Thermodynamical and Dynamical Impacts of an Intense Cyclone on Arctic Sea Ice

Abstract

This study investigates the thermodynamical and dynamical influences of the intense cyclone in the Arctic Ocean in August 2012 on the synoptic-scale sea ice evolution, using the Arctic Ice Ocean Prediction System (ArcIOPS). As it is hard to fully isolate sea ice loss owing to extreme cyclone from that owing to background atmospheric state in most previous studies on this topic, this study introduces a newly developed algorithm to remove the cyclone component in atmospheric forcing, and conducts sea ice and heat budget analyses in two simulations driven by atmospheric forcing with and without the cyclone. Strong impact of the intense cyclone on sea ice locates on the east side of the cyclone's path, that is, Pacific Arctic for this case. The cyclone affects sea ice in two ways. First, cyclone-induced enhancement in ice-ocean interaction leads to increased sea ice basal melt in the Pacific Arctic and part of the Atlantic Arctic, which induces strong sea ice area and volume loss when the cyclone's intensity peaks. Second, as the cyclone strengthens, the increases in air temperature, humidity and wind speed accelerate turbulent heat exchange at the air-ocean and air-ice interfaces, leading to enhanced local sea ice surface melt in the Chukchi Sea and northern Beaufort Sea. The cyclone-induced strong winds stir sea ice leading to enhanced gradients in sea ice velocity field and thus increased sea ice deformation, which further induces strong sea ice area loss. This study also demonstrates a precise atmospheric forcing field is essential for sea ice modeling.

Key Points

• Removing a cyclone from the atmospheric forcing in an Arctic model reveals its contributions to sea ice loss along its trajectory

• Oceanic and turbulent heat fluxes are the key factors driving enhanced ice basal and surface melt during an intense cyclone in the Arctic

• Strong sea ice loss locates on the east side of the trajectory of an intense cyclone

Plain Language Summary

We use a numerical model to study the effects of a strong Arctic cyclone on sea ice. First, we use a newly developed “cyclone removal algorithm” to remove the cyclone from the atmospheric data that is used to force the model. We can then compare how the sea ice changes between the model runs with and without the cyclone. We find that the cyclone has a strong local impact on sea ice, and sea ice losses are different on the two sides of the cyclone's path. The cyclone results in warm, moist air and strong wind, which leads to strong sea ice surface melting. At the same time, the cyclone increases the transfer of heat from the ocean to the sea ice, causing strong sea ice bottom melting. The strong wind also leads to more sea ice ridging and a reduction in sea ice area. Our study indicates that the effects of cyclones on sea ice are complex and that atmospheric data must accurately represent cyclones if we want to correctly model changes in the Arctic sea ice.

1 Introduction

Arctic sea ice extent has declined significantly since 1979 (Parkinson and DiGirolamo, 2016; Simmonds and Li, 2021), with drastic reduction in late summer and early autumn (Cavalieri & Parkinson, 2012). Sea ice thickness and volume have also declined, with the most noteworthy phenomenon being the shrinking of the multiyear ice zone (Bi et al., 2018; Kwok, 2018). Along with the thinning trend in the Arctic multiyear sea ice, the area covered by seasonal sea ice in the Arctic has significantly increased, particularly in the Pacific sector (Hao et al., 2020). Seasonal sea ice, normally with small thickness and low strength, is vulnerable to cyclone activities.

Cyclones affect summertime sea ice through thermodynamics and dynamics (Clancy et al., 2022; Holt & Martin, 2001; Koyama et al., 2017; Kriegsmann & Brümmer, 2014; Lammert et al., 2009; Maslank et al., 1995; Valkonen et al., 2021). Thermodynamically, cyclones typically result in increased cloud cover, which promotes sea ice melting due to the increase of downward longwave radiation (Kay & Gettelman, 2009; Wang et al., 2020) and, at the same time, impede sea ice melting due to the decrease of incoming shortwave radiation (Perovich, 2018). Warm and vapor-rich air brought by cyclones promotes sea ice loss through elevated turbulent heat exchange at the ice surface (Lee et al., 2017; Luo et al., 2017). Precipitation associated with cyclones also affects ice mass balance through modifying ice surface albedo and snow/ice integrated heat conductivity (Screen et al., 2011; Sturm et al., 2002; Webster et al., 2018). Furthermore, cyclones strongly stir the upper ocean and induce large sea ice bottom melting by bringing the warm water in the upper ocean into the surface ocean beneath the ice (Zhang et al., 2013). Dynamically, cyclones promote the fragmentation of pack ice and reduce the size of sea ice floes, leaving the sea ice more susceptible to wind and ocean surface currents (Brümmer & Hoeber, 1999). Some cyclones in the Atlantic sector of the Arctic have been found to reinforce the transpolar drift and thus reduce sea ice volume in the Arctic through pushing sea ice into lower latitudes (Semenov et al., 2019; Wei et al., 2019), while some cyclones in the Pacific sector of the Arctic affect sea ice distribution in the Arctic by preventing sea ice advection from the Canadian Basin to the Beaufort Sea and Chukchi Sea owing to the cyclonic wind anomaly (Semenov et al., 2019).

The historical summer sea ice extent minimum in the satellite era occurred in 2012. Many studies have focused on the effects of “The Great Arctic Cyclone of August 2012” (Simmonds and Rudeva, 2012) on the sea ice loss in summer 2012, and it has been suggested that the increased oceanic heat transport at the ice bottom brought by the cyclone was the main reason for the strong sea ice loss (Lukovich et al., 2021; Parkinson and Comiso, 2013; Stern et al., 2020; Zhang et al., 2013). However, there is no consensus on the lasting time of the cyclone's influence; for example, Zhang et al. (2013) found that the cyclone-induced sea ice reduction lasts for about 2 weeks over a large region, while Stern et al. (2020) suggested that the cyclone-induced sea ice reduction only lasts for a few days and locates in a small region. Lukovich et al. (2021) proposed that the cyclone-induced loss of sea ice extent depends on the time and location of the cyclone, and the sea ice volume loss caused by the extreme cyclone in 2012 is primarily a thermodynamical result. It is noteworthy that the cyclone of August 2012, with strong baroclinicity, was concurrent with a tropopause polar vortex (Simmonds and Rudeva, 2012), and its influence on sea ice loss is more drastic than that due to normal Arctic cyclones.

Some studies have also pointed out that cyclones act as a mechanism in promoting and preserving summertime sea ice extent in the marginal ice zone. Schreiber and Serreze (2020) indicated that sea ice concentration is higher after a region is influenced by a cyclone compared to when it is not. The reason is that the air temperature in the lower troposphere decreases when a cyclone occurs, and the thermodynamic-associated decrement in sea ice loss yields the dynamic-associated increment in sea ice loss. Finocchio et al. (2020) found that cyclones decelerate seasonal loss of sea ice extent in May and June owing to the reduced net shortwave radiation fluxes at the surface, while cyclones no longer decelerate seasonal loss of sea ice extent in July and August because the thermodynamical responses do not play a dominant role in sea ice extent loss in late summer.

The thermodynamical and dynamical influences of extreme cyclones on summertime sea ice evolution are very complicated and intertwined. Most previous studies on the influences of extreme storms on sea ice were based on inter-comparison of sea ice loss in different summers with or without extreme cyclones. However, in such studies it is hard to fully isolate sea ice loss due to extreme cyclones from that due to the background atmospheric state. Detailed analysis based on numerical experiments with a reliable cyclone removal algorithm will help us to further understand the roles of cyclone's thermodynamical and dynamical impacts on the Arctic sea ice during the life of a cyclone, and how these impacts are related to the cyclone-induced anomalies in air temperature, air humidity, clouds, and winds. Such studies will potentially benefit many scientific and social topics, for example, polar ocean biogeochemistry, ship navigation, and polar aviation.

This study uses a newly developed cyclone removal algorithm, an Arctic ice-ocean coupled model with data assimilation capability, and a sea ice budget analysis method to systematically assess the thermodynamical and dynamical responses of sea ice to the extreme cyclone in the summer of 2012. The paper is organized as follows: Section 2 describes the modeling system, the numerical experiments, and the cyclone removal algorithm. Section 3 briefly assesses the model performance against available observations. The thermodynamical and dynamical contributions to sea ice loss during the life of the cyclone are illustrated in Section 4. Section 5 presents heat flux budget analysis related to sea ice loss, followed by the analysis of the stability of upper ocean stratification in Section 6. A discussion and conclusion are given in Section 7.

