this paper, the Bering Sea freeze-up timing product is also retrieved from the above multichannel spaceborne radiometers [11], which are available at the NASA Goddard Space Flight Center. The heat transport (relative to 1.9 °C) across the entire Bering Strait is estimated and adopted from the continuous mooring observations conducted by the Polar Science Center, University of Washington, and the November data are only chosen from 1997 to 2018 [16]. The monthly SST with a resolution of 0.25° × 0.25° is from the National Oceanic and Atmospheric Administration (NOAA) optimum interpolation (OI) SST version 2 dataset [37], which is derived from a combination of in situ observations and Advanced Very High-Resolution Radiometer (AVHRR) satellite infrared data. The selected period spans from January 1979 to December 2021. The European Remote-sensing Satellite (ERS) altimetric data and Topex/Poseidon dynamic topography data are used to calculate the geostrophic current [38]. The data have a grid spacing of 0.25° × 0.25°.

The atmospheric variables used in this paper are from ERA5 reanalysis provided by the European Centre for Medium-Range Forecasts (ECMWF). It is based on the integrated Forecast System (IFS), the main ECMWF global forecasting model. It has a horizontal resolution of approximately 30 km (distributed at 0.25°) [39], and has been available from 1979 to the present. Most variables are acquired monthly, including sea level pressure (SLP), 10 m wind, and 1000 hPa geopotential height. In addition, the climatology mixing-layer depth used is from the GLORYS12V1 product in the Copernicus Marine Service (CMEMS). The spatial resolution is 0.083° × 0.083°. Details can be found on website (https://doi.org/10.48670/moi-00021, accessed on 13 January 2023).

2.2. Methods

2.2.1. Correlation Analysis Method

Correlation analysis is used to compute and measure the strength of a linear relationship between two variables. A high correlation indicates a strong relationship between the two variables, whereas a low correlation indicates that the variables are weakly related. Its computation expression is as follows:

$$r=\frac{n\sum xy-\sum x\sum y}{\sqrt{\left(n\sum x^2-{\left(\sum x\right)}^2\right)\left(n\sum y^2-{\left(\sum y\right)}^2\right)}}$$

(1)

Student t-test is used to evaluate temporal correlations under the hypothesis (i.e., the null hypothesis, p value ≤ 0.05) that no relationship exists between the observed phenomena.

2.2.2. Empirical Orthogonal Function Analysis Approach (EOF)

The December SIC is detrended and then decomposed using the empirical orthogonal function analysis approach (EOF), which identifies the most dominant patterns. Each pattern is defined by a spatial pattern (the so-called EOF) and a time series (principal component, PC). The principal components (PCs) of the most prevalent modes are standardized by using Z-score method and used to the indices. The EOFs of a space–time physical process can represent mutually orthogonal space patterns where the data variance is concentrated, with the first pattern accounting for most of the variance, the second accounting for most of the remaining variance, and so on.

2.2.3. Composite Analysis Approach

According to the specific classification method, the observation variables are classified. In this paper, the value in the time series (PC1) of the leading mode (EOF1) computed from December SIA is what determines the classification. Please refer to Section 3.1 for detailed information. The classification objects set in this paper include sea surface temperature, sea level pressure, and wind field. The composite differences are separately computed for comprehensive analysis. In this process, we evaluated statistical significance using a two-sided Student’s t-test.

