Application of a SeaWinds/QuikSCAT sea ice melt algorithm for assessing melt dynamics in the Canadian Arctic Archipelago


  • Stephen E. L. Howell,

    1. Foothills Climate Analysis Facility, Centre for Alpine and Arctic Climate Research, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • Adrienne Tivy,

    1. Foothills Climate Analysis Facility, Centre for Alpine and Arctic Climate Research, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • John J. Yackel,

    1. Foothills Climate Analysis Facility, Centre for Alpine and Arctic Climate Research, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • Randall K. Scharien

    1. Foothills Climate Analysis Facility, Centre for Alpine and Arctic Climate Research, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author


[1] A remotely sensed sea ice melt algorithm utilizing SeaWinds/QuikSCAT (QuikSCAT) data is developed and applied to sea ice the Canadian Arctic Archipelago (CAA) from 2000 to 2004. The extended advanced very high resolution radiometer Polar Pathfinder (APP-x) data set is used to identify spatially coupled relationships between sea ice melt and radiative forcings. In situ data from the Collaborative Interdisciplinary Cryospheric Experiment (C-ICE) (2000, 2001, and 2002) and the Canadian Arctic Shelf Exchange Study (CASES) (2004) are used to validate APP-x data during the melt period. QuikSCAT-detected maps of melt onset, pond onset, and drainage are created from 2000 to 2004, and results indicate considerable interannual variability of melt dynamics within the CAA. In some years, melt stages are positively spatially autocorrelated, whereas other years exhibit a negative or no spatial autocorrelation. QuikSCAT-detected stages of melt are found to be influenced by interannual varying amounts and timing of radiative forcing making prediction difficult. The spatiotemporal variability of ice melt also influences the distribution of ice within the CAA. The lower-latitude regions of the CAA are shown to have accumulated increasing concentrations of multiyear ice from 2000 to 2005. This paper concludes with a discussion of the interplay between thermodynamic and dynamic sea ice processes likely to have contributed to this trend.

1. Introduction

[2] Changes in Northern Hemisphere sea ice extent and concentration have been reported during the past several decades [Cavalieri et al., 1997; Parkinson et al., 1999; Johannessen et al., 1999; Comiso, 2002; Cavalieri et al., 2003; Serreze et al., 2003; Stroeve et al., 2005] that is reflective of the sensitive nature of sea ice in the climate system. The Arctic Oscillation [Thompson and Wallace, 1998] is widely used to characterize Northern Hemisphere atmospheric circulation and has been linked to observed changes in sea ice extent and concentration [Deser et al., 2000; Drobot and Maslanik, 2003; Rigor and Wallace, 2004; Liu et al., 2004; Belchansky et al., 2005], sea ice motion [Zhang et al., 2000; Rigor et al., 2002], and sea ice melt dynamics [Drobot and Anderson, 2001a; Belchansky et al., 2004a]. The albedo of snow covered sea ice controls the input and absorption of solar radiation and greatly influences observed changes. Changes in sea ice albedo under climate warming scenarios is of paramount concern because decreased albedo causes an increase in solar radiation absorption, thus producing further albedo declines and accelerating sea ice melt, the sea ice–albedo feedback [Curry et al., 1995]. Simulations by nearly all Global Climate Models (GCMs) of future climate conditions project considerable reductions in summer sea ice extent [Walsh and Timlin, 2003] and the sea ice–albedo feedback is one of the primary mechanisms responsible.

[3] The Canadian Arctic Archipelago (CAA) is an intricate series of islands located on the North American continental shelf, separated by the Parry Channel with the Queen Elizabeth Islands (QEI) located to the north (Figure 1). The Northwest Passage (NWP) lies in the middle of this region linking the Atlantic and Pacific oceans. The projected decline in sea ice extent by GCMs has raised many questions about the potential of the NWP becoming a viable shipping route. Considering this route is significantly shorter than both the Panama Canal and Cape Horn, improved spatiotemporal ice melt information within the CAA would better facilitate safer and efficient shipping strategies. Previous studies within the NWP have already noted that sea ice within the region has exhibited marked spatiotemporal changes over the last three decades [Falkingham et al., 2002; Howell and Yackel, 2004].

Figure 1.

Map of the Canadian Arctic Archipelago showing its subregions and the locations of C-ICE and CASES field sampling sites.

