Melt pond covered sea ice is a ubiquitous feature of the summertime Arctic Ocean when meltwater collects in lower-lying areas of ice surfaces. Horizontal transects were conducted during June 2008 above and below landfast sea ice with melt ponds to characterize surface and bottom topography together with variations in transmitted spectral irradiance. We captured a rapid progression from a highly flooded sea ice surface with lateral drainage toward flaws and seal breathing holes to the formation of distinct melt ponds with steep edges. As the mass of the ice cover decreased due to meltwater drainage and rose upward with respect to the seawater level, the high-scattering properties of ice above the water level (i.e., white ice) were continuously regenerated, while pond waters remained transparent compared to underlying ice. The relatively stable albedos observed throughout the study, even as ice thickness decreased, were directly related to these surface processes. Transmission through the ice cover of incident irradiance in the 400–700 nm wave band ranged from 38% to 67% and from 5% to 16% beneath ponded and white ice, respectively. Our results show that this transmission varied not only as a function of surface type (melt ponds or white ice) areal coverage but also in relation to ice thickness and proximity to other surface types through the influence of horizontal spreading of light. Thus, in contrast to albedo, this implies that regional transmittance estimates need to consider melt pond size and shape distributions and variations in optical properties and thickness of the ice cover.
 The positive net surface energy balance of the sea ice in the Arctic Ocean during summer results in the melting of snow and ice, and some of the meltwater collects in lower-lying areas of the ice surface. Enhanced absorption of solar radiation in these flooded areas creates a feedback that further promotes localized melting, eventually forming melt ponds. Ice elevated above the melt ponds, however, continually drains as melt progresses, thereby maintaining a quasi-stable high-scattering surface layer (as liquid inclusions drain and are replaced by void spaces) with a white granular appearance [Perovich et al., 2001]. These processes result in the typical mixture of melt ponds and white ice that characterize the surface of the summertime Arctic sea ice environment [e.g., Fetterer and Untersteiner, 1998].
 Melt ponds emerge shortly after the onset of ice melt that typically occurs in late May or during June in the Arctic Ocean [Markus et al., 2009]. Initially pond surface fractions increase rapidly, briefly reaching as high as 80% coverage on level first-year sea ice typically within weeks [e.g., Langleben, 1971; Barber and Yackel, 1999; Tschudi et al., 2008]. Particularly on level ice, this period is marked by large variation in pond coverage due to diurnal freeze-thaw cycles. Thereafter, as ponds deepen and develop more distinct boundaries, the coverage decreases and stabilizes at 10–40% [Fetterer and Untersteiner, 1998; Tschudi et al., 2008]. On multiyear sea ice, pond coverage is typically below that of first-year sea ice by being confined to depressions between mounds or hummocks with high topographic relief [Eicken et al., 2002; Perovich et al., 2002].
 The disparity of global circulation model-predicted decreases in sea ice extent and concentration with observations of more rapid changes is likely contributed to by feedback mechanisms associated with the interaction of solar radiation with melt ponds that are not appropriately parameterized in current models [Pedersen et al., 2009]. Melt pond formation results in a sharp decrease in surface albedo and an increase in solar radiation absorption in the sea ice and surface seawaters, thereby accelerating sea ice melt, breakup, and the transformation from a domain of ice-related biological production to a marine pelagic domain [Inoue et al., 2008; Tremblay et al., 2008]. The increased transparency of the ice cover to photosynthetically active radiation (PAR) together with enhanced stratification by ice melt may also trigger an under-ice phytoplankton bloom [Mundy et al., 2009].
 In spite of the recognized and increasing importance of ponded sea ice in shaping the light environment in the Arctic Ocean, transmission of solar radiation through melt pond covered and melting seasonal sea ice has not been extensively documented. One of the first spectral transmittance measurements were conducted under melting landfast first-year sea ice by Maykut and Grenfell . Light et al.  used an in-ice optical probe to obtain spectral transmission estimates for bare and ponded multiyear ice and snow covered seasonal sea ice (no ponds) in the Beaufort Sea, while Ehn et al. [2008b] used a similar approach for melting landfast first-year ice in the western Hudson Bay. Both studies reported ice attenuation coefficients similar to those reported by Grenfell and Maykut  but with less wavelength dependence. For multiyear ice, Perovich  inferred from albedo and ice thickness measurements—through the use of a radiative transfer model—that about half of the solar radiation transmitted to the Arctic Ocean during the melt period from June to mid-August was through the sea ice (about 95% ice coverage) out of which about 40% was through melt ponds (about 20% surface fraction). Recently, Nicolaus et al.  collected a comprehensive time series of spectral transmittance at one fixed location beneath multiyear ice as it drifted across the Arctic Ocean during the summer of 2007. No consideration has typically been given to horizontal variability in ice properties (e.g., melt ponds and white ice patch sizes), and transmittance results have been point measurements under specific locations and analyzed under the assumption of horizontal homogeneity [Buckley and Trodahl, 1987].
 Much of the interannual and seasonal variability associated with freezing and melting is due to seasonal sea ice, and as multiyear ice continues to decline with the changing climate, this variability is expected to increase [e.g., Maslanik et al., 2007]. Detailed investigations that aim to improve our understanding of processes that govern the seasonal evolution of sea ice and snow covers are therefore a high priority in polar research. As a part of a diving program during the International Polar Year (IPY) Circumpolar Flaw Lead (CFL) System Study, we measured spectral light transmission along transects beneath pond-covered, landfast sea ice in Darnley Bay [Mundy et al., 2009] and Franklin Bay, two adjacent coastal bays in the Canadian Beaufort Sea (Figure 1). Here we provide insights into how the melting of landfast sea ice progressed toward breakup and how the propagation (vertical and lateral) of shortwave solar radiation in the sea ice cover was affected by ice thickness and surface characteristics such as white ice freeboard and melt pond depth and width.
2. Material and Methods
 Under-ice diving observations were performed on six occasions at four different sites located on the landfast ice about 1 km from the ice edge; the first site in Darnley Bay was on 8, 11 and 13 June 2008 (69.83°N, 123.62°W), the second site on 16 June 2008 in Franklin Bay (69.95°N, 125.86°W), the third site on 18 June 2008 in Darnley Bay (69.82°N, 123.65°W), and finally the fourth site on 21 June 2008 in Franklin Bay (69.97°N, 125.90°W) (Figure 1). All measurements were conducted between 11:30 and 14:00 Local Apparent Time. The melt progression of the sites was almost identical, so it was reasonable to assume that premelt conditions at the sites had been similar in terms of ice formation, structure and thickness. At the time of our sampling, the ice cover had reached an advanced stage of melt with the surface area partitioned into melt ponds and bare white ice. The term “white ice” is used here following, e.g., Maykut and Grenfell , for areas of ice elevated above the melt pond surface level such that their melting surface layer had drained and thus were porous and highly scattering. The first three sampling dates in Darnley Bay were performed at the exact same location with dives through a hole prepared through the ice on 7 June. This hole was located in the center of a large white ice area and did not appear to influence water flow in the adjacent melt ponds. For the latter dives, new sites were selected each time as the ice edge was progressively breaking up previous study sites. On these occasions we utilized naturally occurring ringed seal (Phoca hispida) breathing holes for under-ice access.
