The DT-ASI retrieval algorithm (Hao and Su, 2015), based on FY-3B/MWRI data, includes the following steps.

In the first step, the ASI algorithm based on fixed tie points (Spreen et al., 2008) obtains the initial SIC field C using the following equation:

where P is the polarization difference between the horizontal and vertical BTs at 89 GHz; $ {d}_{3},{d}_{2},{d}_{1},{\rm{and}}\;d_{0} $ are obtained by solving equations for the conditions of the pure water point value $ {P}_{0} $ and pure ice point value $ {P}_{1} $. The initial values are as follows: $ {P}_{0}=47 $ and $ {P}_{1}=11.7 $; when $ P>{P}_{0} $, C is assigned a value of 0% for seawater; when $ P<{P}_{1} $, C is assigned a value of 100% for sea ice.

The second step is to calculate the new daily tie point values. The daily average values of $ {P}_{0} $ and $ {P}_{1} $ under true atmospheric circumstances are calculated without using any weather filter. $ {P}_{0} $ is the average of the 89 GHz polarization difference in all grid points without ice in the region of 53°–75°N; $ {P}_{1} $ is the average of the 89 GHz polarization difference of all grid points where the weather-filtered SIC is greater than 95% in the region of 85°–90°N. Because the MWRI data has a larger hole than AMSR-E in the center of the North Pole, the selection range of ice point values is smaller. To determine the values of the open water tie point, the vertical polarization BT ratios of the 18.7 GHz and 36.5 GHz channels are introduced as a criterion (Zhang, 2012). Using the statistics of $ {P}_{0} $ and $ {P}_{1} $ as initial values, the BT polarization difference P is calculated again using the ASI algorithm, and the new FY-3B/MWRI SICs are obtained based on Eq. (1).

The third step is weather filtering. Due to the influence of abnormal atmospheric processes, some errors will occur on the open water surface, which will lead to false sea ice retrievals. It is necessary to introduce weather filters to deal with these errors. Hao and Su (2015) used two types of weather filters (Gloersen and Cavalieri, 1986).

The gradient ratios of $ {\rm{GR}}\left({36,18}\right) $ and $ {\rm{GR}}\left({23,18}\right) $ were used for judgment, and ${\rm{GR}}(X,Y) = [{\rm{TB}}(X,{\rm{V}})-{\rm{TB}}(Y,{\rm{V}}\left)\right]/ \left[{\rm{TB}}\right(X,{\rm{V}})+ {\rm{TB}}\left(Y,{\rm{V}}\right)]$ can be used to filter the false sea ice generated by clouds and water vapor, where X and Y are different frequency of BT and$ {\rm{V}} $ stands for vertical channel. The threshold values are 0.045 and 0.04 respectively, which at least keeps all SICs at least 15% above. We use these two weather filtering methods (Gloersen and Cavalieri, 1986) and criteria to process the newly obtained FY-3B/MWRI SIC.

Limited by the wide diameter of the antenna, MWRI BT can detect relatively weak ground echo signals (Parkinson, 1987; Zhai et al., 2017). This makes it difficult to distinguish sea ice from land near the shore, which affects SIC products. To solve this problem, NSIDC SIC data has been introduced to identify pixels near the coastline. Liu et al. (2020) used the microwave scanning radiometer BT data from FY-3C to perform SIC retrieval in the polar region and used the minimum concentration template to remove land pollution effects. Similarly, in the FY-3B/DT-ASI SIC obtained by our retrieval, the land coverage and ocean coverage could not be distinguished. A land mask was applied to the ice concentration maps and adopted the technique of land spillover correction to deal with this problem. The errors in sea ice in the marginal area of the FY-3B SIC were corrected according to steps 1–3 of the land spillover correction method issued by the National Snow and Ice Center (Markus and Cavalieri, 2009). The specific steps of the algorithm are as follows:

(1) Classify all meshed pixels according to distance from the shoreline. Water points directly adjacent to the shoreline are set to 1. Outer boundary points are set to 2 and 3, respectively, and open water points and points adjacent to points with values of 2 and 3 are set to 0. Land locations close to the shoreline are set to 4, and points far from the shore are set to 5.

(2) Use a 7×7 window (about 87.5 km×87.5 km) to judge the concentrations of points with values of 1 and 2. Points with a value of 0 or 3 are not included in the detection range.

(3) Determine whether all points with a value of 3 in each window are open water. If they are, then the SIC of the points corresponding to points with values of 1 and 2 in the window are set to 0; otherwise, the concentration values of the corresponding points do not change.

