The National Ice Center sea ice charts in digital format are used for comparison with the Nimbus-5 ESMR v1.1 SIC. During the period when Nimbus-5 ESMR was operated, weekly ice charts were based on satellite data (Nimbus-5 ESMR, NOAA 2, 3, 4 VHRR visual and infrared imagery) combined with aerial reconnaissance data and observations from ships . The different sources of information are combined in a manual analysis in the ice chart . Uncertainties in the sea ice concentration according to of ±5% to ±10%, arise from poor resolution of printed satellite imagery and cloud cover that obscures the surface when using VHRR data . The sea ice concentrations of the ice chart are generally higher than the sea ice concentrations derived from satellite microwave radiometer data . Although this comparison is not totally independent, because Nimbus-5 ESMR data are used in the ice chart creation, we have still chosen to include the comparison here.

The raw N5ESMR data set contains erroneous measurements as a result of instrumental flaws and calibration failure. For v1.0, used various QC filters specifically for each error type; we also follow this approach here. Although they were successful in removing most erroneous data points, the filters also removed a significant amount of valid data, especially during the latter half of the N5ESMR operation after 10 November 1974. The removal of valid data-points has been found to be the result of a systematic change in the instrument noise level, reflected in the housekeeping data used for filtering. This and other filters have been revisited here, and we have therefore created a new QC filtering scheme that relies solely on features present in the TB data and not in the housekeeping data. The new scheme has been tested on all available swaths, and the threshold values are selected by qualitative inspection of data point distributions and outliers.

There are four types of filters in the new scheme, each targeting specific faulty patterns in the data. These are (i) value filters, where thresholds are used to discard data outside physically valid TB intervals; (ii) pixel filters, which evaluate the validity of a pixel based on its neighbors; (iii) sweep filters that remove one or more sequential sweeps; and (iv) swath filters that detect if the entire swath is corrupt.

uses a one-channel SIC algorithm based on dynamical tie points for ice (Tp,ice) and water (Tp,water) to derive SIC, shown in Eq. (), where i and j represent the spatial coordinates and t is time. 5SIC(i,j,t)=TB(i,j,t)-Tp,water(t)Tp,ice(i,j,t)-Tp,water(t) The tie points Tp,ice and Tp,water were the mean TB in the areas covered by the respective surface types in the ERA5 climatology and estimates of the sea ice concentration from the data itself. The 15 d running mean tie points were calculated to reduce day-to-day noise.

The problem with this approach is that it does not account for variations in TB for different types of sea ice. The different ice surfaces show different spectral properties and spatial and temporal dependence of Tp,ice(i,j,t). At 19.35 GHz, there is a significant difference between the radiometrically distinct ice type that we call first-year ice (FYI) and multi-year ice (MYI). Especially in Antarctica, the radiometric difference is caused not only by the ice age but also by the snow cover. MYI has a lower emissivity due to increased scattering from the snow and sea ice microstructure . For v1.0, this resulted in an SIC bias, where SIC is underestimated in MYI regions.

In our new method, Tp,ice(i,j,t) is described as a function of two radiometrically distinct tie points Tp,FYI(t) (typically 235–240 K) and Tp,MYI(t) (typically 215–220 K), together with a ratio parameter cMYI(i,j,t). The ice tie point is a function of cMYI(i,j,t) as shown in Eq. (): 6Tp,ice(i,j,t)=cMYI(i,j,t)Tp,MYI(t)+(1-cMYI(i,j,t))×Tp,FYI(t). This model has two parameters, cMYI(i,j,t) and SIC(i,j,t), to predict, but there is only one observable variable (TB). Modern microwave radiometers (e.g., SMMR, AMSR2) measure more frequencies and polarizations; and, for these, there are algorithms that utilize this multimodality to resolve the ambiguity . Here, we store spatio-temporal information, which can itself be regarded as a distinct modality to resolve the SIC and ice type ambiguity. This is the logic behind the LDTP algorithm.

An example of the effect of all QC filters (excluding the swath filter) individually and together is shown, along with the results of the old filtering scheme from in Fig. .

