SEARCH

SEARCH BY CITATION

Keywords:

  • in situ observations;
  • remote sensing oceanography;
  • sea ice dynamic/thermodynamic

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[1] This paper investigates the influence of solar heating and intermittent cloud cover on the physical and dielectric properties of naturally snow-free, warm (>−2°), first-year sea ice (FYI) in the southeastern margin of the Beaufort Sea during advanced melt. A simple three-layer physical model describing the surface is introduced and copolarized C band microwave signatures are simulated using a multilayer scattering model forced with four sets of measured surface parameters. Modeled backscatter signatures are compared to coincident surface-based C band scatterometer signatures in order to elucidate the signature controlling properties of the ice. Results show that 50 MHz impedance probe dielectric measurements of desalinated upper ice layers exhibit statistically significant diurnal variations due to the link between solar forcing and the availability of free water in brine-free upper ice layers. Enhanced downwelling longwave radiation to the surface from low-level stratus clouds is positively linearly associated (r = 0.709) with volumetric moisture mv detected in upper ice layers. Model results show that desalinated upper ice layers contribute volume scattering from smooth, snow-free FYI under the observed surface mv range. Sustained cloud-free periods result in the formation of a 0.5–2.5 cm granular surface layer, composed of 5.2 mm ice grains, which enhances backscatter under relatively dry conditions. Sensitivity analyses show that layer thickness plays a significant role in scattering due to the increased number density of inclusions which act as discrete scatterers, and sufficient energy may penetrate to, and scatter from, the saline columnar ice layer under relatively dry conditions only (mv < 2%).

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[2] The recent extensive decline in Arctic summer sea ice [Comiso, 2002; Rigor and Wallace, 2004; Stroeve et al., 2007; Nghiem et al., 2007] has prompted the need for an improved understanding of observed losses and their potential impacts on the climate, biogeochemical cycles, and ecological processes within a regional context. Emphasis is being placed on the seasonal sea ice melt season due to linkages between melt season duration, the amount of solar energy absorbed into the ice-ocean system, and successive seasonal declines in sea ice volume through enhanced sea-ice-albedo feedback [Perovich et al., 2007]. Advanced melt, denoted by the occurrence of melt ponds on the ice surface, is a critical component of the melt season as low-albedo melt ponds accelerate melt 2–3 times through enhanced shortwave radiation absorption [Untersteiner, 1961; Maykut and Perovich, 1987]. Snow-free melting ice also transmits 3–15% of incident solar radiation to the ocean [Inoue et al., 2008; Light et al., 2008].

[3] Several techniques have been developed for detecting the onset and duration of the Arctic melt season, primarily using passive and active microwave radiometers and scatterometer data, with demonstrated utility in sea and climate studies [see Barber, 2005; Markus et al., 2009, and references therein]. Despite its importance, comparatively little success in retrieving sea ice geophysical and climate state information during advanced melt has been achieved due to ambiguities associated with the structural and property inhomogeneities of the decaying ice volume and the spatially and temporally dynamic nature of melt ponds, factors which vary on a scale below the resolution of radiometers and scatterometers [Fetterer and Untersteiner, 1998; Markus and Dokken, 2002]. The application of satellite data for understanding spatial and temporal variations in melt pond coverage is primarily limited to estimates using optical methods during cloud-free conditions [Markus et al., 2003; Tschudi et al., 2008]. The enhanced spatial resolution of synthetic aperture radar (SAR) compared to radiometers and scatterometers, as well as the ability of SAR to image the Earth's surface irrespective of cloud cover, has motivated researchers to examine its utility for bridging the gap in our understanding of the evolving ice-ocean-atmosphere system between melt onset and freezeup [Onstott and Gogineni, 1985; Jeffries et al., 1997; Barber and Yackel, 1999; Yackel and Barber, 2000; Hanesiak et al., 2001; Scharien et al., 2007]. These works have primarily focused on the derivation of melt pond fraction and proxy estimates of surface albedo in order to understand the evolution of summer ice albedo, and to evaluate the potential of SAR for aiding the parameterization of sea ice and climate models. Despite SAR showing considerable promise, its utility during advanced melt requires further understanding of linkages between the evolving physical properties of the sea ice cover and observed changes in backscatter. Theoretical backscattering models provide the physical insight required to bridge the gap between the sea ice geophysical properties and empirical observations of microwave scattering by providing a means for elucidating the signature controlling properties of the ice [Winebrenner et al., 1992].

[4] This study focuses on the physical and structural properties of naturally snow-free first-year sea ice (FYI) which influence its C band microwave scattering response during advanced melt. Scattering from snow-covered FYI is controlled by snow physical properties when wet snow is present, and the influence of variable atmospheric forcing on snow wetness, dielectric, and microwave scattering properties is well understood [Livingstone et al., 1987; Barber et al., 1995a]. Diurnal or cloud cover induced increases in water in liquid phase in the snow also increases its complex dielectric constant (ɛ*) and enhances the absorption of microwave energy. Once the snow cover disappears, a seasonal drainage of meltwater and desalination of the ice contributes to an increase in volume scattering and brighter returns in SAR imagery [Holt and Digby, 1985; Drinkwater, 1989; Barber and Yackel, 1999]. However, intraseasonal fluctuations in melt caused by atmospheric forcing and decay-induced retexturing in upper ice layers will also play an important role in controlling the microwave signature of the ice. Though decay-induced properties of sea ice have been studied elsewhere [Eicken et al., 1995; Perovich et al., 2009], though data for FYI are sparse and more work correlating in situ physical measurements with observations of microwave scattering is needed. In concert with previous works detailing the optical properties of summer sea ice [Perovich et al., 2001; Light et al., 2008], this study vertically partitions the ice into three constituent layers: the surface granular layer (SL), the drained layer (DL), collectively referred to as the upper ice layers, and the columnar ice layer (CL). The C band, like- or copolarized, microwave scattering properties of the three layers are evaluated using surface-based scatterometer measurements, with emphasis placed on volume scattering from brine-free upper ice layers. The full treatment of polarimetric scattering is beyond the scope of this paper. The following research questions are addressed: (1) How do diurnal solar heating and intermittent cloud cover affect the physical, structural, and dielectric properties of snow-free FYI during advanced melt? (2) What are the dominant physical mechanisms of the volume which influence C band microwave backscatter from FYI during advanced melt? (3) What is the relative importance of surface and volume scattering to total backscatter?

[5] An overview of field data collection methods (section 2) is followed by an outline of functions for estimating wetness, brine volume, and the C band dielectric properties from field data (section 3). The microwave scattering model used in this study and its parameterization (section 4) and statistical techniques used in this study (section 5) are then detailed. The results of this paper are divided into two parts. First, a description of the changing physical characteristics of the ice and its dielectric properties is provided and evaluated within the context of diurnal solar forcing and cloud cover forcing at the surface (section 6). Next, the influence of ice physical properties on copolarized C band microwave scattering is assessed by combining measured variables with knowledge-based assumptions and initiating a multilayer surface and volume scattering model, and comparing results to surface-based scatterometer measurements (section 7).

