Both the Arctic and Antarctic sea ice extents (SIEs) from 44 coupled models in the Coupled Model Intercomparison Project Phase 6 (CMIP6) are evaluated by comparing them with observations and CMIP5 results. The CMIP6 multimodel mean can adequately reproduce the seasonal cycles of both the Arctic and Antarctic SIE. The observed Arctic September SIE declining trend (−0.82 ± 0.18 million km² per decade) between 1979 and 2014 is slightly underestimated in CMIP6 models (−0.70 ± 0.06 million km² per decade). The observed weak but significant upward trend of the Antarctic SIE is not captured, which was an issue already in the CMIP5 phase. Compared with CMIP5 models, CMIP6 models have lower intermodel spreads in SIE mean values and trends, although their SIE biases are relatively larger. The CMIP6 models did not reproduce the new summer tendencies after 2000, including the faster decline of Arctic SIE and the larger interannual variability in Antarctic SIE.

In data-sparse regions such as the Arctic, atmospheric reanalysis is one of the key tools for understanding rapid climate change at the regional and global scales. The utility of reanalysis datasets based on data assimilation is affected by their accuracy and biases. Therefore, it is important to evaluate their performance. Here, we conduct inter-comparisons of two temperature variables, namely, the 2-m air temperature (Ta) and the surface temperature (Ts), from the widely used ERA-I and ERA5 reanalysis datasets provided by the European Centre for Medium-Range Weather Forecasts (ECMWF) against in situ observations from three international buoy programs (i.e., the International Arctic Buoy Programme (IABP), the Multidisciplinary Drifting Observatory for the Study of Arctic Climate (MOSAiC), and the Cold Regions Research and Engineering Laboratory (CRREL)) during 2010–2020 in the Arctic. Overall, the results show that both the ERA-I and ERA5 were well correlated with the buoy observations, with the highest correlation coefficient reaching 0.98. There were generally warm Ta biases for both ERA-I (2.27 ± 3.33 °C) and ERA5 (2.34 ± 3.22 °C) when compared with more than 3000 matching pairs of daily buoy observations. The warm Ta biases of both reanalysis datasets exhibited seasonal variations, reaching the maximum of 3.73 ± 2.84 °C in April and the minimum of 1.36 ± 2.51 °C in September. For Ts, both ERA-I and ERA5 exhibited good consistencies with the buoy observations, but have higher amplitude biases compared with those for Ta, with generally negative biases of −4.79 ± 4.86 °C for ERA-I and −4.11 ± 3.92 °C for ERA5. For both reanalysis datasets, the largest bias of Ts (−11.18 ± 3.08 °C) occurred in December, while the biases were rather small (less than −3 °C) in the warmer months (April to October). The cold Ts biases for ERA-I and ERA5 were probably overestimated due to the location of the surface temperature sensors on the buoys, which may have been affected by snow cover. Both the Ta and Ts biases varied for different buoy programs and different sea ice concentration conditions, yet they exhibited similar trends.

Analyses of landfast ice in Arctic coastal areas provide a comprehensive understanding of the variations in Arctic sea ice and generate data for studies on the utilization of the Arctic passages. Based on our analysis, Arctic landfast ice mainly appears in January–June and is distributed within the narrow straits of the Canadian Archipelago (nearly 40%), the coastal areas of the East Siberian Sea, the Laptev Sea, and the Kara Sea. From 1976–2018, the landfast ice extent gradually decreased at an average rate of −1.1 ± 0.5 × 10⁴ km²/yr (10.5% per decade), while the rate of decrease for entire Arctic sea ice was −6.0 ± 2.4 × 10⁴ km²/yr (5.2% per decade). The annual maximum extent reached 2.3 × 10⁶ km² in the early 1980s, and by 2018, the maximum extent decreased by 0.6 × 10⁶ km², which is an area approximately equivalent the Laptev Sea. The mean duration of Arctic landfast ice is 44 weeks, which has gradually been reduced at a rate of −0.06 ± 0.03 weeks/yr. Regional landfast ice extent decreases in 16 of the 17 subregions except for the Bering Sea, making it the only subregion where both the extent and duration increases. The maximum mean landfast ice thickness appears in the northern Canadian Archipelago (>2.5 m), with the highest increasing trend (0.1 m/yr). In the Northeast Passage, the mean landfast ice thickness is 1.57 m, with a slight decreasing trend of −1.2 cm/yr, which is smaller than that for entire Arctic sea ice (−5.1 cm/yr). The smaller decreasing trend in the landfast ice extent and thickness suggests that the well-known Arctic sea ice decline largely occurred in the pack ice zone, while the larger relative extent loss indicates a faster ice free future in the landfast ice zone.

