This paper presents the first long-term climate data record of sea ice extents and backscatter derived from intercalibrated satellite scatterometer missions (ERS, QuikSCAT and ASCAT) extending from 1992 to the present date (Verhoef et al., 2018). This record provides a valuable independent account of the evolution of Arctic and Antarctic sea ice extents, one that is in excellent agreement with the passive microwave records during the fall and winter months but shows higher sensitivity to lower concentration and melting sea ice during the spring and summer months. The scatterometer record also provides a depiction of sea ice backscatter at C- and Ku-bands, allowing the separation of seasonal and perennial sea ice in the Arctic and further differentiation between second-year (SY) and older multiyear (MY) ice classes, revealing the emergence of SY ice as the dominant perennial ice type after the historical sea ice loss in 2007 and bearing new evidence on the loss of multiyear ice in the Arctic over the last 25 years. The relative good agreement between the backscatter-based sea ice (FY, SY and older MY) classes and the ice thickness record from Cryosat suggests its applicability as a reliable proxy in the historical reconstruction of sea ice thickness in the Arctic.
Dating as far back as 1978, passive microwave sensors provide the longest
record of sea ice concentration and extents available to date and are
currently established as the sea ice monitoring standard for climate
studies, regardless of well-known difficulties around the detection of lower
concentration and melting sea ice conditions during the summer months (Meier
et al., 2015). The scatterometer sea ice record presented here only dates
back as far as 1992, but proves more sensitive to summer sea ice, its
primary purpose being the conservative detection and removal of ice-contaminated scenes that compromise scatterometer wind retrievals. Previous
long-term scatterometer sea ice records have been developed, spanning the
decade-long QuikSCAT mission from 1999 to 2009 (Remund and Long, 2014), and
extended into 2014 using the Oceansat-2 scatterometer (OSCAT) mission (Hill
and Long, 2017). These precedent scatterometer records (which use maximum
likelihood class discrimination based on the Ku-band pseudo-polarization
HH
Some research groups have opted to blend active and passive microwave observations (i.e., the gradient and polarization ratios from radiometer, along with the C-band anisotropy coefficient from scatterometer data) in a multisensor approach towards a sea ice edge product (Aaboe et al., 2015). The validation of the summer sea ice extents from blended records against operational sea ice charts, which also have arguable accuracy during summer, shows negative biases by up to 30 % (Aaboe et al., 2016), indicating that the distinct sea ice detection skills of scatterometer data may be lost in the blend. Note though, that the primary aim of passive microwave records is the mapping of sea ice concentration, sea ice extent being only a downstream indicator.
In this paper, an independent record of sea ice extents has been produced from intercalibrated scatterometer data: the QuikSCAT mission from 1999 to 2009 (Belmonte Rivas and Stoffelen, 2011), extended forward to the present day with the ASCAT record (Belmonte Rivas et al., 2012) and backwards to 1992 with the ERS mission (Otosaka et al., 2017) using dedicated Bayesian sea ice detection algorithms designed to maximize the skill for ocean–ice discrimination. These algorithms have been tuned to match the passive microwave sea ice extents during the fall and winter months and to remain consistent across the scatterometer overlap periods in 2000 (ERS with QuikSCAT) and 2008 (QuikSCAT with ASCAT). The stability and intercalibration of the ERS, QuikSCAT and ASCAT backscatter records is guaranteed to within 0.1 dB via buoy collocation (QuikSCAT; Verhoef and Stoffelen, 2016), ocean calibration (ASCAT and ERS; Verhoef and Stoffelen, 2017) and nonlinear corrections from cone metrics (ERS; Belmonte Rivas et al., 2017), offering a stable reference to verify the consistency of calibration adjustments made in passive microwave records, which – if not done properly – are known to cause discontinuities and affect long-term trends in sea ice concentration (Eisenman et al., 2014; Titchner and Rayner, 2014).
