Geophysical controls on C band polarimetric backscatter from melt pond covered Arctic first-year sea ice: Assessment using high-resolution scatterometry


  • R. K. Scharien,

    Corresponding author
    1. Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, Winnipeg, Manitoba, Canada
    • Corresponding author: R. K. Scharien, Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, 474 Wallace Bldg., Winnipeg, MT R3T 2N2, Canada. (

    Search for more papers by this author
  • J. J. Yackel,

    1. Cryosphere Climate Research Group, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • D. G. Barber,

    1. Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, Winnipeg, Manitoba, Canada
    Search for more papers by this author
  • M. Asplin,

    1. Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, Winnipeg, Manitoba, Canada
    Search for more papers by this author
  • M. Gupta,

    1. Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, Winnipeg, Manitoba, Canada
    Search for more papers by this author
  • D. Isleifson

    1. Centre for Earth Observation Science, Clayton H. Riddell Faculty of Environment, Earth and Resources, University of Manitoba, Winnipeg, Manitoba, Canada
    Search for more papers by this author


[1] Geophysical controls on C band polarimetric backscatter from the discrete surface cover types which comprise advanced melt first-year sea ice (FYI): snow covered ice, bare ice, and melt pond; are assessed using polarimetric radar scatterometry from test sites representing high Arctic and marginal ice zones in the Canadian Arctic. Surface characterization data is used to evaluate the interaction of polarized radiation with each feature, and dominant scattering mechanisms are assessed in a regional context. High-resolution time series (diurnal) scatterometry and coincident atmospheric boundary layer profile data are used to explain linkages between ice-atmosphere interactions and polarimetric backscatter in a marginal ice zone. The co-polarization ratio for FYI melt ponds is shown to be distinct from snow covered ice or bare ice during early and peak phases of advanced melt, making it a candidate parameter for the unambiguous detection of pond formation and the inversion of melt pond fraction. The ratio displays an increasing trend with radar incidence angle in a manner consistent with Bragg surface scattering theory, though it is not predictable by a Bragg model. Cross-polarization backscatter intensity shows potential for discriminating the onset and duration of freeze events in a marginal ice zone, due to dominant backscatter from the snow cover adjacent to melt ponds. Preliminary results here outline the potential of covariance matrix derived polarimetric measurements for the inversion of advanced melt sea ice geophysical parameters, and provide a basis for the investigation of distributed targets in late season spaceborne polarimetric SAR scenes.

1. Introduction

[2] The reduction in Arctic sea ice extent and thickness during the microwave satellite observational period, beginning in 1979, is well documented [Serreze et al., 2007]. The loss of ice during summer, including a 30-year minimum extent in 2007, is particularly important as it promotes the uptake of solar radiation in the upper ocean and warming of surface air temperatures [Perovich et al., 2007; Francis et al., 2009]. A recent depletion of thick, multiyear ice (MYI) and increased proportion of seasonal or first-year ice (FYI) points to the sensitivity of the Arctic ice cover to solar energy gained by the atmosphere-ice-ocean system during the summer [Maslanik et al., 2007; Kwok et al., 2009].

[3] The Arctic advanced melt season, distinguished by the formation and evolution of melt ponds on sea ice, is a critical component in the magnitude and timing of the seasonal ice decay cycle and, for FYI, break-up and the exposure of the ocean to solar insolation. Melt ponds reduce ice albedo, leading to accelerated ice decay and enhanced transmission of solar radiation into the ocean [Langleblen, 1969; Perovich et al., 2003; Inoue et al., 2008]. Melt pond fraction on FYI (20–60%) is normally greater than on MYI (≤30%) [Eicken et al., 2004], contributing to an earlier ice break up [Hanesiak et al., 2001], greater ocean heat uptake, and a delayed freezeup [Perovich et al., 2007]. Spatiotemporal variations in pond fraction and related atmosphere-ice-ocean interactions are poorly understood at the regional scale due primarily to a lack of observational studies. Remote sensing methods are essential, but currently limited to coarse resolutions and temporally sporadic estimates derived from optical sensors during cloud-free periods [Markus et al., 2003; Tschudi et al., 2008].

[4] The proliferation of spaceborne dual-polarization (dual-pol) and polarimetric microwave synthetic aperture radars (SARs), together with planned constellation missions aimed at data continuity and increasing the frequency of coverage, have motivated further research into utilizing SAR for sea ice geophysical parameter estimation and feature identification in support of regional scale studies. A polarimetric SAR measures the amplitude and phase of a backscattered signal in four linear horizontal (H) and vertical (V) transmit-receive polarization combinations (HH, VV, HV, and VH), whereas a dual-pol SAR is limited to two polarization combinations and their intensity ratios. Current dual-pol SAR missions capture data across much larger swath widths than polarimetric SARs (>100 km compared to <50 km), making them more amenable to regional scale studies. Dual-pol and polarimetric data have potential for alleviating ambiguities associated with single-channel SAR acquired during advanced melt, when dynamic wave roughness on melt pond surfaces and decay-driven snow and ice physical property changes limit image interpretability [Barber and Yackel, 1999; Yackel and Barber, 2000]. Furthermore, the resolution cell of a SAR is typically composed of a composite of sub-scale features during advanced melt – snow covered ice (hereafter referred to as snow), bare ice, and melt ponds. Though a significant body of research characterizing relationships between snow and sea ice physical properties and polarimetric microwave backscatter does exist [see Winebrenner, 1990; Drinkwater et al., 1992; Hallikainen and Winebrenner, 1992; Nghiem et al., 1995; Nghiem and Bertoia, 2001; Scheuchl et al., 2004, and references therein], in situ observations of advanced melt features are rare and need to be evaluated before SAR can be reliably exploited.

[5] This study examines the geophysical controls on in situ C band polarimetric backscatter from FYI advanced melt features by contrasting data from two field studies of landfast FYI in distinct geographical zones: high Arctic (HA) and marginal ice (MI). First, FYI advanced melt features: snow, bare ice, and melt ponds; are defined within a seasonal framework. Second, in situ C band polarimetric scatterometer signatures from these features are evaluated within the established framework, and within the context of the following research questions:

[6] 1. What are the dominant mechanisms which influence C band polarimetric backscatter from snow and bare FYI during advanced melt?

[7] 2. What are the relationships between wind speed (U), fetch length (F), melt pond water depth (D), and C band polarimetric backscatter from FYI melt ponds?

[8] Last, the potential of C band dual-pol and polarimetric SAR for improved Arctic first-year sea ice geophysical parameter estimation during advanced melt is discussed, followed by conclusions and recommendations for further investigation.