2.1. Under-Ice Transmittance Measurements
 At a distance >5 m from the dive holes, 22–24 m transect lines were set up. These were designed to cross a representative melt pond in a north to south direction, from one white ice mound to another. At each end point of the transect line, wooden poles of known length were inserted through vertically drilled holes in the ice. Thereafter, ropes marked at 1 m intervals were tightened between the pole ends, first below the ice by divers and then above the surface. Once the ropes were tight and holding the poles in place by tension, the poles were adjusted so that known distances of the poles protruded above the seawater level.
 Each experiment started with measurements of transmitted spectral downwelling irradiance, Ed(λ), using a Satlantic, Inc., HyperOCR spectroradiometer (350–800 nm spectral range, with a 3.3 nm sampling interval, 10 nm spectral resolution and 0.3 nm spectral accuracy) at 1 m intervals along the under-ice transect line by a SCUBA diver. The spectroradiometer was part of a free-falling optical profiler package manufactured by Satlantic, Inc., that also recorded instrument tilt and depth. The diver taking the measurements along the line used a semi-closed circuit rebreather with 60% O2 in order to minimize exhaled bubbles, which tend to collect in cavities at the ice bottom or to penetrate into the ice cover, where they scatter light, thereby impacting light propagation. A reference sensor was installed above structures on the surface to monitor changes in the incident spectral irradiance, Es(λ). The spectral transmittance was calculated as
where zrope denotes the depth of the rope. This formula thus represents the fraction of incident downwelling spectral irradiance reaching the level of the rope. At the end of the transect lines, the free-falling spectroradiometer was released to record a vertical irradiance profile to a depth of 50–60 m. The profiles collected in Darnley Bay are shown by Mundy et al. . After completion, the diver then measured the distance from the rope to the underside of the ice with a ruler. This distance ranged from 0 to 0.78 m. Together with measurements of the distance from the above-surface rope to the ice surface or melt pond bottom, it was possible to calculate ice draft and freeboard at each mark along the transect lines. On 16 June, a small ice ridge, not detectable from the surface or from a distance under the ice due to low visibility on this day, was included within the transect line. As the rope was tightened between the poles, it came in contact with the ridge. Therefore, to calculate the bottom topography, the ice thickness at the known contact location was measured after the dive and used in the estimation of ice draft at other positions along the transect line.
2.2. Corrections for Light Attenuation in Under-Ice Water
 The irradiance measurements were made at a distance from the sea ice bottom (Table 1). To estimate the transmittance at the ice-ocean interface, it was necessary to account for the optical attenuation in the water layer located between the radiometer and the ice bottom (Figure 2). The thickness of the water layer varied both along and between transects, and the optical properties changed markedly during the experiments. A visibly noticeable feature immediately beneath the sea ice cover was the formation of a ∼0.5 m thick meltwater layer with a sharp interface to the underlying seawater column. On 16 June, and to a lesser extent on 18 and 21 June, this meltwater layer was influenced by waters rich in colored dissolved organic matter (CDOM), probably originating from the Horton and Hornaday rivers (∼25 km from the study site). The influence of these waters was observed through a strong decrease in transmittance at lower wavelengths and a shift in peak transmittance toward longer wavelengths on these dates.
Table 1. Mean Values for Parameters in Figures 2, 5, and 7, With Minimum Photosynthetically Active Radiation (PAR) Transmittance Values for White Ice in Parentheses and Maximum Transmittance Values for Melt Ponds to Show the Effect of Melt Pond Edgesa
8 June Site 1
11 June Site 1
13 June Site 1
16 June Site 2
18 June Site 3
21 June Site 4
Transmittance values obtained with the edge spread function fitting are in between the mean and the minima/maxima (see Figure 7).
[μmol s−1 m−2]
[μmol s−1 m−2]
 We opted here to apply a simple optical model to account inasmuch as possible for the attenuation in the water layer between the ice and the radiometer. The transmittance at the ice-ocean interface T(zio) was calculated as:
where zrope and zio are the depths of the rope and the sea ice bottom, respectively, at location x along the transect. Ksw(λ) is the diffuse attenuation coefficient for the water layer and was expressed as:
 Here, μd is the average cosine of the downwelling light field, and a(λ) is the spectral absorption coefficient, with subscripts “sw,” “ph” and “cdom” denoting pure seawater, phytoplankton and CDOM, respectively. The cosine μd was set to a constant value of 0.8 based on modeling developed by Ehn et al. [2008a] and on comparisons with vertical irradiance profiles in the water column obtained at the end of the transect lines. Equation (3) neglects a contribution by scattering in the seawater, although in the seawater layer it can be considered small compared to absorption and negligible compared to scattering in the sea ice. Pure seawater absorption was taken from Smith and Baker , and aph(λ) was calculated as 0.06 Aph(λ) [chl]0.65 [Morel and Maritorena, 2001], where the normalized absorption coefficient Aph(λ) and the phytoplankton chlorophyll-a concentrations [chl] were obtained from studies by Mundy et al.  and Hop et al. . Measured [chl] for the meltwater layer was 0.40 mg m−3 on 8 June, 0.33 mg m−3 on 11 June (used also for 13 June), 2.25 mg m−3 on 16 June, 1.26 mg m−3 on 18 June, and 1.33 mg m−3 on 21 June. Finally, acdom(λ) was calculated as the exponential curve, acdom(440) exp[−0.018 (λ − 440)], where acdom(440) was estimated from the spectral shape of T(zrope) with the assumption that the melting and continually draining sea ice contained no CDOM and thus that all CDOM was in the seawater layer. Any additional absorption by detrital matter was assumed to be incorporated into the CDOM absorption coefficient. The adopted acdom(440) values were 0.01 m−1 for 8–13 June, 0.5 m−1 for 16 June, 0.3 m−1 for 18 June, and 0.1 m−1 for 21 June.