The scatterometer sea ice record also monitors the evolution of sea ice
backscatter collected at C-band and Ku-band, which are widely applied to
discriminate ice classes, such as first year (FY) and older (second-year, SY
and multiyear, MY) sea ice in the Arctic. It is known that sea ice
backscatter is modulated by surface permittivity, surface roughness and the
presence of volume inhomogeneities, such as air and brine pockets or snow
layers above (Ulaby et al., 1981). The main basis for FY and MY ice
separation lies in older ice types becoming brighter with increased volume
scattering after summer melt, although MY detection may become difficult
because of patches of bright FY ice that may have locally undergone
deformation. To date, the separation between Arctic FY and MY ice types using
active microwaves has relied on fixed backscatter thresholds defined after
visual inspection of stable winter backscatter histograms. For example, Kwok
(2004) established
The introduction of an intercalibrated sea ice extent and backscatter record from multiple scatterometer missions (ERS, QuikSCAT, ASCAT) consistently connected from 1992 to 2016 through dedicated and validated sea ice detection and backscatter normalization algorithms is the object of this contribution. In Sect. 2, we introduce the satellite scatterometer missions and the Bayesian detection algorithms that constitute this record. In Sect. 3, the scatterometer sea ice extents are compared with passive microwave fields, showcasing their agreement and distinct sensitivities. This section also provides an overview of the long-term evolution of sea ice extents and sea ice types afforded by 25 years of scatterometer data, along with a taste of its potential to stimulate new research questions.
The continuous monitoring of the Polar regions with satellite scatterometers began in March 1992 after the launch of the European Remote Sensing (ERS) satellites, which operated a C-band instrument (5.3 GHz, VV polarization) in global mode until January 2001. It was continued on July 1999 by the Quick Scatterometer (QuikSCAT), which operated a Ku-band instrument (13.4 GHz, HH and VV polarization) up until November 2009. The scatterometer record ends with another C-band instrument that collects VV polarized backscatter at 5.3 GHz, the Advanced Scatterometer (ASCAT) on Metop-A, operating from January 2007 to the present date. The temporal spans of the satellite missions that make up the scatterometer record are illustrated in Fig. 1, showing the location of the mission overlap periods (2000 and 2008) used to verify the consistency of the sea ice extents and sea ice types across the C- and Ku-band missions. For reference, the observation geometries of the constitutive ERS, QuikSCAT and ASCAT scatterometers (namely, three single-sided VV polarization fan-beams, single rotating VV and HH polarization pencil beams and three double-sided fan-beams) are shown in Fig. 2. While the QuikSCAT mission observes backscatter at Ku-band from an incidence angle of 54 deg (outer VV-pol beam) and 46 deg (inner HH-pol beam), the ERS and ASCAT missions observe backscatter at C-band from a broad range of incidence angles collected across the swath (from 18 to 64 deg in VV-pol).
Temporal spans of the satellite scatterometer missions used.
The algorithm for sea ice detection with scatterometers is a maximum likelihood class discrimination approach based on probabilistic distances to ocean wind and sea ice geophysical model functions (GMFs). The GMFs describe the behavior of backscatter as a function of observation geometry (i.e., incidence and azimuth angles) and geophysical variables such as wind speed and direction or sea ice type.
The ocean wind GMF, also known as the wind cone, is an empirically derived model used to derive ocean surface wind vectors operationally (see Fig. 3): we presently use CMOD7.1 (Stoffelen et al., 2017) for ERS and ASCAT, and NSCAT-4 (Verhoef and Stoffelen, 2017) for QuikSCAT. The sea ice GMF, also known as the sea ice line, is empirically derived from stable wintertime backscatter levels (Belmonte Rivas and Stoffelen, 2011; Belmonte Rivas et al., 2012; Otosaka et al., 2017). Physically, the discrimination between open water and sea ice classes is based on the separability between surface and volume scattering effects: in the QuikSCAT case, the discrimination relies on polarization and azimuthal anisotropy of backscatter (high for open water; lower for sea ice), while in the ERS ASCAT case, the discrimination relies on backscatter directivity (i.e., the derivative of backscatter with incidence angle) and azimuthal anisotropy (high for open water; lower for sea ice). Previous Bayesian formulations for sea ice detection with scatterometer data, e.g., Remund and Long (2014), have used aggregates such as mean backscatter, polarization ratio and azimuthal anisotropy as class discriminants and empirically adjusted covariances to represent the class dispersions. The advantage of the GMF approach is that the dispersion of measurements around extended class model functions is smaller than around class aggregate means, approaching the limits imposed by the scatterometer noise levels and allowing the Bayesian method to reach its maximum discrimination power (Otosaka et al., 2017).