[4] To date, methods for estimating sea ice melt over broad-scale areas have primarily relied on passive microwave data from the Scanning Multichannel Microwave Radiometer (SMMR) and Special Sensor Microwave Imagery (SSM/I) [e.g., Serreze et al., 1993; Smith, 1998; Drobot and Anderson, 2001b; Belchansky et al., 2004b]. The problem with passive microwave data is its coarse spatial resolution makes it difficult to accurately detect melt within the closely nit islands of the CAA because of pixel-based land contamination. Moreover, although melt onset is an important climatological parameter, more thermodynamically advanced melt information is required operationally for eased ship routing [De Abreu et al., 2001]. Current spaceborne active microwave synthetic aperture radar (SAR) sensors such as the Earth Resources Satellite 1 and 2 (ERS-1/2), RADARSAT-1 and ENVISAT-Advanced SAR (ASAR) offer increased spatial resolution and advanced melt information [e.g., Barber et al., 1995, 2001; Yackel et al., 2001] but are restricted in terms of their small areal coverage. Image size, acquisition demands and processing time requirements also make the use of SAR sensors for the operational monitoring of sea ice melt difficult. SeaWinds/QuikSCAT (QuikSCAT) Scatterometer Image Reconstruction (SIR) active microwave data has large areal coverage (1800 km swath) and high spatial and temporal resolution making it ideally suited for mapping sea ice melt dynamics over broad-scale regions. QuikSCAT data has been used to monitor sea ice extent [Remund and Long, 1999, 2003], detect the onset of snowmelt over the Arctic Ocean Polar Pack [Forster et al., 2001], and provide advanced sea ice melt information within the CAA [Howell et al., 2005].

[5] This study presents the development and application of a QuikSCAT sea ice melt algorithm. We begin by describing the algorithm development followed by a discussion of its application for assessing sea ice melt and decay dynamics in the CAA from 2000 to 2004. We then examine spatial coupling between stages of sea ice melt and radiative forcing using a spatial autocorrelation index, followed by an evaluation of the utility of spatial regression models as a potential ice melt prediction method. Finally, we explore the interplay between ice thermodynamics and dynamics within the CAA from 2000 until freezeup in 2005.

2. Data

[6] QuikSCAT data for the period 2000 to 2004 was obtained from the NASA Scatterometer Climate Record Pathfinder (SCP) project. The SeaWinds scatterometer on board the QuikSCAT satellite is a dual polarized real aperture radar operating a 13.4 GHz (Ku band). QuikSCAT provides normalized cross-section backscatter (σ°) values at fixed incident angles of 46° (HH) and 54.1° (VV) over a swath width of 1800 km with twice daily temporal resolution (i.e., daily ascending and descending passes). QuikSCAT data is available in two image products, eggs and slices, at Scatterometer Image Reconstruction (SIR) enhanced and nonenhanced grid resolution. The spatial resolution for the nonenhanced grid products is 11.125 km for slices and 22.25 km for eggs and the spatial resolution of SIR enhanced is ∼8–10 km and ∼4 km for eggs and slices, respectively [Long, 2000].

[7] The closely nit islands of the CAA make the use of the lower-resolution grid products problematic because of land contamination. Therefore the increased spatial resolution of the QuikSCAT SIR enhanced product is more desirable. The limitation with the SIR product is that the resolution enhancement has a tendency to amplify noise [Early and Long, 2001]. This imposes constraints on the data regarding sampling density, nulls introduced by the aperture function(s), acceptable noise level, and the temporal stability of the study area [Long et al., 1993]. A comparison between SIR enhanced and non-SIR enhanced combined ascending/descending egg and slice products was performed by Howell et al. [2005] for monitoring landfast first-year ice (FYI) melt in the CAA. They suggest that the SIR egg combined pass product is more appropriate because SIR slice products produced noisier results. Therefore the QuikSCAT SIR combined pass egg image product is selected for this analysis to maximize spatial resolution and minimize noise.

[8] The extended advanced very high resolution radiometer (AVHRR) Polar Pathfinder (APP-x) [Wang and Key, 2005a] radiative forcing data is used to identify coupled relationships between QuikSCAT-detected melt stages and radiative forcings in the CAA. The APP-x data set offers a plethora of surface, cloud and radiation parameters over the Arctic at a 25 km spatial resolution. The data has been validated with in situ data from the Surface Heat Balance of the Arctic Ocean (SHEBA) field experiment [Maslanik et al., 2001; Stroeve et al., 2001] and used to assess trends and spatiotemporal variability in the aforementioned parameters for Arctic regions [Wang and Key, 2003, 2005b]. Maslanik et al. [2001] point out that APP-x data set can be used to observe coupled relationships between surface state conditions and atmospheric forcings at a relatively fine spatiotemporal resolution. However, Key et al. [1997] caution that the satellite-derived surface radiation parameters at high latitudes are more appropriate for monthly timescales and investigating processes during the melt season may result in some added uncertainty. In light of this fact, we first compare APP-x retrievals with in situ and QuikSCAT data during the melt period.

[9] The in situ data was collected by an on-ice meteorological tower during the Collaborative Interdisciplinary Cryospheric Experiment (C-ICE) (75°14.6 86′N, 97°09.207′W) in 2000, 2001, and 2002 and during the Canadian Arctic Shelf Exchange Study (CASES) (70°2.516′N, 126° 15.894′W) in 2004 (Figure 1). For C-ICE surface temperature (Ts), incoming shortwave radiation (K↓), outgoing shortwave (K↑), incoming longwave (L↓), and outgoing longwave (L↑) data were obtained at continuously logging 15-min averages from the premelt period (early May) until the melt pond period (approximately mid-June to early July) for a snow covered, smooth landfast first-year sea ice (FYI) site. Only very few Ts data points are available for C-ICE2000; therefore, in order to offer some means of comparison to APP-x data we used 1 m air temperature (Ta) as surrogate. The same meteorological variables collected over the same ice conditions (i.e., smooth landfast FYI site) were obtained at CASES except, the temporal domain ranged from mid-January until late May. A complete description of instruments used to acquire the C-ICE meteorological data is given by Mundy et al. [2000], Mundy and Downie [2001], and Owens and Papakyriakou [2003] for 2000, 2001 and 2002, respectively. For the CASES meteorological instrument descriptions please refer to Papakyriakou et al. [2004].