2 Model and Numerical Experiments

2.1 Arctic Ice Ocean Prediction System

The Arctic Ice Ocean Prediction System (ArcIOPS), operated by the National Marine Environmental Forecasting Center in China, is the operational numerical forecasting system for the Arctic synoptic-scale sea ice evolution. The system constitutes an Arctic sea ice-ocean coupled numerical configuration based on the Massachusetts Institute of Technology General Circulation Model (MITgcm; Marshall et al., 1997), and an Ensemble-based Kalman Filter (EnKF) data assimilation model configured on the Parallel Data Assimilation Framework (PDAF; Nerger et al., 2012), which can assimilate sea ice and ocean observations simultaneously (Liang et al., 2020; Mu et al., 2019). The ensemble size of the ArcIOPS is 12. The average horizontal resolution is 18 km with a horizontal grid of 420 × 384 points. The open boundaries are close to 55°N both in the northern Pacific and northern Atlantic oceans. The sea ice model of the ArcIOPS uses viscous-plastic dynamics and zero-layer snow/ice thermodynamics (Hibler, 1980). The ocean model of the ArcIOPS has 50 vertical levels, with the top 10 layers in the upper 100 m. The K-profile parameterization (KPP) scheme is used to determine vertical mixing in the ocean model (Large et al., 1994). Seven atmospheric parameters are used to drive the system, including 2 m air temperature, 2 m specific humidity, 10 m wind speed components (u and v), downward shortwave and longwave radiation at sea surface, and precipitation. The bulk formula is used to calculate the surface heat budget in the ArcIOPS (Adcroft et al., 2019).

2.2 Numerical Experiments

We choose “The Great Arctic Cyclone of August 2012” (Simmonds and Rudeva, 2012; Figure 1) to study its thermodynamical and dynamical impacts on the Arctic sea ice evolution. According to the 3-hourly Japanese 55-year Reanalysis (JRA-55) fields (Harada et al., 2016; Kobayashi et al., 2015), the cyclone formed on August 2 over Siberia and moved northeastward (Figure 1d), reached the ocean on August 4, and grew to a strong cyclone on August 5 with a central pressure of 987 hPa (Figure 1g). The cyclone moved poleward over the East Siberian Sea and reached the minimum central pressure of 964 hPa at 21 UTC August 6. The maximum wind speed was 19.3 m s⁻¹. It then moved toward the Canadian Arctic Archipelago and weakened gradually. At 00 UTC August 13, the central pressure of the cyclone increased to 1,002 hPa. The cyclone lasted for almost 13 days and died down on August 15 in the Canadian Arctic Archipelago. Arctic cyclones with intensity and longevity similar to this cyclone are rare, particularly in August (Simmonds and Rudeva, 2012; Zhang et al., 2013).

Figure 1

Spatial distributions of (a, b) wind and sea level pressure (SLP), and (d, e) 2 m air temperature at 12UTC August 6, 2012 in the atmospheric forcing fields used in (a, d) the RealExp run, and (b, e) the NoCycloneExp run. (c) The difference between (a) and (b). (f) The difference between (d) and (e). The red lines represent the accurate extent of the cyclone at 12 UTC August 6. In (a, b, c) color shadings denote wind speed, arrows denote wind vectors, and white contours denote the SLP with an interval of 8 hPa. In (d) the purple line denotes the trajectory of the cyclone from 00 UTC August 4 to 00 UTC August 13, the white dots denote the central positions of the cyclone with a time interval of 3 hr, the yellow square and diamond denote the central positions of the cyclone at 00 UTC August 4 and 00 UTC August 13, respectively. (g) Time series of the central pressure of the cyclone. (h) Time series of the mean SLP (blue lines) and wind speed (orange lines) during August 4–12 in the accurate extent of the cyclone in the two atmospheric forcing fields. The solid and dashed lines represent the RealExp and NoCycloneExp runs, respectively.

In this study, we used the ArcIOPS to conduct our analysis. The 3-hourly JRA-55 reanalysis fields were used to force the system as the model parameters had been optimized under the JRA-55 fields. We carried out two steps to perform the numerical modeling.

In the first step, initialized from the twelve ensemble historical restart files of the ArcIOPS on 1 January 2012, the system were integrated twelve times from 1 January 2012 to 28 July 2012 forced by the 3-hourly JRA-55 fields. The twelve ensemble restart files at 00UTC 29 July 2012 were saved to use in the next step. During this step we assimilated satellite-retrieved daily sea ice concentration and sea surface temperature (SST) data in ice-free regions into the system every day. The sea ice concentration observations, obtained from the University of Hamburg, were derived from the Special Sensor Microwave Imager Sounder (SSMIS) brightness temperature data (Cavalieri et al., 2011; Kaleschke et al., 2001). The SST observations, obtained from the United Kingdom Met Office (UKMO), were derived from the Group for High-Resolution SST Multi-Product Ensemble (GMPE) SST data (Martin et al., 2012). We also assimilated two kinds of satellite-retrieved sea ice thickness data in January–April 2012 into the system. Daily sea ice thickness observations in thin ice zones (<1 m), assimilated on a daily basis, were derived from the Soil Moisture Ocean Salinity (SMOS) brightness temperature data (Tian-Kunze et al., 2014). Weekly sea ice thickness observations, assimilated every 7 days, were derived from the European Space Agency satellite mission CryoSat-2 radar altimetric measurements (Laxon et al., 2013; Ricker et al., 2014). We assimilated all the observations into the ArcIOPS by an Ensemble-based Kalman Filter (EnKF) data assimilation scheme (Liang et al., 2020), which improved the reliability of the modeled sea ice and ocean evolution in January-July 2012 and laid a good foundation for the following numerical simulation.

In the second step, initialized from the ensemble mean field of the twelve restart files of the ArcIOPS at 00UTC 29 July 2012, we designed two experiments without any data assimilation to obtain a more realistic simulation result. The two experiments have the same configuration except the atmospheric forcing. One experiment, denoted by RealExp, uses the 3-hourly JRA-55 fields from 00UTC 29 July 2012 to 00UTC 1 September 2012 to drive the ArcIOPS. The other experiment, denoted by NoCycloneExp, uses the same JRA-55 fields from 00UTC 29 July 2012 to 00UTC 1 September 2012 but with the removal of the cyclone component from 00UTC 4 August 2012 to 21UTC 12 August 2012. This period was selected according to the intensity of the cyclone.

2.3 Removal of the Cyclone

In the JRA-55 fields, a real forcing field at a given time can be expressed as:

(1)

where the climatological field represents the large-scale general features of the atmosphere at the given time, while the disturbance field is the deviation of the real field from the climatological field at the given time. In this study, we define the climatological field as the 10-year averaged annual cycle of three-hourly output from JRA-55 for the period 1996–2005. The disturbance field is then taken as the difference between the JRA-55 field for a specific time and the climatological field for the same time. The disturbance field includes the features in a restricted region depicting the analyzed cyclone (called “cyclone component”) and any other non-basic features (called “non-cyclone component”). The non-cyclone component also includes features due to the Arctic climate change between the 1996–2005 climatology and 2012. The JRA-55 field with the removal of the cyclone component is the sum of the climatological field and the non-cyclone component.

We applied the following five procedures to the JRA-55 field at each time step from 00UTC 4 August 2012 to 21UTC 12 August 2012 (Figure S1 in Supporting Information S1):

1. Found the central position of the cyclone by checking the minimum pressure in the affected region of the analyzed cyclone.

2. Determined the approximate extent of the cyclone using the positions of the outermost closed isobar (P_e) of sea level pressure in the affected region of the analyzed cyclone.