2. Seasonal Framework

[9] In concert with previous works [Perovich et al., 2002; Eicken et al., 2004], the FYI advanced melt season is subdivided into three seasonal phases which are separated by distinct geophysical transitions: here termed early, peak, and late. The early phase, forced by above freezing temperatures and a positive surface energy balance, is initiated by the appearance of melt ponds in shallow areas of the isothermal, and rapidly decaying, snow cover [Livingstone et al., 1987; Comiso and Kwok, 1996; Barber and Yackel, 1999]. The funicular snow regime, roughly 7% water by volume, has been reached such that water is no longer bonded between snow crystals and it freely drains to the snow-ice interface. Draining meltwater forms a slush layer at the snow-ice interface and re-crystallizes into superimposed ice [Onstott and Gogineni, 1985; Holt and Digby, 1985; Onstott, 1992]. Decaying snow metamorphoses into large (mm-cm scale), rounded, crystalline aggregates which may be bonded by percolating meltwater to form an ice lattice near the basal superimposed ice layer [Colbeck, 1973; Kim et al., 1984; Drinkwater, 1989]. For smooth FYI, melt ponds formed by snowmelt can flood greater than 50% of the ice surface [Derksen et al., 1997; Fetterer and Untersteiner, 1998; Yackel et al., 2000; Grenfell and Perovich, 2004]. As the snow cover ablates, the upper snow–ice volume is flushed by meltwater and desalinated, while adjacent melt ponds grow in size.

[10] The peak phase begins when the snow cover disappears and the concurrent peak in pond fraction is reached, after which it generally begins to decline. Superimposed ice is rapidly obliterated, leaving bare ice exposed at the surface. The desalinated upper 5–20 cm of the bare ice is composed of two sub-layers: a fragmented surface granular layer (SL), overlying a porous, drained layer (DL). The SL is a feature of the DL which forms during periods when direct solar radiation penetrates the ice surface, causing it to fragment along grain boundaries into a 0.5–3 cm thick layer granular, snow-like layer with grains measuring 1–6 mm [Drinkwater, 1989; Eicken et al., 1995; Perovich et al., 2001; Light et al., 2008; Scharien et al., 2010]. Variable diurnal ice ablation and lateral meltwater drainage rates cause fluctuations which may supersede the supra-diurnal decline in pond fraction which occurs during this phase.

[11] The late phase is initiated once the ice volume becomes permeable to fluid transport, as temperature-enlarged brine pockets become linearly interconnected and mesh with existing drainage channels; melt ponds quickly drain from the surface [Weeks and Ackley, 1982]. The percolation threshold for ice is crossed, promoting nutrient replenishment and enhanced organism growth within the ice [Golden et al., 1998], as well as air-sea gas exchange including CO2 [Gosink et al., 1976; Semiletov et al., 2004; Delille et al., 2007]. This stage is often referred to as the drainage stage due to the rapid, initial, loss of melt ponds, though seawater may refill the remaining negative freeboard regions of the rotten ice lattice prior to complete decay, break-up, or eventual freezeup and transition to second-year ice.

3. Methods

3.1. Data Collection

[12] Data from HA and MI zones were collected in the Canadian Archipelago and southeastern Beaufort Sea, respectively (Figure 1). HA data were collected within a 2 by 2 km site in Allen Bay, adjacent to Resolute, NU from days 26 June to 12 July 2006 as part of the POL-ICE research project [Geldsetzer et al., 2006]. MI data were collected adjacent to the research icebreaker CCGS Amundsen in Darnley Bay, NWT, between days 02 and 21 June 2008 as part of the Circumpolar Flaw Lead System Study (CFL) [Barber et al., 2010].

Figure 1.

(left) Map of the high Arctic (HA) study site in the Canadian Archipelago (2006), at 74°N, and the marginal ice (MI) study site in the southeastern Beaufort Sea (2008), at 69°N. Location of the land-fast ice margin in 2008 is depicted by the dashed yellow line. (right) Oblique photograph from a helicopter showing melt pond covered first-year sea ice on 12 June 2008. The ship is the CCGS Amundsen.

[13] During each project, a sled-mounted, tracker controlled, polarimetric C band scatterometer system was positioned at 2.5 m height over homogeneous cover types snow, bare ice, or melt pond and scans were acquired (Table 1). A scan, which took 13–15 min to complete, comprised complex backscatter (amplitudes and phases) recorded in HH, HV, VV, and VH polarization states across a 20–60° incidence angle (θ) range. A 2° range spacing and 30° azimuth width was used for each scan line, resulting in the collection of 11–33 independent samples (nind) for each line across the 20–60° θ range. Samples for each scan line were averaged in azimuth to reduce fading, and their complex scattering amplitudes were transformed to a covariance matrix. Computed backscatter errors ranged from 2.2–1.8 dB from 20–60° θ, as the signal-to-noise ratio improved with nind. Noise equivalent sigma zero (NESZ) values for VV, HH, and HV channels were measured at both sites and found to be, in the worst case, −28 dB, −24 dB, and −38 dB, respectively. Further details on the scatterometer system including calibration routine, signal processing, near-field correction, derivation of nind, and error determination is found in Geldsetzer et al. [2007]. In total, 48 individual, spatially unique, scatterometer scans of snow, melt ponds, and bare ice were collected at HA. Snow scans at HA were collected from 26–28 June, after which only bare ice and melt pond scans were collected for the remainder of the 16 day study period. Conversely, at MI a single time series of scans was collected for each surface feature. The snow series was collected for 17 h beginning 03 June 07:00 universal time (UT), the melt pond series for 17 h beginning 11 June 20:50 UT, and the bare ice series for 28 h beginning 12 June 15:10.

Table 1. Scatterometer Scan Data From 26 June to 12 July 2006 at the High Arctic (HA) Study Site, and From 03 to 13 June 2008 in 2008 at the Marginal Ice (MI) Study Sitea
  • a

    The abbreviation nscans denotes the number of scans collected; the symbol ‘θ’ represents incidence angle in degrees; Ta is air temperature.

Bare ice2220600.76.8872060−1.05.4
Melt pond2120600.54.08022601.22.8

[14] Variables and instrumentation used to characterize the atmospheric forcing of the ice cover, as well as the geophysical properties of the ice cover and melt ponds, are provided in Table 2. Meteorological variables at HA were logged at 15 min intervals at a tower mounted on the ice and adjacent to the sampling site, whereas at MI meteorological and flux data were logged at 1 min intervals on an automated weather station mounted to the front deck of the Amundsen vessel. At MI, atmospheric profiles of temperature and absolute humidity were used to characterize the thermodynamic structure of the atmospheric boundary layer (ABL) during each scatterometer time series. The profiles were logged at 10 min intervals, and were derived from microwave brightness temperatures using the manufacturer's neutral network (NN) retrievals. The NN was trained using historical radiosonde measurements from Inuvik NWT, and a radiative transfer model [Solheim et al., 1998; Gaffard et al., 2008]. A thorough analysis of the microwave radiometer profiler data collected during the CFL project is available in Raddatz et al. [2011]. Not shown in Table 2 are ancillary manual weather observations at HA and MI, and all-sky photographs of cloud conditions taken every 10 min at MI.