[10] The Canadian Ice Service (CIS) Eastern and Western digital ice charts were used to describe the CAA ice coverage conditions from 2000 to 2004. We selected ice coverage as oppose to ice extent because the former considers reductions in ice concentrations from inside the ice edge while the latter does not. Regional digital ice charts are one of the primary climatological products issued by the CIS, providing information on ice conditions. The regional digital ice charts are derived weekly, from the integration of data from a variety of sources including surface observations and aerial and satellite reconnaissance (with the primary source being RADARSAT-1 SAR since 1996), and represent the best estimate of ice conditions based on all available information at the time [Crocker and Carrieres, 2000]. The CIS digital ice charts for the Eastern and Western Arctic were found to more accurately represent ice coverage compared to passive microwave estimates which can underestimate ice coverage as much as 21.5% during the melt season [Agnew and Howell, 2003].

3. Algorithm Development

[11] The QuikSCAT sea ice melt algorithm for the CAA is based on the temporal evolution of the backscatter coefficient (σ°) that can be thresholded into distinct ice melt states: winter, melt onset (MO), pond onset (PO) and drainage. At C-band (i.e., ERS-1/2 and RADARSAT-1), σ° has been linked to these thresholds for FYI and multiyear ice (MYI) [e.g., Winebrenner et al., 1994; Barber et al., 1995, 1998, 2001; Yackel et al., 2001]. At Ku band, QuikSCAT σ° has been shown to follow a similar seasonal evolution as C band for FYI and can be used to identify these melt stages [Howell et al., 2005]. The QuikSCAT FYI melt transition thresholds are based on numerous site specific σ° temporal evolution plots within the CAA from Howell et al. [2005]. However, the regional icescape of the CAA consists of both FYI and MYI types, the former are typically concentrated in the lower-latitude regions and the latter in the high-latitude regions of the QEI. There are some years when MYI from the QEI and/or the Polar Pack invades the lower-latitude regions late in the summer season and becomes landlocked (Figure 2). When describing the QuikSCAT-detected melt stages and thresholds for FYI, we show that they are also applicable to MYI based on additional temporal evolutions of QuikSCAT σ° at MYI sites within the CAA (Figure 2). According the CIS digital ice charts, premelt ice conditions in the M'Clintock region for 2004 contains homogenous regions of landfasted FYI and MYI in very close proximity to each other (Figure 2). As a result, this area is used as the algorithms primary representation of how temporal evolution QuikSCAT σ° signatures allow for the identification of both FYI and MYI melt information (Figure 3).

Figure 2.

Premelt sea ice conditions in the Canadian Arctic Archipelago for 2004 showing the locations of MYI and FYI temporal evolution QuikSCAT σ° sites. The black regions represent areas of greater 9/10ths MYI. The light grey regions represent areas of greater than 9/10ths FYI.

Figure 3.

Temporal evolution of QuikSCAT σ° and estimated melt stage thresholds for landfast FYI and MYI within the M'Clintock region during 2004. Thresholds applicable to both landfast FYI and MYI.

[12] We acknowledge that a limitation of the algorithm is the moderate spatial resolution of QuikSCAT SIR data. This can introduce within-pixel mixed ice types that do not follow the theoretical σ° evolution of each pure ice type. In addition, regions with significant ice deformation will not likely mimic this evolution. As a solution, we only apply the algorithm thresholds to homogenous FYI and MYI pixels established during winter conditions. An ordinary kriging routine is then applied to fill in ‘pixel gaps’ in order to create a continuous melt surface across the CAA. We also acknowledge that the algorithm is limited to landfast regions of FYI and MYI within the CAA. Consequently, dynamic ice movement may cause erroneous melt information near the shear zones of the Polar Pack in the western CAA and in other regions where recurring polynyas (i.e., Cape Bathurst Polynya in ‘Amundsen’) and early spring ice breakup (i.e., East Barrow) are common.

3.1. Winter

[13] Premelt “winter” conditions, as detected by QuikSCAT (hereinafter referred to as Ku band for the algorithm development section), exhibit stable low- and high-σ° conditions for FYI and MYI, respectively (Figures 3 and 4) . For FYI, this can be attributed to relatively smooth ice surface scattering, whereas the high MYI backscatter is likely attributable to high amounts of volume scattering from air bubbles contained within the upper hummock layers.

Figure 4.

Temporal evolution of QuikSCAT σ° at landfasted MYI sites within the Canadian Arctic Archipelago during 2004.