2. Field Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[6] In situ sea ice measurements were made between days 154 (2 June) and 173 (21 June) 2008, over snow-free first-year fast ice in Franklin Bay and Darnley Bay in the southeastern Beaufort Sea (Figure 1). Data were collected as part of the Circumpolar Flaw Lead System Study (CFL), a Canadian government funded International Polar Year (IPY) project, based aboard the research icebreaker CCGS Amundsen [Barber et al., 2010]. Several ice “raids” were conducted on the ice adjacent to a flaw lead, during which surface geophysical measurements of the evolving snow and sea ice volume and coincident electromagnetic data were collected.

image

Figure 1. Map of first-year sea ice sampling locations in Franklin Bay and Darnley Bay in the southeastern extreme of the Beaufort Sea. Location of the landfast ice margin derived from a Canadian Ice Service regional ice chart for 5 May 2008 is depicted by the dashed line.

Download figure to PowerPoint

[7] Profiles of ice temperature (Ti) and salinity (Si) at 10 cm depth intervals were sampled from ice cores extracted using a 9 cm diameter core barrel (Kovacs© Mark II Coring System). Each core was placed in a holder immediately after extraction, photographed next to a measuring stick, and Ti measurements were made by drilling into the center at each depth interval then inserting a handheld Digi-Sense RTD thermometer with Cole-Parmer RTD probe (±0.02°C). During probe stabilization, a second worker drilled the next hole in depth from the top of the core. The core was then sectioned below each drill hole and 2 cm thick cylinders were sealed and transferred for conductivity measurements of Si. A second core to 1 m depth was extracted, sealed, and transferred to a cold lab aboard the Amundsen where it was cut into vertical thin sections and photographed using naturally transmitted light and cross-polarizing filters for structural analysis. Given that ice microstructure from extracted cores is compromised by refreezing during cold storage as well as brine drainage prior to storage, analysis of vertical thin sections was limited to qualitative assessments of inclusion properties as pertaining to the electromagnetic properties of the ice [Eicken et al., 1995]. When present, disaggregated surface ice grains were photographed against a grid plate with 2 and 3 mm grid spacing. Grain photographs were processed using a purpose built computer code which extracted grain area and structure, with structure defined by major and minor axis fits to an equivalent polygon in two dimensions. An assumed z dimension size of 1 mm was used to estimate the volume of each ice grain and to determine its equivalent spherical diameter (hereafter referred to as grain size). A constant of 1 mm was chosen based on manual observation which indicated the z dimension was lower than values of the projected surface. Given the data set, no measurements of grain volume could be made. To determine grain volume, NIR measurements are required in order to retrieve the optical diameter such as suggested by the model of Kokhanovsky and Zege [2004]. Other field techniques do exist to measure either optical diameter of specific surface area (surface to volume ratio), but they require heavy lab measurements and instruments that were not available at the time. The grain ratio was derived from the ratio of minor to major axes and used as a morphological indicator. Further details of the grain imaging and measurement techniques are given by Langlois et al. [2007].

[8] Dielectric permittivity (ɛ′i) and dielectric loss (ɛ″i) of the ice at the air-ice interface were measured using a Stevens Water Monitoring Systems Hydra Probe. The Hydra Probe uses a tine assembly consisting of a central waveguide and three outer rods, each 4.5 cm in length and 3 mm wide, to measure the impedance of the sample at 50 MHz over a cylindrical area of 5.7 cm in length by 3 cm in diameter [Seyfried and Murdock, 2004; Stevens Water Monitoring System, Inc., 2007]. The sensor was calibrated in the field using isopropyl alcohol for ɛ′ (±0.6%) and a saline solution of known conductivity for ɛ″ (±0.7%) as previously reported by Geldsetzer et al. [2009]. Other examples of the use of this sensor in snow and sea ice studies include Backstrom and Eicken [2006] and Gully et al. [2007]. Samples were obtained by vertically inserting the probe tines into the ice to their maximum depth (Figure 2, left). Insertion of the probe tines into the melting ice volume was expected to cause two sources of sampling error: the lateral drainage of meltwater toward (and in contact with) the probe tines; and the creation of air voids adjacent to the probe tines. Sampling errors were controlled for by making several measurements at 1 m interval along a linear transect (Figure 2, right), and discarding samples beyond ±2σ from each transect. Mean ɛi′ and ɛi″ pairs from 16 transects conducted between year day (YD) 160 and 173, comprising a total of 131 probe samples, were retained for analyses.

image

Figure 2. Schematic of sampling methodology for determining the dielectric permittivity (ɛ′) and loss (ɛ″) of the upper ice layer at 50 MHz. Samples were collected by vertically inserting (left) the probe into the ice to the sensor maximum depth (right) at several points along a transect.

Download figure to PowerPoint

[9] A fully polarimetric, C band (5.5 GHz), sled-mounted microwave scatterometer was deployed on the ice at a height of 1.8–2 m above the surface (Table 1). The portable system scans the surface over an adjustable range of incidence angles (θ) and across a specified azimuth using a tripod-mounted Kipp and Zonen tracker. A scan consists of backscatter samples for scan lines at 2° intervals over a θ range of 20° to 60°. The azimuthal scan width was set at 30°, although this was occasionally modified to maintain the beam within a homogeneous sample. An external calibration of the radar was conducted during each field deployment using a trihedral corner reflector. A complete review of scatterometer specifications, calibration routine, signal processing, near-field correction, and error determination are given by Geldsetzer et al. [2007]. For this study, averages of four contiguous scans are used, centered on the time of in situ data collection. Four scans take approximately 1.2 h and yield 72 independent samples (i.e., range gates) of backscatter from the surface for each θ line in the 20–60° interval. Only like-polarized backscattering coefficients (σ°VV, σ°HH) are retained for this study.

Table 1. C Band Scatterometer Specifications
ParameterSpecification
Frequency5.5 GHz (C band)
Antenna Beam Width5.4°
Bandwidth5–500 MHz
Range Resolution0.30 m
ModesHH, VV, HV, VH
Noise FloorCopolarized: −36 dBm2
External CalibrationTrihedral Corner Reflector

[10] Meteorological conditions were monitored at an automated weather station mounted on the front deck of the Amundsen. Ta (±0.1°C) and RH (±3%) were recorded at 1 min interval using a Vaisala HMP45C212 probe mounted at 14 m ASL. Downwelling shortwave (Kd) and longwave (Ld) radiation flux data were logged at 1 min interval using Eppley model PSP pyranometer (±5% accuracy) and PIR pyrgeometer (±10% accuracy), respectively. Digital photographs taken every 10 min using an “all-sky” camera were used to qualitatively assess weather and cloud conditions over the site.