3. Based on the method described by Kurihara et al. (1993, 1995), we converted the original wind components in the Cartesian coordinate system to wind components in a polar coordinate system with the pole located at the central position of the cyclone, and then we determined the “accurate” extent of the cyclone by testing the radial profiles of the tangential component of the wind (v_tan(r, θ)), in each of 24 directions (15° angle between adjacent two directions) originating from the central position of the cyclone. We tested each profile from the central position of the cyclone to the margin of the cyclone with an increment of 10 km, to determine the accurate cyclone radius on that direction. We expected large changes in atmospheric variables in the region between the cyclone and the adjacent anticyclone (Wernli and Papritz, 2018); therefore, we defined the radius in that direction as the distance between the central position of the cyclone and the point where the conditions of v_tan < 3 m s⁻¹, pressure P > (P_e − 3) hPa and wind speed U < 5 m s⁻¹ were satisfied at the same time. If the pressure at the selected point was less than P_e, then we redefined the radius in that direction as the distance between the central position of the cyclone and the location where P = P_e.

4. Removed the cyclone component from the disturbance field using a 9-point smoothing operator iteratively over each point in the accurate extent of the cyclone (Winterbottom and Chassignet, 2011), which can be expressed as:

   (2)
   where (L, K) represent the respective grid coordinates of the smoothed atmospheric variable (H^t) which is calculated from the atmospheric variable in the previous iteration (H^t−1). Meanwhile, the change in spatial variance in the accurate extent of the cyclone was computed as:
   (3)
   where and represent the spatial variance of the smoothed variable and that of the atmospheric variable in the previous iteration, respectively. The iteration stopped when the Δσ² reached the given threshold values for the respective variables, which were set to 1.0 × 10⁻⁴ for 10 m wind speed components (m s⁻¹) and downward shortwave (W m⁻²) and longwave radiation (W m⁻²) at sea surface, 1.0 × 10⁻⁵ for 2 m air temperature (°C), 1.0 × 10⁻¹⁰ for 2 m specific humidity (kg kg⁻¹) and 1.0 × 10⁻²⁰ for precipitation (m s⁻¹), respectively. The 9-point smoothing operator is much more suitable for calculating the variables in the planetary boundary layer (PBL) than that based on potential vorticity inversion (Arakane & Hsu, 2020). The disturbance field after the iteration excludes the major part of the cyclone component and, therefore, can be regarded as the non-cyclone component.

5. Added the non-cyclone component to the climatological field to obtain the atmospheric forcing field with the removal of the cyclone component.

Figure 1h shows the time series of sea level pressure (SLP) and wind speed averaged in the accurate extent of the cyclone in the two sets of atmospheric forcing fields with and without the cyclone component. The tendencies of the mean SLP and wind speed in the two sets of atmospheric forcing fields are totally different. Associated with the growing-decaying of the cyclone, the mean SLP in the extent of the cyclone in the atmospheric forcing fields used in the RealExp run is generally lower than 1,003 hPa before August 10, while the wind speed increases before 12UTC August 6 and thereafter decreases (Figure 1h). However, the mean SLP in the atmospheric forcing fields used in the NoCycloneExp run is higher than that in the RealExp run by 6 hPa on average during the life of the cyclone, and the mean wind speed is always 2–4 m s⁻¹, which is lower than that in the RealExp run with a magnitude of 5–9 m s⁻¹ (Figure 1h). Figures 1a–1f show the spatial patterns of the SLP, wind speed and 2 m air temperature at 12UTC August 6 in the two atmospheric forcing fields, and their differences. The SLP and wind speed used in the NoCycloneExp run are moderate over the Arctic Ocean (Figure 1b). In the extent of the cyclone the wind speed in the RealExp run is larger than that in the NoCycloneExp run by up to 16 m s⁻¹ (Figures 1a–1c). In the region south of 80°N, the 2 m air temperature used in the NoCycloneExp run is lower than that in the RealExp run (Figure 1f). Furthermore, in the central Arctic the 2 m air temperature used in the NoCycloneExp run is slightly higher than that in the RealExp run (Figure 1f). This pattern is consistent with Schreiber and Serreze (2020) who showed that the air temperature in lower troposphere decreased after experiencing a cyclone in summertime.

In summary, the cyclone removal algorithm reasonably eliminates the major influences of the cyclone in the JRA-55 fields. Therefore, the comparison between the two runs can represent the net impacts of the strong cyclone on the Arctic sea ice evolution. The 6-hourly model results from July 29 to August 31 were output and used in our analysis. We define two study regions (Figure 2) for our analyses. As the selected cyclone in this study is very large and covers the entire Arctic Ocean, we conduct sea ice budget analysis for Region A, encompassing most of the Arctic Ocean. We define a smaller Region B, focused on the Pacific Sector, to highlight the responses of the upper ocean condition to the cyclone, based on evidence that the cyclone-induced enhanced ice-ocean heat flux plays an important role in sea ice reduction and the enhanced ice-ocean heat flux locates in the Pacific Arctic as shown in Sections 4 and 5.

Figure 2

Definition of the study areas, Region A (blue line) and Region B (red line).

3 Validation of the Simulation Result

We compare the modeled sea ice area (SIA), sea ice extent (SIE), and sea ice volume (SIV) in Region A to the observations and reanalysis data. To avoid the uncertainty in the satellite sea ice concentration (SIC) observations, the grid cells with SIC <15% in the corresponding data set are excluded in calculations for both the model output and observations in this section. Two kinds of satellite-retrieved SIC data are used in the comparison, the Ocean and Sea Ice Satellite Application Facility (OSISAF; Tonboe et al., 2016) SIC data obtained from the Norwegian Meteorological Institute, and the Advanced Microwave Scanning Radiometer 2 (AMSR2) SIC data obtained from the University of Bremen (Spreen et al., 2008). The OSISAF data with a spatial resolution of 10 km is derived from the SSMIS brightness temperature data by implementing a hybrid algorithm. The AMSR2 data with a spatial resolution of 6.25 km is also retrieved from brightness temperature data with the ARTIST Sea Ice (ASI) algorithm (Spreen et al., 2008). Sea ice volume reanalysis data is from the Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS; Zhang and Rothrock, 2003) with a mean horizontal resolution of 22 km in the Arctic. Due to the assimilation of the SIC data from the National Snow and Ice Data Center (NSIDC) and the SST data from the NCEP/NCAR Reanalysis, the PIOMAS SIV data are a reliable source to assess the modeled SIV (Schweiger et al., 2011). The SIC data from the PIOMAS are also used in the comparison.

To quantitatively validate the modeled sea ice, we divide the life of the cyclone into three phases according to its intensity: 4–9, 10–17, and 18–31 August (Table 1). During August 4–9 (Phase 1), the cyclone grows and strengthens, and the SIA and SIE reduce quickly in the RealExp run (Figure S2a and S2b in Supporting Information S1). The reduction rates of the SIA and SIE in the RealExp run agree well with the AMSR2 observations. It is noteworthy that the reduction rate of the SIA in the OSISAF data is significantly lower than those from other data sources (Table 1; Text S1 in Supporting Information S1). During August 10–17 (Phase 2), the reductions of the SIA decelerate in the PIOMAS and the RealExp run, which is consistent with the observations. However, the SIA and SIE in the RealExp run reduce faster than those in the PIOMAS and the observations, probably owing to the relatively higher sea ice concentration in the Pacific sector of the Arctic on August 9 (Figure S3 in Supporting Information S1). During August 18–31 (Phase 3), along with the diminishing of the cyclone, both the reduction rates of the SIA and SIE in the RealExp run and in the observations are greatly reduced (Table 1). Similar to the modeled SIA evolution, the modeled SIV in the RealExp run shows rapid reduction during August 4–9, and slow reduction after August 9 (Table 1; Figure S2b in Supporting Information S1). The reduction rate of the SIV in the RealExp run are 128.1, 64.4 , and 32.5 km³ d⁻¹ during the three phases, which close to that in the PIOMAS data. Comparing with the PIOMAS data, the RealExp run produces thicker ice in the pack ice zone north of the Canadian Arctic Archipelago and thinner ice in the Pacific sector of the Arctic on 4, 9, and 17 August (Figure S4 in Supporting Information S1). Overall, our model produces reliable spatial patterns and reduction rate of the Arctic sea ice in response to the extreme cyclone, which sets a stable foundation for the following analysis.