[14] In order to identify stable winter conditions for distinct (i.e., not mixed) pixels of FYI or MYI, σ° for each pixel in the CAA regions is first averaged from year day (YD) 90 to 120. Each pixel in the CAA is then thresholded at less than −18 dB for FYI or greater than −11 dB for MYI. This approach avoids mixed ice type pixels, thus ensures temporal evolution melt stage thresholds are representative. Moreover, by establishing average stable winter conditions, change thresholds in σ° from these stable winter conditions can be used for both FYI and MYI types.

3.2. Melt Onset

[15] FYI Ku band MO is identified by the first sharp upturn in σ° from stable winter conditions (Figure 3). As the snow-ice interface temperature approaches −5°C, brine volumes begin to increase [Assur, 1958], which increases the dielectric constant and thus σ° and contributes to increase volume scattering. With its small wavelength, Ku band reacts readily to the basal brine volume increase but when trace amounts of liquid water become present as snow temperatures approach freezing scattering will dominate above the snow-ice interface a result of the shallow penetration depth of the Ku band [Ulaby et al., 1984; Howell et al., 2005].

[16] MYI MO is identified by a distinct downturn from stable winter conditions (Figures 3 and 4). At C band, surface wetting at the air-snow and snow-ice interfaces do not play a role in the initial decrease in MYI σ°. This is because as liquid water increases within the snowpack (i.e., ∼1% by volume), the larger amount volume scattering from the hummock layer will be masked by snow surface scattering [Winebrenner et al., 1994; Barber et al., 1995]. Ku band σ° evolution of MYI is similar to C band σ° evolution because of its smaller wavelength and reduced penetration depth that masks (even more so than C band) hummock layer volume scattering.

[17] An absolute change of greater than 2 dB for each pixel from stable winder σ° conditions is used to mark the transition to MO for both FYI and MYI. The change in Ku band σ° was found to be greater than 2 dB at MO for FYI in all cases investigated from Howell et al. [2005] and in all MYI cases (Figures 3 and 4). Both are viewed as a conservative estimate of MO.

3.3. Pond Onset

[18] For FYI, the upturn in Ku band σ° continues as increases in the amount of high dielectric water in the snow cover results in greater amounts of snow surface and snow volume scattering (Figure 3). For MYI, the Ku band σ° downturn continues as the lower snow cover scattering (but higher dielectric loss) dominates over the higher air bubble volume scattering observed during winter (Figure 3). The trend in the MYI σ° is followed by a σ° upturn as the MYI snow cover is completely ablated and a high dielectric melt pond cover dominates, causing an increase in σ° (Figures 3, 4a, 4c, and 4d). In some cases, the MYI σ° upturn is not very distinct (Figures 4b and 4e). We speculate that is attributable to low wind speeds not roughening the melt pond surface after complete snow ablation or a thin snow cover which rapidly ablates.

[19] In order to capture the transition into the PO stage, we employ an absolute change threshold from winter conditions for each pixel of greater than 5 dB. Ku band σ° for FYI was found to be greater than 5 dB from winter conditions at PO for all cases investigated by Howell et al. [2005]. Figures 3 and 4 also indicate that a 5 dB change is representative of the PO stage for MYI even in the absence of an upturn.

3.4. Drainage

[20] From the ponding stage and into the drainage Ku band σ° follows a similar evolution for both FYI and MYI. The very high sensitivity of Ku band to liquid water and subsequent reduced penetration depth causes scattering to be dominated by liquid water in or on the surface of the snowpack. This dominance of scattering by liquid water on the sea ice surface is reflected in the how Ku band σ° FYI and MYI temporal evolutions cross each other at the ponding stage (Figure 3).

[21] A gradual decrease in FYI and MYI Ku band σ° coincides with a decrease in pond fraction as melt ponds begin to drain through brine drainage channels, sea holes, and cracks in the ice (Figures 3 and 4). This gradual decrease in pond fraction can be impeded by changes in surface geophysics such as freeze-thaw cycle effects, surface hydraulics, predominant winds and precipitation events that can occur on hourly, diurnal and weekly timescales [Holt and Digby, 1985; Yackel and Barber, 2000; Eicken et al., 2004]. The FYI Ku band σ° downturn after PO has been observed by Howell et al. [2005] at a variety of locations in the CAA. The downturn in Ku band σ° is also present for the MYI sites (Figure 4). On the basis of this, we selected an absolute change threshold of 4 dB from the PO stage to make a conservative transition estimate into the drainage stage.