3. Derived Variables

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

3.1. Wetness (mv) and Brine Volume (Vb/V)

[11] The permittivity of a volume of wet snow ɛ′w is dominated by the presence of even a small amount of water, such that its dielectric behavior acts independently of temperature, frequency, and grain structure across the 10 MHz to 2 GHz range [Tiuri et al., 1984; Denoth, 1989]. Following this rationale for the upper ice layers encountered in this study, the 50 MHz dielectric measurements described above were converted to estimates of wetness or volumetric moisture content (mv). From Drinkwater [1989] a simple relationship Δɛw = ɛw − ɛd was used to determine the change in permittivity of a snow or ice volume caused by water, as an increase over its estimated dry state density (ρd)-dependent permittivity (ɛ′d). Linear models for a dry air-snow mixture were used to determine the ɛ′d [Cumming, 1952; Nyfors, 1982; from Ulaby et al., 1986]

  • equation image

From Denoth [1989] the relationship Δɛw = 0.206mv + 0.0046mv2 was then rearranged to provide an estimate of surface mv. Values of 450 kg m−3 and 600 kg m−3 were used to represent the lower and upper limits of the ρd of upper ice layers, based on independent surface validated data for summer FYI in the Labrador Sea, FYI and MYI in the Eurasian sector of the Arctic Ocean, FYI and MYI across the Arctic basin, and MYI in the Beaufort Sea [Drinkwater, 1989; Eicken et al., 1995; Perovich et al., 2001, 2009]. Level MYI properties were included given that the drainage and retexturing of the brine-free upper portion of FYI during advanced melt makes it negligibly different from level MYI in terms of density [Eicken et al., 1995]. A mean mv at the lower and upper density limits was derived for each of the 16 dielectric transects.

[12] A comparison of the models in (1) to estimates of ɛ′d for a pure ice “host” dielectric with air bubble inclusions using the Polder-Van Santen (PVD), as modified by de Loor, mixing formula [Shokr, 1998] showed the two methods differ by a maximum 0.04 units and agree according to a least squares regression r2 = 0.99 (p = 0.000) over a 450–900 kg m−3ρd interval. The approach in (1) is thus considered appropriate for determining the ɛ′d of upper ice layers encountered here.

[13] Ti and Si profiles from ice cores were used to estimate the bulk volume of brine (Vb/V) of the saline CL according to Frankenstein and Garner [1967]

  • equation image

The temperature range in (2) is appropriate for the CL observed in this study.

3.2. C Band Dielectric Properties

[14] C band (5.5 GHz) ɛ′i and ɛ″i of the SL and DL were estimated by forcing a dielectric mixing formula with mv values from section 3.1 [Mätzler et al., 1984; Drinkwater and Crocker, 1988; Barber et al., 1995b]

  • equation image

where ɛ*d represents the ice-air host and ɛ*w the water inclusions expressed as complex numbers (ɛ* = ɛ′ − jɛ″). Specification of ɛ*d required estimates of the dielectric loss (ɛ″d) in addition to (1). The PVD formula was used to solve for ɛ″d of ice within an air host volume, resulting in a ɛd″ ≈ 0.003 at C band. The input ɛ″i used to initiate the PVD formula is subject to variability in the literature [see Ulaby et al., 1986], though the effect of this error on ɛ″d is expected to be minimal given that ɛ″w ≫ ɛ″d. The temperature and frequency-dependent ɛ*w in (3) was fixed at 65.50 + 36.96i for a temperature of 0°C and frequency of 5.5 GHz according to the Debye formula [Tiuri et al., 1984; Hallikainen and Winebrenner, 1992]. A coupling factor (χ) in (3) accounts for water inclusion shape, modeled here as isotropically oriented oblate spheroids based on pore structure analysis in the freeboard layer of summer sea ice by Eicken et al. [1995]. On this basis χ was set to 2/3 and A0, the dominant depolarization factor, was set as 0.053 [Drinkwater and Crocker, 1988; Barber et al., 1995b].

[15] Estimates of C band ɛ′i and ɛ″i for the saline CL were derived by forcing the PVD mixing model with Vb/V estimates from section 3.1. Only the upper 0.4 m of the CL was used based on two criteria: (1) brine layers near the ice-ocean interface have no influence on the C band microwave scattering properties of the ice other than to effectively reduce the Vb/V estimates and lower its estimated dielectric values; and (2) error due to the temperature equilibration of an extracted ice core with its surrounding environment and drainage of brine during sampling is reduced by taking samples from top to bottom but discarding measurements after a predetermined cutoff point. The CL was modeled as a pure ice host with needle-shaped (z direction) brine inclusions given their expected vertical orientation within the intracrystalline layers of the warming CL as they coalesce to form brine tubes [Weeks and Ackley, 1986; Light et al., 2003; Golden et al., 2007]. This yielded orthogonal sets of ɛ′i and ɛ″i corresponding to parallel and perpendicular orientations of the incident electric field relative to the inclusions. An arithmetic mean of orthogonal components was used based on Shokr [1998], who found the mean to best agree with field measurements of C band ɛ′i and ɛ″i for columnar FYI below −12°C. To our knowledge, direct comparisons between measured and modeled C band dielectric properties for columnar FYI above −12°C have not been made.

4. Overview of the Scattering Model

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[16] C band microwave backscatter (σ°VV,σ°HH) was modeled using a multilayer approach which accounts for the reflection, refraction, absorption, surface scattering and volume scattering contributions of each layer. The total backscattered signal was determined using a formulation similar to that of Kim et al. [1984] and Ulaby et al. [1984] but extended to include as many layers as needed. Surface scattering was approximated by the physical optics, or Kirchoff, model under the scalar approximation for smooth surfaces described by a Gaussian autocorrelation function [Drinkwater, 1989]. The physical optics model is appropriate for surfaces with a RMS slope ms < 0.25 where ms = equation image(h/l) and has shown good agreement with scatterometer and SAR observations of undeformed sea ice [Kim, 1984; Ulaby et al., 1986; Drinkwater, 1989; Barber and Thomas, 1998]. The volume scattering component (σ°vol) of each layer was obtained by applying a two-way loss factor to account for losses associated with incoming and outgoing microwaves [Winebrenner et al., 1992; Kendra et al., 1998]. The σ°vol was also dependent on the number density of layer inclusions and their individual backscatter cross sections [Drinkwater, 1989].

[17] Parameterizing the scattering model involved merging field measured variables with knowledge-based bounds into a three-layer physical model representing snow-free, advanced melt FYI (Figure 3). The surface layer (SL) was treated as a mixture of 5.2 mm ice grains (see section 6.4 for grain size) and water inclusions embedded in an air background. The second layer (DL) comprised a desalinated ice background embedded with water and air inclusions. These two layers overlaid the CL which contained brine inclusions within a pure ice background. Six model simulations were run based on four unique sets of field-derived C band dielectric properties (section 3.2) and layer thickness hi data (Table 2). Two experimental adaptations (S2-2L and S3-2L), where the three-layer system was adapted into a two-layer system via removal of the SL, were also run. The hSL was taken as the median value incurred during the nearest coincident 50 MHz impedance probe transect, while hDL and hCL, where hCL = 0.4 m − (hSL + hDL), were derived from coincident core measurements. TSL and TDL were set to 0°C in the presence of liquid water inclusions. Included in Table 2 are vertical penetration depth (δp) and two-way loss factor (L), which ranges from 0 (high loss) to 1 (low loss), estimates for each input layer [Ulaby et al., 1984].

image

Figure 3. Physical model for scattering calculations.