Table 1. The Changes of the SIA, the SIE, and the SIV During 4–9, 10–17, and 18–31 August in Different Data Sources

Variable	Period	AMSR2	OSISAF	PIOMAS	RealExp	NoCycloneExp
SIA (10^5 km^2  d^−1)	August 4–9	−1.3	−0.3	−0.7	−1.3	−0.9
	August 10–17	−0.3	−0.2	−0.3	−1.0	−1.1
	August 18–31	−0.4	−0.2	−0.2	−0.3	−0.3
SIE (10^5 km^2 d^−1)	August 4–9	−1.2	−1.3	−0.9	−1.2	−0.6
	August 10–17	−0.6	−0.7	−1.0	−1.8	−1.8
	August 18–31	−0.5	−0.4	−0.5	−0.5	−0.7
SIV (km^3 d^−1)	August 4–9	–	–	−146.2	−128.1	−98.9
	August 10–17	–	–	−78.6	−64.4	−74.1
	August 18–31	–	–	−48.2	−32.5	−35.6

4 Sea Ice Budget Analysis

4.1 Sea Ice Area

The sea ice model in ArcIOPS divides each grid cell into two parts: the ice-covered area and the open water area. The ratio of ice-covered area to the grid cell area is sea ice concentration. The change of sea ice concentration in each grid cell is determined by several factors, including sea ice-atmosphere heat flux, oceanic heat flux at ice bottom, atmosphere-open water heat flux, sea ice advection, and sea ice deformation. In the melt season, the atmosphere-ice heat flux leads to ice surface melting and the oceanic heat flux leads to ice basal melting. The atmosphere-open water heat flux will be primarily used to melt the sea ice in the local grid cell, and after the sea ice melts out in the local grid cell, the residual heat will be used to warm the seawater, which then indirectly affects the sea ice or seawater in the surrounding grid cells by the advection and diffusion in the ocean. Besides the weak heat exchange between the surface mixed layer and the deeper layer, the entrainment of warm water into the surface mixed layer from below, normally occurred under strong cyclones' influence, contributes to changes in the surface mixed layer temperature, which impacts the oceanic heat flux at the ice bottom. As described in Liang et al. (2022a, 2022b), the change of the SIA in a given region with area S over a period t can be described as:

(4)

in which, is the total change of the SIA over the period, , , and represent the rates of change of sea ice concentration due to the oceanic heat flux at ice bottom, sea ice-atmosphere heat flux, and atmosphere-open water heat flux, respectively. The first three terms on the right side of Equation (4) are the thermodynamical impacts of ocean and atmosphere on the SIA, and the last two terms generally represent the dynamical impacts of ocean currents and winds on the SIA. The subscripts (x, y) represent the two orthogonal axes, are the components of advection of sea ice concentration, and represents the integral over the given region and the period, that is, , in which dS = dxdy is the area element of the integration, and dt is the time element of the integration. Moreover, represents the change of the SIA due to the net advection of sea ice area across the boundary of the region. As this study uses many symbols, a list of the symbols and their descriptions are given in Appendix A for convenience. In our model setting, variables (, , , , ) are directly saved by the model every 6 hr, thus dt = 21,600 s. is calculated as the residual term, which mainly includes the contributions from sea ice ridging processes and small computational errors owing to the discretization of the continuous equations.

The sea ice area budget analysis of the two runs, and their differences, are shown in Figure 3. In the RealExp run (Figure 3a), the 6-hourly reduction of the SIA () increased significantly during August 4–14, and the largest rate of reduction of 1.7 × 10⁵ km² d⁻¹ occurred on August 8, while the reduction rate of the SIA in the rest of our study period was less than 1.0 × 10⁵ km² d⁻¹. This result means that enhanced sea ice area loss occurred when the cyclone passed through the Arctic Ocean. The shape of the evolution of is similar to that of during August 4–12. The sea ice area loss due to oceanic heat flux at ice bottom, that is, , increased rapidly during August 4–7, with the largest rate of reduction of 3.6 × 10⁴ km² d⁻¹ on August 7, which implies that the enhancement of Ekman pumping and turbulent mixing in the ocean boundary layer induced by the cyclone leads to strong sea ice basal melt. The contribution from oceanic heat flux at the ice bottom accounted for 24.3% of on August 7. The sea ice area loss due to atmosphere-open water heat flux, i.e., , also increased during August 4–7 with the largest reduction rate of 5.6 × 10⁴ km² d⁻¹ on August 7, which suggests that the local sea ice melted rapidly when the cyclone moved across the Arctic Ocean, owing to the increased atmospheric heat flux entering the ocean through sea ice leads. The contribution from atmosphere-open water heat flux accounted for 37.7% of on August 7. The sea ice area loss due to sea ice-atmosphere heat flux, that is, , reduced from the value of 2.8 × 10⁴ km² d⁻¹ on August 4 to the value of 1.6 × 10⁴ km² d⁻¹ on August 12 without enhancement when the intensity of the cyclone peaks. was quite small compared with other terms, indicating that sea ice area loss due to sea ice advection out of the Region A was almost negligible. The sea ice area loss due to the residual processes, that is, , increased gradually during August 4–8, and then stayed at a high level with a contribution of about 40% to during August 8–16, and thereafter reduced to the previous level until August 25. This result shows that, due to the acceleration of sea ice movement during the developing period of the cyclone, sea ice ridging is substantially enhanced which leads to strong sea ice area loss, and this effect lasts for a long time even after the cyclone has died down.

Figure 3

The terms of the 6-hourly SIA changes within Region A in (a) the RealExp run, (b) the NoCycloneExp run, and (c) the differences between (a) and (b). The black, blue, orange, yellow, purple and green lines represent the , , , , , and terms, respectively. Negative values in (c) mean that sea ice area loss in the RealExp run is stronger than that in the NoCycloneExp run. The period when the cyclone is removed is shaded by light gray.

In the NoCycloneExp run during August 4–12 (Figure 3b), and were always nearly zero, suggesting that their contributions to were negligible in the absence of the cyclone, and almost remained unchanged. The sea ice area loss corresponding to decreased at a rate higher than that of the RealExp run. The changes of and during the cyclone period were relatively small in the NoCycloneExp run comparing to the RealExp run. The comparison of each term during the life of the cyclone between the two runs (Figure 3c) shows that, and are the two terms with significant differences, the differences in and are relatively small, and there is almost no difference in terms.

In conclusion, the thermodynamical impacts of the cyclone on the sea ice area, as revealed from the , , and terms, have large differences during the life of the cyclone, and decrease quickly when the cyclone dissipates. The dynamical impact due to sea ice ridging gradually increases during the life of the cyclone, and its influence persists for a longer time than that induced by thermodynamics. Note that the sea ice area loss due to the residual processes (R_A) in the NoCycloneExp run increased quickly after August 10, and peaked on August 17. This result partly arose from a cyclone occurring in middle August, which drove the enhanced sea ice ridging process as also revealed in the RealExp run. During August 16–19, the difference in the R_A dominated the difference in between the two runs.

The three thermodynamical terms have large differences during August 6–8 (Figure 3c). In the RealExp run, the strong cyclone covered most regions in the Arctic Ocean during this period. Sea ice concentration in the Pacific sector of the Arctic was substantially lower than that in the Atlantic sector of the Arctic both in the RealExp and NoCycloneExp runs (Figures 4a–4c). The SLP in the NoCycloneExp run was fairly homogeneous in the Arctic Ocean (Figure 4b), indicating the removal of the cyclone. The sea ice concentration in the Pacific sector of the Arctic in the NoCycloneExp run was much higher than that in the RealExp run (Figure 4c), which implies that the main affected area of the cyclone is the Pacific sector of the Arctic.

Figure 4

Spatial distributions of the SLP and sea ice concentration (SIC) (first row), the 6-hourly (second row), (third row), (fourth row), (fifth row), and R_A (sixth row) terms averaged during August 6–8 for each grid cell. The left, central, right columns denote the RealExp run, the NoCycloneExp run, and the differences between the RealExp and NoCycloneExp runs. The black contours in (a) and (b) represent the SLP in the RealExp run and the NoCycloneExp run, respectively. The interval of the SLP contour is 4 hPa. The purple lines represent the mean sea ice edge during August 6–8 in the RealExp run. The black line in (c) represents the trajectory of the cyclone. The differences in the SIC, , , , , and R_A terms between the RealExp and NoCycloneExp during August 6–8 runs at a 95% confidence level are shaded in the right column based on independent-sample student t-test. The negative value means SIA loss in the grid cell. Note the changed color ranges in different panels.