Table 2. Summary of Variables or Samples From High Arctic (HA) and Marginal Ice (MA) Sites Used to Aid in the Interpretation of C Band Scattering Data, the Instruments Used to Collect Each Variable or Sample, and the Accuracy of Each Variable Where Applicable
Variable (Symbol, Units) or SampleaInstrumentAccuracyProject
  • a

    Abbreviations: “temp.” for “temperature”; “n/a” for “not available”; and “abs.” for “absolute.”

1 m air temp. (Ta1, °C); relative humidity (RH1, %)Veriteq SP-20000.15°C; 2%HA
2 m wind speed (U2, m s−1); direction (dir, °)Onset S-WCA-M0030.5 m s−1; 5°HA
1 m wind speed (U1, m s−1); direction (dir, °)Onset S-WCA-M0030.5 m s−1; 5°HA
Dielectric Permittivity (ε′) at 20 MHzDenoth PSI-0432%HA
Ice core, 9 cm diameter sampleKovacs© Mark IIn/aHA, MI
Snow sample, 100 cm3 sampleSnow-Hydro Cuttern/aHA, MI
Snow density (ρs, kg m−3)Digital scale40 kg m−3HA, MI
Snow salinity (Ss ‰); Ice salinity (Si ‰)WTW Cond 330i0.05%HA, MI
Snow temp. (Ts); ice temp. (Ti)Cole-Parmer RTD probe0.02°CHA, MI
14 m air temp. (Ta14, °C); relative humidity (RH14, %)Vaisala HMP45C2120.1°C; 3%MI
14 m wind speed (U14, m s−1); direction (dir, °)R.M. Young 051030.3 m s−1; 3°MI
14 m air pressure (P, mbar)R.M. Young 61205V2%MI
Downwelling shortwave radiation (Kd, W m−2)Eppley PSP pyranometer5%MI
Downwelling longwave radiation (Ld, W m−2)Eppley PIR pyrgeometer10%MI
Dielectric permittivity (ε′); dielectric loss (ε″) at 50 MHzStevens Hydra Probe II0.6%; 0.7%MI
Root-mean square surface roughness (hrms, m)Riegl LD3100VHS-GF 905 nm Laser Profilern/aMI
Atmospheric air temp. (Tatm, °C); abs. humidity (dv, g m−3) to 1 km heightRadiometrics TP/WVP 3000 Microwave Profiler1.8°C; 0.3 g m−3MI

[15] Snow properties were derived at 2 cm vertical intervals in a snow pit made during each snow scan at HA, and at the start/stop points of the snow scan series at MI. Following Denoth [1989], measurements of ε′ and ρs were used to estimate volumetric moisture content (mv). Estimates of mv from the bare ice surface granular layer were also made during the bare ice scatterometer series at MI by vertically inserting the probe into the ice surface for measurement of ε′, and by assuming an ice density of 450 kg m−3 for the mv inversion [Scharien et al., 2010]. Ice core properties were measured at 10–15 cm vertical intervals from cores extracted throughout each study, and ice thickness, hi was recorded during the extraction of each core. Drainage of brine from sampled cores and bias toward lower salinities are expected during advanced melt as a result of the sampling process [Eicken et al., 1995; Timco and Frederking, 1996].

[16] For melt pond scans at HA, coincident 1 m wind speed (U1) and direction were logged at 1 s intervals using a sensor mounted adjacent to, but not interfering with, the scatterometer. Pond depth (D) was determined as the sample mean from a transect taken along the pond major axis at 0.5 m spacing, and the upwind fetch distance (F) was taken at the center of the radar footprint. At MI, hrms was measured by mounting the laser profiler at 0.5 m height on a tetrapod in the pond and adjacent to the radar footprint. Vertical range measurements to the air-water interface were sampled at 5 Hz and logged at 2 s intervals. Though the laser was mounted in the pond, wave diffraction from tetrapod legs was observed to be negligible, and reflected waves rapidly dissipated into the predominant wave pattern. Ancillary estimates of pond fraction were made from transect surveys (length = 200 m; n = 8), bisecting the predominant major axes of melt ponds (HA) and using helicopter-based aerial photography (MI).

3.2. Backscatter Parameters

[17] Co-polarized (co-pol) and cross-polarized (cross-pol) backscatter coefficients σ°vv, σ°hh, and σ°hv were derived from the complex covariance matrix of each scan line and specified in decibel (dB) format. Reciprocity of cross-pol channels was enforced and σ°vh was discarded. Analyzed polarimetric parameters were: the span or total power, co-pol ratio (rco), cross-pol ratio (rcr), co-pol phase difference (ϕvv-hh), and the co-pol correlation coefficient (ρco). The ϕvv-hh and ρco observations from melt ponds were not retained, as the scatterometer system did not compensate for the time delay between VV and HH transmit-receive measurements induced by the motion of the surface in the radial direction. The reader is referred to Ulaby and Elachi [1990] and Geldsetzer et al. [2007] for details on the derivation of these parameters; for a better understanding of their application in sea ice studies, a brief review is provided.