4. Results and Discussion

4.1. Evaluation of APP-x Data

[22] APP-x Ts values compared to C-ICE and CASES all exhibit significant positive correlations within the 99% confidence level (Table 1). For C-ICE2000 Ta values were also significantly correlated to APP-x Ts values (Table 1). To assess the Ta substitution for Ts, C-ICE2000, C-ICE2001, C-ICE2002, and CASES Ta values were correlated with APP-x Ts data and all were found to be significant at the 99% confidence interval with virtually the same Pearson correlation coefficient (not shown). Both in situ and APP-x Ts values at C-ICE and CASES increased with the transition to QuikSCAT-detected MO (Figures 5 and 6) . APP-x Ts values continued to increase at C-ICE for QuikSCAT-detected PO and into the drainage stage (Figure 5). The limited temporal availability of in situ data at CASES restricted comparison beyond QuikSCAT-detected MO (YD145) (Figure 6). The extreme negative fluctuations of APP-x Ts values during the melt season at C-ICE coincide with in situ observed local storm and/or recent meteoric precipitation events which likely complicate APP-x retrievals. For the entire CAA, MO was detected by QuikSCAT at mean APP-x Ts values between −4.3°C and −7.7°C (Table 2). For PO and drainage melt stages the entire CAA exhibits mean values between 0.5°C and −5.4°C (Table 2). 2001 and 2002 experience more Ts underestimations at C-ICE and the entire CAA compared to other years for all QuikSCAT-detected stages of melt (Table 2). At C band σ°–detected MO, Barber et al. [1995] observed Ts values near −5°C for FYI and MYI sea ice sites. For the same sites they also observed Ts values near 0°C at C band–detected PO and drainage. Considering QuikSCAT's ultra sensitively to high dielectric liquid water [Howell et al., 2005], combined with the reported 1.98 K root mean square error (RMSE) of APP-x Ts values at SHEBA [Wang and Key, 2005a] we suggest reasonable confidence can be placed in APP-x Ts values for the QuikSCAT-detected stages of melt.

Figure 5.

Time series QuikSCAT σ° values with APP-x and in situ surface temperatures at C-ICE2000, C-ICE 2001, and C-ICE 2002 (1 m air temperature at C-ICE2000 plotted as a surrogate for surface temperature).

Figure 6.

Time series QuikSCAT σ° values with APP-x and in situ surface temperatures and surface radiative fluxes at CASES 2004.

Table 1. APP-x Surface Temperature and Surface Flux Data Pearson Correlation Coefficients Compared to C-ICE and CASES in Situ Meteorological Data
  • a

    Significant at 99% confidence level with in situ 1 m air temperature.

  • b

    Significant at the 99% confidence level.

Table 2. Mean and Standard Deviation (in Parentheses) APP-x Daily Average Surface Temperature and Surface Radiative Flux Values at QuikSCAT-Detected MO, PO, and Drainage for the CAA
Melt Onset
2000−4.3 (4.1)332 (64)430 (60)240 (28)295 (18)
2001−7.7 (5.1)265 (43)408 (42)225 (25)282 (19)
2002−7.3 (4.4)267 (43)419 (37)221 (25)283 (18)
2003−4.5 (3.3)249 (61)393 (51)248 (24)295 (14)
2004−5.0 (3.8)266 (49)404 (50)245 (28)293 (16)
Pond Onset
2000−0.8 (3.2)267 (90)429 (62)246 (30)310 (15)
2001−5.4 (4.8)257 (49)424 (37)226 (24)291 (18)
2002−4.9 (3.9)246 (61)420 (51)231 (26)293 (16)
2003−2.0 (2.7)216 (66)385 (57)261 (24)306 (12)
2004−2.4 (2.6)239 (60)403 (48)257 (24)304 (11)
20000.5 (3.5)151 (80)361 (80)272 (31)317 (17)
2001−5.2 (4.7)229 (64)416 (47)230 (28)292 (18)
2002−5.0 (5.1)184 (72)397 (62)240 (29)294 (20)
2003−1.1 (3.1)167 (72)378 (59)261 (25)310 (14)
2004−1.2 (2.8)169 (73)377 (57)265 (25)309 (13)

[23] APP-x K↓ correlations were found to be significant within the 99% confidence level for C-ICE2001, C-ICE2002, and CASES although C-ICE2002 exhibited a low correlation coefficient (Table 1). For all years, APP-x K↓ values generally overestimate in situ values but were found to increase into the QuikSCAT-detected melt period (Figures 6 and 7) . APP-x K↑ values were found only to be significant within the 99% confidence level for C-ICE2001 and CASES (Table 1) but all years exhibit a marked decrease in K↑ after QuikSCAT-detected PO (Table 2 and Figure 7). The inconsistency between in situ K↑ and APP-x K↑ values is likely related to scale. The APP-x K↑ values are integrating a melt pond fraction over a 25 km area and in situ K↑ values tend to be measuring the reflective shortwave from a more highly reflective snow patch because of the need to occasionally relocate the meteorological tower to remnant snow and drained ice patches situated among the flooded and/or ponded surface after PO. For all years, both C-ICE and CASES in situ L↓ and L↑ values were found to be positively correlated with APP-x values and significant at the 99% confidence level (Table 1). CAA APP-x L↓ and L↑ values increased with the transition into QuikSCAT-detected PO and drainage (Table 2 and Figure 8) and compared to APP-x shortwave data, the longwave data exhibit less variability (Table 2).

Figure 7.

Time series QuikSCAT σ° values with APP-x and in situ shortwave fluxes at C-ICE2000, C-ICE 2001, and C-ICE 2002.

Figure 8.

Time series QuikSCAT σ° values with APP-x and in situ longwave fluxes at C-ICE2000, C-ICE 2001, and C-ICE 2002.

