In light of a rapidly changing Arctic sea ice cover, measuring sea ice thickness remains an important task [1
]. Since the launch of ESA’s Soil Moisture and Ocean Salinity (SMOS) mission in 2009, it has been demonstrated that passive microwave measurements at L-band are well suited for sea ice thickness retrieval [2
]. Having a large sensitivity to thin sea ice of 0.5 to 1 m, L-band measurements nicely complement Cryosat-2 altimeter data, which show increasingly large relative uncertainties for thin ice below 1 m [6
]. Due to the daily coverage of polar regions, SMOS data can also be used to monitor and analyze sea ice thickness changes on short time scales.
After the second L-band sensor Aquarius [7
] stopped delivering data in June 2015, there is currently a third L-band sensor in orbit: NASA’s Soil Moisture Active Passive (SMAP) mission delivers data since April 2015 and carries a passive microwave radiometer measuring at the same frequency as SMOS and with a comparable footprint size. This is an excellent opportunity to compare brightness temperature (TB) measurements from SMOS and SMAP and to analyze the benefits of a combined product for sea ice thickness retrieval. While SMOS TBs are multi-angular, covering an incidence angle range of 0
to about 70
, SMAP measures at a fixed incidence angle of 40
. Thus, as a first step for a brightness temperature comparison, SMOS TBs have to be fitted to the SMAP incidence angle.
Different methods have been suggested to fit SMOS data to specific incidence angles. A simple approach is the averaging of the multi-angular TBs in incidence angle bins of a fixed width. A bin width of 5
has been chosen for the CATDS (Centre Aval de Traitement des Données SMOS) L3 TB dataset [8
], for example. Other studies fitted a quadratic function [9
], a third order polynomial function [10
] or an exponential function [11
] to the vertically and horizontally polarized TBs separately. In addition, De Lannoy et al. [9
] weighted the TBs at each incidence angle depending on the radiometric error. Zhao et al. [12
] proposed a more sophisticated two-step regression fitting function to refine the multi-angular SMOS TB measurement and to reduce their uncertainties.
However, none of these studies systematically compared the performance of the different methods. This is why, in the first part of this study, we test different fitting methods and analyze which one is most suitable for a comparison of SMOS and SMAP TBs over high latitude ocean regions poleward of 60. We construct a synthetical dataset to evaluate the performance of different fitting methods and choose one method for the following SMOS and SMAP brightness temperature comparison.
Only a few studies comparing SMOS and SMAP brightness temperature measurements have currently been published. Bindlish et al. [13
] compared SMOS and SMAP TBs from simultaneous overpasses within a maximum time window of 30 minutes and with footprint distances of less than 1 km. They found a very good agreement with high correlations and a small bias of less than 0.4 K over ocean areas for TBs at the top of the atmosphere (TOA). Over land, SMOS observations showed a cold bias of about 2.7 K as compared to SMOS for both polarizations. Al-Yaari et al. [14
] compared SMOS and SMAP TBs over land to produce a consistent soil moisture data set. They found cold biases of SMAP with respect to SMOS of about 2–4 K over most land-areas but large warm biases of up to 10 K over high latitude regions. They attributed a part of the bias to corrections of galactic and atmospheric effects [9
] and water-body corrections applied for SMAP but not in the SMOS product. Huntemann et al. [15
] compared daily averaged TBs in the Arctic over open ocean and sea ice. They found good agreement over sea ice but larger differences of about 5 K over the polar oceans and suggested a linear fit to account for the differences between SMOS and SMAP TBs in a combined product.
In this study, we focus on comparing measurements from the two sensors over the polar ocean regions, which are regions of interest for the retrieval of sea ice thickness. We also compare daily TB products, but, unlike Huntemann et al. [15
], who simply compared L1C data products, we use TOA TBs for both sensors. In addition, Huntemann et al. [15
] only had three months of data available, while, in this study, we compare two years of overlapping data using the latest data versions of both products. In addition, we perform a much more detailed TB comparison over different surface types, such as the polar oceans, thin first-year-ice, and thick multi-year-ice in both hemispheres.
As a next step, we derive sea ice thicknesses from SMOS and SMAP TBs using the algorithm developed at the University of Hamburg (UHH) [3
] to analyze how observed TB differences translate to ice thickness differences. Since the UHH algorithm is based on TBs averaged over the incidence angle range from 0
, it has to be adapted to the SMAP incidence angle of 40
first. Finally, we discuss the advantages and disadvantages of a combined SMOS and SMAP sea ice thickness product.
The SMOS and SMAP data products are described in Section 2.1
. An evaluation of the different fitting methods is presented in Section 2.2
and a brief description of the adapted sea ice thickness retrieval method is given in Section 2.3
. In the following sections, we present the results for the TB comparison over a stable target (Section 3.1
) and over polar ocean and sea ice regions (Section 3.2
), as well as the sea ice thickness comparison (Section 3.3
). Finally, all results are summarized and discussed in Section 4
To obtain brightness temperatures at a specific incidence angle from the multi-angular SMOS measurements, we evaluated the performance of six different fitting methods based on their accuracy and computing time. Even though the bin mean is the fastest method, it only yields accurate results if the averaging interval is optimized and if more than 50 measurements are available. The optimal interval decreases from about 20
near-nadir to about 3
, indicating that a fixed interval width of 5
as in the CATDS dataset [8
] is not an optimal choice. The Zhao fit including a weighting by the radiometric accuracy showed the overall best performance, even for a small number of 15 to 50 measurements. These results are in line with those by Zhao et al. [12
] who found that the fitted TBs over land areas are more consistent with results from a radiative transfer model than the CATDS data.
