Analysis and comparison of heat flux of landfast ice during 2016 in the Prydz Bay, Antarctica

1. Introduction

Sea ice is an important component in the climate system, which strongly affects the energy balance through the ice albedo feedback mechanism (Elders and Pegion, 2017). The Antarctic sea ice cover has a strong influence on the atmosphere and ocean (Valkonen et al., 2008). A previous study showed the sea ice extent increased with a rate of 1.7% per decade during the 1979–2015 period (Comiso et al., 2017), but the sea ice cover began to decrease in 2014. In the Antarctic, the sea ice is on average thinner, at lower concentration and, is located at lower latitudes than in the Arctic, and is often impacted by the katabatic winds from cold continental areas (Vihma et al., 2009), such as Zhongshan Station, which was affected by the overnight katabatic winds almost every day.

The change of sea ice cover will affect the surface energy flux. Surface energy balance is important to the interaction between the ocean and the atmosphere, which also affects the formation and ablation of the sea ice (Wendler and Worby, 2001). The surface energy balance includes radiative fluxes, turbulent heat fluxes, which have been carried out in the polar regions to learn about its climate processes (Perovich et al., 2002; Persson et al., 2002; Perovich and Polashenski, 2012; Walden et al., 2017; Yu et al., 2017). However, it was difficult to accurately estimate the local turbulent surface flux due to the lack of observations.

Direct observations of surface turbulent and radiative flux measurements over sea ice have rarely been conducted, especially in the Antarctic (Allison et al., 1982; Wendler et al., 1997; Wendler and Worby, 2001; Vihma et al., 2009; Van Den Broeke et al., 2005), although several studies have investigated the variability of surface radiation fluxes over the Antarctic land surface (Välisuo et al., 2014; King et al., 2015; Yang et al., 2016). Liu et al. (2020) found that the strong winds in the katabatic wind zone enhance the downward sensible heat flux (SH) in Antarctica. The limited observations have resulted in most surface energy budgets over Antarctic sea ice having been conducted by numerical models (Bintanja and Van Den Broeke, 1995; King et al., 2001). Reanalysis data are widely used in the Antarctic. The European Centre for Medium-Range Weather Forecasts Interim Re-analysis (ERA-Interim), National Centers for Environmental Prediction Reanalysis2 (NCEP R2), and Japanese 55-year Reanalysis (JRA55) datasets cover a long period of time, which is beneficial to research of Antarctica. However, reanalysis data reveals large differences amongst them as well as the in situ data (Shu et al., 2012). It is thus worthwhile and imperative to obtain more in situ data to improve modeling results. The recent version of the Los Alamos sea ice model (CICE) included the column ice model, which can be used in the study of landfast ice.

The Zhongshan Station is located in the Prydz Bay, East Antarctica, close to the Ice Sheet. The in situ observations were gathered at a nearby station. However, the observations and simulations of surface energy balance were sparse, especially for the in situ data. The surface energy balance were measured on the Antarctic Plateau (Ding et al., 2019). The diurnal cycle and monthly variation of surface energy balance on landfast ice under the conventional meteorological conditions were analyzed (Liu et al., 2020), which showed the surface energy balance of this region and indicated that latent heat flux (LE) and net heat flux (R_n) were balanced by SH and ground heat flux (G) before October and LE was the only surface heat sink. The surface energy balance process is not well represented in current climate models. Thus, observations and model results are both important and necessary for understanding the mechanism of snow/ice among ice, ocean, and atmosphere.

Nonetheless, the reanalysis datasets and model simulations are still important to compare Antarctic sea ice change to in situ data, but need to be evaluated by more in situ data. In this paper, the data used will be presented in Section 2. In Section 3, this study will first briefly analyze the observation data on the radiative and turbulent surface fluxes over the landfast ice nearby the Zhongshan Station in austral spring to early summer, from 8 April to 26 November 2016. Then, this study will concentrate on the reanalysis data of ERA-Interim, NCEP R2, JRA55, and the CICE output result, which will be evaluated using the observation data in the Zhongshan Station. The discussion and conclusions are drawn in the last section.