Observation geometries of the satellite scatterometers.
Ocean wind and sea ice geophysical model functions (GMFs) in the
scatterometer measurement space: QuikSCAT
To calculate the Bayesian sea ice probability, the algorithm computes the
minimum normalized squared distance (or maximum likelihood estimator, MLE)
from observations
The number of degrees of freedom is given by the difference between the size
of the measurement space (
Our Bayesian approach affords two parameters for algorithmic tuning: one is
the tolerance factor
In order to build a uniform record of sea ice backscatter, all the C-band measurements must be normalized to a standard incidence angle (chosen at 52.8 deg, set in the middle of the ASCAT mid-beam swath, and closest to the QuikSCAT VV-pol incidence) using a model that describes the dependence of C-band sea ice backscatter on incidence angle. In the present version, the normalization assumes a linear relation between C-band backscatter and incidence angle using sea-ice-type-dependent coefficients (Ezraty and Cavanie, 1999). A refined incidence angle correction based on the empirical C-band sea ice backscatter model developed in Otosaka et al. (2017) is planned for a future release. The largest obstacle, though, arises from the presence of composite C- and Ku-band observations in a single-backscatter record, since their sensitivities to sea ice type differ. Both frequencies are similarly responsive to surface roughness, e.g., over deformed first-year sea ice, but Ku-band is more responsive to volume scattering in multiyear ice (Ezraty and Cavanie, 1999). As a result, the separability between deformed FY and MY ice classes is better at Ku-band. At C-band, the disambiguation between deformed FY and MY ice classes is more difficult in terms of backscatter alone, although recourse can always be made to additional information, such as the monitoring of backscatter derivatives or the introduction of geographical constraints, such as a marginal sea mask.
The time series of Arctic and Antarctic sea ice extents observed by the ERS, ASCAT and QuikSCAT scatterometers from 1992 to 2016 are shown in Fig. 4, along with the differences to the sea ice extents from the SSMI(S) passive microwave sea ice concentration (15 % threshold) algorithms from the NSIDC-0051 (Cavalieri et al., 2015) and the OSI SAF's latest major reprocessing release (OSI SAF-409a as in Tonboe et al. (2016) extended into 2016 with OSI SAF-430) in Fig. 5. The sea ice extents are constrained by a unique land mask built from the union of all active and passive sensors' land masks. No significant long-term trends are observed in the active-to-passive differences, other than a slight increase in the variability of the Arctic sea ice extent biases from 1992 to 1996, which has been attributed to data gaps in the ERS-1 mission due to SAR operations (Otosaka et al., 2017).
Time series of monthly Arctic (continuous) and Antarctic (dashed) scatterometer sea ice extents from 1992 to 2016.
Time series of scatterometer sea ice extent differences to passive
microwave (SSMI-based) products from NSIDC-0051 and OSI SAF-409a from 1992 to
2016 for the Arctic
The correspondence between the QuikSCAT scatterometer and passive microwave
sea ice extents from the NSIDC (NT-based) algorithms was extensively
verified using coincident SAR imagery (Belmonte Rivas and Stoffelen, 2011)
to reveal excellent agreement during the winter and fall seasons, and
persistent differences during the spring and summer months. Figure 6
illustrates the seasonal pattern of active-to-passive sea ice extent
differences over the ERS–QuikSCAT and QuikSCAT–ASCAT overlap periods in 2000
and 2008. The agreement between the overlapping C- and Ku-band scatterometer
sea ice extents is very good all year round, with differences within
0.25 million km
Comparison of ERS, QuikSCAT, ASCAT and passive microwave
(OSI SAF-409a and NSIDC-0051) sea ice extents over 2000
Comparison of summer sea ice extents on 15 September 2016
from passive microwaves (NSIDC-0051
Surface wetness and melt ponding are thought to be responsible for large errors in passive microwave sea ice concentrations during spring and summer (Comiso and Kwok, 1996; Kern et al., 2016), and these errors affect the ocean heat contents and associated surface fluxes when assimilated into ocean and atmosphere reanalyses (Hirahara et al., 2016). In this context, the scatterometer record nicely complements the passive microwave products in monitoring the expanse and evolution of the lower concentration and water-saturated (rotten) late spring and summer sea ice classes. It is the different degree of inclusion of these mixed sea ice and open ocean conditions that is mainly responsible for the sea ice extent differences observed between scatterometers and passive microwaves in the spring and summer months. The reader is referred to (Belmonte Rivas and Stoffelen, 2009) for a more extended collection of collocated SAR and MODIS plates illustrating the nature of these differences, including a variety of scenes dominated by water-saturated (brash) ice, decaying floes, ice bands and mixtures thereof.