[18] The behavior of C band σ°vv and σ°hh from sea ice are comparatively well understood for all seasons including advanced melt, when dynamic wind-wave roughness on pond surfaces is known to cause considerable variation [Carsey, 1992]. The σ°hv is known to increase when enhanced depolarization occurs, which for sea ice is caused by either (or both) volume scattering, or multiple surface scattering from roughness features which are large relative to the incidence wavelength (5.6 cm at C band) [Drinkwater et al., 1991; Nghiem et al., 1995]. This behavior has been exploited for ice edge (ice/water) discrimination due to much greater depolarization caused by ice relative to water, as well as ice type identification based on internal structure and surface topography [Scheuchl et al., 2004; Livingstone, 1994]. Scheuchl et al. suggest the potential of σ°hv for discriminating melt ponds on sea ice due to negligible depolarization by ponds relative to ice, though this has not been rigorously examined to date. The rcr provides further indication of depolarization or multiple scattering effects, as σ°hv shows a greater increase relative to σ°hh when these events occur [Fung, 1994]. The rco has been the studied using theoretical and observational approaches by several authors. Hajnsek et al. [2003] showed that, for surface scattering from very smooth surfaces relative to the incident wavelength, small perturbation model (SPM) or Bragg scattering theory applies whereby rco is independent of surface roughness and increases with θ at a rate dependent only on the complex permittivity (ε*) of the target. However, most natural surfaces are likely rougher than the validity range of the SPM allows at C-band, and rco is better described by the Integral Equation Model (IEM) theory which shows that rco increases with greater θ and ε*, but also decreases with increasing surface roughness [Drinkwater et al., 1991; Fung, 1994]. In either case, the rco tends toward zero when volume scattering occurs. C-band rco has been shown to be effective for separating FYI and MYI from young ice, leads, and open water [Thomsen et al., 1998; Scheuchl et al., 2004]; separating FYI and MYI from open water [Nghiem and Bertoia, 2001; Geldsetzer and Yackel, 2009]; and proxy ice thickness estimates [Zabel et al., 1996; Nakamura et al., 2009]. The ρco is a measure of the proportion of backscatter which is polarized; when σ°vv and σ°hh are perfectly correlated, ρco is unity and σ°hv is null. At C band the ρco from predominantly surface scattering FYI decreases with increasing θ at a rate determined by salinity, and it is generally lower for MYI than FYI due to depolarization [Drinkwater et al., 1992; Geldsetzer et al., 2007]. Nghiem et al. [1995] found low ρco due to volume scattering from non-spherical scatterers i.e., those with preferential orientation, while Thomsen et al. [1998] attributed low ρco to a mixture of ice types or scattering mechanisms within a resolution cell. Ship-based ρco observations show a relationship to new ice thickness [Isleifson et al., 2010]. The ϕvv-hh, a measure in degrees of the phase angle difference between co-pol channels, is negative for FYI and closer to null for MYI [Drinkwater et al., 1992; Drinkwater et al., 1992; Geldsetzer et al., 2007].

4. Results

4.1. The High Arctic (HA) Site

4.1.1. Weather and Ice Conditions

[19] A timeline of meteorological conditions from HA, including a discretization of advanced melt phases (see Section 2), is shown in Figure 2a. Generally, the advanced melt period was characteristic of a high Arctic environment, where the spatiotemporal evolution of the ice cover is strongly coupled to the availability of solar energy and characterized by uniform phase transitions. The early phase at HA (26–28 June), during which pond fraction remained above 50%, was dominated by persistent low-level stratus cloud cover and high relative humidity. Air temperature at 1 m height remained close to the melting point, as energy at the surface was used to melt the snow cover. The peak phase (28 June–11 July) was distinguished by brief, cloud-free, periods with warmer temperatures and enhanced horizontal mixing in the boundary layer. Peak phase pond fraction was 20–40%, with maxima following warming periods on 05 and 08 July. The late phase was made apparent by a visibly reduced pond fraction (<20%) and remnant low salinity bare ice patches. Ice salinity and thickness data taken over the duration of HA are shown in Figure 2b. The hi ranged from 140–164 cm, bulk Si from 2.1–3.2‰, and the upper 10 cm remained saline (0.3–0.6‰) throughout the sampling period. Bare ice cores were composed of a 7–14 cm thick DL overlying columnar ice to maximum hi, with the DL surface occasionally disaggregated into a SL of 0.5–1 cm thickness. An additional late phase ice core, taken from ice flushed by a fully drained melt pond, had lower hi (94 cm) and bulk Si (0.7‰). Data from this core is exemplary; no coincident scatterometer samples were taken this late into the season. No superimposed ice layers were observed within extracted ice cores during any phase.

Figure 2.

(a) Meteorological conditions at HA from 26 June to 12 July 2006. Horizontal bars in the Ta1 series denote clear sky events derived from cloud observations. (b) Salinity profiles of first-year sea ice from cores extracted during HA, with the time axis at the top corresponding to the time series ice core thickness measurements shown in red. Salinity data from remnant ice flushed by a drained melt pond on 11 July is shown in blue; the thickness measurement on 11 July corresponds to the non-flushed ice core. Snow on first-year sea ice stratigraphy, density, and wetness from data collected on (c) 26 June and (d) 28 June.

[20] Stratigraphic sections and physical properties of two unique snow conditions from HA are shown in Figures 2c and 2d. Significant melt-freeze metamorphism of snow grains was evident for each condition, with rounded grains having formed tightly packed clusters of 0.5–1 cm diameter. The snow cover was isothermal and devoid of brine, with grain shape and mv similar to previous observations of snow on FYI during advanced melt [Drinkwater, 1989; Barber et al., 2001]. A rough (cm-scale) superimposed ice layer at the snow-ice interface, formed from the recrystallization of meltwater at the base of the snow, was also observed for each condition. Between samples, the snow changed from a consolidated granular lattice within an air/liquid water background, to grains freely interspersed within the air/liquid water background. Snow thickness reduced by half and mv increased from 4–6%.

4.1.2. Dual-Pol Backscatter at HA

[21] Exemplative σ°vv, σ°hh, σ°hv, rco, and rcr for snow, bare ice, and melt pond cover types at HA are shown in Figure 3. Snow signatures a and b in Figure 3 (left) are derived from individual scans corresponding to snow stratigraphies in Figures 2c and 2d, respectively, whereas bare ice and melt pond signatures are averages of all scans of those features (see Table 1), with error bars denoting sample standard deviations. Following the equations of Drinkwater [1989], the estimated vertical penetration depth of C band energy in the snow was reduced from 7 to 4 cm as mv increased by 2%. A loss in snow backscatter by 3–7 dB also occurred, and was strongest across the near to middle range θ (≤40°). Greater backscatter losses are expected due to attenuation from within the wetter snow, as found by Drinkwater [1989] over FYI in the Labrador Sea during spring. Despite the occurrence of such a large change over a short period, snow rco remained close to zero. Comparing snow to bare ice in Figure 3, σ°vv and σ°hh from bare ice is more closely matched to snow case a and is much greater than snow case b. Regardless of snow condition a or b, σ°hv and rcr from bare ice averaged 3 dB greater across all θ, an indication of enhanced volume scattering from the desalinated upper bare ice layer. A small positive trend in bare ice rco with θ was observed, though the maximum of 1.5 dB is generally low and falls within previous observations of both FYI and MYI from other seasons [Scheuchl et al., 2004; Geldsetzer et al., 2007; Winebrenner, 1990].

Figure 3.

Exemplative co-pol and cross-pol backscatter, as well as co-pol and cross-pol ratios, from snow, bare ice, and melt pond cover types at HA. Snow signatures a and b correspond to the stratigraphies shown in Figures 2c (blue lines) and 2d (red lines). Fit lines are quadratic.