2.3. Characterizing the Effect of Horizontal Spreading on Transmittance
 An edge spread function (ESF) is often used to assess the spatial resolution of an imaging system by expressing its response to a sharp edge [e.g., González and Woods, 1992]. The edge in our case is the white ice/melt pond edge where the white ice surface layer strongly attenuates radiation, mainly through scattering, while the melt pond water is weakly attenuating, resulting in substantially higher transmission through the melt pond covered ice. As radiation propagates within the ice cover, it is subject to scattering that redirects photons in new directions, thereby reducing the ability to detect the sharp melt pond edge from below the ice cover. To statistically characterize the lateral component of the radiation that was transmitted through the white ice and melt pond covered sea ice, we applied a modified ESF whose derivative produces a Gaussian point spread function across each side of the melt pond:
where Tmp and Twi are characteristic transmittances for the melt pond and the white ice, respectively; erf is the error function; x0 is the center of the melt pond; is the distance from x0 to where T(x) = (Tmp + Twi)/2; and σ is the standard deviation describing the average distance that radiation spreads laterally across the white ice/melt pond edge before emerging at the ice bottom. The above five unknowns were obtained by fitting the ESF to the ice-ocean interface transmittance transect T(zio, x), obtained using equation (2), at a specific wavelength and then integrating over a wavelength range. Thus, x0 ± coincides with the locations of the melt pond edges (as seen from the bottom), 2 equals the melt pond width, and a distance of about 3σ indicates how far from the edges it is necessary to sample in order to obtain representative transmittance measurements that are minimally influenced by adjacent surface types.
2.4. Spectral Surface Albedo
 The shortwave spectral albedo for wavelengths between 350 and 2200 nm was measured at numerous representative melt pond and white ice locations nearby and at the diving sites, both shortly after completion of under-ice light recordings as well as on selected days without dives. Spectral albedo was calculated as the ratio between upwelling and downwelling irradiance spectra that were measured consecutively with a cosine collector of a FieldSpec FR spectroradiometer (Analytical Spectral Devices, Inc.) [Ehn et al., 2008b].
2.5. White Ice Temperature and Salinity
 Sea ice temperature and salinity profile measurements were conducted in Darnley Bay at white ice locations in close proximity to the diving sites but typically a day before or after dive recordings due to time constraints. Ice cores were excavated using a MARK II ice corer (Kovacs Enterprises, Inc.). Interior ice temperatures were measured at selected depths immediately after the retrieval of ice cores by drilling a 2 mm diameter hole into the center of the core and inserting a temperature probe (Traceable Digital Thermometer, Model 4000). The minimum temperature was recorded after readings had stabilized because both the air and surface temperatures were higher than the interior ice temperatures. An additional core was cut with a handsaw into 5 cm sections that were then placed, as quickly as possible, into airtight buckets for melting. Salinity was determined from the ice melt using a handheld conductivity meter (WTW 330i). Salinity of sea ice is difficult to measure accurately in melting conditions since brine tends to drain immediately as the ice core is removed from the ice cover [Ehn et al., 2008a]. For this reason, our salinity measurements may be considered as low estimates.
3. Results and Discussion
3.1. Characteristics and Development of the Melting Ice Cover
 Melt ponds had formed and the snow cover had fully melted on the landfast ice in Franklin Bay and Darnley Bay by the beginning of June 2008. The relatively low reflectance signal from the landfast ice in Figure 1 shows that the pond coverage was extensive throughout the region. Horizontal surface meltwater flow was evident before and at the beginning of the CFL diving campaign, as channels connecting melt ponds had formed and with meltwater flowing toward cracks and holes originally formed by seals. The presence of seal breathing holes on an otherwise level ice cover may thus significantly affect the melt pond coverage [Holt and Digby, 1985]. Nevertheless, surface flow requires that the melt pond water surface (i.e., hydraulic head) is above the seawater level [Eicken et al., 2002, 2004]. For this to occur, ablation of bare white ice surface layers that are above the seawater level must exceed vertical percolation of meltwater through the ice cover. The hydraulic head will then be established by a balance between the two aforementioned processes as well as horizontal flow of surface meltwater into and out of an area. Additionally, mass loss from the ice surface will raise the ice cover to achieve a buoyancy-determined isostatic equilibrium thereby promoting additional meltwater flow.
 Profiles of the surface and bottom topography for the four diving sites are summarized in Figure 2 together with PAR collected along the rope below the ice (see also Table 1). On 8 June, the pond depth was unfortunately not measured. However, since relatively strong east-west meltwater flow was observed toward a seal hole located about 30 m away along a channel, we adopted 0.03 m as a constant value for the pond hydraulic head, which enabled us to estimate the pond depth using the ice topography measurements and knowledge of the melt pond edge locations. This value was taken from the highest value observed on 11 June (Figure 2b). By 13 June the water level in the ponds had dropped to the level near that of the seawater, and consequently horizontal flow was not evident during the rest of the diving campaign. During these first days, a transformation thus occurred at site 1 from 1) a flooded melt pond stage, when meltwater on top of the surface contributed to the weight of the ice cover, to 2) a drained melt pond stage, when meltwater obtained hydrostatic equilibrium with the seawater, thereby ceasing to affect the weight of the ice cover. Large diurnal variations in melt pond water levels, coverage, and flow were noticeable prior to this transformation with maximum values reached in the afternoon. This variability was likely produced first by a diurnal freeze-thaw cycle prior to substantial lateral flow of meltwater, then by the development of a distinct pond surface topography that was influenced by flow rates. Diurnal variations became largely absent as ponds and channels were confined to established surface depressions and connected to the underlying ocean via large drainage channels. A similar evolution has been discussed in detail by Eicken et al.  and Perovich et al. .
 In response to loss of sea ice melt mass from the ice surface via drainage, the ice cover rose with respect to the seawater surface level during the study, i.e., the ice freeboard increased. On 8 June the freeboard on white ice was 0.07 ± 0.02 m. On the southern side, 18 m into the transect line, a 1 m stretch of the white ice surface appeared slushy, as if in the early stages of forming a melt pond similar to those observed by Ehn et al. [2008b]. On 11 and 13 June the white ice freeboard had increased to an average of about 0.16 m, even as the transect-averaged ice thickness decreased from 1.24 m on 8 June to 1.13 and 1.10 m on 11 and 13 June, respectively (Figures 2a–2c). The previously slushy area was also indistinguishable from its surroundings. A decrease in ice thickness would lead to a decrease in freeboard, unless accompanied by significant changes in surface properties. Therefore, an increase in freeboard would be related to increased porosity (lower density) due to drainage of melted ice and smaller amounts of meltwater remaining above the seawater level. Our transect lines do not represent a sufficiently large sample of the area to explain this change in the ice cover's isostatic balance. We note, however, that melt pond coverage in the area decreased significantly from 8 to 11 June, including an observed pond width reduction from about 9 to 7 m (Figures 2a–2b).
