In recent years, much attention has been paid to the behavior of passive microwave sea ice concentration (SIC) products for marginal ice zones. Based on the definition of ice edges from ship observations, we identified pseudo-ship observations (PSO) and generated PSO ice edges from twelve cloud-free moderate-resolution imaging spectroradiometer (MODIS) images. Two SIC products of the advanced microwave scanning radiometer 2 (AMSR2) were compared at the PSO ice edges: ARTIST (arctic radiation

more »

... turbulence interaction study) sea ice (ASI-SIC) and bootstrap (BST-SIC). The mean values of ASI-SIC pixels located at ice edges were 10.5% and 10.3% for the Arctic and the Antarctic, respectively, and are below the commonly applied 15% threshold, whereas the mean values of corresponding BST-SIC pixels were 23.6% and 27.3%, respectively. The mean values of both ASI-SIC and BST-SIC were lower in summer than in winter. The spatial gaps among the 15% ASI-SIC ice edge, the 15% BST-SIC ice edge and the PSO ice edge were mostly within 35 km, whereas the 15% ASI-SIC ice edge matched better with the PSO ice edge. Results also show that the ice edges were located in the thin ice region, with a mean ice thickness of around 5-8 cm. We conclude that the 15% threshold well determines the ice edge from passive microwave SIC in both the Arctic and the Antarctic. provides brightness and temperature at a relatively fine resolution. ASI and BST algorithms have been applied on AMSR2 and the daily ASI-SIC and BST-SIC map based on AMSR2 has been accessible since August 2012. Comparisons have been made between different SIC algorithms. On the one hand, a specific algorithm has been tested on different satellite sensors. Comiso and Parkinson [7] evaluated the SIC, sea ice extent and area derived by BST and NT2 from AMSR-E and SSM/I at the Arctic and the Antarctic. They found that the difference of BST on different sensors was much smaller than the difference of BST and NT2 on one sensor. On the other hand, comparisons have been done among different algorithms. Spreen et al. [8] compared ASI, NT2 and BST on AMSR-E with ship-based observation (OBS) and found correlations equal to 0.80, 0.79, and 0.81, respectively. Heygster et al. [9] compared ASI, NT2 and BST on AMSR-E in the Arctic and Antarctic, and found biases of SIC below 2% and root-mean-square errors between 7% and 11%. Ozsoy-Cicek et al. [10] compared NT2 and BST on AMSR-E with OBS in the Antarctic and found that the SIC derived from NT2 had slightly higher correlation with OBS-SIC than with BST. Beitsch et al. [11] compared AMSR-E ASI, BST and NT2 with OBS and found that ASI-SIC and BST-SIC had an insignificant bias and that BST-SIC had the lowest root mean square deviation (RMSD) (<13%), whereas NT2-SIC had the largest bias among the three algorithms in all seasons. Beitsch et al. [12] found that BST and ASI of AMSR-E are more in accordance with OBS, as compared with BST and ASI applied to SSM/I data. Year-round RMSD values of BST and ASI are 13.2% and 14.3% for SSM/I, and 11.6% and 13.3% for AMSR-E. Ivanova et al. [13] compared thirteen sea ice algorithms, and found ASI performs well over ice, but is sensitive to cloud liquid water over the marginal ice zone (MIZ). In general, algorithms tend to underestimate the ice concentration in the region of thin ice as well as in the vicinity of the ice edge, especially if the MIZ is diffuse [14, 15] . The accuracy of ice concentration in the MIZ attracts the interest of researchers, and in-depth study would be of great value. Ozsoy-Cicek et al. [2] compared the sea ice edge of AMSR-E-based SIC using NT2 and the BST retrieval algorithm with OBS. They found a good correlation inside the ice pack; however, in the MIZ, the correlation decreased. Both NT2 and BST tend to underestimate low ice concentration. The 15% ice concentration threshold has been used to determine the sea ice extent in climatology analysis. This threshold, however, is not always consistent in different studies. For the Arctic, the threshold has been set between 15% and 30% [16, 17] . For the Antarctic, the threshold even reached 40% [18] . As for in-situ validation, the ship observations were recorded using the protocol specified by the Antarctic Sea Ice Processes and Climate (ASPeCt) program [19] . Worby and Comiso [6] evaluated the ice edge location with OBS in the Antarctic and found that PM ice edges were 1-2 • of latitude south of the OBS. Heinrichs et.al [20] found that the ice edges determined from AMSR-E data and SAR data were within one AMSR-E grid square (12.5 km). Ozsoy-Cicek et al. [2] compared AMSR-E sea ice edges with NIC (the U.S. National Ice Center) sea ice edges as well as OBS in the Antarctic. Their results indicated that the sea ice edge detected by AMSR-E was further south and that AMSR-E was ineffective in the detection of low SIC. Because the number of ship observations located at the sea ice edge is limited, Zhao et al. [21] proposed a method to generate PSO from optical satellite images, and used PSO to assess the quality of AMSR-E ASI SIC products at the ice edge in the Antarctic. They found that the mean ASI-SIC at the ice edge was 13%, i.e. close to the 15% threshold, and that the correlations between ASI-SIC and PSO crossing the boundary were low. In this paper, we modified this method and applied it to evaluate AMSR2 ASI and BST SIC at the ice edge in both the Arctic and the Antarctic, as well as to further assess the reliability of a 15% SIC threshold for both poles. * Statistically significant at the 0.05 level.