Figure 8 shows the long-term evolution and interannual variability of the
Arctic and Antarctic minimum and maximum sea ice extents from the
scatterometer and passive microwave records. These figures attest to the
coincident emergence of significant events, such as the Arctic summer minima
in 2007 and 2012, or the Antarctic wintertime maxima in 2014, on top of
long-term trends that bear witness to Arctic sea ice decline and a modest
increase in Antarctic sea ice extents. Note that, while the NSIDC algorithm
ranked the summer of 2016 as second lowest in Arctic sea ice extent, tied
with 2007 (NSIDC, 2016), the scatterometer record observes a somewhat slower
trend in the decline of Arctic summer ice over the last 5 years and only
ranks 2016 as the fifth lowest, with 4.5 (3.9) million km
The monitoring of sea ice backscatter may be used to discriminate Arctic FY and MY sea ice types, but it also can be applied to estimate sea ice motion by feature tracking (Zhao et al., 2002; Lavergne et al., 2010), characterize Antarctic sea ice types (Morris et al., 1998; Haas, 2001; Willmes et al., 2011) or estimate the onset and duration of melt (Drinkwater and Liu, 2000; Howell et al., 2008). As already noted, the discrimination between Arctic FY and MY ice types using active microwaves is not without difficulty, its main hindrances being the seasonal variability of backscatter, including the effects of surface deformation, ice/snow metamorphism and a developing snow cover or the appearance of summer signatures that are more dependent on surface weather via processes such as wet snow attenuation and changes in brine temperature (Barber and Thomas, 1998). The annual cycles of MY ice coverage in the Arctic Ocean were estimated using the QuikSCAT record (1999–2009) by Kwok et al. (2009) with a fixed backscatter threshold from January to April and the record by Swan and Long (2012) with a seasonally dependent backscatter threshold from November to April. A multi-mission record was produced that extended forward in time to 2014 using Ku-band Oceansat-2 scatterometer (OSCAT) data (Lindell and Long, 2016). In order to avoid the high-backscatter FY ice in the marginal ice zone (MIZ) being classified as MY ice, Kwok et al. (2009) introduced a static geographical mask, while Lindell and Long (2016) applied a MIZ correction algorithm based on the temporal persistence of the MY signature, along with a 40 % sea ice concentration mask from passive microwave data.
Evolution of minimum and maximum monthly sea ice extents from
scatterometers (black line) and passive microwaves (continuous and dashed red
lines) from 1992 to 2016 for the Arctic
For the determination of the time series of Arctic MY ice coverage, we adopt the single-backscatter threshold approach. To avoid dealing with seasonal variability, we only use stable wintertime (March) backscatter maps, assuming that the backscatter signatures of the reference winter sea ice classes do not change with time. We also introduce a geographical mask to screen the high-backscatter response from MIZ sea ice, which has been attributed to surface deformation by compression and irreversible snow/ice metamorphism after melt-freeze events (Voss et al., 2003; Willmes et al., 2011). The geographical mask delimits the Arctic Basin (see red contours in Fig. 11) across the Fram Strait and Svalbard, to Severnaya Zemlya through Franz Josef Land (Kwok et al., 1999). An additional line from Point Barrow to Wrangel Island also excludes the Chukchi Sea from the MY area estimations. The geographical mask omits the ubiquitous presence of multiyear ice in the Greenland Sea or its episodic incursions into the marginal Chukchi, Barents and Kara seas.