 As of 11 June, melt pond boundaries were distinctly marked by steep and, at places, overhanging walls such as those presented in Fetterer and Untersteiner . These small-scale features were not captured by our 1 m horizontal measurement resolution (Figure 2). The white ice freeboard along the transect lines remained at 0.16 ± 0.05 m until 21 June, at site 4 (Figure 2f), which had a smaller white ice freeboard of 0.13 ± 0.06 m, perhaps indicating more porous ice at this site. The measured melt pond depths were about 0.11 m on average on 8 June and decreased to about 0.06–0.07 m on 11–13 June (Figures 2a–2c). On 16 June (Figure 2d), average melt pond depths were 0.20 m, which was the deepest recorded during the dive campaign and coincided with the thickest ice cover, while on 18 and 21 June (Figures 2e–2f) the depths were down to 0.11 and 0.13 m, respectively. Increased absorption of solar radiation results in more rapid bottom ice melting beneath melt ponds compared to white ice and the formation of dome-like elevations on the ice underside. The bottom topography revealed that these domes were not as distinct as the melt ponds but were broader (Figure 2), probably reflecting the lateral spreading of radiation, incident on melt ponds, across pond edges (see section 3.2.4). We also noted that the ice-ocean interface of the melting ice was not completely smooth, as is commonly the case in growing sea ice, but showed a pattern of what can be characterized as pockmarks. These pockmarks were shallow (less than 0.05 m) and round-shaped with diameters on the order of 0.1–0.2 m. The similar size and distribution with brine channel spacing at the ice bottom reported by Mundy et al.  leads us to speculate that these pockmarks were related to brine channels and that their structures, once formed, were accentuated by under-ice turbulence.
 The sea ice in Darnley Bay melted rapidly during the dive program. Ice temperatures were highest at the surface and the bottom (Figure 3a). Salinities were ∼0 above the water level, and decreased from as high as 6 to about 3 within the inner parts of the ice as melt progressed (Figure 3b). The lower temperatures remaining within the inner parts of the ice would have caused refreezing of the fresher surface meltwater, hindering vertical drainage through the white ice. Thus, it was likely that meltwater from white ice would flow predominately in horizontal directions toward ponded areas. We did not measure temperature and salinities in the ice under the melt ponds, but given the higher solar energy absorption and the more advanced stage of melt of the melt pond ice compared to white ice, it can be assumed that temperatures in the ice below melt ponds were higher than those shown in Figure 3 for white ice, resulting in higher porosity. It is therefore probable that a larger portion of meltwater pathways through the sea ice would pass through the more permeable pond ice. The flow of warmer surface meltwater would further promote additional melting in melt pond ice above that caused by direct solar absorption. The salinity of the melt pond water increased from ∼0 to 1 on 8–16 June (sites 1 and 2), to 5.1 on 18 June (site 3), and to 2.5 on 21 June (site 4). This implies that some seawater was percolating upward through the ice cover (perhaps through larger seal breathing holes) even as meltwater was forming at the ice surface.
 Ice breakup in the two bays progressed gradually southward by waves breaking the ice cover along melt ponds and channels, where the ice cover was thinnest and weakest, and forming what were essentially white ice floes. Any meltwater present on these floes quickly drained past floe edges, leaving behind a white high-scattering surface. As ice floes broke off from the landfast ice edge, they were transported away by winds and currents, as shown in Figure 1. Both bays were completely ice free by the end of June.
3.2. Light Interactions With the Melt Pond Covered Sea Ice
3.2.1. Spectral Albedo
Figure 4 summarizes the spectral shortwave albedos for the duration of the diving program for both white ice and melt pond surface types. Maximum albedo values were observed at 500–570 nm for white ice and 450–500 nm for melt ponds. Through integration between 400 to 700 nm, PAR albedos were calculated to be 0.70 ± 0.06 and 0.22 ± 0.04 for white ice and melt ponds, respectively. Similarly, the total shortwave albedos were 0.57 ± 0.05 and 0.15 ± 0.03, respectively. These values were in agreement with those observed by Grenfell and Perovich  off Point Barrow, Alaska, during the same time of the year and under comparable ice conditions.
 Interestingly, white ice and melt pond albedos during the diving program were very stable, showing no significant temporal trends. This can be explained by our observations, which indicated that as the ice cover became less massive due to meltwater drainage and rose upward with respect to the seawater level, the white ice surface layer of the mounds was continuously regenerated and remained highly scattering even as the ice thickness decreased [e.g., Maykut and Grenfell, 1975; Ehn et al., 2008b; Light et al., 2008]. Similarly, scattering within melt pond water remained relatively constant and low compared to the underlying sea ice even as its thickness changed. Therefore, melt pond albedos were controlled by the underlying sea ice and not by the depth the melt pond water. Typical albedos can thus be based on ice surface type alone, and regional albedos can be estimated from observations of the surface fractions of the ice surface types, as suggested previously by, e.g., Perovich et al. .
3.2.2. Surface Incident Irradiance
 An example of the spectral incident irradiance Es(λ) is shown in Figure 4. Within visible or PAR wavelengths, the shape of the Es(λ) spectra did not vary notably during the dive campaign from changes in sky conditions. Thus, the temporal change in Es(λ) measured during dive transects is summarized in Figure 5 by the incident irradiance integrated over PAR wavelengths, i.e., Es(PAR). The associated transmitted irradiance Ed(PAR, zrope), at the level of the rope zrope is additionally shown in Figure 5. Sky conditions, affecting irradiance, remained approximately constant during dives, except on 11 June, when the cloud cover was variable. During the dives on 8, 16, and 21 June, the sky was cloud free, while unvarying overcast conditions persisted on 13 and 18 June, yet with the solar disk visible through the clouds. Reflection from the clouds together with direct radiation resulted in short durations of high maximum levels of incident irradiance, which also influenced transmittance levels, particularly below the melt pond (Figure 2b). The solar zenith angles of the incident solar radiation during the dives ranged from 47° to 50°.
3.2.3. Sea Ice Transmittance Along the Transect
 The spectral transmittance T(λ, zrope), i.e., the portion of incident downwelling spectral irradiance transmitted to the level of the rope zrope, illustrates the effect of surface properties and ice thickness (difference between high transmission through melt ponds and low transmission through white ice) on light penetration (Figure 6a). Furthermore, the shapes of the transmittance spectra are influenced considerably by optically active impurities, such as absorption by algae, CDOM, and detrital matter located primarily in the seawater layer between the rope and the ice bottom. Absorbing organic constituents affected the spectra, particularly at wavelengths below about 500 nm and below the melt pond on 16 June, when the seawater layer between the rope and the ice bottom was over 0.5 m thick. Typically, near infrared radiation is assumed to be absorbed within the first few centimeters of the surface and therefore assumed not to contribute significantly to the heating of interior ice and the water column. However, we observed a notable increase in levels of irradiance at wavelengths above 750 nm transmitted through the melt pond on 18 June.