2.2.4. Poleward Oceanic Heat Transport with Finite Depth

Using the following formula [20], poleward oceanic heat transport is computed:

$$PHT={{\displaystyle \int }}_{{\lambda }_1}^{{\lambda }_2}{{\displaystyle \int }}_{-H}^0\rho c_{p}vTa\cos \varphi \mathrm{d}z\mathrm{d}\lambda$$

(2)

where $\rho =1022.95{\mathrm{kg}\mathrm{m}}^{-3}$ is the seawater density; $c_p=3900 {\mathrm{J}\mathrm{kg}}^{-1}°{\mathrm{C}}^{-1}$ is the specific heat capacity of seawater; $T$ is the seawater temperature; $v$ is the meridional velocity; $\varphi$ is the latitude; $a=6371 \mathrm{km}$ is the radius of the Earth; and $\lambda$ is the longitude. Because $\rho$, $c_p$, and $a$ are constant, PHT is proportional to $vT{H}_{MLD}\cos \varphi$. T is the SST as the mean temperature within the depth of the MLD (the mixed layer depth). The surface current ($\overrightarrow{V}$) can be considered as the sum of the geostrophic current (${\overrightarrow{V}}_{ge}$) and the Ekman transport (${\overrightarrow{V}}_{ek}$). Since the depth on the Bering Sea shelf is less than 150 m, the Ekman transport with finite depth is used here:

$$\overrightarrow{V}={\overrightarrow{V}}_{ge}+{\overrightarrow{V}}_{ek}$$

(3)

$$\overrightarrow{V_{ek}}=\left(U_e, V_e\right) and \overrightarrow{\tau }={\rho }_0{C}_D\left|\overrightarrow{u_s}\right|\overrightarrow{u_s}=\left({\tau }_x,{\tau }_y\right)$$

(4)

$$U_e=S_{yx}+S_{xy}$$

(5)

$$V_e=S_{yy}-S_{xx}$$

(6)

$${\epsilon }_i=\frac{{\tau }_i}{a\rho A_z}·\frac{1}{\cosh 2aD+\cos 2aD}, i=x,y$$

(7)

$$\left\{\begin{array}{c}{S}_{ix}=\frac{{\epsilon }_i}{a}\left[\frac{1}{2}\left(\cosh 2aD+\cos 2aD\right)-\cosh aD·\cos aD\right]\\ S_{iy}=\frac{{\epsilon }_i}{a}\sinh aD·\sin aD\end{array}\right.$$

(8)

$$a=\sqrt{\frac{f}{2A_z}}$$

(9)

$${\overrightarrow{V}}_{ge}=\frac{g}{f}\left(-\frac{\partial h}{\partial y},\frac{\partial h}{\partial x}\right)$$

(10)

where ${\rho }_0=1.25 \mathrm{kg} {\mathrm{m}}^{-3}$ is the surface air density, $C_D=0.00125$ is the drag coefficient of wind on seawater, $f$ is the Coriolis parameter, g = 9.8 m/s² is the acceleration of gravity, $A_z=0.01 {\mathrm{m}}^2 {\mathrm{s}}^{-1}$ is turbulent viscosity coefficient, D denotes the bathymetry, and h is the dynamic topography. The zonal and meridional directions are indicated by the subscript ‘x’ and ‘y’, respectively. As mentioned above, $\overrightarrow{u_s}$ is used from the wind data collected at the height of 10 m above sea level, and $v$ in Equation (2) corresponds to the meridional component of the sea surface velocity ($\overrightarrow{V}$).

3. Results

In this section, we first examine the climatology of sea ice cover in December and its spatiotemporal patterns. Next, a composite analysis for SST, SLP, and wind is performed. We explore the correlations between climate variables and current instances in places with significant changes in sea ice cover to investigate the connecting mechanism of the relationships. Additionally, because Wang et al. (2022) concentrated on the changes in sea ice area in January, this section also compares the findings of Wang et al. (2022) in order to examine the potential changes in the driving factors of sea ice changes in different months.