2. Data

The Automatic Weather Station (AWS) and buoys were setup at the in situ sites. The in situ sites were on landfast ice, in the coastal area off the Zhongshan Station (69°22′S, 76°22′E), Prydz Bay, East Antarctica, and were measured continuously from 8 April to 26 November 2016. The observation site was off the nearby islands and free from obstructions in all directions. The AWS includes a series of instruments: the air temperature sensor by probe 107, SI-111, and HMP 155 for both air temperature (T_a), surface temperature (T_s), and humidity. Also there was a three-dimensional sonic anemometer CSAT3B, and an in situ, open-path, mid-infrared gas (CO₂/H₂O) analyzer integrated with a 3D sonic anemometer IRGASON, of which more details can be found in Liu et al. (2020). All the data were converted to a daily mean and quality control was carried as described in Liu et al. (2020). The sea ice temperature and thickness were measured by one nearby sea ice buoy containing 300 temperature sensors, which had temperature sensors per centimeter with a resolution of 0.062 5°C and accuracy of 0.5°C. There were four buoys employed and the one nearest the AWS was used. The calculated sea ice thickness by using buoys with errors of ±7 cm, were evaluated by using the manned drilling hole observed sea ice thickness. The sea ice temperature was also used to calculate the energy exchange between the ice and ocean. The snow depth was manually measured by a stainless-steel ruler every week before 20 September and thereafter almost every day until 26 November, and had an accuracy of 0.1 cm. The cloud fraction was recorded at the Zhongshan (WMO No. 89573) manned weather station, located 1 km inland from the sea ice observation site. The longwave and shortwave radiation were measured by the CNR4 manufactured by Kipp&Zonen (Delft, The Netherlands). The surface turbulent fluxes, SH and LE, were calculated by a data logger using the eddy covariance method (Blanken et al., 2000).

Reanalysis data (ERA-Interim, NCEP R2, and JRA55) will be evaluated by the in situ observation data. ERA-Interim is produced with the observation fields, the forecast model, and a four-dimensional variational assimilation system (4DVAR) with a spectral model integrated (Simmons et al., 2007). The outputs of ERA-Interim were bilinearly interpolated to 10 various resolutions from 0.125° to 3°. The ERA-Interim synoptic of forecast accumulations 8 times per day at the grid resolution of 0.125° were downloaded from the European Center for Medium-Range Weather Forecasts (ECMWF) website. The NCEP R2 dataset provides global reanalysis fields of atmospheric data with a spatial resolution of 1.875° and 4 times per day. NCEP R2 is a second limited version of NCEP-R1 (Kanamitsu et al., 2002). The JRA55 data extended back to 1958 and is the second Japanese global atmospheric reanalysis project. JRA55 also employs 4DVAR data assimilation with a grid resolution of 1.25° (Ebita et al., 2011). The time resolution is 8 times per day. In this study, the daily averaged values of surface LE, SH, T_a, wind speed, relative humidity, S_net (net shortwave radiation) and L_net (net longwave radiation), T_s, and cloud fraction from the above three reanalysis datasets were used.

The CICE model, developed by the Los Alamos National Laboratory, USA, is an efficient sea ice component for a fully coupled atmosphere–land–ocean–ice global climate model (Hunke et al., 2018). It is widely used in climate system models due to the complex parameterization. The CICE consists of a thermodynamic component, an ice dynamics component, and a transport component. Displaced-pole grids were used in CICE with a horizontal resolution of 1°×1°, seven ice layers, and one snow layer in the vertical. The numerical experiments in this article are based on CICE version 6. The CICE output results (SH, LE, upward and downward longwave radiation, downward shortwave radiation, and ice temperature) based on version 6 will also be assessed by using the in situ observations. More details can be found in Section 3.3.