Comparing with the NoCycloneExp run, sea ice concentration increased near the sea ice edge in the Laptev Sea in the RealExp run, resulting from the movement of sea ice toward the above-mentioned areas caused by the cyclone-induced northeastward wind anomaly. Sea ice area loss due to in the RealExp run was substantially higher than that in the NoCycloneExp run (Figure 4f), and the activity of the cyclone primarily resulted in the increase in sea ice area loss due to in the Pacific sector of the Arctic (Figure 4d). Moreover, it is evident that sea ice area loss due to in the marginal ice zone, where sea ice concentration is lower than 0.8, was larger than that in the pack ice zone in the NoCycloneExp run (Figure 4e): this was because the upper ocean mixed layer in the marginal ice zone was warmer than that in the pack ice zone and the turbulent upward mixing of seawater also benefited from the faster sea ice drift in the marginal ice zone (with a mean velocity of ∼5.8 cm s⁻¹ in the marginal ice zone compared with ∼3.1 cm s⁻¹ in the pack ice zone), which induced larger heat transport from the ocean to the overlaying sea ice. Sea ice area loss due to was also larger in the marginal ice zone than that in the pack ice zone (Figures 4g and 4h), while the significantly enhanced sea ice area loss due to mainly occurred near the sea ice edge and in part area of the central Arctic when the cyclone passed (Figure 4i). The negative value at the sea ice edge in the Pacific sector of the Arctic in Figure 4i arose from the fact that sea ice remained in those locations in the NoCycloneExp run but melts out in the RealExp run. Sea ice area loss due to was relatively weak compared to other thermodynamical terms, but it was also enhanced in the Chukchi Sea and northern Beaufort Sea in the RealExp run (Figures 4j–4l). The sea ice area loss due to dynamic processes was strengthened in the marginal ice zone in the Pacific sector of the Arctic in the RealExp run (Figures 4m–4r), mainly owing to the strong winds induced by the cyclone. The differences in , , and between the NoCycloneExp run and the RealExp run were not significant in the central Arctic and the Atlantic sector of the Arctic where sea ice concentration was large, while the difference in sea ice concentration was significant at the 95% confidence level in the entire Arctic Ocean. This means that the sum of sea ice area loss due to , , , and the dynamic terms is significant in these regions, despite the individual contribution to the SIA loss is small and statistically insignificant.

4.2 Sea Ice Volume

Similar to Equation (4), the change of SIV in a given region over a period can be expressed as (Liang et al., 2022a, 2022b):

(5)

where is the total change of the SIV over the period, and , , and are the rates of change in cell-mean sea ice thickness due to oceanic heat flux at ice bottom, sea ice-atmosphere heat flux, and atmosphere-open water heat flux, respectively. represents the contribution of snow flooding to the change in cell-mean sea ice thickness. When snow thickness on sea ice continuously increases, part of the snow submerges below the sea level due to the instability of buoyancy and converts to ice. are the components of advection of cell-mean sea ice thickness. The symbol , which is the same as that in Equation (4). Variables (, , , , , ) are also directly recorded by the model. is the residual term. As sea ice ridging process changes sea ice area but not volume, R_V only includes the small computational errors which are always near zero (Figure 5).

Figure 5

The terms of the 6-hourly SIV change within Region A in (a) the RealExp run, (b) the NoCycloneExp run, and (c) the differences between (a) and (b). The black, blue, orange, yellow, magenta, purple and green represent the , , , , , , and , respectively. Negative values in (c) mean that sea ice volume loss in the RealExp run is stronger than that in the NoCycloneExp run. The period when the cyclone is removed is shaded by light gray.

The sea ice volume budget analysis for the two runs, and their differences, are shown in Figure 5. Sea ice volume loss due to was larger than the other terms in the two runs. In the RealExp run (Figure 5a), sea ice volume losses due to thermodynamical terms, that is, , and , increased noticeably when the cyclone passed, with changing substantially. From 4 to 7 August, sea ice volume loss due to increased from 7.4 km³ d⁻¹ or 6.1% of to 35.6 km³ d⁻¹ or 26.1% of , and then decreased to the normal value on August 9. The loss of the total SIV due to sea ice-ocean heat flux reached 131.8 km³ during August 4–9, accounting for 17.6% of the total SIV change. Because the snow flooding process in the Arctic is weak (Graham et al., 2019), was close to 0. also contributed only a small amount to the total sea ice volume loss. In the NoCycloneExp run (Figure 5b), the 6-hourly SIV loss () decreased during the whole of August owing to the trend of and , and the other terms had no significant changes. The comparison between the two runs (Figure 5c) shows that the difference in sea ice volume loss generated by was larger than that by and by before August 10. During August 4–9, the loss of the SIV due to the cyclone reached 152.4 km³, in which , , and contributed 95.5 km³, 30.9 km³, and 30.7 km³, respectively (Figure 5c).

In conclusion, the three thermodynamical terms have different responses to the cyclone, with the most substantial response in . Due to the larger rate of sea ice area loss in the RealExp run during August 4–12, the sea ice area and extent on August 13 was smaller in the RealExp run than those in the NoCycloneExp run (Figure S2c in Supporting Information S1). The larger sea ice volume reduction after August 12 in the NoCycloneExp run with respect to the RealExp run was mainly due to and (Figure 5c) and because a larger sea-ice zone received more energy from atmosphere and ocean than a smaller one.

Figure 6 shows the spatial distributions of the cell-mean sea ice thickness and sea ice motion, the 6-hourly , , , , and R_V terms averaged during August 6–8 in the two runs, and their differences. In the RealExp run, sea ice movement in the Pacific sector of the Arctic was cyclonic, and the ice moved faster than that in the Atlantic sector (Figure 6a). On a basin scale, the sea ice drift speed in the RealExp was significantly larger than in the NoCycloneExp run (Figure 6c). Sea ice thickness distributions were similar in the two runs, with the thick sea ice being located at the north side of the Canadian Arctic Archipelago and thin ice located in the Pacific sector of the Arctic (Figures 6a and 6b). However, the sea ice in the RealExp run was substantially thinner than that in the NoCycloneExp run, especially in Chukchi and Beaufort seas (Figure 6c). Sea ice volume loss due to showed large values in the Pacific sector of the Arctic in the RealExp run (Figure 6f), while no obvious contribution from the was found in the NoCycloneExp run (Figure 6e). Sea ice volume loss due to was relatively large in the marginal sea ice zone in both runs (Figures 6g and 6h). Due to the strong wind induced by the cyclone, enhanced SIV loss due to mainly occurred near the sea ice edge in the Pacific sector of the Arctic (Figure 6i). Sea ice volume loss due to happened almost in the entire Arctic Ocean in both runs (Figures 6j and 6k), but was also strongly enhanced in the RealExp run, especially in the Beaufort and Chukchi seas (Figure 6l). The strong sea ice volume loss due to advection occurred in the Beaufort Sea and Chukchi Sea, especially in the marginal ice zone (Figures 6m–6r), which was related to the strong winds and low sea ice concentration. The 95% confidence intervals for the differences in sea ice thickness, , , and between the RealExp run and the NoCycloneExp run are similar to that of the differences in sea ice concentration, , , and , meaning that the significant sea ice thickness change is associated with the combination of all terms in Equation (5).

Figure 6

Spatial distributions of the grid cell-mean sea ice thickness and sea ice motion (first row), the 6-hourly (second row), (third row), (fourth row), (fifth row), and R_V (sixth row) terms averaged during August 6–8 for each grid cell. The left, central, and right columns denote the RealExp run, the NoCycloneExp run, and the differences between the RealExp and NoCycloneExp runs. The purple lines represent the mean sea ice edge during August 6–8 in the RealExp run. The black line in (c) represents the trajectory of the cyclone. The differences in sea ice thickness, the , , , and R_V terms between the NoCycloneExp and RealExp runs at a 95% confidence level are shaded in the right column based on independent-sample student t-test. The negative value means SIV loss in the grid cell. Note the changed color ranges in different panels.