[22] The range of U1, F, and D conditions coincident to melt pond signatures in Figure 3 are provided in Table 3. The minimum U1 was 1.9 m s−1, below which pond scans were discarded due to low SNR. This cut-off point is consistent with Donelan and Pierson [1987], who found that at 0°C and U10 ≤ 2.2 m s−1, viscous dissipative forces are greater than the force of the wind required to generate water waves large enough to backscatter C band energy. Collectively, the findings here and by Donelan and Pierson indicate a wind speed threshold of approximately 2 m s−1, below which C band backscatter from FYI melt ponds is predominantly specular. Above this threshold, σ°vv and σ°hh in Figure 3 are consistent with open water observations which are subject to the dynamic influence of wind-wave surface roughness. Pond rco is unique, and increases with θ in a manner consistent with the aforementioned Bragg theory of independence from surface roughness [Hajnsek et al., 2003]. The effect of U1, F, and D on backscatter parameters is examined in the following section, while the comparison of pond rco to Bragg theory is provided in section 5 and includes data from the MI study site.

Table 3. Descriptive Statistics of Surface Variables Associated With Melt Pond Samples From the High Arctic (HA) Study Sitea
  • a

    The symbol U1 is 1 m wind speed, F is fetch, and D is pond depth; abbreviations Min., Max., and SD denote minimum, maximum, and standard deviation, respectively.

U1, m s−11.911.38.22.6
F, m12.044.523.010.1
D, cm3.

4.1.3. Melt Pond Statistical Analysis

[23] Dual-pol parameters σ°vv, σ°hh, σ°hv, rco, and rcr from spatially distributed melt pond scans at HA were used to evaluate research question 2 (see Section 1). Linear regression relationships between backscatter parameters at θ of 20°, 40°, and 56° – representing near, middle, and far range θ – and explanatory surface variables U1, F, and D were tested. Two further constraints were imposed on the scatterometer data set, in addition to the SNR cut-off mentioned in the previous section. Scans not taken with the scatterometer oriented in the upwind direction were removed, in order to eliminate the azimuthal effect on backscatter caused by varying the orientation of the radar relative to surface wave patterns [Ulaby et al., 1986]; and individual parameters observed to fall within the NESZ of a given channel, which occurred at far range only, were removed. These constraints reduced number of scans to 14 in the near and middle range θ, without modifying the range of explanatory surface variables in Table 3. At the far range θ only σ°vv was retained for regression testing; the NESZ constraint severely limited all other parameters to U1 ≥ 6.4 m s−1 which precluded their testing using this method. Due to small sample sizes, regression testing here was used to provide a means by which to evaluate the functional relationships between variables, and not to establish models for future prediction. Changes in the coefficient of determination (Δr2) were used to assess the relative contribution of each significant explanatory variable in a given regression model.

[24] Significant regression models (p-value < 0.01) reveal a strong dependency of backscatter on D, in addition to U1, and in only one case was F determined to be an explanatory variable (Table 4). At the near range θ, backscatter intensity from roughened water surfaces is known to be highly variable and predominantly sensitive to the wind-induced roughness state. As such it is perhaps not surprising that vertically polarized backscatter was dependent on all surface parameters at the near range θ. However, the contribution of D in explaining variation in significant parameters, particularly at middle range θ, is notable. Combining these results with qualitative observations indicating that melt ponds were generally smoother near shallow edges, raises the possibility that bottom friction may be a damper of melt pond waves [Lin, 2008]. This effect is less apparent at the far range θ where σ°vv, the only tested parameter due to system noise constraints, was weakly dependent on D. The rco and rcr could not be explained by the input surface variables. Sensitivity analyses were conducted on multivariate models in order to evaluate the effect of U1 on backscatter at the minimum (3 cm) and maximum (11 cm) observed D (Figure 4). Included in Figure 4 are approximate NESZ reference lines for the RADARSAT-2 satellite operating in ScanSAR Wide and Fine Quad-Pol modes, contrasting modes in terms of resolution, 50 m compared to 5 m, and swath width, 500 km compared to 25 km (MacDonald, Dettwiler and Associates Ltd., RADARSAT-2 Product Description RN-SP-52-1238, 2011). These lines provide a means by which to evaluate the likelihood of a parameter being detectable within a scene acquired by a conventional spaceborne SAR. The contribution of D as a function of θ is exemplified in Figure 4, with the variations in backscatter due to D the strongest at steeper θ. At the middle range θ, changes in σ°hv would likely be undetectable using a ScanSAR Wide mode. The total power (not shown) follows closely to σ°vv, an indication that the total backscatter from melt ponds comes predominantly from vertically polarized waves.

Table 4. Significant Results From Stepwise Multiple Linear Regression Tests Between Independent Surface Variables and Dependent C Band Scattering Parametersa
Scattering ParameternVar.r2p-Valuer2(1)r2(2)r2(3)
  • a

    Significance level α = 0.01. Explanatory variables are Var, the regression coefficient of determination is r2, the proportion of variability explained by each Var is Δr2 given in order that they appear, near-range is NR, mid-range is MR, and far-range is FR.

σ°vv14D, U1, F0.8170.0010.5160.1860.115
σ°hv14D, U10.8020.0000.5240.279
σ°vv14D, U10.7460.0010.4570.289
σ°hh14D, U10.8190.0000.4540.365
σ°hv14D, U10.7610.0000.4730.288
σ°vv14U1, D0.8490.0000.7810.068
Figure 4.

Modeled wind speed-to-backscatter relationships at low (3 cm) and high (11 cm) melt pond depths for co-dependent parameters at (a) near, (b) middle (mid), and (c) far range θ, derived from spatially distributed melt pond scans at HA. Grey reference lines indicate the nominal noise equivalent sigma zero for the RADARSAT-2 satellite operating in ScanSAR Wide (dashed line) and Fine Quad-Pol (solid line) modes.

4.2. The Marginal Ice (MI) Site

4.2.1. Weather and Ice Conditions

[25] A general description of seasonal weather and ice conditions is given, followed in the next paragraph by a description of geophysical properties of each surface feature. Detailed meteorological and atmospheric forcing data associated with the time series scatterometer scans of each feature are given in Section 4.2.3. In contrast to the HA site, the advanced melt period at MI was characteristic of a marginal ice zone [Anderson and Drobot, 2002]. Upon arrival at MI, the snow and ice cover showed evidence of amorphous phase transitions caused by variable weather patterns and freeze-melt cycling. The snow cover was desalinated and contained superimposed fresh ice layers. Freeze-melt cycling continued during the study, and melt pond surfaces were frequently covered by a thin skim of ice. Observed MI ice core data are shown in Figure 5a. The seasonal hi ranged from 125–170 cm, with lower bulk salinities observed at MI (1.9–2.1‰) compared to HA (2.1–3.2‰). Unlike at HA, where the upper 10 cm of the ice remained saline throughout the sampling period, the top 10 cm at MI was reduced to 0‰ by 10 June (the bare ice time series was collected on 12 June). Once the snow ablated, bare ice surfaces were either consolidated or composed of a disaggregated SL (0.5–2.5 cm), and below the surface were DL (0–12 cm) and columnar ice (to maximum hi).