3. Results

3.1 Observation and meteorological conditions

Daily averaged surface pressure, wind speed, wind direction, cloud cover fraction, T_a, and T_s from 8 April to 26 November 2016 are shown in Fig. 1. According to the statistics reported by a previous study (Liu et al., 2020), the wind speed during the observation period was (4.2±2.3) m/s, and the wind direction was mainly eastern, which can also been seen in Figs 1b and c. It also showed diurnal variation with east-south-easterlies in the morning to east-north-easterlies in the afternoon. The observation site was frequently impacted by a cyclone from the south. In addition to the cyclone, the site was also impacted by katabatic winds from night to early noon, which also contributed to enhancing the diurnal variation of air temperature from the cold katabatic flow from the ice sheet. In this analysis, sea level pressure (SLP) in the Zhongshan Station was measured at (978±10) hPa during 8 April to 26 November 2016 (Fig. 1a). A strong cyclone occurred on 10 September, which fell over 24 hPa within 24 h (the cyclone deepening rate ≥1 hPa/h) and resulted in the largest instantaneous wind speed of 17.5 m/s (daily average wind speed, 11.7 m/s, Fig. 1b). The time series of cloudiness (Fig. 1d) shows 42% of occurrences of clear sky conditions (cloud cover ≤20%) and 45% of occurrences of overcast conditions (cloud cover ≥80%, Fig. 1d), which is similar to the short time (about 2 months) statistical result (Hao et al., 2020). During the observation period, the lowest daily mean T_a at 2 m and T_s were −27.5°C and −30.6°C, respectively (Fig. 1e), and occurred in the first half of September (Fig. 1e). The cyclone caused the air temperature to increase, which continued after the cyclone ended.

Figure  1.  Daily average sea level pressure (SLP) (a), wind speed (b), wind direction (c), cloud cover fraction (d), and T_a and T_s (e). T_a: air temperature, T _s: surface temperature.

DownLoad: Full-Size Img PowerPoint

3.2 Assessment of reanalysis data using in situ data

In this study, the net shortwave radiation flux (S_net), net longwave radiation (L_net), the surface net radiation (R_n), and the oceanic heat flux F_w (Persson et al., 2002) are determined as

where $ {S \downarrow } $, $ {S\uparrow } $, $ {L \downarrow } $, and $ {L\uparrow } $ are the downward shortwave radiation flux, upward shortwave radiation flux, downward longwave radiation flux, and upward longwave radiation flux, respectively. R_n is the sum of S_net and L_net, which can be partitioned into LE, SH, and G (Zhou and Wang, 2016). In Eqs (1) and (2), a positive S_net and L_net indicates the surface gains energy.

The environmental parameters (T_s, wind speed, and relative humidity) from the ERA-Interim, NCEP R2, and JRA55 are validated by in situ observed data in the Zhongshan Station in this section (Fig. 2). The reanalysis data showed similar seasonal variation as the observation data. The T_a and wind speed of NCEP R2 data showed the largest differences from the observation, with biases of −7.1°C and 7.3°C and mean absolute error (MAE) of 4.2 m/s and 4.3 m/s (Fig. 2a). The result demonstrated the lower air temperature and higher wind speed from the NCEP R2 data than the observation. The bias and MAE among the reanalysis data and the observations are shown in Table 1. The T_a, T_s, and cloud fraction (Figs 2a, d and e) of ERA-Interim showed the least amount of bias and MAE in comparison to the observation data. The wind speed and relative humidity (Figs 2b and c) perform best with the minimum bias and MAE of JRA55. The T_a and T_s also show good agreement with the observation, but they are not better than the result of ERA-Interim. The cloud fraction of JRA55 had the largest bias and MAE. However, it needs stress that the observed cloudiness in the Zhongshan Station was obtained by different observers and with subjectivity to some extent. The T_s of NCEP R2 data showed the largest bias and MAE of −5.3°C and 5.8°C and demonstrated larger range during the temperature-fall period than the observed data and the other two datasets (Fig. 2d), which could impact the simulation of outgoing longwave radiation.

Figure  2.  Daily average T_a (a), wind speed (b), relative humidity (RH) (c), T_s (d), and cloud fraction (e) of in situ observations and reanalysis data.