Through the land spillover correction, the errors in sea ice near the shoreline were corrected to a major degree, and the SIE was also significantly corrected. This is important for shipping and resource utilization in the Arctic region, e. g., the Northwest Passage, which runs from Greenland through the Canadian Arctic Archipelago to the northern coast of Alaska, can greatly shorten the route between the Atlantic and the Pacific and has important economic value. Sea ice of the Canadian Arctic Archipelago is the key to the navigation time in summer, but PM products are subject to deviations due to low resolution. So, it is necessary to apply the land spillover method in these areas. Figure 1 shows the regional SIC distribution in the Canadian Arctic Archipelago before and after land spillover correction and the local satellite imagery on August 8, 2016. It shows that the land spillover method does not change the distribution of sea ice in large areas and only corrects the sea ice in small areas located at the edge of land (Figs 1a and b). Comparisons with satellite imagery (Fig. 1c) show that there is little sea ice for navigation in the western and southern ocean passages, consistent with the results in Fig. 1b. Most of the SIC in this area was greater than 15%, which would affect the calculation of SIE. Then, we calculated that the SIE values of this region (66°–76°N, 70°–130°W) were $ 17.7\times {10}^{4}\;{\rm{km}}^{2} $ and $ 14.3 \times {10}^{4}\;{\rm{km}}^{2} $ respectively before and after using this method, which decreased by 19.2%. Therefore, the land spillover method can effectively correct a large number of misjudged sea ice, downgrade the differences between different data sets, and guide the determination of navigation time in the local sea area. Meier et al. (2015) pointed out the land spillover issue can add false ice along the coast, particularly during summer. In the Arctic Ocean during wintertime, the landfast ice covers almost the entire coastal area, so we improved to use the land spillover method from July 1st to December 1st every year, because most of the sea ice is not connected to land during this period.

In this section, we compare our results with SIC products obtained by the University of Bremen (UB) and the Ocean University of China (OUC), expressed as AMSR2/ASI and AMSR2/DT-ASI. Both data have a resolution of 6.25 km.

All three sets of results show a similar seasonal cycle of the SIC. We compare the monthly average Arctic SIC of the three products for eight years. For example, Fig. 2 and Fig. 3 give the SIC distributions of the three datasets for March and September. In all three sets of results for both March and September, the monthly average distribution of SIC is very similar and the values are very close, especially the two sets of results of AMSR2. The differences between FY-3B/DT-ASI and the two AMSR2 results appear mainly in the ice edge area. For March, the differences appear primarily in the Sea of Okhotsk and the ice margin zone of the North Atlantic, with smaller values in the FY-3B data. However, in September, FY-3B/DT-ASI is higher than AMSR2/DT-ASI, especially in the East Greenland Sea, and lower than AMSR2/ASI. The main differences are still in the ice margin zone along the Northeast Passage, which facilitates navigation of the Arctic in summer. Compared with the two AMSR2 products, FY-3B/DT-ASI can describe the lower SIC in the edge and the hole at the North Pole is bigger.

In summary, the three types of SIC obtained by similar ASI algorithms resemble each other better in March than in September. The main reason for the difference between FY-3B/DT-ASI and the two AMSR2 SIC products is the difference in the size of footprints.

In this section, FY-3B/DT-ASI SICs are compared with the SIC products distributed by NSMC, which are based on the NT2 algorithm, referred to here as FY-3B/NT2. In most cases, our results are lower than those from FY-3B/NT2. For example, Fig. 4 shows that the two SICs differ greatly in the sea ice margin area and at the edge of the land. SIC from FY-3B/DT-ASI is higher in the north and east of the Beaufort Sea and the marginal ice zone of the Barents Sea and lower in most of the Arctic Ocean, especially in the Greenland Sea and the Baffin Bay. The large difference in SIE values between the two sets of FY-3B data ($ 11.19 \times {10}^{6}\;{\rm{km}}^{2} $ from FY-3B/DT-ASI and $ 13.07\times {10}^{6}\;{\rm{km}}^{2} $ from FY-3B/NT2) comes mostly from the region close to the coast.

The daily spatial average SIC difference and the monthly spatial average SIC difference between FY-3B/DT-ASI and FY-3B/NT2 are shown in Figs 5a and b. Overall, the daily spatial average SIC from our results is smaller than that from FY-3B/NT2, with an average of −2.31%. Several abnormally large differences are due to the SIC polluted by the lack or excessive orbital data (Fig. 5a). From Fig. 5b, it is not difficult to find that the difference of monthly averaged SIC from 2011 to 2018 is larger in the first half-year than in the second half-year.