5 Heat Flux Budget Analysis

We study the causes of sea ice loss in more detail by analyzing heat flux budgets. In ArcIOPS, heat flux budget in the sea ice zone can be expressed as:

(6)

where is the total net heat flux in the sea ice zone, is the net shortwave radiation heat flux, is the net longwave radiation heat flux, is the latent heat flux, is the sensible heat flux, and is the oceanic heat flux in the sea ice zone. Here, the sea ice zone includes the ice-covered area and the open water area between ice floes, that is, the region within the sea ice edge. As mentioned in Section 4.1, away from the complex processes in the real world, in the melt season the atmosphere-open water heat flux in our model will first be used to melt local sea ice in the grid cell when there is ice in the grid cell. After the sea ice melts out, the residual heat is used to warm the seawater. The heat fluxes at each time step are calculated for only those grid cells including sea ice. As revealed in Figure 5, the dynamical contribution to sea ice volume loss is small compared to the thermodynamic terms with being primarily related to , and being directly linked to .

Figure 7 shows the time evolution of each term in sea ice zone within Region A in the two runs, and the differences between them. The shape of the evolution of agrees well with in both runs. The two runs show large differences in all the terms during the life span of the cyclone, and the in the RealExp run is substantially higher than that in the NoCycloneExp run (Figure 7c). The used for the sea ice basal melt significantly increased due to the enhancement in cyclone-induced upper ocean turbulent mixing when the cyclone passed through the Arctic Ocean (Figure 7a). The in the RealExp run increased from 1.84 × 10¹⁸ J d⁻¹ on August 4 to 1.01 × 10¹⁹ J d⁻¹ on August 6, and its contribution to increased from 4.76% to 21.96% during this period (Figure 7a). However, the in the NoCycloneExp stayed nearly constant at about 1.28 × 10¹⁸ J d⁻¹ (Figure 7b). During August 5–9, the difference in accounted for about 67% of the difference in between the two runs. The daily averaged net shortwave radiation and longwave radiation within sea ice zone were 2.17 × 10¹⁹ J and −6.35 × 10¹⁸ J in the RealExp run, and 2.34 × 10¹⁹ J and −6.78 × 10¹⁸ J in the NoCycloneExp run in August, respectively (Figures 7a and 7b). Shortwave radiation is the most important source of energy for sea ice and ocean, thus the dominated the evolution of in both the two runs. The is closely related to sky condition; the increase of cloud amount induced by the cyclone reduces the shortwave radiation arriving at the ocean surface, thus the was smaller in the RealExp run than that in the NoCycloneExp run. The is also greatly affected by cloud amount. When the cloud amount increases, the downward longwave radiation increases, leading to the weakening of radiative cooling on the ice surface. The transfers heat from ocean and sea ice to atmosphere. When the cyclone occurred, net longwave radiative heat flux released by the sea ice zone decreased. The difference in the between the two runs was fairly constant near ∼2.60 × 10¹⁸ J d⁻¹ during August 5–8, and the value reduced quickly to near zero along with the decay of the cyclone (Figure 7c). Due to the warm and moist air brought by the cyclone from lower latitudes, the and also transfer heat from the atmosphere to the sea ice zone and lead to the melt of sea ice. The difference in between the two runs was large before August 9, and that in was large before August 12 (Figure 7c). The difference in , which increased during August 4–7 and decreased thereafter, changed with the intensity of the cyclone. The difference in shows comparable values to that in along with the intensifying of the cyclone. The differences in and between the two runs attribute to not only the changes in wind speed and temperature difference between air and the underlying surface, but the differences in SIA.

Figure 7

The terms of the 6-hourly heat flux budget in the sea ice zone within Region A in (a) the RealExp run, (b) the NoCycloneExp run, and (c) the differences between (a) and (b). The black lines represent the terms, the blue lines represent the terms, the orange lines represent the terms, the yellow lines represent the terms, the purple lines represent the terms, and the green lines represent the terms. The period when the cyclone is removed is shaded by light gray. Positive value means heat absorbed by sea ice/open water. Positive values in (c) mean that the heat absorbed by sea ice and open water in sea ice zone in the RealExp run is larger than that in the NoCycloneExp run.

Although the and can explain the evolution of and , the processes involved in the change of the and are not clear. Hence, we conduct separate heat flux budget analyses for ice-covered and open water areas. The heat flux budget over sea ice and open water can be written as:

(7)

The terms in Equation (7) have the same meaning as in Equation (6), except the subscripts “ice” and “ocn” are used to denote whether the term is computed over ice-covered or open water areas.

Sea ice extent, ice-covered area and open water area in the sea ice zone within Region A in the NoCycloneExp run were larger than those in the RealExp run when the cyclone passed through (Figures 8a–8c). In the ice-covered area, the difference in was the largest term dominating the difference in between the two runs (Figure 8f). The difference in between the two runs was large during August 4–8 (Figure 5c). The mean value of the difference in the turbulent heat fluxes (sensible plus latent heat fluxes) during the same period between the two runs was 2.36 × 10¹⁸ J d⁻¹, and the value of the radiative heat fluxes (shortwave plus longwave heat fluxes) was 0.56 × 10¹⁸ J d⁻¹. Thus, the change of was primarily dominated by the turbulent heat fluxes; that is, the strong wind disturbed the atmospheric boundary layer, resulting in the increase of the turbulent heat transfer from air to ice, further enhancing ice surface melt. The evolution of was closely related to in the two runs (Figures 8d and 8e). The difference in between the two runs reached the largest value during August 7–8 (Figure 8f). In the RealExp run, the change in was consistent with that in ; heat released by open water area due to longwave emission decreased when the cyclone strengthened and increased when the cyclone weakened (Figure 8d). Affected by the cloud amount, the in the RealExp run was less than that in the NoCycloneExp run. However, the magnitude of the change of without and with the cyclone exceeded that of (Figure 8f). As a result, the radiative heat flux absorbed by the open water area in the RealExp run was much smaller than that in the NoCycloneExp run. Meanwhile, when the cyclone passed through Arctic Ocean, both the and absorbed by the open water grew substantially, particularly the changes in the (Figure 8f). Considering all the heat flux budget terms in the open water area together, during the life span of the cyclone the increase in turbulent heat flux absorbed by open water area overcame the decrease in radiative heat flux entering the ocean, and thus led to enhanced , which is consistent with Stern et al. (2020).

Figure 8

Time series of sea ice extent, ice-covered area and open water area (top panels), the terms of the 6-hourly heat budget over ice-covered area (middle panels) and open water area (bottom panels) in the sea ice zone within Region A. The left, middle, and right panels show the RealExp run, the NoCycloneExp run, and the differences between the RealExp and NoCycloneExp runs. In the middle panels, the black lines denote the terms, the blue lines denote the terms, the orange lines denote the terms, the yellow lines denote the terms, the purple lines denote the terms, and the green lines denote the terms. In the bottom panels, the black lines denote the terms, the blue lines denote the terms, the orange lines denote the terms, the yellow lines denote the terms, the purple lines denote the terms. Positive values in (c) mean that the heat absorbed by sea ice in the RealExp run is larger than that in the NoCycloneExp run. Positive values in (f) mean that the heat absorbed by open water in sea ice zone in the RealExp run is larger than that in the NoCycloneExp run. The period when the cyclone is removed is shaded by light gray.