DownLoad: Full-Size Img PowerPoint

Table  1.  The bias and MAE of atmospheric factors between AWS and the reanalysis data

		Ta/°C	WS/(m·s−1)	RH/%	Ts/°C	Cloud/%
ERA-Interim	Bias	−2.1	2.4	16.8	−0.5	10.9
	MAE	2.8	2.5	17.6	1.9	22.6
NCEP R2	Bias	−7.1	4.2	16.2	−5.3	−13.7
	MAE	7.3	4.3	18.5	5.8	21.9
JRA55	Bias	−2.0	−0.1	8.7	−1.3	−19.7
	MAE	3.3	2.1	8.9	2.9	26.0
Note: WS represents wind speed; RH represents relative humidity.

 | Show Table

DownLoad: CSV

Positive SH and LE indicated sea ice energy gain, as shown in Fig. 3. The surface gained more downward shortwave radiation as the daily noon solar elevation angle increased. According to the statistics reported by Liu et al. (2020), the ice or snow surfaces gain energy from net shortwave radiation flux and SH, and they lose energy through LE and L_net, which is also shown in Fig. 3 during the observation period. The daily mean S_net (Fig. 3c) was above zero from the beginning of September and increasing rapidly, which was significant after the polar night and demonstrated seasonal variation caused by the solar elevation angle variation (Fig. 3c). The L_net also has shown inner-seasonal variation and obvious seasonal oscillation of SH shown (Fig. 3b), which was mainly caused by the temperature impact of the synoptic process (Fig. 3d). After the middle of October, the L_net demonstrated a decreasing trend, which signified more outgoing energy of the surface.

Figure  3.  Daily average LE (a), SH (b), S_net (c), L_net (d), and R_n (e) of in situ observations and reanalysis data.

DownLoad: Full-Size Img PowerPoint

The daily mean turbulent fluxes (SH and LE) calculated from the in situ data and radiation fluxes (S_net, L_net and R_n; Figs 3c-e) were used to validate those obtained from the three reanalyzed datasets, ERA-Interim, NCEP R2, and JRA55 in the Zhongshan Station, Antarctic. Figure 3 shows that all of the reanalysis data demonstrated the seasonal variation similarly to the observation, except the SH of NCEP R2, but still show larger differences for the amplitude.

The LE of reanalysis data showed larger differences from the observations from the second half of October, especially for the NCEP R2 data, with melting onset as shown in Fig. 3a. The bias and MAE among the reanalysis data and the observations is shown in Table 2. The SH of ERA-Interim and JRA55 show the same fluctuation; however, that of NCEP R2 showed larger differences (Fig. 3b). SH of NECP R2 data also showed larger values and bigger variation, with a bias and MAE of 26.1 W/m² and 39.3 W/m². The wind speed also showed bias, which impacted the result of SH and LE. Before November, the S_net of NCEP R2 showed larger differences, and the JRA55 showed larger differences after November. The S_net of ERA-Interim data demonstrated more fluctuation compared with other data (Fig. 3c). The L_net of ERA-Interim data showed larger negative values than the other datasets, which indicated the surface lost more energy through the thermal radiation process, which may be caused by the differences in T_s and cloudiness fraction (Fig. 3d). The L_net of ERA-Interim shows smaller bias (−41.2 W/m²) than the observations and a larger difference in MAE (41.4 W/m²), which also leads to almost double the difference of R_n than JRA55. The S_net also leads to a difference for R_n (Fig. 3e). For the period when shortwave radiation increased due to solar elevation angle increases in the JRA55 and NCEP R2 datasets, the difference of S_net also led to a larger difference of R_n. As the positive and negative bias compensated for the overall deviations, the R_n of NCEP R2 demonstrated the smallest bias and larger MAE than JRA55. Among the three reanalysis sets, the JRA55 data showed the smallest differences from the observations in all the variables in the Zhongshan Station.