 Acknowledging the limits of our assumption of a homogeneous seawater layer beneath the ice, the effect of seawater, algae, CDOM, and detritus attenuation was removed using equation (2) to obtain the spectral transmittance at the ice bottom T(λ, zio, x). We recommend that future studies measure T(λ, zio, x) directly at the ice bottom to eliminate the need for this correction, although we note that this might be difficult when conducting under-ice surveys using remotely or autonomously operated vehicles at a fixed depth. The ESF (equation 4) was then applied to the spectral T(λ, zio, x) data to obtain Tmp(λ) and Twi(λ) (Figure 6b). For Twi(λ) we considered only the southern portions of the transects because of data availability (Figure 2). The spectral transmission through melt ponds at wavelengths around 470–520 nm ranged from 48% to 50% of the incident irradiance on 21 and 16 June, to 78% on 18 June (Figure 6b). The corresponding Twi(λ) for white ice ranged from 8% on 16 June to 20% on 18 June. However, we point out that observed peak transmittances (Figure 2) were still higher with a maximum value of 82% obtained on 18 June near the center of the melt pond. An increase in values at 750 nm at the ice-ocean interface is noticeable for transmittances through melt ponds in comparison to values measured at zrope (Figure 6a) but not for the white ice, where they remained close to zero. For 18 June, the transmittance at 750 nm almost doubled from 5% to 10% when seawater attenuation was removed. The ESF fitting did not return reasonable values for the low transmittances above 750 nm.
 The transmittance to the depth of the rope zrope along the north-south transects lines are summarized in Figure 2 by the integrated PAR transmittances:
Sea ice PAR transmittances T(PAR, zio), which were first corrected for the attenuation of the seawater layer (see equation 2) and then integrated over wavelength as in equation (5), are shown in Figure 7. Apart from the general difference in magnitude of transmittance under melt pond ice and white ice, an effect of changing ice surface properties on transmitted PAR can be seen by comparing observations for site 1 (Darnley Bay) between 8 and 11–13 June, during which the flooded ice surface evolved into a state where melt ponds had drained. During this process, the ice rose with respect to the seawater level such that the white ice freeboard increased by about 0.07 m and average pond depth decreased from 0.11 m to about 0.07 m. These changes resulted in somewhat reduced transmittance levels below the ice even though the ice thickness had decreased by >0.1 m (Figures 7a–7c). The decrease in transmission as the ice cover rose was due to drainage of ice melt or brine-filled inclusions that subsequently became air-filled voids in layers above the freeboard, thereby significantly increasing scattering, and also due to a decrease in the surface area covered by melt ponds. The melt pond water depth appeared, however, to be of little importance in determining transmittance compared to the much more highly attenuating sea ice below the pond.
 The transmittance maxima seen in transects (Figures 2 and 7) about 1 m north (left) of the center of the melt pond indicate that a direct or highly forward scattered component of the incident radiation propagated through the ice. Reflection from the northern side white ice pond bank may have additionally contributed to the peak values. The transmittance peak is particularly pronounced for the transect on 11 June (Figure 7b), when it coincided with a temporary increase in surface incident irradiance. Conversely, we speculate that the shading from the white ice bank to the south may have caused the relative decrease in transmittance seen on the southern side (right) of the ponds. These features were present on all transects, except on 8 June, when freeboard was low and pond banks were less distinct. If only highly scattered light was transmitted through the ice, we would expect a smooth curve similar to the ESF fitting, with symmetrical decreases from the maximum at the middle of the melt pond. Since the solar zenith angle of the incident solar radiation during the dives was ∼47°–50°, the angle of refraction θr through the melt pond surface was about 35°. Following this direction through the ice to the rope would suggest a northward shift of radiation by zrope tan(θr), or roughly 1 m, relative to the surface, which corroborates our observations. A strong direct component could indicate low scattering, although the high-scattering nature of sea ice implies that most of the radiation propagating through the ice would be subject to scattering interactions. However, the inclusions in melting sea ice that dominate the scattering of light are mainly brine and meltwater pockets and channels with refractive indices very close to that of ice, so that scattering is highly forward directed. The high porosity of the melting sea ice would tend to let air bubbles (with a high relative index of refraction) escape to the surface, thereby leaving brine and ice melt inclusions alone to determine the scattering in ice layers below the seawater surface level.
3.2.4. Influence of Horizontal Spreading on Transmittance
 The smooth increase in transmittance when crossing from white ice to melt pond ice suggests that the adjacent ice type affects the transmission through propagation of radiation in horizontal directions. Thus, transmittance results presented for melt ponds and white ice in this study cannot be said to fully represent a specific surface type without the influence of boundaries. Only on 18 June, when the thickness of the ice beneath the melt pond was less than 0.7 m and the pond width exceeded 10 m, did irradiance levels approach stable values across the melt pond, indicating a limited influence from the white ice sides. For the other sites, transmittances through the melt ponds were representative of typical ponds with characteristic widths observed in the area.
 Variations in transmittance levels between dive sites can also be explained by differences in ice thickness. The thicker the sea ice, the more the light is scattered and absorbed as it passes through the ice. Multiple scattering within the ice cover causes diffusion of the light field and thereby works to lower the overall transmittance, but it does so with a distribution that is more horizontally even. The influence of ice thickness on the horizontal spreading of light can be noted in Figures 2 and 7 by observing the increase in steepness of change in transmittance when transitioning from below white ice toward melt ponds as the ice cover gets thinner. However, the horizontal diffusion of light is not a linear function of ice thickness, since multiple scattering in the ice will also cause the light field to become increasingly downward directed with depth [Buckley and Trodahl, 1987; Ehn et al., 2008b]. The fact that changes in irradiance levels along the transects are collocated so precisely with the distinct edges of the melt ponds indicates that the light field is highly downward directed.
 In general, the ESF fitting (equation 4) closely matches the range of T(zio, x) observations to produce representative melt pond and white ice transmittance transects both for PAR (Figure 7) and spectrally within the PAR range. Effects of measurement errors or outliers, such as the peak in transmittance on 11 June, are reduced by the fitting procedure. The derivative of the ESF describes the change in T(zio, x) when transitioning between surface types. As dT(x)/dx approaches zero, the influence of the adjacent ice type becomes smaller. However, it is important to note that the derivative of the ESF will always become zero at the maximum and minimum, i.e., Tmp and Twi, and thus these values are representative only in the context of the geometrical setting being considered. With the exception of 18 June, the distance below melt ponds where the ESF derivative is zero was short, covering only a single measurement point. Hence, influences from adjacent surface types cannot be ruled out. On 18 June, however, about six measurement points under the melt pond were within the region where the ESF derivative was approaching zero. Therefore, Tmp at this site was representative of the measured melt pond ice type alone, without the influence of neighboring sea ice, and was at the same time significantly larger than other Tmp values observed.