When comparing “All” and “Old” in Fig. , it is observed that significantly fewer valid data points are removed with the new scheme. Some “miscalibration” errors still occur, as seen around sweep 200 for “All”, but they are less prominent, and the new filters do detect some errors not detected by the old filters, e.g., near sweep 250.

In general, the old QC filters removed significantly more data than the new QC filters, especially after 10 November 1974, as seen in Table . The temporal pattern for data removal is further examined in Fig. , which compares the fraction of data removed for all swaths using the old and new filters. The figure clearly shows a pattern change around 10 November 1974 for the old QC filters, which is not seen in the new QC filters.

A peak in errors is observed around January 1977, which we found to be the result of a periodic calibration error where the mean signal changes approximately every twentieth sweep; thus, the new filter detects this error as intended.

Figure  illustrates the tie point and TB differences between the and the LDTP algorithms for a pixel in area (b) Baffin Bay of Fig. .

Baffin Bay cycles between open water during summer and consolidated FYI during winter, as shown in Fig. . The ice tie points are consistently lower than the tie points of the LDTP algorithms, and the LDTP tie points follow the TB closely. However, LDTP tie points closely follow the tie points by the end of 1973, 1976, and 1977, due to a mixture of a long melt season and missing data, which means that the age limit of the tie point is activated and the old, hemispheric tie points are used. Figure  shows that the new tie points in Baffin Bay replicate the TB variability.

Figure  shows the spatial performance of SIC (A, G), TB (B, H), local ice tie points (C, I), SIC standard error (D, J), rSD (E, K), and cell updates (F, L) for both the Arctic and Antarctic near the maximum extent of the ice. We see that the ice tie points for the Arctic (Fig. c) are distinct for FYI on the Siberian shelves, Baffin Bay and Hudson Bay, and MYI in the central Arctic. The ocean has a uniform signature, which is explained by the fact that the hemispheric open water tie point is selected. MYI is no longer distinguishable on the SIC map as in . As seen in Fig. e, rSD is a stability measure with low values near 3 in the central Arctic, smaller regions in the Ross Sea and Weddell Sea, and high values near the ice edge. Updates are more frequent in the Arctic than in Antarctica (Fig. f and l), which could be due to the fact that the consolidated MYI has a lower rSD than FYI (see Sect. ) and that there is little MYI in Antarctica compared to the Arctic. As seen in Fig. , MYI has a median rSD slightly lower than FYI. which means that the algorithm accepts more MYI tie points than FYI tie points. This, contrary to , means that the algorithm performs better on MYI than on FYI.

It is clear that the requirement for the tie points, namely that the tie point signature is that of 100 % SIC, is very difficult to achieve in summer when cracks and leads remain open and melt ponds form at the surface. This issue remains even in modern sea ice concentration data sets such as those of . During summer melt in the Arctic, melt ponds start to form on the surface of the ice, and at the same time, the emissivity of the snow and ice between the melt ponds changes. Eventually, late in the melt season, some of the melt ponds will melt through the floe and come into direct contact with the ocean below. However, because of the very shallow penetration depth (δp) of microwaves in water δp<1mm @ 19 GHz,, the melt ponds will have a signature similar to that of the water in the cracks, and the leads in-between the floes and the SIC mapped with microwave radiometers are a measure of the ice surface fraction . In Antarctica, melt ponds are rare, but there are still melt-induced emissivity changes of snow and ice . Although SIC estimates have high uncertainties during the summer, the algorithm still recovers quickly when the melt ends. Melt-induced changes in emissivity affect the SIC estimates and the sea ice area, while the SIE estimates are mainly affected by the level of open water noise .

The new daily v1.1 SIC is compared with v1.0 and the Arctic ice charts from the National Ice Center (NIC) in Fig. . We have used the zones of Fig.  to evaluate the biases over different surface types between the different products.

The Arctic SICv1.1-SICv1.0 difference plot in Fig. a shows that the LDTP algorithm correctly has SIC near 100 % over MYI, with an average difference of 9.0 % and an average difference over FYI of -2.0%. The positive difference SICv1.1-SICv1.0 in Arctic MYI is a manifestation of the overestimated MYI tie point in and vice versa for FYI. This issue has been resolved in the new LDTP processing, as illustrated by the difference SICv1.1-SICNIC in Fig. c.