Table  2.  The bias and MAE of flux between AWS and the reanalysis data (unit: W/m²)

		LE	SH	Snet	Lnet	Rn
ERA-Interim	Bias	−6.5	10.2	15.6	−41.2	−25.6
	MAE	7.6	14.3	15.7	41.4	29.7
NCEP R2	Bias	−7.9	26.1	24.8	−24.0	0.8
	MAE	11.3	39.3	24.8	25.7	27.7
JRA55	Bias	0.7	−1.1	12.7	−21.6	−8.9
	MAE	6.8	13.5	13.8	22.1	25.9

 | Show Table

DownLoad: CSV

3.3 Ocean heat fluxes based on observed ice surface temperature and ice thickness

Apart from the heat flux exchange between ice and air, the ice also gains energy from the ocean (Fig. 4). The energy budget process (Mcphee and Untersteiner, 1982; Perovich and Elder, 2002; Lei et al., 2010) is described as follows:

Figure  4.  Conductive heat flux at ice base F_C (a), equivalent latent heat flux F_L (b), specific heat flux F_S (c), oceanic heat flux F_W (d), and monthly mean value of the estimated oceanic heat flux with errors (e). All lines represent a 7-d moving average.

DownLoad: Full-Size Img PowerPoint

where F_C is the conductive heat fluxes $\left( {{F}}_{{\rm C}}{=}{{k}}_{{\rm {si}}}\dfrac{{\partial }{{T}}_{{\rm {si}}}}{{\partial }{{Z}}_{{\rm {si}}}}\right)$ at ice-sea interface, and F_L and F_S are the equivalent latent heat flux $\left( {{F}}_{{\rm L}}{=}{{-}\rho }_{{\rm {si}}}{L}_{\rm f}\dfrac{{\partial }{{Z}}_{{\rm {si}}}}{{\partial }{t}}\right)$ and specific heat flux $\left( {{F}}_{{\rm S}}{=}{\rho }_{{\rm {si}}}{c}_{\rm {si}}\dfrac{{\partial }{{T}}_{{\rm {si}}}}{{\partial }{t}}\right)$ in the ice bottom. Additionally, ${{k}}_{{\rm {si}}}$ is the sea ice thermal conductivity, and $\dfrac{{\partial }{{T}}_{{\rm {si}}}}{{\partial }{{Z}}_{{\rm {si}}}}$ is the vertical ice temperature gradient. Sea ice density is given by ${\rho }_{{\rm {si}}}$, ${L}_{\rm f}$ is the sea ice latent heat of fusion, $\dfrac{{\partial }{{Z}}_{{\rm {si}}}}{{\partial }{t}}$ is the ice growth rate, ${c}_{\rm {si}}$ is the sea ice specific heat at the basal layer, and $\dfrac{{\partial }{{T}}_{{\rm {si}}}}{{\partial }{t}}$ is the ice temperature variation rate. The sea ice temperature recorded by the buoy and the calculated sea ice thickness were used to calculated the energy budget process. The oceanic heat flux ${{F}}_{{\rm W}}$ was calculated using the method described by Lei et al. (2010). The precision of the oceanic heat flux was affected by the ice thickness, temperature, and the specified salinity at the ice base. The positive F_W indicates ice gain energy from the ocean.

The F_C and F_L were the main portions of the ice-ocean energy exchange, which showed opposite signs (Figs 4a and b). The F_S demonstrated small magnitudes and around 0 for the whole period (Fig. 4c). The average F_C, F_L, F_S, and F_W were 27.8 W/m², −12.6 W/m², 0.2 W/m², and 15.4 W/m², respectively, during the evaluated period. The F_W demonstrated obvious seasonal change (Figs 4d and e). Before August, the air temperature decreased and ice thickness increased with the decrease in solar radiance. The monthly mean F_W varied within a range of 19–23 W/m² (Fig. 4e), with an average value of 21 W/m². The heat available in the surface ocean reduced from August, with a range from 8 W/m² to 13 W/m², and reached a minimum in August and October, with an average value of 11 W/m², which was consistent with the findings of previous studies (Lei et al., 2010; Zhao et al., 2019), but showed different seasonal variation, which may be due to errors of sea ice thickness measurements derived from the buoys. The calculated F_W was dependent on the sea ice growth rate. The monthly result shows a maximum deviation of ±7 W/m² in July.