Download figure to PowerPoint

Table 2. Scattering Model Inputsa
SimulationLayerhi (cm)Ti (°C)Vi/V (%)ɛ′iɛ″iδp (m)L
  • a

    Three-layer and two-layer specifications of the same simulation are labeled “3L” and “2L,” respectively. Parameters are as follows: ice layer thickness is hi, ice layer temperature is Ti, volume fraction of layer inclusion is Vi/V, ice layer dielectric permittivity is ɛ′i, ice layer dielectric loss is ɛ″i, penetration depth is δp, and two-way loss factor is L.

1SL0.50.01.42.060.0400.190.95
 DL5.00.00.32.240.0024.120.98
 CL34.5−0.618.09.293.1400.0030.00
2 (3L)SL1.00.03.92.460.1100.070.76
 DL5.00.02.32.660.0800.100.36
 CL34.0−0.618.09.293.1400.0030.00
2 (2L)DL6.00.02.52.630.0900.090.26
 CL34.0−0.618.09.293.1400.0030.00
3 (3L)SL0.50.01.42.060.0400.190.95
 DL12.00.00.32.240.0024.120.94
 CL27.5−1.18.05.801.3500.0060.00
3 (2L)DL12.50.00.02.210.0024.120.94
 CL27.5−1.18.05.801.3500.0060.00
4DL7.00.02.72.730.1000.080.17
 CL33.0−0.79.06.161.5300.0050.00

[18] Model assumptions included the use of a 2 mm diameter and spherical shape to describe inclusions in all layers except for SL ice grains. For air and water inclusions, these assumptions agree with bounds defined for water in wet snow [Barber and LeDrew, 1994] and air in FYI [Eicken et al., 1995; Light et al., 2003]. For brine inclusions in the CL, little is known about their size, number density, and distribution in warm FYI (>−2°C), though they are known to enlarge, connect, and elongate with a rise in temperature [Perovich and Gow, 1996; Barber and Nghiem, 1999; Light et al., 2003]. Given the high volume of brine inclusions in the CL, a dielectric mismatch causing predominantly surface scattering is expected at the DL-CL interface. On the basis of surface scattering, the effect of prescribing a single brine inclusion size is likely eclipsed by the large variance associated with ɛ′i and ɛ″i estimates at warm temperatures [Morey et al., 1984; Eicken et al., 1995; Backstrom and Eicken, 2006]. Model assumptions also included ice layer density estimates, set at 450 kg m−3 (SL), 600 kg m−3 (DL), and 910 kg m−3 (CL) based on the references provided in section 3.1. Layer surface roughness was assumed fixed at h = 0.5 cm and l = 3 cm, which falls within the bounds reported for smooth FYI during advanced melt, as well as within the ms validity requirement of the physical optics model [Onstott, 1992; Tucker et al., 1992; Carlström and Ulander, 1995].

5. Statistical Analyses

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[19] A general linear model (GLM) was used to assess the influence of the diurnal cycle on the dielectric properties of the snow-free desalinated ice surface. Null hypotheses (Ho) of no significant difference in 50 MHz ɛ′i and ɛ″i were tested as a function of local “time of day” (TOD) factor. The TOD test factor had three levels: morning (ID = 1; n = 49) defined by transects conducted between 0800 and 1100 LST, afternoon (ID = 2; n = 28) between 1300 and 1600 LST, and evening (ID = 3; n = 54) between 1800 and 2100 LST. The Type III method for calculating the model sum of squares was used due to an unbalanced GLM design (i.e., uneven number of samples between factor levels). Post hoc difference-of-means tests evaluated pair-wise relationships between levels based here on marginal (i.e., unweighted) means to account for the unbalanced GLM design. The Levene's statistic validated the appropriateness of GLM design based on the assumption of equality of variances after square root transforms were conducted on ɛ′i and ɛ″i.

[20] An increase in the dielectric constant of the saline snow basal layer and ice surface, induced by warming from enhanced longwave radiative flux to the surface under cloudy conditions, is known to increase C band microwave scattering from smooth, snow covered FYI during spring [Barber and Thomas, 1998]. In the warm and brine-free surface conditions encountered here, an increase in the longwave radiative flux to the surface (Ld) contributes to the melting of ice at the surface. A bivariate regression test was conducted to determine the strength of linear association between Ld and mv during above freezing air temperatures, i.e., to assess the effect of cloud cover on the presence of in situ liquid water caused by surface melt. Each transect mean mv from section 3.1 was paired with a 2 h mean Ld centered on the acquisition time of the coincident mv transect. For the 16 transects from which mean mv and coincident Ld estimates were derived, n = 14 were used in the regression test. Two samples were removed due to below freezing temperatures during sampling. Both mv and Ld were considered normally distributed according to a Shapiro-Wilk normality test.

[21] While it is known that meteoric snow grain size and shape evolves as a function of radiation, temperature, and wind, and that these changes affect snow microwave backscatter and emission [Langlois and Barber, 2007], the properties of disaggregated FYI ice grains that comprise the SL have not been fully examined. A GLM approach was used to assess the evolution of ice grain size and grain ratio over the data collection period of this study. The null hypothesis (Ho) of no significant difference in grain size and grain ratio based on sampling sessions as fixed factors was tested. Five factors (n = 206) over the YD 164 to 173 period were evaluated. A Levene's test for homogeneity of variances indicated that grain size variances for the 5 factors were equal and that the parametric GLM was appropriate. The preconditions of the parametric GLM were not met for grain ratio, thus necessitating the use of the nonparametric Kruskal-Wallis (K-W) test.

6. Results and Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

6.1. Ice Properties

[22] Ice core characteristics are summarized in Table 3 and vertical profiles of Ti, Si, and Vb/V are given in Figure 4. Note that ice core data in Figure 4 is only shown for snow-free FYI, i.e., a shallow meteoric snow cover was observed for C1.1. A reversal in the Ti profile from a winter trend, and low near-surface Si, indicated that substantial melt and flushing of brine occurred prior to sampling. Bulk Si ∼2 ‰ and Ti ∼−1°C also indicated the ice was in an advanced stage of decay. Flushing of brine from the upper 0.4 m with time is apparent in Table 3, though overall hi and Ti trends were biased by sampling spatial variability. A repositioning of the Amundsen into stable ice conditions to accommodate late season data collection meant that some samples were taken kilometers apart. Layer thicknesses are shown in Table 3 according to the nomenclature of layers described in section 1, with hCL = hi − (hSL+hDL). The snow cover present during C1.1 meant hSL = 0, after which hSL varied between 0.5 and 2.5 cm when present. Examination of time-lapse sky photographs and downwelling radiative fluxes showed that the presence of the SL was related to the occurrence of persistent solar radiation. Daily averages of Kd and Ld are shown in Figure 5 along with vertical lines indicating ice core sample times. No SL was observed at C2.4 and C3.1, during periods of sustained, low-level stratus cloud cover as indicated by low daily Kd (<200 W m−2) and increased daily Ld. Observations of the presence/absence of the SL are consistent with Perovich et al. [2001] and Light et al. [2008], who attributed the formation of the SL to the penetration of solar radiation into the ice and fragmentation of ice grains along grain boundaries, and the ablation of the SL to overcast and humid days, when ablation is greater and condensation is able to accumulate at the surface. Though the two maximum DL thicknesses occurred during periods of sustained cloud cover, there is insufficient data to substantiate a link between cloud cover and the thickness of the DL.

image

Figure 4. Vertical profiles of salinity, temperature, and brine volume from snow-free first-year sea ice during advanced melt.