Summertime internal melting of Antarctic sea ice is common due to the penetration of solar radiation below the snow and ice surface. We focus on the role of internal melting and heat conduction in generating gap layers within the ice. These often occur approximately 0.1 m below the ice surface. In a small-scale survey over land-fast sea ice in Prydz Bay, East Antarctica, we observed, for the first time, gap layers 0.6–1.0 m below the surface for both first-year ice and multi-year ice. A 1D snow/ice thermodynamic model successfully simulated snow and ice mass balance and the evolution of the gap layers. Their spatial distribution was largely controlled by snow thickness and ice thickness. A C-shaped ice temperature profile with the lowest values in the middle of the ice layer resulted in heat flux convergence causing downward progression of the internal melt layer. Multidecadal (1979–2019) seasonal simulations showed decreasing air temperature favored a postposed internal melting onset, reduced total internal melt, and delayed potential ice breakup, which indicated a higher chance for local coastal ice to be shifted from first-year ice to multi-year ice.

Convolutional neural networks (CNNs) have achieved great progress in the classification of surface objects with hyperspectral data, but due to the limitations of convolutional operations, CNNs cannot effectively interact with contextual information. Transformer succeeds in solving this problem, and thus has been widely used to classify hyperspectral surface objects in recent years. However, the huge computational load of Transformer poses a challenge in hyperspectral semantic segmentation tasks. In addition, the use of single Transformer discards the local correlation, making it ineffective for remote sensing tasks with small datasets. Therefore, we propose a new Transformer layered architecture that combines Transformer with CNN, adopts a feature dimensionality reduction module and a Transformer-style CNN module to extract shallow features and construct texture constraints, and employs the original Transformer Encoder to extract deep features. Furthermore, we also designed a simple Decoder to process shallow spatial detail information and deep semantic features separately. Experimental results based on three publicly available hyperspectral datasets show that our proposed method has significant advantages compared with other traditional CNN, Transformer-type models.

High-frequency observations of the ice–ocean interaction and high-precision estimation of the ice–ocean heat exchange are critical to understanding the thermodynamics of the landfast ice mass balance in Antarctica. To investigate the oceanic contribution to the evolution of the landfast ice, an integrated ocean observation system, including an acoustic Doppler velocimeter (ADV), conductivity–temperature–depth (CTD) sensors, and a sea ice mass balance array (SIMBA), was deployed on the landfast ice near the Chinese Zhongshan Station in Prydz Bay, East Antarctica, from April to November 2021. The CTD sensors recorded the ocean temperature and salinity. The ocean temperature experienced a rapid increase in late April, from −1.62 to the maximum of −1.30 ^∘C, and then it gradually decreased to −1.75 ^∘C in May and remained at this temperature until November. The seawater salinity and density exhibited similar increasing trends during April and May, with mean rates of 0.04 psu d⁻¹ and 0.03 kg m⁻³ d⁻¹, respectively, which was related to the strong salt rejection caused by freezing of the landfast ice. The ocean current observed by the ADV had mean horizontal and vertical velocities of 9.5 ± 3.9 and 0.2 ± 0.8 cm s⁻¹, respectively. The domain current direction was ESE (120^∘)–WSW (240^∘), and the domain velocity (79 %) was 5–15 cm s⁻¹. The oceanic heat flux (F_w) estimated using the residual method reached a peak of 41.3 ± 9.8 W m⁻² in April, and then it gradually decreased to a stable level of 7.8 ± 2.9 W m⁻² from June to October. The F_w values calculated using three different bulk parameterizations exhibited similar trends with different magnitudes due to the uncertainties of the empirical friction velocity. The spectral analysis results suggest that all of the observed ocean variables exhibited a typical half-day period, indicating the strong diurnal influence of the local tidal oscillations. The large-scale sea ice distribution and ocean circulation contributed to the seasonal variations in the ocean variables, revealing the important relationship between the large-scale and local phenomena. The high-frequency and cross-seasonal observations of oceanic variables obtained in this study allow us to deeply investigate their diurnal and seasonal variations and to evaluate their influences on the landfast ice evolution.