 The ESF fitting reveals a pattern of larger horizontal spread of radiation on the northern side of melt ponds. As with the noncentrally located melt pond transmission maxima (Figure 7), we attribute this to a direct or highly forward scattered component of the incident solar radiation. Differences are illustrated by separate fittings of the north and south sides of the transects (Figure 7). The largest difference was observed on 11 June, when the standard deviation σ (describes the steepness of the slope of the horizontal transmittance distribution in equation (4)) of the northern side was 38% of that for the more steep southern side. The fitting was affected by a 26% higher Tmp value on the northern side and possibly as well by fewer measurement points below the white ice on the northern side because of the way the transect lines were set up. For the other dates, σ from the fitting of the northern side ranged from 51% (8 June) to 64% (21 June) of the southern side, whereas Tmp values were 1% (8 June) to 24% (16 June) higher than the corresponding southern side values. On 18 June, however, the Tmp difference between the sides was negligible as the ice thickness below the melt pond decreased below 0.7 m, the melt pond width was over 10 m, and a comparable number of data points were measured under white ice on both sides.
 As the ice becomes thinner, the path traveled by photons within the sea ice becomes shorter, partly due to fewer scattering events, thereby also resulting in a reduction in horizontal distance propagated before exiting the ice bottom. Up to 81% of the variability in σ in our data set was explained by thickness changes of ice below the melt ponds (Figure 8). Therefore, it is the “solid” ice below the melt ponds that most strongly influences σ, which can be explained by the fact that the high-scattering white ice surface layer, with high optical thickness, would overshadow any scattering from additional ice thickness increase. Consequently, the effect of ice thickness on σ would be weaker under under white ice than it is under melt ponds.
 Local changes in ice thickness become important when estimating a regionally averaged transmission through ice covers. We suggest that, even if the overall pond covered surface fraction remains constant, a larger average transmission will occur through ice covers with larger melt ponds. This is because irradiance levels in the ice directly beneath large ponds are higher than below small ponds that are affected more by their surrounding white ice. Locally higher irradiance levels will then promote more melting, and ice thickness will be reduced and porosities increased over time more below larger melt ponds compared to smaller ones. Thus, we may expect ice covers with relatively larger melt ponds to develop both deeper ponds and more distinct ice bottom domes, i.e., thinner ice cover and therefore higher average transmittance. Moreover, the fact that surface albedo is determined mainly by surface type areal fractions [Perovich et al., 2002] and not by individual size implies that relatively more solar energy is absorbed by an ice cover with numerous small ponds compared to an ice cover with a few large ponds. We speculate that this may result in solar radiation being partitioned so that relatively more melting occurs at the surface than occurs at the bottom as pond sizes decrease.
3.2.5. Irradiance Attenuation in the Melting Sea Ice
 The bulk ice downwelling diffuse attenuation coefficients associated with melt pond ice and white ice transmittances from the ESF fittings, Tmp(λ) and Twi(λ) (Figure 6b), were calculated respectively as:
where hsi is the ice thickness (excluding melt pond water in case of melt ponds), hmp is the melt pond depth, and R is the surface specular reflection set to 0.05 (Figure 9a). The associated bulk PAR diffuse attenuation coefficients were Kmp(PAR) = 0.73 ± 0.18 m−1 and Kwi(PAR) = 1.56 ± 0.12 m−1. Note that we used the fitting for the south side of the transect to obtain Twi(λ). The diffuse attenuation coefficient of melt pond water, Kpond(λ), was calculated as Ksw(λ), with assumption of no CDOM or particulate matter. Kmp(λ) and Kwi(λ) more directly reflect the optical properties of the sea ice by reducing some of the variation due to ice thickness differences seen in the transmittance spectra. However, as described above, any changes in the radiation field between the sea ice surface and bottom have a direct impact on these values. An important factor affecting the radiation field is the proximity to other ice types, which in the context of this study are mostly characterized by the contrast in near-surface properties between white ice and melt ponds. The results shown in Figure 9 thus implicitly incorporate effects from changes in the radiation field but are strictly valid only for the white ice and melt pond geometries encountered during the diving observations (Figure 2).
 The change in PAR transmission between the different dive sites as a function of ice thickness is shown in Figure 9b. If values on 8 June are rejected as outliers, the ice transmittances both for white ice and melt pond ice are well described by the law of exponential decay where the decay constants are Kmp(PAR, Δhsi) and Kwi(PAR, Δhsi) for melt pond ice and white ice, respectively. These constants represent downwelling diffuse attenuation coefficients; they also include and are affected by any changes in surface geometries and optical properties in the ice cover that have occurred together with the thickness change. The high transmittances on 8 June with respect to ice thickness compared to other observations were likely the result of the lower white ice freeboard on that day as the surface was flooded by meltwater (see section 3.1). Thus, by excluding 8 June, Kmp(PAR, Δhsi) and Kwi(PAR, Δhsi) represent sea ice during the drained melt pond stage, when the porosity in the ice below the ponds was sufficient to keep melt ponds in hydrostatic equilibrium with the seawater. Estimated Kmp(PAR, Δhsi) and Kwi(PAR, Δhsi) were 1.27 m−1 and 1.60 m−1, respectively (Figure 9b). Such a parameterization as a function of ice thickness is useful when assessing the portions of incident solar radiation being absorbed in the ice and the underlying seawater, which have direct impacts on ice melting and heating of the water column as well as primary production at times when the ice melt progression is at its most intense stage of the year.
 The average of the bulk spectral diffuse attenuation estimates from Figure 9a (i.e., Kmp(λ) and Kwi(λ)) and those from thickness variations between sites (Kmp(λ, Δhsi) and Kwi(λ, Δhsi)), calculated in the same way that PAR values are calculated in Figure 9b, are shown in Figure 9c together with commonly used spectra of similar magnitude by Grenfell and Maykut  (hereafter GM77). Our bulk attenuation values are on both sides of what was reported for bare first-year sea ice by GM77, larger than white ice with a surface scattering layer and smaller than melt pond ice. Although we note that these GM77 spectra are for interior ice measured at depths between 0.1 and 0.5 m from the ice surface. Interestingly, the Kwi values were in good agreement with GM77's spectral extinction coefficients for the interior portion (below 0.1 m depth) of bare multiyear ice, except for more wavelength dependence at λ > 600 nm in their spectra. Similarly, the Kmp spectra agreed well with GM77's ponded multiyear ice. On the other hand, Kmp(λ, Δhsi) was higher by about 0.5 m−1 compared to the average of the bulk Kmp(λ). For white ice, Kwi(λ, Δhsi) was consistent with bulk Kwi(λ) at λ < 550 nm, however, and showed a greater increase with wavelength at λ > 600 nm. We note that bulk attenuation coefficients in GM77 are calculated based on net irradiance, i.e., using (1-albedo) rather than (1-R) in equations (6a) and (6b) and assuming negligible upwelling at the ice-ocean interface; however, using this approach with our data resulted in unrealistically small or even negative Kmp and Kwi values. This may be related to neglecting potentially significant upwelling irradiance at the ice bottom and thus overestimating the transmitted net irradiance. Our diffuse attenuation coefficients are therefore based on downwelling irradiance. However, for the interior portions of sea ice where the shape of light field does not change much with depth (i.e., within the asymptotic zone), net and downwelling attenuation coefficients should be comparable.