In Antarctica, the stable signatures, illustrated by the local ice tie points in Fig. i, have low values (215–220 K) near the ice edge in the Weddell and Ross Seas and in the central part of the Amundsen Sea, and higher values (230–240 K) elsewhere. The low value regions are also coincident with the regions where the SICv1.1 is higher than SICv1.0, as illustrated in Fig. e. This can be the result of a low rSD occurring despite less than 100 % SIC for the area, violating the main assumption of the LDTP algorithm presented in Eq. (). A lower rSDmax might alleviate that, but it would result in even fewer tie point updates for the Antarctic. In all other regions in Antarctica, SICv1.1 is smaller than SICv1.0. In general, the running standard deviation rSD is higher in ice covered regions in Antarctica (Fig. e) than in the Arctic (Fig. k), and this affects the frequency of tie point updates (Fig. f and l).

The total SIC standard error in Fig. d and j is dominated by the smearing uncertainty along the ice edge. Because the zone with intermediate concentrations near the ice edge is much wider in Antarctica than in the Arctic, the zone with high total standard error (20 %–30 %) in Antarctica is much wider than in the Arctic.

In the Arctic SICv1.1 and SICNIC comparison (Fig. c), the differences are related to the spatial resolution of ESMR and navigational ice charts in the Greenland Sea and along the ice edge in general. Because of the coarse resolution of ESMR, even sharp ice will appear gradual in SICv1.1, and comparing it with the sharp ice edge in the ice chart, the differences (SICv1.1-SICNIC) will be negative over ice near the ice edge and positive over open water near the ice edge, as seen in Fig. c. Over the central Arctic Ocean pack ice, the SICv1.1 is near 100 % as expected, while the ice chart is assigned a constant SIC of 95 % as an indication that the real SIC is between 90 % and 100 % . Keeping this 5 % bias in mind, we find an average SICv1.1-SICNIC bias over MYI of 2.5 % and 2.9 % for FYI, which means that SICv1.1 on average predicts that the SIC is 2.7 % higher than the 95 % SIC from the digitized ice chart. Ice charts are not available for Antarctica. We do not have ice charts covering Antarctica and there has not been an independent study comparing the Antarctic SIC with Landsat or other high resolution optical data similar to . The validation of was almost exclusively in Arctic FYI and the validation shows that SICv1.1 is virtually unbiased with a median difference of 0.0 %. We believe this is also an indication that SICv1.1 in Antarctica is less biased than in v1.0.

The TB of thin ice/new ice is between ice and open water TB and SIC and the negative SICv1.1-SICNIC difference in the Greenland Sea is probably caused by the presence of new ice e.g., or an overestimation of SIC in the ice charts (or both). The ice charts from the 1970's do not have information on ice type (except a mapping of fast ice), but an inspection of the ice charts after 1995, when ice type information became available in the charts, shows that the Greenland Sea and the Barents Sea have large fractions of new ice/ thin ice during mid-winter. In open water far from the ice edge, the SIC is overestimated compared to the ice chart due to geophysical noise that affects the SIC of the microwave radiometer .

The SIE derived for both hemispheres using v1.0 and v1.1 is shown in Fig. . Tabular versions of the new data are found in Appendix B. The extent is calculated as the total area with an SIC greater than 30 %, and only months with more than 99 % spatial coverage in the 25 km EASE 2.0 grid are used.

As expected, the two schemes are very similar, as the LDTP algorithm of v1.1 mainly affects multi-year ice SIC, which is not detected at SIC<30%. The main difference between the algorithms for computing SIE is the temporal coverage, which is improved by the new approach as a result of the updated QC filtering scheme. Some small differences can be seen in the latter half of 1974, likely a result of summer melt, which violates the assumptions behind both algorithms; that is, that a stable signature for 100 % SIC (the tie point) can be estimated. The summer period sea ice area and SIC estimates should therefore be used with caution.