Download figure to PowerPoint

image

Figure 5. Time series points showing daily average incoming shortwave (Kd) and longwave (Ld) radiation fluxes during the CFL study. Smoothed lines between points are for illustration only. Vertical lines correspond to ice core sample times and are labeled according to core ID as per Table 2 and the text.

Download figure to PowerPoint

Table 3. Ice Core Summarya
Year DayIce Core IDhi (m)Si (‰)Si0–40 (‰)Ti (°C)Layer Depth (cm)
SLDL
  • a

    A dash denotes missing data. Parameters are as follows: ice thickness is hi, bulk ice salinity is Si, salinity of the top 40 cm is Si0–40, bulk ice temperature is Ti, the surface layer is SL, and drained ice layer is DL.

154C1.11.702.11.1−1.90.07.0
160C2.11.362.11.3−1.32.010.0
162C2.21.292.11.5−1.4
164C2.31.251.91.0−0.71.05.0
166C2.41.401.90.9−1.20.025.0
171C3.11.001.20.7−0.70.013.0
173C4.11.751.70.6−1.22.55.0

6.2. The 50 MHz Dielectric Properties of Upper Ice Layers

[23] Measured upper ice ɛ′i and ɛ″i at 50 MHz are shown in Figure 6 grouped by TOD factors 1–3. The ionic conductivity of all samples was 0 S m−1 (not shown), indicating brine-free ice. Figure 6 also shows the estimated ɛ′d range corresponding to the density range of 450–600 kg m−3 for cold ice. A dashed line in Figure 6 represents the theoretical maximum ɛ″i for cold ice at its maximum density of 917 kg m−3, derived from the Debye relaxation equation for pure ice at 50 MHz frequency (f) [Debye, 1929]

  • equation image

based on a relaxation frequency fio = 7.23 KHz at 0°C, a static dielectric constant ɛi0 = 91.5, and a high-frequency limit ɛi = ɛ′i = 3.15. Observed ɛ′i and ɛ″i means are above estimated values for cold ice, indicating the presence of in situ meltwater. Error bars for ɛ′i fall within the estimated cold ice bounds, suggesting the absence of in situ water in the upper ice layers for some samples.

image

Figure 6. Measured 50 MHz dielectric permittivity (ɛ′) and loss (ɛ″) of upper ice layers grouped by local time of day ID: morning (point 1), afternoon (point 2), and evening (point 3). Computed ɛ′ (left axis) for cold, dry ice over the 450–600 kg m−3 density range is shown as the shaded region and ɛ″ (right axis) of ice at its density maximum of 917 kg m−3 are given as a reference.

Download figure to PowerPoint

[24] For ɛ′i, Ho was safely rejected at the 0.05 level (p = 0.045), indicating a statistically significant difference in ɛ′i as a function of TOD. Post hoc analysis showed a significant pair-wise mean difference for ɛ′i between morning and afternoon (Table 4). Ho was also safely rejected at the 0.05 level for ɛ″i (p = 0.028) and post hoc analysis showed a significant pair-wise mean difference for ɛi″ between morning and both afternoon and evening. No significant pair-wise mean difference between afternoon and evening was observed for either ɛ′i or ɛ″i. The upper ice layers exhibit unique dielectric properties during local morning, when the sun is low in the sky and incoming solar radiation is at its minimum. The similarity in ɛ″i for the afternoon and evening is attributed to the sensitivity of ɛ″i to meltwater retained at the surface from the afternoon peak diurnal melt period.

Table 4. Results of Post Hoc Pair-Wise Difference of Means Tests Between Time of Day Factor Levels 1–3 for Dielectric Permittivity equation image′ and Dielectric Loss equation image
 FactorMean DifferenceaSignificance
  • a

    Mean differences are determined using estimated marginal means and tested at the 0.05 significance level.

  • b

    Significant mean difference.

equation image1–2−0.11b0.014
 1–3−0.030.474
 2–3−0.080.056
equation image1–2−0.51b0.039
 1–3−0.51b0.014
 2–30.000.987

6.3. Surface Wetness

[25] Derived mv from each transect, based on upper and lower ρd limits of 450 and 600 kg m−3, respectively, is shown in Figure 7. Given that actively decaying ice is subject to spatial variability in ρd as a function of increasing scale, it can be said that the ranges provided for each transect in Figure 7 represent a best case first estimate of the in situ mv. Uncertainty in mv associated with the assumed ρd range is ±1.8% through (1). Minimization of this error term requires the specification of the ρd for each transect sample, a process which, using conventional sampling techniques, is subject to large error due to the loss or displacement of water, air and brine from actively melting cores [Timco and Frederking, 1996]. The mv results in Figure 7 reflect the observed diurnal properties of ɛ′i in section 6.2, as the lowest values occur in the morning. Figure 8 reveals exceptions to the diurnal characteristics of mv for YD 160-1 and 171-1. On YD 160-1, a high mv of 4–5% was observed despite clear and relatively dry conditions (RH = 86%) for this season. A 2.5 cm thick SL that was notably slushier than subsequent SL observations was present. Analysis of ice microstructure data from YD 160 confirmed the presence of a superimposed ice layer which likely restricted the vertical transport of meltwater beyond the SL and contributed to the higher observed mv. Superimposed ice, which forms from the melt of meteoric snow [Eicken et al., 1994; Haas et al., 2001], has a large polygonal structure [Granskog et al., 2004] that is relatively impermeable to gas and fluid transport until it is obliterated by melt at the surface [Tison et al., 2008]. YD 171-1 was associated with the highest recorded mv of 6–7%, taken during a period of enhanced longwave radiation flux to the surface and fog. Analysis of all-sky photographs showed the advection of a low-level stratus cloud cover into the region occurred 14 h prior to sampling. Coincident Ld (325 W m−2) was greater than Kd (170 W m−2) and RH ≈ 100%. The loss of shortwave radiation during this period was compensated for by strong cloud radiative flux to the surface and this, combined with the increased facility for vapor to transfer to the surface, acted to increase mv.

image

Figure 7. Estimates of the surface wetness (mv, %) present in the upper layer of first-year sea ice surface for densities 450 and 600 kg m−3. Downwelling longwave radiation flux (Ld) associated with each time of day ID are shown as triangles.

Download figure to PowerPoint

image

Figure 8. Comparison of downwelling (top) longwave radiation flux (Ld) and (bottom) relative humidity (RH) to surface wetness (mv) of upper ice layers. Best-fit least squares linear regression line is indicated. The two largest regression residuals (Figure 8 (top)) are labeled by their year day and time of day ID.