 These differences in attenuation estimates, shown in Figure 9, illustrate the difficulty in obtaining representative transmittance values for melting sea ice covers in the Arctic with typically varying surface properties. We hypothesize that the following two melt-related processes together affect the transmittance through the ponded ice cover and help to explain the high Kmp(λ, Δhsi) in particular. First, as discussed above, an increased width of melt ponds allows for more radiation to pass through the ice below melt ponds without being influenced by the white ice surface layer. The highest transmittance through melt pond ice was observed at site 3 on 18 June when the melt pond was wide enough (over 6 × σ) so that adjacent white ice did not affect Tmp (Figure 7e). During all other dives, melt ponds were not wide enough, and thus adjacent white ice affected the Tmp estimates. The result was a steeper slope in Figure 9b and a higher Kmp(λ, Δhsi) value compared to bulk Kmp(λ) estimates with equation (6a). Since the area covered by white ice was significantly larger than for melt ponds (as is generally true, especially once melt ponds have drained), Twi estimates may be expected to be less influenced by light spreading from melt ponds, which may thus reflect the more equal Kwi estimates in Figure 9c. Second, we hypothesize that the sea ice cover becomes more transparent to light as the melt progresses. This is because sea ice portions with higher absorption due to, e.g., a higher load of particulates will melt faster than their purer surroundings. We speculate that this selective melting, together with flushing of meltwater through the ice cover, will remove impurities from the sea ice. Light scattering in the ice may also decrease as the ice cover becomes more porous and inclusions become larger or merge; however, this effect has never been verified through observations. Removal of particles and a decrease in scattering as a function of ice melting and thickness decrease would thus also tend to increase Kmp(λ, Δhsi) relative to bulk Kmp(λ) estimates.
 If we nevertheless assume that the optical properties of the sea ice below melt ponds did not change over time, i.e., that observed variations in Kmp(λ) were caused solely by differences in the widths of the melt ponds, we can use Kmp(λ) for 18 June in equation (6a) to estimate what the surface type specific Tmp would have been for the other transects with differing hsi and hmp thicknesses. Such Tmp(PAR) estimates are shown by plus signs in Figure 9b. These results indicate decreases in Tmp(PAR), attributable to the influence of the adjacent white ice, by about 3%, 19%, 27%, 30%, and 35% for 8, 11, 13, 16, and 21 June, respectively. These potentially large reductions in Tmp highlight the importance of considering melt pond geometries when estimating transmittance.
 Furthermore, we can similarly assume that light attenuation in white ice below the seawater level (Figure 2) can be represented by Kmp(λ) for 18 June and that white ice layers above the seawater level are the sole cause of higher attenuation in white ice [Maykut and Grenfell, 1975]. This allows for an estimate of Twi(λ, 0 m), i.e., the transmittance at the seawater surface level below the high-scattering white ice surface layer, as Twi(λ, zio) exp[Kmp,18June(λ) dsi], where dsi is the white ice draft. The integrated Twi(PAR, 0 m), shown by crosses in Figure 9b, indicates that 86%–70% of the incident PAR is attenuated within the white ice surface layer and that it varies less as a function of ice thickness compared to Twi(PAR, zio) (i.e., exponential slope of −1.18 m−1 versus −1.60 m−1). This level of attenuation in the white ice surface layer can be considered an upper limit because most likely the diffuse attenuation coefficient for interior white ice is higher than that of ice below melt ponds. We note that Twi(PAR, 0 m) is similar to the surface transmission parameter, i0, defined by Grenfell and Maykut , except that we use downwelling rather than net irradiance (see also Light et al. ). For first-year sea ice with a similar albedo as ours, Light et al.  found i0 > 0.9, while our downwelling PAR estimates range from 0.14 to 0.3 depending on white ice thickness (Figure 9b). For melt pond covered ice, we concur with Light et al.  that i0 for PAR is close to 1; in fact, it is just below 0.95 in our case owing to the specular reflection of 5% at the melt pond surface.
 Further work that combines field experimentation and optical modeling is needed to confidently establish transmittance relationships with ice thickness and surface type descriptions in order to better understand the partitioning of solar radiation absorption in the ice cover versus the ocean and its impacts on the melt progression. Here we have presented only a very limited data set. Future studies could also establish useful empirical relationships that relate Tmp and Twi to ice thickness and surface characteristics (e.g., pond fractions and descriptions of pond size and shape distributions) for sea ice types under various stages of melt, using a combination of aerial surveys [Prinsenberg et al., 1996; Hanesiak et al., 2001] and under-ice surveys of transmitted irradiance with remotely or autonomously operated vehicles to increase the number of data points.
4. Summary and Conclusions
 Our results have shown that PAR transmission through melt ponds was typically about 4 to 5 times higher than through white ice. Tmp(PAR) and Twi(PAR) ranged from 38% to 67% and 5% to 16%, respectively (Figure 7). These measurements resulted in average bulk PAR diffuse attenuation coefficients of 0.73 m−1 for ice beneath melt ponds and 1.56 m−1 for the white ice. The smallest Kmp(PAR) of 0.46 m−1 was measured below the melt pond on 18 June 2008. The maximum transmission of 82% centered on 480 nm was also obtained on the same day below the center of the melt pond. Significant levels of radiation compared to surface incident values were also transmitted through the sea ice cover at wavelengths above the PAR range (Figure 6). As ice thickness decreased below melt ponds, transmission of radiation above 750 nm increased (Figure 6b). Since melt ponds typically cover around 10%–30% of the seasonal sea ice cover during June–July, melt ponds are likely to be as important or more important regionally than the rest of the ice cover in transmitting solar energy to the water column.