Figure 5.

(a) Salinity profiles of first-year sea ice from cores extracted during MI, with the time axis at the top corresponding to the time series ice core thickness measurements shown in red. Snow on first-year sea ice stratigraphy, density, and wetness from MI on 03 June 2008, at (b) commencement and (c) termination of the snow scan time series. Density and wetness measurements were unattainable for Figure 5c due to consolidation within the snow.

[26] Snow conditions corresponding to the start and stop times of the snow scatterometer time series on 03 June are shown in Figures 5b and 5c. Initially, the snow cover was cold (Ts = −0.4°C) and made up of two layers of superimposed fresh ice separated by large, well metamorphosed, 5–10 mm diameter snow grain clusters freely dispersed within an air-only background. During the series a freeze-melt transition occurred, during which the snow became wet and isothermal, and grains were bonded by meltwater into a fragile but consolidated lattice. The fragility of the lattice prohibited density sampling and the subsequent inversion of mv. However, a measured dielectric constant of ε* = 2.29 + 0.2i at 50 MHz, which is much greater than that expected for a low density ice-air mixture, indicated the presence of liquid water in the snow [Tiuri et al., 1984; Ulaby et al., 1986]. A rough (cm-scale) layer of superimposed ice was observed at the snow-ice interface. Ice core data coincident to the snow series on 02 June revealed desalinated ice to 5 cm depth, with salinity ranging from 0.5–1.0‰ across the 5–10 cm depth interval. During the bare ice time series on 12 June 2008, the DL was 12 cm thick and salinity was first detected at 15 cm depth. The mv of the uppermost bare ice layer ranged from 1.4–3.9%, and the thickness of the SL varied from 0.5–2 cm.

4.2.2. Dual-Pol Backscatter at MI

[27] Exemplative σ°vv, σ°hh, σ°hv, rco, and rcr for snow, bare ice, and melt pond cover types at MI are shown in Figure 6. Snow signatures a and b in Figure 6 correspond to the individual snow stratigraphies in Figures 5b and 5c. Bare ice and melt pond signatures a and b in Figure 6 each correspond to averages of all scans (see Table 1) but are separated by melt-freeze conditions on the basis of Ta14 (bare ice) or liquid and ice lens covered states (melt pond). The snow at MI was relatively dry and stratified compared to HA, resulting in co-pol and cross-pol backscatter several dB higher. For snow devoid of brine or moisture, C band penetration depth is much greater than hs, such that strong surface scattering from the rough snow-ice interface is expected to have contributed to observed backscatter. A volume scattering contribution from the upper portion of the ice is also expected, given that the salinity of the upper 10 cm was low (0.3‰) thereby indicating drainage and replacement by air inclusions. With melting conditions, co-pol and cross-pol backscatter across all measured θ reduced by 3–6 dB, which points to strong attenuation of the primary ice scattering signature by water inclusions within the snow. Variability in rco and rcr across the minimum to maximum θ range corresponds to the presence of inhomogeneities within the scatterometer footprint at a given time. MI bare ice signatures are notably consistent with those from HA (see Figure 3), though during freezing conditions the ice was a more efficient scatterer which raised co-pol and cross-pol backscatter each by as much as 5 dB. Bare ice σ°hv and rcr remained strong during either melt or freeze conditions, however in contrast to HA, the magnitudes of these parameters at MI are notably less than that of snow. Melt pond signatures at MI are similar to HA when the pond is in liquid form, with the exception that σ°hh at MI is stronger over the middle to far θ range, and it follows that rco is less over the equivalent θ range. The appearance of an ice skim on the pond surface was met with a concomitant reduction in co-pol backscatter by several dB across the near to middle range θ. The rco was reduced by 2–5 dB across the middle to far range θ as the ratio is no longer sensitive to the high dielectric liquid pond surface, but rather the low dielectric surface ice skim. Caution is ascribed to the interpretation of σ°hv and rcr signatures from melt ponds due to low backscatter magnitudes relative to the NESZ of the scatterometer. As mentioned, a comparison of melt ponds to Bragg scattering theory is further addressed in section 5.

Figure 6.

Exemplative co-pol and cross-pol backscatter, as well as co-pol and cross-pol ratios, from snow covered FYI, bare ice, and melt pond cover types at MI. Snow signatures a and b correspond to the stratigraphies shown in Figures 5b (blue lines) and 5c (red lines). Bare ice and melt pond signatures a and b are delineated for bare ice in freeze (blue) and melt (red) states, and ponds in liquid (red lines) and ice lens covered (blue lines) states, respectively. Fit lines are quadratic.

4.2.3. Time Series Backscatter at MI

[28] Time series plots of σ°hv, rco, ρco and coincident data characterizing atmospheric forcing at the surface are shown for snow (left) and bare ice (right) at MI in Figure 7. The snow series followed a 17 h, local night-to-day, clear-sky, freeze-melt transition, with peak Kd (737 W m−2) observed at solar noon and peak Ld (274 W m−2) observed 3 h later. ABL profiles indicated frontal warm air advection aloft and increasing absolute humidity near the surface over the duration of the series. Adjacent melt ponds were capped by a 3 cm thick ice lens at all times, and local pond fraction was 40%. Conversely, the bare ice series followed a 28 h, clear-sky to cloud cover, melt-freeze transition, caused by a ridge of high pressure which advected cold air, low-level cloud (100 m), and fog over the site. Radiative fluxes followed a local clear-sky diurnal pattern until the cloud cover caused an increase in Ld by ∼50 W m−2. Winds remained light (2.6 ± 1.3 m s−1) and Ta14 > −1°C during the freeze event which, combined with stable absolute humidity values, suggests a slow rate of water vapor deposition on the ice surface. Looking at the backscatter parameters, variability across the observed θ (e.g., see rco) is indication of the heterogeneity of scattering mechanisms within the scatterometer footprint. These features, which may be isolated areas concentrated with meltwater during warming periods, for example, could not be individually accounted for. Instead, general trends associated with the changing conditions for each case are examined. Snow σ°hv dramatically decreased during melt, with the greatest decrease (15 dB) at far-range θ (54°). The rco shifted from predominantly ≤0 to ≥0 with melt, and the magnitude of change was not dependent on θ. Snow ϕvv-hh and ρco were most affected by melt at θ ≥ 40° where, on average, ρco increased by 0.4 and ϕvv-hh decreased by 48°. Turning to bare ice, σ°hv decreased during melt, though the greatest decrease (9 dB at θ ≥ 46°) was not as large as for the snow cover. Like snow, bare ice rco was also ≥0 with melt, and tended to 0 with freeze. The ρco was predominantly higher (>0.8) than for snow and relatively unchanged by the melt-freeze transition. Similarly, bare ice ϕvv-hh was typically greater than snow and relatively unaffected by the melt-freeze transition.