Two parameters that proved a quantitative evaluation of the large-scale changes and trends of sea ice are SIE and SIA. The usual method for calculating SIE is the sum of the areas of all grid cells with SIC greater than 15%, and SIA is the sum of the products of the area of each grid unit and the corresponding SIC. Here, we compare daily SIEs obtained from SICs from five datasets (FY-3B/NT2, NSMC; FY-3B/DT-ASI, OUC; AMSR2/DT-ASI, OUC; AMSR2/ASI, UB; SSMI/NT2, NSIDC). For SIA, we compare results from four datasets, excluding SSMI/NT2, which does not provide SIA data.

A comparison of climatological daily SIEs for 2012–2018 is presented in Fig. 6a. It can be seen directly that the SIE time series of all these datasets exhibit obvious seasonal changes, and some SIE fluctuations due to changes in weather scale can also be observed. The yearly average SIEs are $ 11.94\times {10}^{6}\;{\rm{km}}^{2} $, $ 10.52\times {10}^{6}\;{\rm{km}}^{2} $, $ 10.51\times {10}^{6}\;{\rm{km}}^{2} $, $ 10.29\times {10}^{6}\;{\rm{km}}^{2} $, $ 10.5\times {10}^{6}\;{\rm{km}}^{2} $, and $ 10.3\times {10}^{6}\;{\rm{km}}^{2} $ for FY-3B/NT2, AMSR2/DT-ASI, SSMI/NT2, AMSR2/ASI, FY-3B/DT-ASI(V0), and FY-3B/DT-ASI, respectively. FY-3B/DT-ASI(V0) stands for our SIC result without using the land spillover method. Therefore, the reduction of SIE by the algorithm accounted for 87.9%, and the land spillover accounted for 12.1%. The SIE from the FY-3B/NT2 data is much greater than that from the other data. Among the three datasets of results obtained using the ASI and DT-ASI algorithm, in comparison with SSMI/NT2, the absolute biases of the SIE are all within $ 0.8\times {10}^{6}\;{\rm{km}}^{2} $(Fig. 6b). From January to July, their SIEs are less than those from SSMI/NT2 (with the bias between AMSR2/DT-ASI and SSMI/NT2 being the smallest), whereas from September to November, larger than SSMI/NT2, so the deviation varies with the seasons and exhibits similar fluctuations. As shown in Fig. 6c, the SIE of FY-3B/DT-ASI is smaller than AMSR2/ASI, and the SIE of AMSR2/DT-ASI is higher than AMSR2/ASI. Overall, the three SIE results of the ASI algorithm are very close in summer.

A comparison of SIAs is shown in Figs 6d and e. The time series of SIA from the four SIC datasets are similar to the SIE time series. The SIA from the FY-3B/NT2 data is much greater and the FY-3B/DT-ASI results are closer to the other two AMSR2 datasets. The yearly average SIAs are $ 10.60\times {10}^{6}\;{\rm{km}}^{2} $, $ 9.49\times {10}^{6}\;{\rm{km}}^{2} $, $ 9.38\times {10}^{6}\;{\rm{km}}^{2} $, $ 9.51{\times 10}^{6}\;{\rm{km}}^{2} $, and $ 9.42{\times 10}^{6}\;{\rm{km}}^{2} $ for FY-3B/NT2, AMSR2/DT-ASI, AMSR2/ASI, FY-3B/DT-ASI(V0), and FY-3B/DT-ASI, respectively. The reduction of SIA by the algorithm accounted for 92.4%, and the land spillover accounted for 7.6%. Comparison of FY-3B/DT-ASI and AMSR2/DT-ASI with AMSR2/ASI shows that the differences between $ -0.5\times {10}^{6}\;{\rm{km}}^{2} $ and $ 0.5\times {10}^{6}\;{\rm{km}}^{2} $. The changing trends of the two biases are consistent, which is the opposite of SIE.

Figures 7a-d give monthly average SIE and SIA time series from four SIC datasets for March and September from 2011 to 2018. The inter-annual changes of SIAs and SIEs in the four SIC datasets in March and September were basically the same. Both reached their lowest on record in September 2012. The SIEs and SIAs from FY-3B/NT2 are clearly larger than those from the other three SIC datasets. The SIEs from three used ASI algorithm datasets are lower for March and higher for September than SSMI/NT2. The SIE and SIA for March from the FY-3B/DT-ASI are the smallest, and very close to AMSR2/ASI in September.