3.1. 43-Year December SIC and Its Time-Space Patterns

Figure 2 depicts the spatially averaged and standard deviation of sea ice concentration (SIC) from 1979 to 2021. In December, the primary freeze-up regions over the Bering Sea are in Norton Sound, Anadyr Bay, and the north of St. Lawrence Island. The most substantial standard deviations are around St. Lawrence Island and Anadyr Bay in Figure 2b, suggesting that the sea ice over these regions is the most affected by ocean and atmospheric processes. This paper emphasizes these regions as the focus of its discussion. From the perspective of the time series of SIA in Figure 1b, the December SIA is accelerating to decline. Estimates show that SIA expanded in 1979–2000, with an increased rate of 1.7 × 10³ km² per year. In the preceding 20 years of the 21st century, SIA has decreased by 5.4 × 10³ km² per year, especially in the past ten years, reaching 12.9 × 10³ km² per year. In the past five years, the Bering Sea has been ice-free on several occasions in December [4,32]. The massive sea ice anomaly has caused the elimination of the cold pool in summer, which has a tremendous influence on the local marine ecology [2,40].

We used the EOF analysis approach to decompose the monthly mean sea ice concentration in December and extract the primary spatial modes and their time series of sea ice. In Figure 3, the first two spatial modes identified can be used to explain 0.68 and 0.10, respectively, of the fluctuation in the December SIA. The autocorrelation is also very small, indicating that the two spatiotemporal modes are independent [41]. In EOF1 (Figure 3a), the sea ice fluctuation over the Bering Sea is synchronized. The primary region of sea ice change is found around St. Lawrence Island, reaching Nunivak Island in the south and the Bering Strait in the north. Its 2.5 isoline is consistent with the 20% isoline of SIC standard deviation in Figure 2b. In EOF2, sea ice is characterized by a “seesaw” fluctuation along the east and west coasts of the Bering Sea. The primary regions of sea ice fluctuation are in the southern and northern parts of Nunivak Island and Anadyr Bay, respectively. Because the explained variance in EOF2 is substantially smaller than that of EOF1, this paper primarily focuses on the discussion of the time series of EOF1. In PC1, the sea ice changes into a notable interannual characteristic. The time series of PC1 and December SIA are synchronous with a correlation coefficient of 0.99 (p value < 0.01).

To further examine the cause of the interannual variation in the December SIA, we separate the observed period in Figure 3c into heavy-ice years (HI, PC1 > 1), normal years (NM, −1 ≤ PC1 ≤ 1), and light-ice years (LI, PC1 < −1). The influence of atmospheric and oceanic processes in November on ice formation in the Bering Sea is discussed by the composite analysis approach.

In addition, it is noteworthy that the EOF1 of the December SIC is consistent with the EOF2 pattern of ΔSIA in January, as shown by Wang et al. (2022). We speculate that this is the result of a delayed freeze-up. The sea ice originally produced in December is delayed due to the constraints of meteorological and oceanic conditions. As the local climate environment approaches the need for freezing, seawater continues to freeze. The consistency of locally produced sea ice in spatial mode appears to imply that their influencing factors are consistent, but the correlation between their time series is only 0.1.

3.2. Composite Analysis of SSTs and 10 m Wind Field in November

Here, we use the composite analysis approach to examine the changes in SST, SLP, and wind field in November based on the classification of sea ice light–heavy years mentioned above. In the years of HI (Figure 4a), the SSTs in the northern portions of St. Lawrence Island, as well as the southern and northern portions of Nunivak Island, are 0.8 degrees Celsius lower than they would normally be. This is favorable for seawater freezing. In November, there is an unusual low-pressure center in the Gulf of Alaska and abnormally high pressure in the Chukchi Sea. This generates a northerly anomaly above the unusually cold water, causing a large amount of cold air to move southward and speed up the formation of sea ice. The atmospheric and oceanic conditions in the LI years (Figure 4b) differ significantly from those in the HI year. Warm water with an abnormally high SST of more than 1 degree Celsius can be found in the north of St. Lawrence Island, as well as south and north of Nunivak Island. The center of the anomalous low-pressure system is currently situated over the southern Bering Sea, causing a southerly anomaly that pushes southern warm air northward and prevents the formation of sea ice (Figure 4b).