3.4 Comparison with CICE version 6 result

In this study, the atmospheric force data used ERA-Interim reanalysis data from ECMWF, including wind speed, air temperature, dew-point temperature, precipitation, snowfall, total cloud, SLP, and downward oceanic heat flux. The radiation data were simulated using the ERA-Interim forcing data. The ocean force data were obtained from the Community Climate System Model (CCSM), including the climatological monthly surface tilt, sea surface velocity, heat flux of mixed layer bottom from the output of the CCSM control run and sea surface temperature, and sea surface salinity of Polar science center Hydrographic Climatology (PHC). The integration time was from 1979 to 2016. To compare with the observations at the Zhongshan Station, the nearest grid (69.6046°S,76.4375°E ) of the CICE output with sea ice thickness more than 0.01 m was chosen from April to November 2016.

The results from CICE version 6 were compared using the observed data at the Zhongshan Station. The results show that the value of SH was smaller and LE was larger than the observation data. The turbulent heat flux of the CICE (Fig. 5) result demonstrates the smaller range and do not shows the remarkable amplitude of seasonal variation, with SH bias and MAE of −21.4 W/m², 22.3 W/m² and LE bias and MAE of 3.1 W/m², 5.9 W/m². These results were better than the NCEP R2 compared with the AWS, but showed larger bias than the ERA-Interim. The bias and MAE of SH compare with ERA-Interim are −31.1 W/m², −31.2 W/m², and the bias and MAE of LE compare with ERA-Interim are 9.4 W/m², 9.5 W/m², which indicated larger turbulence flux than the forcing data. The radiation of the CICE output was calculated by the model parameterized scheme using the forcing data. The upward longwave radiation (Fig. 6a) was consistent with observations. The downward longwave radiation (Fig. 6b) demonstrated seasonal variation, but with a smaller range. The larger difference in downward longwave radiation caused the larger differences of L_net (Fig. 6c). The downward shortwave radiation (Fig. 6d) showed an increase with the increase of solar elevation angle, but did not show the variation affected by cloud or synoptic process. The radiation (upward longwave radiation, downward longwave radiation, L_net, and downward shortwave radiation) showed bias of 3.9 W/m², 4.7 W/m², 0.8 W/m², and −1.5 W/m² and MAE of 9.4 W/m², 24.0 W/m², 21.6 W/m² and 12.9 W/m² compare with the observations. The radiation shows larger bias of −106.5 W/m², −64.7 W/m², 41.8 W/m², and −40.0 W/m², and MAE of 106.5 W/m², 64.7 W/m², 46.6 W/m² and 43.4 W/m² with the ERA-Interim data, which indicated the better stimulation of radiation in CICE than the ERA-Interim radiation. The L_net of CICE does not show a large fluctuation from day to day. The simulated ice temperature was assessed with respect to the temperature observed by ice buoys. The ice temperature (Fig. 7) from CICE was reported in seven levels for five different ice categories, which were averaged and then compared with the observations. The bias and MAE were 0.9°C and 1.0°C, respectively. The simulated ice temperature also showed the temperature increase process from September. The simulated ice temperature can also demonstrate rapidly decreasing processes caused by the air temperature decrease, such as the process around 22 June and 7 July, but did not show the amplitude.

Figure  5.  Daily average SH (a) and LE (b). The sign convention is such that upward heat fluxes are positive and downward heat fluxes are negative.

DownLoad: Full-Size Img PowerPoint

Figure  6.  Daily average upward longwave radiation (a), downward longwave radiation (b), net longwave radiation (c), and downward shortwave radiation (d).

DownLoad: Full-Size Img PowerPoint

Figure  7.  Daily average ice temperature of buoy and CICE.

DownLoad: Full-Size Img PowerPoint

4. Discussion and conclusions

This study analyzed the flux variation, evaluated the reanalysis datasets and the CICE results based on the in situ data observed on landfast ice nearby the Zhongshan Station from 8 April to 26 November 2016. According to the results of Liu et al. (2020), the in situ data demonstrated that the strong katabatic wind enhanced the downward SH in Antarctica and the downward radiation, with significant differences in the monthly mean diurnal variation and demonstrated that LE was the only heat sink of the surface. In this study, the daily average turbulence flux and F_W were briefly analyzed. The reanalysis data and CICE 6 output results were evaluated by using the in situ data. The reanalysis data showed large differences compared with the observation data, The CICE 6 output of SH, LE, and the longwave radiation also showed larger errors and still needs to be improved.