For the consistency of the record, the backscatter thresholds for MY ice
detection at Ku- and C-bands are matched according to their joint backscatter
distributions and resulting spatial boundaries. Figure 9a and b show the
joint backscatter distributions of Arctic sea ice at C-band and Ku-band for
the month of March in 2000 and 2008, before application of the geographical
mask. Before masking, the joint distributions of wintertime sea ice
backscatter are characterized by two elongated clusters: an upper cluster
corresponding to perennial (MY) ice and a lower one corresponding to seasonal
(FY) ice (Ezraty and Cavanie, 1999). The cluster elongation accounts for
geophysical variability, with perennial ice types getting brighter as they
accumulate summer conditions and seasonal ice types becoming brighter with
surface deformation and/or metamorphism. Note that the range of backscatter
variability associated with deformation and/or metamorphism in the lower
seasonal ice cluster (
Joint distributions of wintertime (March) sea ice backscatter at
C-band (
Figure 10a, c show the marginal distributions of sea ice backscatter at Ku-band (a) and C-band (c) that correspond to the geographically masked joint distributions for the year 2000. The marginal backscatter distributions are characterized by two well-defined modes, associated with FY and MY sea ice types and connected by a transition range. Using spatially collocated Ku- and C-band backscatter measurements, sea ice types in the transition range between the FY and MY modes may be further separated (see joint distribution in Fig. 9c) into deformed FY (in the high range of the seasonal sea ice cluster), SY ice (in the low range of the perennial sea ice cluster) and FY–MY mixtures (along the path connecting the seasonal and perennial clusters). We note that an optimal threshold for MY detection should guarantee that most of the rough FY is removed from the MY category, while collecting various fractions of SY, MY and FY–SY–MY mixtures.
Marginal distributions of wintertime (March) sea ice backscatter
collected at Ku-band
After the anomalously large loss of Arctic sea ice that occurred in the
summer of 2007, the shape of the wintertime sea ice backscatter histograms
were remarkable altered. The earlier bimodal (FY and MY) histograms were
replaced by trimodal distributions, featuring a smaller MY mode, and a new
mode corresponding to SY ice emerging in the low range of the perennial sea
ice cluster (Fig. 9d). The emergence of the new SY mode is also evident in
the marginal distribution of Ku-band backscatter for the year 2008
(Fig. 10b), though more difficult to see in the marginal distribution of
C-band data for the same year (Fig. 10d, around
The spatial distributions of the FY, SY and old MY classes that result from applying the single threshold approach on Ku- and C-band backscatter images are displayed in Fig. 11, along with the average sea ice age from the EASE-Grid data set NSIDC-0611 from Tschudi et al. (2016) for that period. The spatial distributions of the total MY ice class (defined as the sum of SY and old MY classes) from the scatterometer and the Lagrangian sea ice age analyses are in general good agreement, although their depictions of the SY ice class differ somewhat. We note that the old MY sea ice class has a larger geographical spread in the Lagrangian data set, particularly over areas where MY ice is exposed to strong shear stress, such as in the Beaufort Sea. From the analysis of joint backscatter distributions, we know that the scatterometer SY class is bound to contain varying amounts of FY–SY–MY mixtures (and probably some deformed FY too), thus an inherent ambiguity remains regarding the dominance of pure SY ice versus mixed FY–MY combinations in a cell labeled SY, particularly before 2007. On the other hand, the Lagrangian data set is monitoring the age of the oldest ice in a cell, regardless of its weight over other ice fractions, probably biasing this product towards a larger spread of old MY ice in the Arctic. Outside of the red contour that delineates the Arctic basin mask in Fig. 11, we cannot register older ice reliably because of the strong backscatter from deformed MIZ ice.
Geographical boundaries of wintertime FY (dark blue), SY (light
blue) and MY (white) sea ice classes from Ku-band backscatter (left panel),
C-band backscatter (middle panel) and NSIDC sea ice age (right panel) for
2000
The time series of the total MY sea ice extents is shown in Fig. 12, along with the extents of
the separate SY and old MY class contributions calculated using the
backscatter threshold approach on wintertime (March) data collected within
the geographically masked Arctic Basin. All estimates
exclude a common polar gap extent of 0.354 million km
Time series of monthly wintertime (March) total multiyear sea ice extents (segmented into SY and older MY classes) within the Arctic Basin from the scatterometer (black) and the NSIDC sea ice age (red) records. Error bars are representative of the scatterometer class extent errors associated with a fixed backscatter threshold uncertainty of 0.1 dB (i.e., calibration accuracy).