Sea ice freezing is directly caused by a drop in sea surface temperature (SST). Therefore, we determined the spatial distribution of the correlation coefficient between the SST and PC1 in November, as depicted in Figure 5. A strong negative correlation is exhibited in the shelf and southeast regions of the Bering Sea. The correlation coefficient is usually greater than 0.5 in the region where the SIC changed significantly in December. In fact, the strength of SST in November cannot be used to interpolate the extent of sea ice. For instance, sea ice may also be suppressed in a specific region with a low SST where continuous heat transport exists. In addition, the anomalous wind field in November can also hasten latent heat evaporation of surface seawater and cause cooling. From the contribution of advection, air–sea flux, and diffusion to SST given by Wang et al. (2022), warm advection in November modulates the annual variation in SST over the Bering Sea shelf. The northward heat transport has a significant impact on how strongly the SST rises. As a result, it is critical to discuss the impact of poleward heat transport on December sea ice.

3.3. The Influence of Poleward Heat Transport on Sea Ice

In fact, both atmospheric and oceanic forcings together determine the extent of sea ice in the Bering Sea [20,23,42]. The Bering Sea is dominated by northerly or northeasterly winds in the winter, which drag the sea ice southward continuously; however, the heat content of southern warm water limits the advance of sea ice. Because of the slow Bering Sea shelf current, the northward heat transport is frequently considered to be low, and it is often mentioned that the winter SIA is primarily caused by atmospheric forcing [22,30,33]. Stabeno and Bell (2019) mentioned that sea ice advances are primarily dependent upon atmospheric forcing and, to a smaller degree, on ocean temperatures. However, Wang et al. (2022) reported that the sea ice area increment in January is modulated by poleward heat transport.

Since most of the Bering Sea is still ice-free in November, the effect of the wind dragging sea ice southward will be absent. In this section, we compute the correlation between PHT and PC1 in November, as depicted in Figure 6. It even reaches −0.5, indicating a strong inverse correlation between the two, and implying that the December SIA decreases as PHT rises. The regions with a value of less than −0.4 are spatially very consistent with the obvious change in sea ice in Figure 2. In contrast to Figure 5, there is no significant high correlation between PHT and PC1 in the southern Bering Sea, where we cannot reasonably explain the high correlation between SST and PC1.

In the calculation of Formula (2), surface velocity and SST have a large influence on the PHT. It is necessary to discuss their contribution to the interannual variation in PHT. Here, we decompose $vT$ as follows:

$$\begin{array}{c}vT=\left(v^{\prime }+\overline{v}\right)\left(T^{\prime }+\overline{T}\right)=v^{\prime }{T}^{\prime }+\overline{v}{T}^{\prime }+v^{\prime }\overline{T}+\overline{v}\overline{T}\\ EV\left(v^{\prime }\overline{T}\right)=\frac{Var\left(v^{\prime }\overline{T}\right)}{Var\left(vT-\overline{v}\overline{T}\right)}\\ EV\left(\overline{v}{T}^{\prime }\right)=\frac{Var\left(\overline{v}{T}^{\prime }\right)}{Var\left(vT-\overline{v}\overline{T}\right)}\\ EV\left(v^{\prime }{T}^{\prime }\right)=\frac{Var\left(v^{\prime }{T}^{\prime }\right)}{Var\left(vT-\overline{v}\overline{T}\right)}\\ EV\left(Cov\right)=\frac{2\times Cov\left(v^{\prime }\overline{T},\overline{v}{T}^{\prime }\right)+2\times Cov\left(v^{\prime }\overline{T},v^{\prime }{T}^{\prime }\right)+2\times Cov\left(v^{\prime }{T}^{\prime },\overline{v}{T}^{\prime }\right)}{Var\left(vT-\overline{v}\overline{T}\right)}\end{array}$$

where the superscript represents the anomaly value. $\overline{v}{T}^{\prime }$, $v^{\prime }\overline{T}$, $v^{\prime }{T}^{\prime }$, and $\overline{v}\overline{T}$ represent the SST anomaly term, sea surface current anomaly term, disturbance term, and average term, respectively. Var denotes the variance corresponding to the specific terms, and Cov denotes the covariances between each term. The area inside the white box in Figure 6 is computed, and the corresponding Explaining Variances (EVs) on the PHT anomaly term ($vT-\overline{v}\overline{T})$ are shown in Figure 7.