Download figure to PowerPoint

[26] The bivariate regression analysis between surface mv and Ld showed a positive correlation r = 0.709 and a regression r2 = 0.502 with a slope significantly different from 0 (ANOVA, F = 0.005) (Figure 8, top). To this it can be said that 50% of the observed variability in mv can be explained by Ld. As expected, the largest model residuals corresponded to the morning case studies on YD 171-1 (+2.1%) and YD 160-1 (+1.6%) discussed above. Though RH is considered a candidate explanatory variable for the observed variability in mv, (Figure 8, bottom), this variable displayed collinearity with Ld and was excluded from the regression analysis.

6.4. Grain Size and Morphology

[27] Calculated SL grain size was 5.2 mm, and GLM results yielded no significant difference in grain size over the study period (p = 0.073; significance level α = 0.05). Calculated grain ratio was 0.63, and the K-W test showed no significance difference in grain ratio over the study period (p = 0.092; significance level α = 0.05). Grain size and morphology for the SL were invariant, leading to the assertion that SL grains originate from an upper ice layer of relatively uniform crystalline structure. Manual observations of grain photographs revealed a mix of granular aggregates and elongated, or candled, grains (Figure 9). Anomalously large (∼2 cm by 1 cm) bonded ice clusters were observed; however, these had been preferentially selected for photographic records and were excluded from quantitative analyses.

image

Figure 9. Photographs of surface granular layer grains taken on (a and b) YD 166, (c) YD 170, and (d) YD 172.

Download figure to PowerPoint

7. Observed and Modeled Scattering

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

7.1. Conditions and Observations

[28] Observed and modeled like-polarized backscatter (referred to as σ°obs and σ°mod, respectively) across the 26–60° θ range for S-1 to S-4 are shown in Figure 10. Truncation below 26° was due to a lack of supportive data to assist in interpretation of the near-range dependency of σ°obs on small-scale surface roughness. A secondary scale of scattering within the scatterometer footprint was apparent in σ°obs, likely caused by localized variations in θ and volume scattering inhomogeneities within the radar footprint [Fung, 1994; Nghiem et al., 1995]. To smooth observed variations, comparisons were made of block-averaged σ°obs and σ°mod based on near-range (NR; 26 ≤ θ ≤ 30), midrange (MR; 32 ≤ θ ≤ 48), and far-range (FR; 50 ≤ θ ≤ 60) intervals. Prior to assessing model performance, an overview of conditions and σ°obs associated with each simulation are given.

image

Figure 10. Observed and modeled C band microwave backscatter from first-year sea ice for scenarios 1–4. Note the scale difference for the S-4 y axis.

Download figure to PowerPoint

[29] Observed conditions associated with simulations are provided in Table 5. S-1 to S-3 were based on three spatially coincident sampling sessions spanning a 25h period, with ice cores taken at the beginning and end of the period. S-1 represents local morning; S-2 (+7 h) represents afternoon; and S-3 (+25 h) marks a return to local morning. Clear skies, above freezing Ta, and calm conditions occurred during S-1, and a 1–2 cm thick skim of ice on adjacent melt ponds suggested a negative surface heat balance prevailed. From S-1 to S-2 calm, clear-sky conditions prevailed, Ta increased from 3.2 to 4.7°C, RH decreased by 5%, and adjacent melt ponds became ice free. SL and DL mv increased and persistent direct solar radiation coincided with a doubling of hSL to 0.01 m over the 7 h period. From S2 to S3, Ta cooled to the freezing point (−0.1°C ± 0.1) and the dew point temperature was reached. Low-level stratus cloud advected into the study site and Kd and Ld merged to near-equivalent values. Below freezing temperatures also prevailed for several hours leading up to S-3. Despite changes in forcing conditions between S-1 and S-3, measured mv, hence the estimated ɛ′i and ɛ″i of SL and DL, were identical. S-4, 7 days later, represents afternoon conditions similar to S-2 but taken at a different site much later in the season. The hSL was null, the DL 2 cm thicker, and a loss of Vb/V in the CL by half compared to S-2 had occurred. Though S-4 was modeled as a two-layer system, partial fragmentation of ice grains and mm-cm scale voids at the air-ice interface were qualitatively observed within the scatterometer footprint.

Table 5. Conditions Associated With Model Simulations and Coincident Scatterometer Observationsa
SimulationYear DayTime (UT)Ice Core IDhSL (cm)Ta (°C)RH (%)Kd (W m−2)Ld (W m−2)
  • a

    Parameters are as follows: surface layer thickness is hSL, air temperature is Ta, relative humidity is RH, downwelling shortwave radiation is Kd, and downwelling longwave radiation is Ld.

11641530C2.30.53.2 ± 0.283436260
21642230C2.31.04.7 ± 0.378583274
31651630C2.40.5−0.1 ± 0.1100323316
41722200C4.10.01.8 ± 0.192492312

[30] Comparisons of σ°obs between S-1 to S-4 are given as the observed change in σ°VV and σ°HH for S-2 to S-4 relative to S-1 in Figure 11. The ability of C band energy to penetrate into each layer and the loss of energy in the layer was evaluated through δp and L (see Table 2). S-1 corresponded to relatively dry conditions, low L, and high δp in the upper layers. Scattering from irregular boundaries and interior dielectric inhomogeneities, as well as surface scattering due to the dielectric mismatch at the brine-filled CL, was expected to occur under these conditions. A decrease in σ°obs between S-1 and S-2 was associated with a mv increase, much greater L, and a dampened δp due to water inclusions. The decrease was more prominent for HH. Spatial variability in the angular trend for S-2 is apparent in Figure 10 and suggests inhomogeneity in meltwater production and drainage at the surface. S-3 σ°obs is similar to S-1 in the FR but, like S-2, significantly weaker in the NR and MR. The lowest overall σ°obs occurs during S-4, under relatively wet afternoon conditions and the absence of a SL. This observation coincided with the lowest L in the DL (0.17). Compared to S-2, the other afternoon sample, σ°obs for S-4 is 4–5 dB less in the NR and 1–2 dB less in the FR for VV and HH. This difference is attributed to enhanced scattering from large ice grains in the low-loss SL present during S-2.

image

Figure 11. Difference in σ°obs for S-2 to S-4 relative to S-1 for (top) VV and (bottom) HH. Differences are given according to near-range, midrange, and far-range incidence angle groupings.