Thus far, we have justified the definition of a separate SY ice class after the emergence of a SY mode with an entity of its own in the scatterometer backscatter histograms. The differentiation of SY and lower concentration of MY using a single frequency remains an open question. By construction, the scatterometer SY class will accommodate various fractions of deformed FY and FY–SY–MY mixtures, which we suggest may be differentiated from the homogenous SY ice signature by recourse to dual Ku-band and C-band observations. In this context, the reprocessed Ku-band Oceansat-2 record spanning the period from 2009 to 2014, also available in our scatterometer record, affords a new opportunity to resolve this ambiguity. We note that the scatterometer record may be helpful for checking currently developing algorithms for MY ice concentration based on satellite passive microwave or blended data, given that none of the latter products uses a separate tie point for SY ice, leaving the SY ice signature to be effectively interpreted as lower concentration MY ice. As an illustration, Fig. 13 shows the spatial distribution of MY ice according to a selection of state-of-the-art products for the month of March 2016, including sea ice age from the scatterometer record, the NSIDC record of Tschudi et al., 2016, and the SICCI record (Korosov et al., 2018), MY ice concentration from the OSI SAF-403 (Aaboe et al., 2016), the University of Bremen algorithm (Ye et al., 2016) and the SICCI algorithm (Korosov et al., 2018) and sea ice thickness from the AWI Cryosat-2 data set (Ricker et al., 2014). The monthly averaged MY ice concentration is calculated directly over the daily MY ice concentrations from the SICCI and Bremen products. Note that the OSI SAF-403 is not a sea ice concentration product but a FY or MY classification. In this case, a daily MY concentration is defined (100 % for the MY class, 50 % for the ambiguous class and 0 % for the FY and OW classes) and a monthly average MY concentration is calculated as above. The monthly averaged sea ice age is calculated over the weekly NSIDC grids (using weeks 9 to 12) and over the daily SICCI grids, and the SY ice class is defined for a monthly average sea ice age between 1.5 and 2.5 years.
Even though the general representation of MY ice is similar across all products, there are remarkable differences as well, mainly regarding the distribution of the old MY ice class north of the Canadian Arctic Archipelago (CAA, with large variations across the sea ice age records) and the presence of MY ice north of the Beaufort and Chukchi seas (with notable differences between the MY ice concentration records). The ice thickness product is revealing in that the thickest sea ice (more than 3 m thick and most likely associated with old MY ice) appears mostly confined to a thin strip along the CAA shore (see label A in Fig. 13g, in agreement with the scatterometer old MY ice class) and shows no traces of thick ice north of the Beaufort and Chukchi seas (see label B in Fig. 13g, in disagreement with some of the MY concentrations and the NSIDC sea ice age record). Further, we note a large extension of very thick ice north of Greenland (more than 3 m thick; see label C in Fig. 13g), which is labeled as SY ice in the scatterometer record (probably ridged SY ice converging into Fram Strait), which effectively appears as low concentration MY ice in the University of Bremen and SICCI algorithms, suggesting problems with the tie point definition in MY ice concentration algorithms (not in the OSI SAF-403 data set, because it is reports MY presence, not concentration). Finally, we find relative good agreement between the scatterometer SY ice class and the 2.0 m isoline from the ice thickness record, suggesting the utilization of the backscatter record as a reliable proxy for the estimation of thick sea ice thickness in the Arctic, in much the same way as Tschudi et al. (2016a) propose relating the NSIDC sea ice age to ice thickness. Another interesting feature refers to the thin tongue of older ice extending across the Arctic Basin towards the New Siberian Islands (see label D in Fig. 13g), which is seen by all products, even faintly in the AWI sea ice thickness, but falls below the SY threshold in the scatterometer-based MY ice classification. We cannot offer an explanation for this feature at the moment, other than acknowledging that efforts towards ensuring the consistency among MY ice products in the Arctic should warrant further research.
Different observation-based products for the representation of
multiyear ice in March 2016.
Noting the lack of extensive in situ validation sources for satellite-based data sets, one should rely on consistency among products as the best approach for checking retrievals. Yet, the differences just noted in this section make it clear that further effort towards the optimal integration of active and passive microwaves is necessary, not only for the classification of sea ice types, but for the determination of summer sea ice edge and concentrations.