Among all EVs in Figure 7, the sea surface current anomaly term contributes the most to the interannual variation in the PHT, accounting for 42.7%, followed by the covariance term, accounting for 36.9%. The SST anomaly term only accounts for 9.7% of the interannual variation in the PHT, which is far less than the sea surface current anomaly term. In covariance terms, the covariance composed of $v^{\prime }\overline{T}$ and $v^{\prime }{T}^{\prime }$ contributes the most to PHT, accounting for 23% of the total. These EVs indicate that the interannual variation in the PHT is determined by the sea surface current in November.

As the Ekman drift driven by wind is an important factor in controlling the surface current, we find that the average wind-driven drift in November is more than 10 times the geostrophic flow, meaning that the heat transport of the wind-driven drift in November is significantly higher than that of the geostrophic heat transport.

Poleward heat transport not only has a significant inhibitory effect on sea ice advance in November but also has the same effect on the SIA increment in January (Wang et al. (2022)). The wind field determines the freeze-up process of seawater in December by affecting poleward heat transport, which is consistent with the leading factor affecting January freeze-up. Meanwhile, these findings mean that the drag effect of the wind field on sea ice is negligible in the early winter.

3.4. Mechanism of Aleutian Low Anomaly to December Sea Ice

In the winter, the Aleutian low is the most important atmospheric phenomenon in the North Pacific [33,34]. Its intensity and position change have a substantial impact on atmospheric circulation, air–sea heat flux, mixed-layer depth, and sea ice extent. Because ocean circulation is low in the Bering Sea shelf, winter atmospheric circulation is the primary driving force for the local interannual change in the marine environment [22,30,33]. Previous work has found that warmer/colder-than-normal winters in the Bering Sea tend to be associated with an anomalous Aleutian low [22,43].

According to the above findings, the Bering Sea wind field anomaly is the key factor in early ice formation. The local wind field anomaly is closely related to the change in the Aleutian Low. Rodionov et al. (2005) reported that changes in the position of the Aleutian low are shown to be more important than changes in its central pressure. Figure 4 shows that the low-pressure center of the SLP anomaly is in the north of the Gulf of Alaska during the HI year, and in the south of the Bering Sea near the Aleutian Islands during the LI year. As a result, we use 1000 hPa geopotential height anomaly data to identify the location of the low-pressure center of the SLP anomaly (identification area is 45~65°N, 150~220°E). Figure 8 depicts the relationship between the location of the SLP anomaly center and PC1. The Bering Sea has an abnormally large SIA in the early freeze-up period when the SLP anomaly center is in the Gulf of Alaska and the northern Bering Sea, while the Bering Sea has an abnormally low SIA when the SLP anomaly center is in the eastern Bering and the Aleutian Islands. Statistics show in Figure 8 that the anomaly center of the Aleutian low, corresponding to most stations meeting PC1 > 0.8, is in the Gulf of Alaska. In contrast, for most stations meeting PC1 < −0.8, it is in the western and southern parts of the Bering Sea. These two types of Aleutian low anomalies correspond to the W1-type and C1-type patterns identified by Rodionov et al. (2005). Two years are exceptions, 1992 (southwest of the Bering Sea, with 1.74 of PC1) and 2016 (south of the Gulf of Alaska, with −1.30 of PC1). There is no obvious rule governing the distribution of stations meeting −0.8 ≤ PC1 ≤ 0.8, and their anomalous low-pressure centers are dispersed over the Bering Sea. More detailed research on the relationship between PC1 and Aleutian low anomalies will be required in the future, using a specific classification technique described by Rodionov et al. (2005).