The oceanic heat flux at the ice base, turbulent heat fluxes in the ice-covered area, and open water area are the key factors showing large differences when the cyclone occurs. Figure 9 shows the spatial distributions of the 6-hourly , , , and averaged during August 6–8 in the two runs, and their differences. The in the RealExp run was larger than that in the NoCycloneExp run, and the largest difference occurred in the regions south of 80°N in the Pacific sector of the Arctic, especially in the Beaufort Sea (Figures 9a–9c). The term increased significantly in the northwestern Beaufort Sea during the life of the cyclone, while the value was quite small and homogeneous in the NoCycloneExp run (Figures 9d–9f). The enhancement was more spatially localized than the turbulent heat flux enhancement by the cyclone (Figures 9f, 9i, and 9l). The largest enhancement located in the northern Beaufort Sea with relative low sea ice concentration, where the cyclone-associated winds strongly stirred the upper ocean. The distributions of the and the were similar, with relatively larger values in the Pacific sector of the Arctic when the cyclone occurred. The difference in between the two model runs was larger than that in (Figures 9i and 9l), mainly due to the smaller difference in open water area compared to that in sea ice area (Figures 9a–9c). The largest difference in occurred in the northern Beaufort Sea and Chukchi Sea, while the largest difference in was found in the northeastern Beaufort Sea, which was related to the strong wind speed area and the changes of temperature and specific humidity.

Figure 9

Spatial distributions of the 6-hourly (first row), (second row), (third row) and (fourth row) terms averaged during August 6–8. The left, central, and right columns denote the RealExp run, the NoCycloneExp run, and the differences between the RealExp and NoCycloneExp runs, respectively. The blue lines represent the mean sea ice edge during August 6–8 in the RealExp run. The black line in (c) represents the trajectory of the cyclone. The differences between the RealExp and NoCycloneExp runs at a 95% confidence level are shaded in the right column based on independent-sample student t-test. Note the changed color ranges in different panels.

6 Stability of Upper Ocean Stratification

The ocean heat flux in the Pacific sector of the Arctic increased significantly during the life of the cyclone, which played an important role in sea ice loss. We use Brunt-Väisälä frequency to characterize the stability of ocean stratification. In the NoCycloneExp run, the upper ocean was strongly stratified with small variation in N² (Figure 10a) and nearly constant potential temperature and potential density (Figure 10d), and the heat contents in the upper 3 vertical levels increased smoothly (Figure 10b). However, beginning on August 5, N² at 10 m depth in the RealExp run decreased rapidly and N² at 20 m depth increased rapidly until August 9, and then slowly recovered. N² at 20 m depth was larger than that at 10 m depth between August 7 and August 15. Meanwhile, the potential temperature and potential density at the upper 20 m reduced during August 5–9 (Figure 10c). The upper ocean tended to be stably stratified after August 16 with the peak stratification went back to 10 m depth (Figure 10a). Moreover, the heat content of the second ocean level (10–20 m) decreased rapidly during August 5–9, and then increased gradually. This result suggests that during the life of the cyclone, the cyclone-induced strong winds stir the upper ocean, reduce the upper ocean stratification and deepen the mixed layer, which further amplify the heat conductivity between ocean mixed layer and the ice bottom and lead to the acceleration of sea ice melting.

Figure 10

The time series of (a) Brunt-Väisälä frequency and (b) heat content in the top three oceanic model layers in the two runs. Values are averaged over sea ice zone in Region B. The solid and dashed lines represent the RealExp run and NoCycloneExp run, respectively. The period when the cyclone is removed is shaded by light gray. The evolution of potential temperature and density in the upper 65 m ocean in the (c) RealExp and (d) NoCycloneExp runs. Values are averaged over sea ice zone in Region B. The contours with color and value denote the potential temperature and density, respectively.

7 Discussion and Conclusion

This paper focuses on the impact of the extreme cyclone that occurred in August 2012 on the Arctic sea ice evolution based on the Arctic Ice Ocean Prediction System (ArcIOPS). Comparison with satellite observations indicates that the ArcIOPS with an averaged horizontal resolution of 18 km can reasonably simulate the response of sea ice to the extreme cyclone. By developing a vortex isolation and removal algorithm, we removed the cyclone component from atmospheric reanalysis fields. Although it is impossible to entirely remove the cyclone component, the kinematic and thermodynamic atmospheric states without the cyclone are estimated as reasonably as possible. The thermodynamical and dynamical influences of the extreme cyclone on the Arctic sea ice evolution were studied separately based on two simulations, one driven by real atmospheric forcing and the other by non-cyclone atmospheric forcing. The differences between model were assessed with a quantitative sea ice budget analysis method.

A schematic of heat flux budget terms at the ice-ocean-atmosphere interfaces in the sea ice zone for the two simulations is shown in Figure 11. In general, the atmosphere-open water heat flux dominates the total heat which is available for sea ice melt in both simulations, and these terms in the two simulations are nearly the same. The largest heat flux difference between the two simulations is the oceanic heat flux, and the value in the RealExp run is about three times larger than in the NoCycloneExp run. The sea ice-atmosphere heat flux in the RealExp run is also slightly larger than that in the NoCycloneExp run by about 0.9 × 10¹⁹ J. Moreover, comparing with the NoCycloneExp run, the RealExp run generates larger turbulent while smaller radiative heat fluxes entering the sea ice zone, and the positive deviation of the turbulent heat flux yields the negative deviation of the radiative heat flux.

Figure 11

Schematic of heat flux budget terms at the ice-ocean-atmosphere interfaces in the sea ice zone for the two simulations. Values are integrated over 4–12 August 2012, for the area of Region A shown in Figure 2. The red and blue values denote the RealExp and NoCycloneExp runs, respectively. Values are given in units of 10¹⁹ J.

When the extreme cyclone passed through the Arctic Ocean (track shown in Figure 1d), its influence on sea ice and surface heat budget varies substantially in different regions: for this cyclone, the largest effects are found in the marginal ice zone in the Pacific sector of the Arctic, particularly in the Beaufort Sea. This localization effect partly agrees with the result in Kriegsmann and Brümmer (2014), who used statistical analyses to show that the pronounced effects of Arctic cyclones on sea ice were in the Eurasian marginal seas. Clancy et al. (2022) proposed that larger sea ice loss occurs on the east side than the west side of a cyclone. This phenomenon is also revealed in our study that regions in the Beaufort Sea with drastic sea ice loss locate on the east side of the trajectory of the cyclone center (Figures 4c and 6c).

Although our model can not accurately reproduce the near-surface temperature maximum layer owing to the relative coarse vertical resolution, the cyclone-associated strong winds strongly stirred the upper ocean to increase oceanic vertical mixing, which led to enhanced heat exchange between the cool ocean surface and the warm subsurface. Enhanced sea ice basal melting owing to the cyclone occurred in the Pacific sector of the Arctic, and part of the Atlantic sector of the Arctic (Figures 4f and 6f). Sea ice area loss caused by the oceanic heat flux accounted for a quarter of the total sea ice area loss when the intensity of the cyclone peaks. Jackson et al. (2012) found that heat from the near-surface temperature maximum layer can be entrained into the surface mixed layer by diffusion or by the erosion of the summer halocline in winter during cyclone periods, which can warm the mixed layer by an average of 0.06°C. Zhang et al. (2013) also demonstrated that the warm water at the near-surface temperature maximum layer entered into the mixed layer when the extreme cyclone occurred, which resulted in increases in oceanic heat flux and sea ice bottom melt rate. Our results are generally in line with the previous studies (Stern et al., 2020; Zhang et al., 2013).

Sea ice surface melting is also promoted owing to the increase in turbulent heat exchange between atmosphere and ice surface, which originates from the cyclone-induced anomalies in air temperature, air humidity and wind speeds. Enhanced sea ice loss due to the increased air-ocean heat flux mainly locates in the regions near sea ice edge in the Pacific sector of the Arctic, particularly in the Chukchi and northern Beaufort Seas. The cyclone brings more cloudy sky conditions. On one hand, it leads to the enhanced downward longwave radiation and the weakened radiative cooling on ice surface, and thus strengthened ice surface melt. On the other hand, it leads to the reduced shortwave radiation arriving at open-ocean/ice surface and thus weakened sea ice melt. Considering these radiative heat fluxes and turbulent heat fluxes together, the extreme cyclone strongly promotes sea ice area and volume loss, mainly because the influence of the turbulent heat flux is strong when during the intensity of the cyclone peaks. Note that a sea ice-ocean coupled simulation with prescribed atmospheric forcing lacks the responses of the atmosphere to the ocean and sea ice, thus our model may dampen the complex feedbacks at the ice-ocean-atmosphere interfaces. Moreover, in generating the initial condition for the two simulations, satellite-observed sea ice concentration, sea ice thickness and sea surface temperature data were assimilated into the ArcIOPS. During this procedure, inconsistencies between the atmospheric forcing and ocean-ice state are also introduced into the model trajectory (Griffies et al., 2009). Using a fully coupled sea ice-ocean-atmosphere model can simulate more reasonable heat fluxes at the ice-ocean-atmosphere interfaces; however, the proposed cyclone removal algorithm is applicable to prescribed atmospheric forcing, not to the simulated atmosphere component in a coupled sea ice-ocean-atmosphere model. Furthermore, note that the JRA-55 reanalysis uses a simplified definition of the surface boundary, where regions with satellite-observed sea ice concentration exceeding 0.55 are considered to be completely covered by sea ice and lower values are interpreted as open ocean (Kobayashi et al., 2015). This simplification leads to a warmer and moister air condition over marginal ice zone in the JRA-55 reanalysis. Replacing the JRA-55 forcing by another atmospheric reanalysis may result in different values in Figure 11; however, we believe that the main conclusion in this study will not be changed.