We compared two years of daily polarized brightness temperatures from SMOS and SMAP over the sea-ice-covered polar oceans and found a good overall agreement. However, we found a cold bias of SMAP compared to SMOS TBs of 1.4 K for vertical and about 4 K for horizontal polarizations over sea ice. Over water, the bias for horizontal polarization even exceeds 6 K. These numbers result in a bias of the TB intensity of about 2.7 K for both ice and ocean surfaces. In the following, we discuss different possible reasons for the TB bias, which have been partly suggested in the literature.
Differences in corrections for galactic and atmospheric effects were found to cause TB biases between SMAP and Aquarius over land and over the Antarctic continent [29
]. However, since we compared uncorrected TOA brightness temperatures for both SMOS and SMAP, we can certainly exclude this reason. We can also exclude RFI contamination of SMOS data as a main reason for the bias. RFI is not an issue around Antarctica, but we found biases there as well, which are consistent with values over the northern polar oceans. Orbital differences could also play a role since SMOS and SMAP have different equator crossing times at 6:00 and 18:00, respectively. However, this would result in bias values and signs changing with time and region, which is not the case as our analysis has shown.
Even though we did not find large differences of the biases observed in different regions over Arctic and Antarctic sea ice and ocean regions, results might be different in lower latitude regions. Other studies with co-located SMOS and SMAP overpasses showed a negligible bias of about 0.5 K over oceans far from the coast, but a larger bias of about 2.7 K over land for both horizontal and vertical polarizations [13
]. This land bias is very similar to the TB intensity bias of about 2.7 K we found over multi-year ice. An initial analysis by Peng et al. [31
] of a re-calibrated version of the SMAP brightness temperatures—which is planned to be released in 2018—showed a reduced bias over land. Thus, this bias is probably related to the calibration of the SMAP data.
This does not explain the different findings over ocean in Bindlish et al. [13
] and in this study, however. A possible reason for this bias is the so-called land-sea contamination (LSC), which is a known issue for SMOS brightness temperatures. Though the exact reasons for LSC are still under discussion [32
], it is known that large TB contrast between land (or ice) and ocean surfaces result in artificially increased TBs over the oceans in coastal regions. This bias can exceed 3 K for TB intensity in the polar regions [33
], which is in line with our TB intensity biases over oceans of about 2.5–3.0 K.
We presented a comparison of brightness temperatures and derived sea ice thicknesses from SMOS and SMAP over the polar ocean regions. As a first step, the multi-angular SMOS measurements were fitted to the SMAP incidence angle of 40
. For this purpose, we tested different fitting methods, such as a bin mean or a linear fit with fixed or optimized bin widths and a more sophisticated fitting function using a two-step regression method by Zhao et al. [12
]. To evaluate the accuracy of the different fitting methods, we constructed a synthetical dataset based on SMOS characteristics in terms of incidence angle distribution and radiometric accuracies in high latitudes. We found that the weighted Zhao fit performs best, yielding a high accuracy even for a small number of measurements of only 15. We could also demonstrate this good performance of the weighted Zhao fit over a stable target. TB time series over the Ross and Ronne ice shelves showed a smaller day-to-day variability for the weighted Zhao fit compared to the bin mean method.
Daily values of fitted SMOS TBs were then compared to those from SMAP over the polar oceans, over thin first-year-ice, and over thicker multi-year-ice. Generally, the time series agreed very well, with correlations over sea ice exceeding 0.99. However, we found a cold bias of SMAP compared to SMOS TBs of 2.7 K for TB intensity over both ice and ocean surfaces. This bias is even larger for horizontal polarization, with about 4 K over sea ice and more than 6 K over water. Since this bias is mostly constant in time and space, it can be adjusted using a linear fitting function.
The TB intensity bias also translates to a sea ice thickness bias. Averaged over one winter season, SMOS thicknesses are 7 cm larger than SMAP thicknesses with an RMSD of 12 cm. After adjustment of the TB bias, the RMSD values for brightness temperature over thick ice decrease to about 1 K, which is within the sensor accuracies.
We also evaluated the changes caused by adapting the SMOS retrieval algorithm from near-nadir incidence angles [3
] to the SMAP incidence angle of 40
. Since the penetration depth of L-band radiation in sea ice decreases with increasing incidence angle, the maximum retrievable thickness is on average 4 cm smaller at 40
than for the near-nadir retrieval. This results in an increase of the number of saturated pixels of about 10%.
We conclude that it is feasible to combine SMOS and SMAP brightness temperatures for an improved retrieval of sea ice thickness. The overall data loss—mostly due to RFI contamination of SMOS measurements—can be largely reduced. In addition, the region of daily coverage is extended equatorwards resulting in a better coverage of the sea ice covered areas around Antarctica. Most importantly, a combined dataset from the two sensors ensures a consistent continuation of the L-band sea ice thickness timeseries in the future even if one of the sensors stops delivering data. This improved data coverage and availability outweighs in our opinion the slight increase of saturated pixels caused by the adaption of the retrieval algorithm from near-nadir incidence angles to the SMAP incidence angle.
With two sensors in orbit and multiple overpasses per day in high latitudes, it is now also possible to look at short-time-changes of the sea ice thickness in a specific region. Such a swath based analysis opens up new possibilities for case studies or for assimilation of L-band sea ice thickness data into forecast models.