 Light transmission through the ice cover was not only a function of surface properties and ice thickness but also of the areal size covered by a particular ice surface type [Perovich, 2005]. The proximity to other surface types on the ice cover was shown to significantly affect transmission estimates due to changes in the light field related to horizontal spreading of light across surface type boundaries. As far as we are aware, such influences by boundaries have not been taken into account in past studies. We are unable at this stage to theoretically asses the importance of melt pond width and edge effects on under-ice transmittance levels without using a 3-dimensional radiative transfer model. Except for those sampled on 18 June, the melt ponds sampled during this study were too narrow (≤9 m wide) to conclusively determine, using the ESF analysis, the distance from pond edges after which transmittance values were unaffected by the adjacent ice type. Consequently, our transmittance results incorporate boundary effects and can be considered representative of the white ice and melt pond surface geometries and the ice thicknesses encountered during this case study. Only at the sampling site on 18 June, when the ice had melted down to thicknesses of less than 0.7 m below melt ponds and of ∼1 m for white ice, was horizontal spreading reduced sufficiently to permit estimates of melt-pond-specific and white-ice-specific transmittances at distances further than 2 m (>3 × σ) from the pond edge (Figure 7e).
 Although horizontal spreading affected light transmission under all surface types, the higher values of transmission through melt ponds compared to white ice generally lead to an increase in light levels under white ice within some distance from the pond edge and, conversely, to a reduction in light levels under melt ponds. Our results therefore implied that point measurements below melt ponds and white ice may not be representative of transmittance for the respective surface type at another location with a differing distance to a boundary or another nearby similar surface type of a different size. Thus, when regional estimates of transmittance for ponded ice covers are attempted based on point measurements, care must be taken when interpreting and applying observations over areas with differing surface type geometries, even if these surfaces have identical underlying thickness distributions and surface type areal fractions. This is particularly true for surface types with contrasting scattering properties such as the white ice and melt ponds in this study. However, we show that the horizontal transmittance distribution (Figure 7) resulting from spreading of light across the melt pond edges is well described by the ESF (equation 4), so that increases in light levels below white ice are compensated by a comparable reduction below melt ponds. This implies that regional transmittance estimates are possible based on characteristic Tmp and Twi values but that these must be obtained separately for melt ponds and white ice with different sizes, shapes, and underlying ice thicknesses. However, once the size of these surface features cross a certain threshold where their widths exceed ∼6 × σ (Figure 7), their characteristic Tmp and Twi ceases to be a function of their size. We expect this to be more common for white ice, which covers a larger portion of the surface. For melt ponds in our study, this threshold was passed only on 18 June when the pond width was >10 m and σ was 0.7 m (Figure 7e).
 In contrast to transmittance, which was strongly affected by ice thickness, the albedo was related more directly to the near-surface ice properties and thus influenced very little by horizontal spreading. The near-surface layer of white ice was continually regenerated as melt progressed, thereby maintaining its high-scattering properties as noted by, e.g., Perovich et al. . This resulted in little variations in the albedo. Similarly, melt pond albedos displayed little variation because of small changes in the optical properties of pond water and the underlying interior ice. The changes in ice thickness likewise produced no discernible effect on the albedo. The implication of this lack of variation is that a regional albedo estimate can be obtained reasonably well for this late-stage melt period from melt pond surface fraction calculations using, e.g., remote sensing imagery [Hanesiak et al., 2001] and representative albedos for surface types. Knowledge of the regional albedo is the requirement for estimations of how much solar energy is absorbed below the surface [Perovich et al., 2002, 2007]. However, in order to calculate the partitioning between absorption in ice and in the ocean, it is necessary to establish what portion of radiation is transmitted through the ice cover.
 We note that the retrieval of ice thickness of melting sea ice during summertime conditions from satellite altimeter freeboard data requires detailed information of both melt ponds and white ice that are very difficult to obtain or estimate. Our limited data set shows no correlation between ice thickness and white ice freeboard. Instead, we found that the white ice freeboard depended on factors such as melt pond coverage and depths, porosity of ice both above and below the seawater level, the rate of ice melt, and whether permeability was high enough for meltwater to be in hydrostatic balance with the ocean. However, an empirical approach to estimate ice thickness or transmittance may take advantage of the typical melt pond stages that are identified in this study and that are in agreement with sea ice melt evolution phases defined by others [e.g., Eicken et al., 2002; Perovich et al., 2002; Nicolaus et al., 2010]:
 1) The flooded melt pond stage commences soon after melt onset. This stage is characterized by intense surface melting that results in surface ponds. A diurnal solar insolation cycle produces large variations in surface melt rates and in fluctuations in the pond covered surface area, particularly on level ice. Combined with low meltwater drainage through the ice, since interior ice temperatures remain sufficiently low to keep the ice porosities low, surface melting (of white ice above the pond water level) results in pond surface that is above freeboard. This provides the forcing for horizontal flow of meltwater toward floe edges or flaws in the ice cover and, consequently, the formation of channel-like and river-like networks on the ice surface. Our diving program started toward the end of this stage with sampling on 8 June.
 2) At some point interior ice porosity increases above a threshold that allows warmer surface meltwater to drain through the ice. This marks the onset of the drained melt pond stage, when pond water levels equal the seawater surface level. During this stage, ponds deepen within the confines of their established borders, and no lateral flow is observed apart from that caused by wind-forcing. Ice thickness continues to decrease through surface and bottom melting. Intrusion of seawater may increase the salinity of the melt ponds, possibly to levels approaching that of the seawater beneath the ice cover. This stage ends either in the breakup of the ice cover or freezeup.
 By using a known sequence of the melt evolution that includes statistics of melt pond sizes, shapes, and areal coverage [e.g., Scott and Feltham, 2010] of the Arctic ice cover, combined with satellite-derived estimates of melt and freeze onset dates [Markus et al., 2009] and thickness [Maslanik et al., 2007], we may be able to establish a basis for future Arctic-wide estimates of albedo and transmittance. We may then be able to evaluate what portion of the incident solar radiation is absorbed in sea ice cover and what portion is transmitted as heat to the ocean and light to be utilized by primary producers.
 This work is a contribution to the International Polar Year-Circumpolar Flaw Lead System Study (IPY-CFL 2008). Funding for IPY-CFL was provided by the Canadian IPY Federal Program Office, the Natural Sciences and Engineering Research Council (NSERC), and the Canada Research Chairs (CRC) programs through grants to D.G.B. Postdoctoral fellowship support was provided to J.K.E. from the Centre National d'Études Spatiales (CNES) and from the NASA Cryosphere Program Award NNX07AR20G to R. Reynolds and D. Stramski, and to C.J.M. from the Fonds québécois de la recherche sur la nature et les technologies (FQRNT). The participation of H. H. in the CFL-project was jointly facilitated by ARCTOS and ArcticNet within the IPY-project PanAME funded by the Research Council of Norway. Thanks to N.-X. Geilfus, J. DeLaronde, and N. Asselin for their help in the field, and to the officers and crew of the CCGS Amundsen for logistical support.