[24] This comparison generally shows good agreement between in situ and APP-x derived Ts and longwave flux data. This is encouraging for linking QuikSCAT-detected stages of melt to radiative forcings within the CAA following Yackel et al. [2001], who found that L↓ and Ts explained the most variation in C band σ°–detected MO. Unfortunately, as shown by Maslanik et al. [2001], the APP-x shortwave flux data experiences more discrepancies. As a result, we exercise caution in interpreting the shortwave factors that influence ice melt, particularly for 2000 and 2002. Overall, the APP-x data set likely provides the best available satellite-based estimate of surface energy balance parameters with a large areal extent and high spatial resolution to represent the CAA and complement QuikSCAT-detected melt stages.

4.2. Ice Decay Spatial Variability and Coupled Radiative Forcings

[25] The spatial dependence (i.e., the relatedness of the data in space) of QuikSCAT-detected sea ice melt stages toAPP-x radiative forcing was statistically linked using Moran's I [Moran, 1950]. Moran's I is a spatial autocorrelation index that differs from a typical correlation index because it also considers the correlation of a variable with respect to its spatial location. A positive index indicates that variables which are similar in location also tend to have similar attributes. A negative index indicates variables at neighboring locations have dissimilar values and random patterns exhibit no spatial autocorrelation. Moran's I allows for the joint exploration of coupling between ice melt and radiative forcing both with respect to location.

[26] The date of MO was significantly positively spatially autocorrelated within the CAA for 2000 and 2004 (Table 3). Examination of the MO maps reveals a ‘distinct’ and clustered MO distribution is present for 2000 and 2004 whereas, 2002 and 2003 exhibit a more random MO pattern and reveal no significant spatial autocorrelation (Figure 9). The 2000 MO dates gradually diffuse outward from the QEI and Viscount-Melville regions with earlier occurring dates near peripheral (e.g., M'Clure and Lancaster Mouth) and southerly (e.g., Amundsen, Coronation-Maud and Larsen Sound) regions. The spatial distribution of MO dates is similar in 2004 with the exception of the M'Clure region reaching MO much later and more spatially extensive melt taking place in the regions surrounding Barrow (Figure 9). MO dates within the CAA for 2001 were found to be significantly negatively spatially autocorrelated indicating that neighboring values have dissimilar values and characteristic of a checkerboard distribution (Table 3). This is particularly evident within the CAA for 2001 where earlier MO dates neighbor later MO dates within numerous regions (Figure 9). The 2000 MO distribution is positively spatially autocorrelated with K↑, K↓ and L↓; however, only K↑ is significant. High K↑ values are present for the QEI and Viscount-Melville and lower values for peripheral regions (not shown). Negative spatial autocorrelations are also present with Ts and L↑ suggesting competition between the forcings but insignificant at a reasonable confidence level (Table 3). No radiative forcing variables where found to be significantly spatially autocorrelated for 2001 and all radiative forcing variables were found to be significantly spatially positive autocorrelated at MO for 2004 (Table 3).

Figure 9.

QuikSCAT σ°–detected melt onset in the Canadian Arctic Archipelago, 2000–2004. Legend is in year day (YD).

Table 3. Moran's I p Values for Potential APP-x Radiative Forcings Affecting QuikSCAT-Detected MO, PO, and Drainage Spatially in the CAAa
  • a

    All p values are positively spatially autocorrelated unless followed by a minus sign indicating a negative spatial autocorrelation.

  • b

    Significant within the 99% confidence level.

  • c

    Significant within the 95% confidence level.

Melt Onset
Pond Onset

[27] Significant negative spatial autocorrelations are present at PO for 2001 and 2004 (Table 3) with late and earlier PO transition dates clearly present within regions of the CAA for these years (Figure 10). 2001 exhibits a much earlier PO transition compared to 2004, predominantly the regions south of the Parry Channel (Figure 10). K↓ and L↓ are significantly negatively spatially autocorrelated with PO for 2004 and Ts and L↑ for 2001 (Table 3). Average Ts values at PO for the CAA in 2004 are cooler than 2001 (Table 2) as a result we speculate that in certain regions the early PO dates in 2001 are additionally influenced by K↓ and L↓ despite lower confidence levels of 80% and 86%, respectively.

Figure 10.

QuikSCAT σ°–detected pond onset in the Canadian Arctic Archipelago, 2000–2004. Legend is in year day (YD).

[28] At the drainage stage, 2000 and 2004 exhibit significant positive spatial autocorrelations and 2001 and 2003 exhibit significant negative autocorrelations (Table 3). This is reflected in the more homogeneous distribution of drainage within the CAA for the former and the patchy distribution for the latter (Figure 11). For 2000 and 2004 only K↓ in 2004 is not spatially autocorrelated at QuikSCAT-detected drainage (Table 3). The negative spatial autocorrelation in QuikSCAT-detected drainage for 2001 is significantly negative spatial autocorrelated with Ts, K↓, L↑ and L↓ (Table 3). 2003 exhibited a negative spatial autocorrelation for drainage but is significantly positively autocorrelated with Ts, L↑ and L↓ suggesting these forcings are not spatially competitive (Table 3).

Figure 11.

QuikSCAT σ°–detected drainage in the Canadian Arctic Archipelago, 2000–2004. Legend is in year day (YD).