The solar radiation demonstrated both seasonal and inner-seasonal variation, which was mainly caused by the solar elevation angle variation and the temperature impact by the synoptic process. In this study, the SH and L_net showed remarkable variation as mentioned above. The ice surface lost energy mainly through the outgoing longwave radiation from the middle of October. The ice-ocean energy also demonstrated seasonal variation. In this result, the ice growth rate varied between 0 and −0.3 m/d, which showed slight melt in the middle of November and similar with the 0–1.7 cm/d show in Lei et al. (2010). The calculated F_W demonstrated seasonal change with a range of 19–23 W/m² and 8–13 W/m² before and after August, respectfully; and the averaged F_W was 21 W/m² and 11 W/m² for these two periods, respectively. The F_W was dependent on the growth rate of ice, which can induce larger errors. The result of F_W was consistent with the previous experiment (Lei et al., 2010; Zhao et al., 2019), but show different seasonal variation, which may cause by the errors of sea ice grow rate in different seasons.

By choosing the nearest grid with the in situ site, the reanalysis data from NCEP R2, ERA-Interim, and JRA55 were evaluated. The comparison results show that the JRA55 dataset demonstrate the smallest bias and MAE with the observation data. The bias and MAE of SH were 0.7 W/m² and 6.8 W/m² and those of LE were −1.1 W/m² and 13.5 W/m². The NCEP R2 data show the largest difference with the observation data. The bias and MAE of SH were −7.9 W/m² and 11.3 W/m², and those of LE were 26.1 W/m² and 39.3 W/m². The ERA-Interim data show moderate differences for the SH and LE. The bias and MAE of SH were −6.5 W/m² and 7.6 W/m²and those of LE were 10.2 W/m² and 14.3 W/m², although it has the highest spatial regulation. The SH and LE for the ERA-Interim were presented by the forecasted element, which also induced more errors.

The CICE results were forced by the ERA-Interim atmospheric data. The radiation was simulated by CICE by using the ERA-Interim forcing data. The radiation results were consistent with the observed data, but did not demonstrate the amplitude of inner seasonal variation, with SH bias and MAE of −21.4 W/m² and 22.3 W/m², and LE bias and MAE of 3.1 W/m² and 5.9 W/m². These turbulent heat fluxes from CICE output were better than the NCEP R2 results compared with the in situ results and showed larger differences with the ERA-Interim data. The radiations (upward longwave radiation, downward longwave radiation, downward shortwave radiation) stimulated by CICE based on ERA-Interim data were better than the results of ERA-Interim. The averaged sea ice temperature can show the rapidly decreasing process caused by air temperature decreasing, such as the temperature decrease fall period in June and July, the local minimum temperature occurred on 22 June, and 7 July, and the temperature increase period from September. However, the amplitude was largely different, but was close to the observation result from the buoys, with bias and MAE of 0.9°C and 1.0°C, respectively.

The variation of surface flux on sea ice was briefly analyzed, and the model results were evaluated in this study. However, the observations were limited to the landfast ice site near the Zhongshan station. More observations are needed to cover a wider ice range and time scales to understand the surface energy balance, evaluate the model results, and improve the parameterization scheme. The results from this study also indicated that the JRA55 can be the forcing data used in CICE in the future.

Acknowledgements

We show our great appreciation to wintering team of the 32nd CHINARE for the field observations supporting. ERA-Interim data are from https://apps.ecmwf.int/datasets/data/, NCEP reanalysis2 data are from https://rda.ucar.edu/ and JRA55 data are from https://jra.kishou.go.jp/JRA-55. We thank Ray Miller for editing the English language. We also thank the Chinese Arctic and Antarctic Administration and Polar Research Institute of China for the logistic support. This is a contribution to the Year of Polar Prediction (YOPP) of the Polar Prediction Project (PPP) by the World Weather Research Programmer (WWRP) of the World Meteorological Organization (WMO).