Schreiber and Serreze (2020) suggested that sea ice concentration is higher after the region is influenced by a cyclone compared to when it is not. Our study finds that the cyclone induces broad lower concentrated sea ice zones while the cyclone-induced higher concentrated sea ice zones are very narrow, which is mainly attributed to the wind-driven sea ice convergence. Clancy et al. (2022) suggested that the dynamical and thermodynamical impacts of cyclones on sea ice are comparable in magnitude. Lukovich et al. (2021) found that the thermodynamical impact of the extreme cyclone in summer 2012 dominates the sea ice reduction in the Pacific sector of the Arctic. Our result is generally in line with Lukovich et al. (2021), possibly because that both studies focused on the extreme cyclone in summer 2012 while Clancy et al. (2022) considered many Arctic cyclones. We have also conducted two additional experiments that are not reported in detail here, one using the atmospheric forcing with the removal of the cyclone-associated wind anomaly, the other using the atmospheric forcing with the removal of the cyclone-associated anomalies in the variables except wind. The former run generates similar sea ice evolution to the NoCycloneExp run, while the latter run has similar sea ice evolution to the RealExp run (not shown). This result indicates that the cyclone-associated wind anomaly plays a crucial role in sea ice loss, highlighting the key processes of turbulent heat exchange and sea ice ridging under the influence of an extreme cyclone.

The sea ice reduction after August 13 in the NoCycloneExp run is significantly larger than that in the RealExp run and the differences in sea ice extent, area and volume between the two runs shrink quickly after the extreme cyclone decays (Text S2 and Figure S2 in Supporting Information S1). As a result, sea ice area and volume on August 31 in the NoCycloneExp run are close to those in the RealExp run. This result means that, although the cyclone induces enhanced sea ice melting and some regions become ice free on August 13, these regions would still become ice free in late August even without the occurrence of the cyclone in early August. The net effect of the extreme cyclone on the Arctic sea ice decline seems to be greatly attenuated on multi-weekly and longer time scales. We also hypothesize that several continuous-occurring normal/weak cyclones may have greater impacts on the Arctic sea ice loss than a single extreme cyclone does. Based on our study involving an ice-ocean model with prescribed atmospheric forcing, we suggest that “The Great Arctic Cyclone of August 2012” seems to play an insignificant role in the 2012 record-low sea ice extent. This cyclone is an extreme case, and cyclones with similar intensity and longevity are rare in the Arctic summertime, while normal Arctic cyclones with lower intensity and higher occurrence frequency are common (Vessey et al., 2020). In our study, the relatively large sea ice loss around August 22 is probably owing to a normal Arctic cyclone. Due to the relatively weak intensity of normal Arctic cyclones, the oceanic responses and sea ice loss are not as strong as those due to extreme Arctic cyclones. The horizontal scales of normal Arctic cyclones are also smaller than extreme cyclones, thus the influences of normal Arctic cyclones on Arctic sea ice evolution are further locally constrained. There is no doubt that, to systematically study the effects of different cyclones on sea ice evolution, more long-term field observations (e.g., Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC)), precise atmospheric forcing field and numerical studies with high resolution are needed (Zhang, 2021).

Acknowledgments

The authors sincerely thank Dr. Laurence Padman for the constructive and editorial comments in improving this final version, and the three anonymous reviewers for the constructive comments during the review process. This work is supported by the National Natural Science Foundation of China under Contract 41706223, 42276250 and the National Key Research and Development Program of China under Contract 2018YFA0605902, 2019YFE0105700.

Appendix A: Symbols and Corresponding Short Descriptions

Table A1 shows a list of symbols and corresponding short descriptions used in this study.

Table A1. Symbols and Short Descriptions

Symbols	Short descriptions
	Total change of sea ice area
	The change of sea ice area due to oceanic heat flux at ice bottom
	The change of sea ice area due to sea ice-atmosphere heat flux
	The change of sea ice area due to atmosphere-open water heat flux
	The net advection of sea ice area across the boundary of the region
	The residual term of the change of sea ice area
	Total change of sea ice volume
	The change of sea ice volume due to oceanic heat flux at ice bottom
	The change of sea ice volume due to sea ice-atmosphere heat flux
	The change of sea ice volume due to atmosphere-open water heat flux
	The change of sea ice volume due to snow flooding process
	The net advection of sea ice volume across the boundary of the region
	The residual term of the change of sea ice volume
	Total net heat flux in sea ice zone
	Net shortwave radiation heat flux in sea ice zone
	Net longwave radiation heat flux in sea ice zone
	Latent heat flux in sea ice zone
	Sensible heat flux in sea ice zone
	Oceanic heat flux at ice bottom
	Total net heat flux over ice-covered area
	Net shortwave radiation heat flux over ice-covered area
	Incoming shortwave radiation heat flux arriving ice-covered area
	Reflected shortwave radiation heat flux by ice-covered area
	Penetrated shortwave radiation heat flux entering the ocean from ice
	Net longwave radiation heat flux over ice-covered area
	Incoming longwave radiation heat flux arriving ice-covered area
	Emitted longwave radiation heat flux by ice-covered area
	Latent heat flux over ice-covered area
	Sensible heat flux over ice-covered area
	Total net heat flux over open water area in sea ice zone
	Net shortwave radiation heat flux over open water area in sea ice zone
	Incoming shortwave radiation heat flux arriving open water area in sea ice zone
	Reflected shortwave radiation heat flux by open water area in sea ice zone
	Net longwave radiation heat flux over open water area in sea ice zone
	Incoming longwave radiation heat flux arriving open water area in sea ice zone
	Emitted longwave radiation heat flux by open water area in sea ice zone
	Latent heat flux over open water area in sea ice zone
	Sensible heat flux over open water area in sea ice zone
	Atmosphere-open water heat flux in sea ice zone
	Sea ice-atmosphere heat flux in sea ice zone
	Oceanic heat flux at ice bottom

Open Research

Data Availability Statement

The authors thank the Japan Meteorological Agency for JRA-55 data (http://search.diasjp.net/en/dataset/JRA55), the University of Hamburg for SSMI (https://icdc.cen.uni-hamburg.de/thredds/aggregationSsmiAsiCatalog.html) and SMOS (http://icdc.cen.uni-hamburg.de/1/daten/cryosphere/l3c-smos-sit.html) data, the Alfred-Wegener-Institut, Helmholtz Zentrum für Polar- und Meeresforschung for the CryoSat-2 data (http://data.meereisportal.de/data/cryosat2/version2.0/), the Copernicus Marine Environment Monitoring Service for the GMPE SST data (http://marine.copernicus.eu/), the Norwegian Meteorological Institute for the OSISAF data (http://www.osi-saf.org/?q=content/sea-ice-products), the University of Bremen for AMSR2 data (http://data.meereisportal.de/data/iup/hdf/), and the University of Washington for the PIOMAS data (http://psc.apl.uw.edu/research/projects/arctic-sea-ice-volume-anomaly/data/). The numerical configuration of the ArcIOPS is available at https://github.com/oucliangxi/ArcticModel18km_MITGCM. The simulation data is available upon reasonable request from the authors.