[29] Spatial regression models employing a maximum likelihood estimator were applied to the APP-x data to explore potential prediction variables that influence the timing of ice melt transitions for the CAA. As oppose to linear regression, spatial regression takes into account the spatial dependence of the data within the CAA and thus allows for exploration and establishment of functional relationships among variables (i.e., melt dates and radiative forcing) located in space (i.e., CAA). Spatial regression also maximizes the reliability of the model by lowering the variance. Although the maximum likelihood spatial regression estimator does not provide any real indication of the goodness of fit for a spatial model [Rogerson, 2001] we computed a pseudo-R2 as suggested by Anselin [1988] in order to evaluate each spatial model.

[30] The results of spatial regression models for predicting MO, PO and drainage with the APP-x data for 2000 to 2004 are listed in Table 4. The variance explained by the spatial regression models ranged between 0.32 to 0.67 for MO, between 0.17 to 0.48 for PO, and 0.22 to 0.48 for drainage. The 2000 MO model resulted in the highest pseudo-R2 of 0.67. Comparison of the 2000 MO spatial model against the QuikSCAT-detected MO shows that the model does detect MO occurring first in the regions of Lancaster Mouth, Amundsen, south Prince Regent, and east Coronation–Maud (Figure 12). The spatial model also predicts later MO occurring in Balantyne-Hazen and Kellet-Crozier regions (Figure 12). Residual plots show larger underestimations for Lancaster Mouth, Amundsen, Coronation–Maud, and the M'Clure regions and overestimation taking place Perry-Sverdrup, Gustaf-Adolph, and KCI West regions (Figure 12). We speculate that the spatial variability of snow thickness distributions on both FYI and MYI could be playing a significant role in the coupling between QuikSCAT-detected dates of MO, PO and drainage and radiation forcing, and is an obvious avenue for further research into the explaining such spatial regression residual results.

Figure 12.

Spatial regression prediction map for (top) MO 2000, (middle) QuikSCAT σ°–detected MO 2000, and (bottom) residuals. Legend is in year day (YD).

Table 4. Spatial Regression Model p Values and Pseudo-R2 of APP-x Radiative Forcings Predicting MO, PO and Drainage in the CAA
  • a

    Significant within the 99% confidence level.

  • b

    Significant within the 95% confidence level.

Melt Onset
Pond Onset

4.3. Sea Ice Thermodynamic and Dynamic Interplay in the Canadian Arctic Archipelago

[31] Although sea ice within the CAA is predominately landfast, its seasonal make up (i.e., type, concentration and coverage) is determined by the interplay between dynamics and thermodynamics. All regions of the CAA reach the same annual maximum (winter) total ice coverage but the total minimum ice coverage varies inter-annually for most regions (Figures 1316). This can be attributed to a specific region not fully breaking up and/or wind dynamics driving ice from one region to another. FYI tends to dominate the icescape for southerly regions of the CAA, but recently these southerly regions are experiencing increases in MYI (Figure 17). This is an important consideration because MYI is thicker than FYI and will take longer to breakup.

Figure 13.

Total ice (black) and MYI (grey) coverage for southwestern regions of the Canadian Arctic Archipelago from 2000 to 2005.

Figure 14.

Total ice (black) and MYI (grey) coverage for western QEI of the Canadian Arctic Archipelago from 2000 to 2005.

Figure 15.

Total ice (black) and MYI (grey) coverage for southeastern regions of the Canadian Arctic Archipelago from 2000 to 2005.

Figure 16.

Total ice (black) and MYI (grey) coverage for the eastern QEI of the Canadian Arctic Archipelago regions from 2000 to 2005.

Figure 17.

MYI concentrations in the Canadian Arctic Archipelago on 1 January from 2000 to 2005. Ice concentration in tenths.

[32] For regions located in the southwestern part of the CAA, MO timing is relatively consistent from 2000 to 2003 (Figure 9). The more southerly regions (especially the Cape Bathurst Polynya in the Amundsen region) reach MO earlier compared to 2004 which experienced a slightly longer transition to MO (Figure 9). With respect to dynamics, the M'Clure Strait, Viscount-Melville, M'Clintock Channel, and portions of Larsen Sound all exhibit marked MYI increases from 2000 to 2005 whereas the Amundsen and Coronation-Maud remained virtually unchanged (Figure 13). Howell and Yackel [2004] suggest that the recent increases in the amount of FYI in the Beaufort Sea (corresponding to the retreat of Polar Pack) (Figure 17) breaks up earlier making the pack more susceptible to wind-driven movement. This results in more open water that facilitates the eastward pack ice movement into the M'Clure Strait. This type of ice motion in the Canadian Western Arctic has been associated with positive phase of the Arctic Oscillation [Zhang et al., 2000; Rigor et al., 2002]. Despite MO occurring early in the M'Clure Strait region, atmospheric forcing is likely insufficient to facilitate complete breakup of the MYI. Instead, the ice becomes mobile moving southward as the FYI regions to the south melt out earlier. As a result, since 2000, the MYI in entering the M'Clure Strait and into Viscount-Melville and the M'Clintock likely continues to flow southward toward Larsen Sound and Coronation-Maud with virtually the entire M'Clintock channel being composed of MYI in 2005 (Figure 17).