Figure 7.

Time series atmospheric conditions, and incidence angle dependencies of coincident backscatter and polarimetric parameters from the (a–j) snow covered ice and (k–t) bare ice time series at MI. A gap in parameters from the bare ice time series was caused by a scatterometer power failure.

[29] Time series plots of σ°hv, rco, and coincident data characterizing atmospheric forcing at the surface are shown for a melt pond in Figure 8, and data describing the melt pond are provided in Table 5. The series occurred during a 17 h, local afternoon to morning diurnal cycle, during which a very thin ice lens formed on the pond surface. The lens, which was 1–2 cm thick at the end of the series, formed with the Kd minimum (15 W m−2) and, though the Ta measured aloft was >0°C, freezing occurred at the surface. The pond fraction was 30%, with ponds wave roughened by strong winds (5.2–9.6 m s−1) prior to freezing over. Looking at the melt pond backscatter parameters, the dichotomy of scattering regimes is immediately apparent. The σ°hv was very low, as expected, though it was much higher over the near to middle θ range when the pond was wind-wave roughened. The rco showed a strong dependency on θ when in liquid phase, and a much reduced dependency when covered by an ice lens. During lens formation, the ratio initially became negative (−3 ≥ rco ≤ 0) then varied for the remainder of the series due to variable ice lens and open water areas forming within the radar footprint.

Figure 8.

(a–f) Time series atmospheric conditions, and incidence angle dependencies of coincident (g) cross-pol backscatter and (h) co-pol ratio from the melt pond at MI.

Table 5. Descriptive Statistics of Surface Variables Associated With the Melt Pond Sampled at the Marginal Ice (MI) Study Sitea
  • a

    The symbol U1 is 1 m wind speed, F is fetch, D is pond depth, and hrms is root mean square height; abbreviations Min., Max., and SD denote minimum, maximum, and standard deviation, respectively.

U14, m s−
F, m16.016.0
D, cm8.
hrms, cm1.

5. Discussion

[30] The previous section summarized the geophysical controls on C band backscatter from advanced melt FYI features in high Arctic and marginal ice zones. The utilization of satellite C band SAR for inverting geophysical information must account for the relative fraction of these features within a resolution cell, as they are often sub-scale. In the context of regional sea ice climatological and heat budget modeling studies, it is the timing of pond formation and the evolution of pond fraction which is of primary interest. Salient results from the previous section are here discussed within this context.

[31] Further studies aimed at understanding the spatiotemporal evolution of sea ice melt pond fraction at the regional scale are required in order to improve physically based melt pond albedo parameterizations when simulating the annual cycle of sea ice [Curry et al., 2001; Pedersen et al., 2009]. C band rco was demonstrated to have potential for unambiguous detection of FYI melt pond formation and pond fraction as it is unique from the background snow or bare ice rco as long as ponds are in liquid state. The rco from ponds at HA and MI were compared to the Bragg model and a model derived from open ocean observations (Figure 9). Included in Figure 9 are: (a) empirical model fits derived from measured melt pond rco from HA and MI; (b) a Bragg model rco estimate based on a melt pond ε* of 67.03 + 35.96i, i.e., determined using the frequency-dependent Deybe equation for pure water at 0.5°C temperature [Barber and Yackel, 1999] with additional bounds based on a ±50% variation in the estimated melt pond ε* included; and (c) rco estimates from an empirical model derived from C band backscatter observations of the ocean in RADARSAT-2 fine quad mode data [Vachon and Wolfe, 2011]. Observed melt pond rco deviates from the Bragg model estimate, through a Bragg-like dependency on θ occurs which is in closer agreement with the empirical ocean model. The Bragg model, considered valid for surfaces which are slightly rough relative to the incident wavelength, has a validity region defined according to the radar wave number k and surface roughness hrms, where khrms < 0.3 [Fung, 1994]. Based on the hrms measured when the pond at MI was in liquid form (see Table 5), the Bragg limit for C band was exceeded by the roughness of pond waves. Though surface roughness data were not collected at HA, the difference in measured rco at MI relative to HA beyond θ > 45° may be attributed to dissimilar sampling strategies employed at each site. Whereas several spatially distributed melt ponds were sampled at HA, including ponds with much lower surface roughness conditions and hence greater average rco, a generally deeper and rougher pond was sampled at MI. From the statistical results in Section 4.1.3, it is apparent that D also plays an important role in modifying the surface roughness of ponds. It is thus conceivable that the aforementioned effect of bottom friction, regarded as important for influencing wave amplitudes in shallow melt ponds at HA (as low as 3 cm), was not a factor for the single, 8–16 cm (see Table 5), pond sampled at MI.

Figure 9.

C band co-pol backscatter ratio (rco) from melt ponds as estimated using the Bragg scattering model compared to empirically derived models derived from melt pond scans at HA and MI (models are specified in the figure). Also shown is the estimated co-pol ratio of open-ocean determined from the empirical model of Vachon and Wolfe [2011].

[32] On the basis of the results in Figure 9, it is suggested that an rco threshold can be utilized to unambiguously detect the formation of melt ponds on FYI from a C band SAR operating at middle to far range θ. Utilizing such a threshold would necessitate sufficient melt pond coverage within the SAR resolution cell to cause a detectable perturbation in rco beyond that expected for snow covered or bare FYI, while also accounting for reductions in rco associated with periods of high pond surface roughness. At a fixed θ, the magnitude of increase in rco from the snow or bare ice reference would be based on the relative fraction of ponds. However, the presence of a surface ice lens on melt ponds, as observed at MI, would yield an rco that is unlikely to be distinguishable from the FYI background. Finally, the detection of pond formation or melt pond fraction on FYI must consider the NESZ of the SAR at far range θ. For example, C band co-pol signatures of melt ponds are above the NESZ of the RADARSAT-2 SAR operating in either ScanSAR or Fine Quad-Pol modes, provided U1 > 1.8 m s−1 and the θ is less than approximately 54°.