We present the first intercalibrated long-term record of sea ice extents and
backscatter derived from satellite scatterometer missions (ERS, QuikSCAT and
ASCAT) extending from 1992 to the present date. The scatterometer record, the
continuation of which is guaranteed by the Metop ASCAT (B and C) and EPS-SG
series into the future, provides a valuable independent account of the state
of Arctic and Antarctic sea ice cover, with daily sea ice extent and
backscatter maps available at
The scatterometer sea ice extents show excellent agreement with passive
microwave fields in the fall and winter seasons, with differences within
0.25 million km
The scatterometer backscatter maps also provide enhanced means to differentiate between sea ice types. Our study of the evolution of the wintertime seasonal (FY) and perennial (MY) ice classes in the Arctic Basin from 1992 to the present day shows, in good agreement with the NSIDC sea ice age data set, a MY ice pack that begins to lose balance around 2005, after several consecutive years of decline, and finally collapses into a record loss in 2007. The scatterometer maps also reveal the emergence of a new mode in the backscatter histograms after the record sea ice loss in 2007, bearing striking resemblance in both temporal evolution and spatial distribution with the SY ice class of the NSIDC sea ice age data set. Monitoring the evolution of the complementary SY and old MY ice classes shows that the decline in the total MY ice extent observed in the Arctic was driven by the loss of older MY ice, while a more steady production of SY ice acted to stabilize the losses and contributed to later recovery events such as that observed in 2014.
We note that the differentiation between SY, deformed FY and lower concentration (but older) MY ice may be difficult using single-frequency backscatter. The simultaneous combination of C-band and Ku-band backscatter measurements allows further differentiation of sea ice types into deformed FY (high C-band, low Ku band), SY ice (low C-band, high Ku-band) and FY–MY mixtures (moderate Ku- and C-band responses), suggesting new approaches to their disambiguation. As such, the combination of coincident Ku-band and C-band missions (such as during the QuikSCAT overlap in 2000 and 2008 or the Oceansat-2 overlap from 2009 to 2014), along with the use of contemporaneous multi-frequency passive microwave data, affords renewed potential towards the generation of combined time series of areal estimates of different ice classes, as demonstrated in Remund et al. (2000), which can be used to guide reanalyses.
The scatterometer backscatter record is helpful for checking currently developing algorithms for MY ice concentration based on satellite passive microwave or blended data, given that none of the latter products use a separate tie point for SY ice, leaving the SY ice signature to be effectively interpreted as lower concentration MY ice. The comparison of a selection of state-of-the-art data sets for the representation of MY ice (including sea ice age, MY ice concentration and ice thickness estimates) in the Arctic reveals some notable inconsistencies, mainly regarding the ambiguity between compact SY and lower MY ice fractions, the spatial distribution of old MY ice in the sea ice age records and the apparently spurious presence of MY ice in the central Arctic in some of the MY concentration records derived from satellite passive microwaves. The relative good agreement between the backscatter-based sea ice (FY, SY and older MY) classes and the ice thickness record from Cryosat suggests its applicability as a reliable proxy in the historical reconstruction of sea ice thickness in the Arctic.
Noting the lack of extensive in situ validation sources for satellite-based data sets, one should rely on consistency among products as the best approach for checking retrievals. Yet, the differences among state-of-the-art products noted in this paper make it clear that further effort towards the optimal integration of active and passive microwaves is necessary, not only for the classification of sea ice types, but for the determination of summer sea ice edge and concentrations.
The scatterometer record of sea ice extents and backscatter
used to conduct this research is available at
MBR and AS developed the methodology for sea ice detection using scatterometer data. MBR, IO and AV processed the scatterometer sea ice record. MBR, IO and AS interpreted the results. MBR wrote the manuscript.
The authors declare that they have no conflict of interest.
The authors would like to acknowledge the financial support of the ESA
Scirocco project, the EUMETSAT OSI SAF for the provision of backscatter
data, our KNMI colleague Jeroen Verspeek for his insights, and the NASA and
NSIDC public data archives as essential towards the completion of this
activity. Processing of the AWI CryoSat-2 (PARAMETER) ice thickness is
funded by the German Ministry of Economics Affairs and Energy (grant
50EE1008) and data obtained from