Download figure to PowerPoint

[31] Figure 12 shows the estimated δp for upper ice layers as a function of increasing mv over the assumed density range 450–600 kg m−3. The curves converge at approximately 2% mv, beyond which the role of density on δp is negligible. δp is rapidly reduced from > 3 m at 0% mv, to 0.21–0.15 m for 2–3% mv. For the maximum observed mv of 6–7% from this study (see section 6.3), δp = 0.09−0.07 m. These estimates point to significant penetration of C band energy into upper ice layers and volume scattering contributions to the total backscattered signal. Results are consistent with the findings of Carlström and Ulander [1995], who used a first-order radiative transfer model to show significant volume scattering from air bubbles within relatively warm, low-salinity (<0.2‰) level FYI in the Baltic Sea. When brine is present in warm ice (−5°C to 0°C), δp is known to rapidly approach zero [Hallikainen et al., 1988; Drinkwater, 1989]. For the brine-affected CL in this study, Vb/V is near its maximum and δp is < 1 cm as shown by the dashed line in Figure 12.

image

Figure 12. Penetration depth (δd) of C band microwave energy into upper ice layers of advanced melt first-year sea ice surface as a function of increasing volumetric water content mv. The dashed line near 0 m indicates the estimated minimum to maximum δd into the saline columnar ice layers encountered in this study.

Download figure to PowerPoint

7.2. Model Performance

[32] Referring back to Figure 10, mode performance is summarized as follows:

[33] 1. S-1 shows good agreement for angular trend and σ°obs up to 50°, after which it overestimates σ° obs. The model indicates two dominant scattering processes: strong volume scattering from the SL; and NR surface scattering from the DL-CL interface (Figure 13).

image

Figure 13. Modeled VV polarization surface scattering from the columnar layer and volume scattering from the surface granular layer as a function of incidence angle for S-1 and S-2.

Download figure to PowerPoint

[34] 2. S-2 (three-layer) overestimates σ°obs beyond approximately 34° due to strong volume scattering from the SL, which overwhelms the attenuation of energy by liquid water. Ulander et al. [1992] noted that ice wetness measurement errors contribute to significant errors in attenuation estimates, this must be considered here. S-2 (two-layer) underestimates σ°obs, which points to the importance of the volume scattering contribution of the SL to the total backscattered signal.

[35] 3. S-3 (three-layer) shows behavior similar to S-1 although: (1) the 7 cm increase in DL thickness leads to a 4 dB increase in the NR-FR volume scattering contribution; and (2) a 10% decrease in CL Vb/V reduces the NR surface scattering contribution from this layer by 3–3.5 dB for HH-VV. S-3 (two-layer) shows better agreement with σ°obs and captures the angular trend, which suggests a reduced role of the SL to the total backscattered signal under low mv and large δp conditions such as S-3.

[36] 4. S-4 underestimates σ°obs in the volume scattering regime beyond 30°. This offset was attributed to either the aforementioned occurrence of voids (i.e., partially fragmented ice grains) or model underestimation of the number density of air inclusions. Vertical thin section photographs in Figure 14 give support for the latter. At the start of the experiment (Figures 14a and 14c) there were fewer discrete volume scatterers in the DL than in the overlapping depth range much later in the experiment and coincident to S-4 (Figures 14b and 14d).

image

Figure 14. Vertical thin section photographs, taken using naturally transmitted light and cross-polarized light, of the 0–20 cm depth range from cores (a) C2.2 on YD 161 and (b) C3.1 on YD 171. Close-up photographs of the 3–10 cm range from (c) C2.2 0.5–2.5 cm and (d) C3.1 are shown. Spherical air bubbles are visible in Figures 14c and 14d and small, ellipsoidal remnants of brine pockets are faintly visible in Figure 14c only.

Download figure to PowerPoint

7.3. Model Sensitivity Analyses

7.3.1. SL Grain Size and Thickness

[37] Simulation of the SL scattering signature involved the assumption of spherical ice grains, thus necessitating an estimate of equivalent spherical diameter to invert a grain size from 2D grain photographs. A 1 mm z dimension size was used for the conversion of 2D grain measurements to spherical diameter. Manual observations of grain structure, combined with an observed grain ratio of 0.63, suggested that grains were elongated and that values for z should be lower than values for the projected surface (1 mm being the lower value). Though some bias is expected, the model demonstrates that a 50% increase in z yields a NR to FR increase in modeled σ°VV of 0.7–1.7 dB for S-2, which is considered acceptable as a compromise considering observed grains are not spherical (see Figure 9).

[38] The effect of increasing hSL from 0.005 to 0.03 m on σ°VV from S-2 is shown for discrete θ intervals in Figure 15. A thicker hSL yields more ice grains which act as discrete scatterers and increase scattering. The gain is equitable for VV and HH (not shown) and is manifest in the MR to FR, where volume scattering dominates.

image

Figure 15. Sensitivity of VV-polarized backscatter at θ = 25°, 30°, 40°, and 60° to thickness of the SL. Input parameters correspond to modeling simulation S-2.

Download figure to PowerPoint

7.3.2. DL Air Inclusion Size

[39] Previous work has shown that the idealization of a fixed air inclusion size, combined with the strong sensitivity of backscatter from desalinated ice to inclusion size, produces uncertainty in model outputs [Winebrenner et al., 1989]. Figure 16 shows the effect of DL air inclusion size on σ°VV and σ°HH using input parameters from the two-layer S-4. A decrease in air inclusion diameter from 2 mm to 1 mm yields a drop in σ° by as much 10 dB. A doubling of inclusion diameter to 4 mm increases σ° by 5–9 dB at θ ≥ 30°. These changes both provide unrealistic model output relative to σ°obs for S-4, which provides support for the 2 mm inclusion size used in this study. The vertical thin section photographs in Figure 14 provide qualitative support for this assumption as a reasonable generality for modeling purposes. Previous model results using a dense medium radiative transfer (DMRT) model for a cold, desalinated upper layer of MYI demonstrated that using a single air inclusion size results in an underestimation of σ° at 10 GHz due to the presence of fewer, larger inclusions that contribute greatly to overall scattering but are not accounted for by the model [Winebrenner et al., 1992]. The results in Figures 14 and 16 point to the need for a robust quantitative analysis of inclusion size and distribution with depth, as well the evolution of these parameters in space and time.

image

Figure 16. Sensitivity of VV- and HH-polarized backscatter to air inclusion size. Input parameters correspond to modeling simulation S-4.

Download figure to PowerPoint

7.3.3. CL Brine Volume

[40] The warming and drainage/desalination process of the CL holds a twofold importance: it affects the ability of the ice to accommodate gas and fluid transport [Golden et al., 1998] as well as its mechanical strength and thus its susceptibility to break up under dynamic forcing [Kovacs, 1996]. According to the model simulations, strong NR surface scattering occurs from the CL under relatively dry conditions (mv < 2%), i.e., when energy is transmitted through the upper layers and backscatters from the DL-CL interface. The strength of the CL backscattered signal, under fixed surface roughness, is determined by Vb/V. The effect on σ°VV and σ°HH from the three-layer S-1 for 4–16% Vb/V range is shown in Figure 17. An increase in Vb/V from 4 to 16% increases σ° by 4 dB at 25°, 2 dB at 30°, and only a negligible amount at 40° under the relatively dry and low-loss (L) conditions for S-1 (see Table 2). For typical mv > 2%, absorption losses make the contribution to the total backscattered signal from the CL negligible. Within this framework it is evident that NR C band σ° is sensitive to variations in the bulk Vb/V only during cold, relatively dry conditions during advanced melt.

image

Figure 17. Sensitivity of VV- and HH-polarized backscatter at θ = 25°, 30°, and 40° to brine volume. Input parameters correspond to modeling simulation S-1.