[33] Results from Section 4 also point to the potential of C band SAR for discriminating climatologically significant events in a marginal ice zone on the basis of dominant backscatter variations caused by the snow covered FYI adjacent to melt ponds. A freezing event in the desalinated and inhomogeneous snow cover observed in a marginal ice zone produced the strongest co-pol and cross-pol backscatter observed in this study. Looking specifically at σ°hv, it was significantly greater than an adjacent melt pond (regardless of pond extremes) across middle to far range θ during a freeze event (Figure 10). With a return to melting conditions, σ°hv from each type are similar. While the volume scattering contribution to σ°hv of desalinated sea ice has been well documented [Carsey, 1992], the direct comparison to melt ponds during advanced melt here outline the potential utility of a σ°hv threshold for detecting freeze events and adjusting melt season length (MSL) calculations for a marginal ice zone which undergoes variable melt-freeze cycles. Despite the significant r2 of 0.80 and 0.76 describing the variation in σ°hv from roughened melt ponds explained by D and U1 at near to middle range θ (see Table 4), the observed range of σ°hv would fall very near or within the NESZ of the RADARSAT-2 SAR operating in ScanSAR and Fine Quad-Pol modes, such that significant increases above the NESZ should closely match the occurrence of freeze event anomalies in a SAR time series. It should be noted that the dynamic influence of melt pond surface roughness on σ°hh limits the utility of rcr for detecting these events. Furthermore, co-pol backscatter maxima during advanced melt have been previously utilized in C band SAR and Ku band scatterometer algorithms to denote pond formation on FYI [Howell et al., 2006; Yackel et al., 2007]. These algorithms were primarily developed on the basis of data characteristic of the high Arctic zone where the expected co-pol backscatter maxima during advanced melt time series is attributed to wind-wave roughened ponds and therefore indicates their presence. Examples here demonstrate that co-pol maxima may indicate atmospherically forced freeze events, and that the rco is better suited for detection of melt ponds.

Figure 10.

(left) Cross-pol backscatter of snow covered first-year sea ice during a freezing event, compared to a wind-roughened melt pond (U14 = 8.8 ± 0.2 m s−1) and a melt pond covered by a 1–2 cm skim of ice. (right) Co-pol backscatter from the same cover types. Fit lines are quadratic.

[34] The rco and σ°hv are attainable using conventional dual-pol SARs, though the RADARSAT-2 satellite is restricted to dual cross-pol acquisitions (HV and HH, or VH and VV transmit-receive channels) when operating in ScanSAR Wide Swath mode. However, planned SAR missions will be capable of providing dual co-pol data (VV and HH transmit-receive channels) in swath widths comparable to that of RADARSAT-2 operating in ScanSAR mode. In polarimetric mode, narrow swath widths of 25 km in conventional SARs impose coverage limitations, however the utility of this mode for local scale sea ice studies during advanced melt warrants further investigation. The ρco, which increases at far range θ when snow covered ice changes from freezing to melting states in a marginal ice zone (see Figure 7), points to its potential utility as an alternate or complimentary parameter for mapping freeze-melt cycling related processes or the passing of cold fronts over sea ice during advanced melt. A full assessment of the utility of these, and the other potentially exploitable parameters described in this paper, necessitates full evaluation of their manifestation at the SAR image scale. Such an evaluation would require complementary information on the relative fractions of the surface features to coincident to SAR acquisitions.

6. Conclusions

[35] The first research question asked: What are the dominant mechanisms which influence C band polarimetric backscatter from snow and bare FYI during advanced melt? As is well understood on the basis of previous studies characterizing the early melt period (prior to pond formation), the increased moisture content of melting snow is related to the attenuation of C band polarimetric backscatter such that the contribution to backscatter from the underlying ice is effectively masked. Results from the high Arctic site confirm this for the FYI advanced melt season with snow volumetric moisture content of 4–6%. However, in a marginal ice zone the onset of a freeze event is linked to strong rough surface scattering from the snow-ice interface, volume scattering from the desalinated upper ice cover, and expected contributions from within the stratified and coarse-grained snow cover. The largest observed C band co-pol and cross-pol backscatter intensities in this study were observed for cold, snow covered FYI during a freeze event. For bare ice, surface mv fluctuations and freeze-melt cycling (marginal ice zone) cause variability in the volume scattering contribution from the desalinated upper ice layer. However, the effect of these variations on C band co-pol and cross-pol backscatter intensities is negligible when compared to that of snow covered ice, particularly in the marginal ice zone. The second research question asked: What are the relationships between wind speed (U), fetch length (F), melt pond water depth (D), and C band polarimetric backscatter from FYI melt ponds? Results from a multivariate linear regression analysis of U, F, and D indicated that D is nearly equitable to U in explaining variability in C band co-pol and cross-pol backscatter intensities, particularly at middle range θ. In only once case was F determined a weak but significant explanatory variable. The rco and rcr are invariant to surface variables, with the rco displaying a Bragg-like increase with θ.

[36] Given the rapid fluctuations in temperature and phase changes which occur at the sea ice surface during advanced melt, approaches for successful geophysical information extraction from satellite data must utilize multiparameter and multisensor approaches where available. An adaptive information extraction technique, such as an artificial neural network (ANN), offers perhaps the best approach for handling large volumes of input satellite and ancillary data, while also facilitating complex training data such as simulations from microwave backscatter models. The inclusion of spaceborne C-band dual-pol or polarimetric SAR data as a primary or secondary source for these adaptive techniques, in the context of advanced melt sea ice geophysical information extraction, first requires an examination of the variability in backscatter and polarimetric parameters over which the dominant, sub-resolution, surface cover types are manifest. The preliminary results presented here, with focus on these sub-resolution features, provide a basis for an evaluation of dual-pol and polarimetric C band SAR for the regional scale monitoring and understanding advanced melt sea ice geophysical processes. Further work involves documenting the complete advanced melt seasonal cycle through its drainage stage, as well as contrasting the advanced melt period for different ice types and regions.


[37] The authors would like to acknowledge Torsten Geldsetzer at Environment Canada and Jim Mead at Pro Sensing, Inc. for their discussions and insights. Special thanks to the Polar Continental Shelf Project (PCSP) and members of the POL-ICE 2006 field project in Resolute Bay, NU, in particular Brent Else, Bin Cheng and Adrienne Tivy. Officers and crew of the NGCC Amundsen, fellow researchers onboard during legs 8 and 9 of the CFL project in 2008, Chris Fuller, Lauren Candlish, Klaus Hochheim, Mukesh Gupta, and Natalie Asselin for their assistance in the field; and Tim Papakyriakou for making available meteorological data in support of this project. Funding for the CFL project was provided by the Canadian International Polar Year (IPY) program office, the Natural Sciences and Engineering Research Council (NSERC), the Canada Research Chairs (CRC) Program, Canada Foundation for Innovation (CFI), and numerous international partner organizations. Support to Randall Scharien was provided by Canadian Northern Studies Trust (CNST), Maritime Awards Society of Canada Graduate Scholarships, and Margaret P. Hess Graduate Scholarship awards; and a Northern Scientific Training Program (NSTP) grant. Support for the polarimetric scatterometer was provided by an NSERC Discovery Grant and Canada Foundation for Innovation New Opportunities Award to John Yackel.