Based on the findings, we can conclude that the east–west swing of the anomalous low-pressure center governs the change in the local wind field in the Bering Sea, which subsequently affects poleward heat transport, hence regulating the early freeze-up. Sea surface temperature has no influence on the freeze-up in this process.

4. Discussion and Conclusions

Affected by local atmospheric and oceanic processes, the early freeze-up onset of the Bering Sea has been delayed, which has a fundamental impact on the air–sea-ice coupling interaction of the Bering Sea in the later period. Estimates have shown that the correlation coefficient between the EFO and December SIA in the past 30 years reaches −0.69, and there exists an inverse correlation between the EFO and the following January SIA of −0.49 (p value < 0.01). In this paper, the poleward heat transport generated by the surface current is regarded as an essential regulatory mechanism for the early freeze-up in the Bering Sea. In addition, because surface currents in the northern Bering Sea flow northward, heat transport across the Bering Strait can also be used to qualitatively describe the heat transport over the northern Bering Sea. Its correlation coefficient with PC1 is computed to be −0.71 (p value < 0.01), which further validates the findings.

Cooling at the ice edge and growth/melt in sea ice influence ocean stability and heat transport in the ocean, making it difficult to establish a causal relationship between oceanic heat transport and sea-ice extent [44], resulting in less research on the effect of ocean heat transport. Furthermore, the flow in the Bering Sea shelf is relatively slow. It is widely assumed that the heat transport caused by shelf currents in the Being Sea has little effect on sea ice. Based on the findings of the paper and Wang et al. (2022), while the Bering Sea ice continues to advance southward under the influence of the wind field in December and January, its area is primarily governed by warm advection. The influence of the wind field on the interannual fluctuation of SIA may be less essential than being generally considered the primary driving mechanism for the southward expansion of sea ice.

Ekman drift generated by wind field anomalies may modify heat transport to the north, which may be evidence for the traditional assumption that wind fields modulate sea ice areas. The SIA during early winter is primarily affected by poleward heat transport, but the influence areas in December and January are not coincidental, implying that the influence area is ephemeral. The sea ice area, on the other hand, is the result of long-term buildup. The mismatch of the two-time scales also contributes to the inability to precisely determine the causal relationship between SIA and PHT throughout the winter.

In conclusion, poleward heat transport of warm water in the northern Bering Sea and the east–west movement of the Aleutian low anomaly appear to be the dominating processes controlling early freeze-up. In November, the eastward (westward) movement of the low-pressure anomaly center leads to the northerly (southeasterly) anomaly over the northern Bering Sea, which suppresses (enhances) local seawater warming. On the other hand, the northerly (southeasterly) anomaly weakens (enhances) poleward heat transport, accelerating (slowing) the freeze-up of seawater in the northern Bering Sea. These findings add to our understanding of the research on ice–air–ocean coupling interactions. We further discuss the roles of warm advection and the abnormal wind field in the interannual variation in SIA in December and January. Our findings suggest that the dragging effect of wind on sea ice in the Bering Sea may be less significant than previously understood, and the true role of the wind field is more likely to generate Ekman drift, drive warm water northward, and thereby affect sea ice.

The Bering Sea ice change is the outcome of interactions between atmospheric and oceanic forcing. It is insufficient to investigate sea ice change as a result of atmospheric forcing alone. Due to the unique characteristics of oceanic forcing in the Bering Sea, the simple and straightforward linear relationship appears incapable of directly describing the coupling interaction between sea ice and warm advection in winter. It is necessary to develop a new methodology for characterizing the high noncoincidence in the monthly PHT influence area, as well as to thoroughly investigate the effect process of atmospheric and oceanic forcing on sea ice.