Download figure to PowerPoint

7.3.4. Wetness

[41] The effect of water in liquid phase on σ°mod for S-1 is shown in Figure 17. A change in mv of 1%, 3%, and 5% for both the SL and DL relative to dry (mv = 0.5%) conditions was tested. In the NR to MR regime, an increase in mv reduces σ°mod due to absorption losses. At approximately 35° the trend reverses and follows a logistic function whose rate of growth and saturation is a function of mv (provided mv > 1%). The magnitude of scattering increase also displays polarization sensitivity due to the Brewster angle effect. The mechanism responsible for the MR-FR scattering gain appears to be an increased contribution from the relatively large, water-wetted ice grains in the SL. The expected NR scattering loss from increased mv and its polarization sensitivity is consistent with scatterometer observations of S-2 (mv = 3.9%) relative to S-1 (mv = 1.4%) as shown in Figure 11. However, the results in Figure 18 suggest the model may not adequately account for absorption losses in the MR to FR regimes.

image

Figure 18. Sensitivity of VV- and HH-polarized backscatter to volumetric liquid water content mv, labeled as a percentage in black for VV curves and gray for HH curves. Values on the y axis represent the change in backscatter from a three-layer system with and initial mv = 0 0.5% in the SL and DL. CL inputs correspond to modeling scenario S-1.

Download figure to PowerPoint

8. Summary and Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[42] First-year sea ice is a layered medium which undergoes changes in its structure and physical properties during advanced melt that supersede its seasonally evolving decay state in terms of its C band microwave scattering properties. In this study the ice cover was partitioned into its three constituent parts: the surface granular layer (SL), the drained layer (DL), collectively referred to as the upper ice layers, and the columnar ice layer (CL); and their physical and C band microwave scattering characteristics were examined.

[43] The first research question asked: How do diurnal solar heating and intermittent cloud cover affect the physical, structural, and dielectric properties of snow-free FYI during advanced melt? The occurrence of strong solar energy input to the surface was linked to occurrence of a 0.5–2.5 cm thick surface granular layer composed of large (grain size = 5.2 mm) ice grains with slightly elongated, or candled, morphology (grain ratio = 0.63). No statistically distinguishable change in grain size or morphology occurred over the study period. Dielectric measurements (50 MHz) provided evidence for in situ water in brine-free upper ice layers, with unique permittivity (ɛ′i) and loss (ɛ″i) between morning and afternoon attributed to diurnal warming and melt, and indistinguishable ɛi″ between afternoon and evening attributed to the retention of meltwater at the surface. The wetness of upper ice layers followed a diurnal pattern, though wetness maxima occurred when the drainage of meltwater was impeded by a superimposed ice layer (4–5% mv) and during periods of persistent low-level stratus cloud cover (6–7% mv). A significant linear relationship between downwelling longwave radiation from low-level stratus clouds and wetness in upper ice layers was found.

[44] The second research question asked: What are the dominant physical mechanisms of the volume which influence C band microwave backscatter from FYI during advanced melt? C band scattering model predictions, combined with in situ scatterometer observations, showed that the backscattered signal from the ice was dominated by: (1) the occurrence of a surface granular layer (SL); and (2) variations in the wetness of desalinated upper ice layers. Under wet conditions (mv > 2%), due to peak diurnal melt or enhanced melt during periods of stratus cloud cover, scattering from discrete air inclusions in brine-free upper ice layers was masked. When the SL was present, wet conditions caused a 1–2 dB increase in modeled far-range (θ > 35°) scattering due to large, water-wetted ice grains. Dry conditions (mv < 2%) facilitated volume scattering from ice grains in the SL and air inclusions in the drained ice layer below the SL, such that increases in layer depths caused significant increases in the backscattered signal due to the greater number density of discrete scattering inclusions. The brine volume in the columnar layer was found to contribute to near-range scattering under relatively dry conditions (mv < 2%) only.

[45] Research question 3 asked: What is the relative importance of surface and volume scattering to the total backscatter? Observed scattering from within a spatially coincident scatterometer footprint changed by several dB over a 25 h period. Given the relatively static nature of surface roughness for a short period, volume scattering processes are considered of significant importance to the observed backscatter change. Modeled penetration depths showed sufficient penetration of energy into the brine-free upper ice layers, such that a significant volume scattering contribution to total backscatter is made possible. Multilayer scattering model simulations demonstrated volume scattering from ice grains in the SL, and air inclusions in drained ice, as the dominant scattering mechanism for incidence angles of ≥30°. These results suggest that the C band backscatter properties are transforming to those associated with MYI during this period, a concept introduced with respect to ice physical properties by Eicken et al. [1995]. Assessment of absolute C band backscatter magnitudes, e.g., for satellite-scale remote sensing studies, requires further work incorporating measurements of surface roughness and its evolution during advanced melt.

[46] Our understanding of the physics of an actively decaying sea ice volume is impeded by the relative lack of reliable and comprehensive monitoring techniques. This issue exists at all scales of observation, from the microscale (e.g., observing changes in dielectric inclusion properties, determining gas volume in upper ice layers, etc.) to the macroscale (e.g., determining melt pond fraction or ice decay state from space). Combining empirical observations of microwave scattering with model predictions aids our understanding of the changing 1-D physics of the system and provides a knowledge base for determining the 2-D manifestation of these changes as distributed targets in microwave imagery. Future work will assess the multifrequency (C, L, and X band) and polarimetric behavior of the advanced melt FYI including melt ponds.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

[47] The authors wish to thank the officers and crew of the NGCC Amundsen, whose dedication and excellence made the CFL project a unique milestone in polar marine science. We are indebted to Klaus Hochheim, Mukesh Gupta, Natalie Asselin, and Brent Else for their assistance in the field, and Tim Papakyriakou and Matthew Asplin for making meteorological data in support of this project available. 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 an NSERC Postgraduate Scholarship (PGS) and a Canadian Northern Studies Trust (CNST) award. Support for the polarimetric scatterometer was provided by an NSERC Discovery Grant and Canada Foundation for Innovation New Opportunities Award to John Yackel.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Field Methods
  5. 3. Derived Variables
  6. 4. Overview of the Scattering Model
  7. 5. Statistical Analyses
  8. 6. Results and Discussion
  9. 7. Observed and Modeled Scattering
  10. 8. Summary and Conclusions
  11. Acknowledgments
  12. References
  13. Supporting Information
FilenameFormatSizeDescription
jgrc11734-sup-0001-t01.txtplain text document0KTab-delimited Table 1.
jgrc11734-sup-0002-t02.txtplain text document1KTab-delimited Table 2.
jgrc11734-sup-0003-t03.txtplain text document1KTab-delimited Table 3.
jgrc11734-sup-0004-t04.txtplain text document1KTab-delimited Table 4.
jgrc11734-sup-0005-t05.txtplain text document1KTab-delimited Table 5.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.