[33] While the regions of east Viscount-Melville are beyond the reach of the MYI tongue from the Polar Pack, they are subject to MYI encroachment from the QEI. The QEI regions are subject to large-scale sea ice dynamics where Polar Pack ice is continually forced up against them by the predominately anticyclonic circumpolar gyre creating some of the oldest and thickest sea ice in the world [Agnew et al., 2001]. MYI is the main ice type in the QEI and takes considerably longer to breakup than FYI. However, when breakup of MYI in the QEI occurs, it can be transported southward into the southern regions of the CAA. Moreover, the most northerly islands of the QEI act as a barrier, or plug, to MYI from the Arctic Basin, impeding MYI transport into the QEI and subsequently into the central and southern Archipelago [Agnew et al., 2001]. Should warming reach and fracture these high-latitude channel plugs, the lower regions of the CAA would be subject to an increased flux of MYI.

[34] Anomalously warm air temperatures in 1998 were observed in the northern regions of the QEI, causing a substantial loss of MYI [Agnew et al., 2001; Jeffers et al., 2001]. Evidence of this occurrence can be seen for the high-latitude regions of KCI East, KCI West and Peary-Sverdrop that exhibit low amounts of MYI in 2000 that increase thereafter as they recover from the 1998 melt event (Figures 14 and 16). The loss of MYI from the QEI greatly impacts the icescape of the CAA because it implies increased amounts of FYI within the QEI that will breakup earlier. Relatedly, as MYI begins to reform in these regions, it may not be as thick as pre-1998 ice, thus warming perturbations may make it more susceptible to breakup and increased mobility. Evidence for some of this mechanism at work is realized through the latitudinal flux southward of MYI through the CAA. From 2002 to 2004, the lower-latitude regions of Peel Sound, West Barrow, East Barrow Prince Regent exhibit an increase in MYI while decreases are observed in the QEI regions of Norwegian Bay, Penny, Jones Sound (Figures 15 and 16). The loss of MYI in Perry-Sverdrup and KCI in 2003 likely corresponds to increases in MYI for Penny, Wellington, and MacDougall for 2004 (Figure 16).

[35] It is apparent that the spatiotemporal variability in ice melt impacts the distribution and transportation of MYI within regions of the CAA. Southerly regions in the CAA are prone to increases in MYI from the Polar Pack and/or the QEI. This is particularly evident in the M'Clintock and Viscount-Melville regions that are acting as a drain trap to MYI (Figure 17). This result addresses a popular misconception whereby summertime reductions in sea ice concentration and coverage within the CAA will not necessarily ease ship navigation risk because these regions appear susceptible to increases in MYI.

5. Conclusions

[36] This study presented the development of a QuikSCAT sea ice melt algorithm applicable to landfast FYI and MYI within the CAA. From 2000 to 2004 the algorithm estimated MO, PO and drainage melt states for the CAA. Spatial autocorrelation and regression analyses results point out the extreme spatiotemporal variability of sea ice melt timing transitions within the CAA. Some years did yield significant positive or negative spatial autocorrelations that were spatially linked with radiative forcing data. Specifically, the years that exhibited both positive and negative spatial autocorrelations indices for QuikSCAT-detected melt stages and radiative forcing suggest competition between the forcing parameters. The variability of sea ice melt coupled with varying radiative forcing makes it difficult to isolate casual relationships between the two; hence spatial regression prediction models were not consistent. Other variables (i.e., snow thickness) need to be considered to better isolate melt-forcing relationships within regions of the CAA.

[37] When examining the interplay between ice thermodynamics and dynamics we illustrated that more southerly regions of the CAA have received annual increases in MYI over the past 5 years. This further substantiates the notion that low ice concentrations and early melt transitions within the CAA make certain regions subject to increases of MYI [e.g., Falkingham et al., 2002; Melling, 2002; Howell and Yackel, 2004]. Therefore caution should be taken with regard to operational (and subsequent economic) implications of the warming perturbations predicted by GCMs especially with regard to the CAA and utilization of the NWP. GCMs may predict lighter ice conditions because of earlier break up and later freezeup; however, this may also bring about an enhanced flux of MYI into the CAA, resulting in only a minor lengthening of the shipping season.


[38] The authors wish to thank Steve McCourt (Canadian Ice Service) for providing the Canadian Arctic Archipelago subregion files and Stefania Bertazzon (University of Calgary) for contextual information about spatial statistics. We also would like to thank all field personnel who participated in C-ICE2000, C-ICE2001, and C-ICE2002 and CASES 2003–2004. This research was supported by the Natural Sciences and Engineering Research Council (NSERC) PGS-B grant to S. Howell and a NSERC Discovery grant and Cryospheric System in Canada (CRYSYS) project (B. Goodison, PI) grant to J. Yackel. Appreciation is extended to the Polar Continental Shelf Project (PCSP) and the Northern Scientific Training Program (NSTP) for their technical, logistical, and financial support. We also wish to thank the two anonymous reviewers who greatly improved the quality of this manuscript.