Surface energy budget of landfast sea ice during the transitions from winter to snowmelt and melt pond onset: The importance of net longwave radiation and cyclone forcings

Authors

  • B. G. T. Else,

    Corresponding author
    1. Clayton H. Riddell Faculty of Environment, Earth, and Resources, Centre for Earth Observation Science, University of Manitoba, Winnipeg, Manitoba, Canada
    2. Now at Faculty of Arts, Department of Geography, University of Calgary, Calgary, Alberta, Canada
    Search for more papers by this author
  • T. N. Papakyriakou,

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

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

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

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

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

    1. Clayton H. Riddell Faculty of Environment, Earth, and Resources, Centre for Earth Observation Science, University of Manitoba, Winnipeg, Manitoba, Canada
    2. Now at International Institute for Sustainable Development, Winnipeg, Manitoba, Canada
    Search for more papers by this author
  • S. Rysgaard

    1. Clayton H. Riddell Faculty of Environment, Earth, and Resources, Centre for Earth Observation Science, University of Manitoba, Winnipeg, Manitoba, Canada
    2. Arctic Research Centre, Aarhus University, Aarhus, Denmark
    3. Department of Geological Sciences, University of Manitoba, Winnipeg, Manitoba, Canada
    4. Greenland Climate Research Centre, Greenland Institute of Natural Resources, Nuuk, Greenland
    Search for more papers by this author

Abstract

Relatively few sea ice energy balance studies have successfully captured the transition season of warming, snowmelt, and melt pond formation. In this paper, we report a surface energy budget for landfast sea ice that captures this important period. The study was conducted in the Canadian Arctic Archipelago from 10 May to 20 June 2010. Over the first 20 days of the study, we found that short periods (1–3 days) of increased net radiation associated with low longwave loss provided most of the energy required to warm the snowpack from winter conditions. An extended period of low longwave loss (5 days) combined with the seasonal increase in incoming shortwave radiation then triggered snowmelt onset. Melt progressed with a rapid reduction in albedo and attendant increases in shortwave energy absorption, resulting in melt pond formation 8 days later. The key role of longwave radiation in initiating melt onset supports past findings, and confirms the importance of clouds and water vapor associated with synoptic weather systems. However, we also observed a period of strong turbulent energy exchange associated with the passage of a cyclone. The cyclone event occurred shortly after melt pond formation, but it delivered enough energy to significantly hasten melt onset had it occurred earlier in the season. Changes in the frequency, duration, and timing of synoptic-scale weather events that deliver clouds and/or strong turbulent heat fluxes may be important in explaining observed changes in sea ice melt onset timing.

1 Introduction

In recent decades, the Arctic icescape has undergone dramatic changes. The summer minimum ice extent is decreasing rapidly [Stroeve et al., 2007; Comiso et al., 2008; Parkinson and Comiso, 2013]; ice is becoming thinner [Kwok and Rothrock, 2009], and younger [Maslanik et al., 2011]; the marginal ice zone has become wider during the melt season [Strong and Rigor, 2013]; and the dates of melt onset and freeze-up have shifted, resulting in longer melt seasons [Howell et al., 2009; Markus et al., 2009; Wang et al., 2013]. Many of these changes are linked through various feedbacks, but of particular interest is the migration of melt onset toward earlier dates. Earlier melt onset allows significant increases in solar radiation absorption at the sea ice and/or ocean surface [Perovich et al., 2007], which in turn leads to enhanced ice melt and enhanced energy storage in the ocean. Therefore, understanding the shift toward earlier melt onset dates is crucial for establishing the linkages between sea ice change and global climate change.

Ideally, the processes controlling melt onset would be investigated by widespread and detailed observations of the surface energy budget of sea ice during the winter-to-spring transition. Unfortunately, such studies are extremely rare. The SHEBA project successfully monitored the surface energy budget of multiyear ice over an entire year [Persson et al., 2002], with the melt onset and freeze-up periods described in detail by Persson [2012]. Annual energy budgets for multiyear ice are also available from modeling studies coupled with observations from Russian North Pole drifting stations [Lindsay, 1998; Jordan et al., 1999], but without detailed descriptions of winter-to-spring transitions. For seasonal ice, the only available observations of melt onset appear to be those of Papakyriakou [1999], who reported the surface energy budget for landfast ice over 3 years in the Canadian Arctic Archipelago. Numerous other studies have directly investigated the surface energy budget of sea ice during the melt season [e.g., Wendler et al., 1997; Vihma et al., 2009; Sedlar et al., 2011; Hudson et al., 2013], or the winter season [e.g., Steffen and DeMaria, 1996; Muller et al., 2012], but without capturing the winter-to-spring transition.

Despite the scarcity of direct observations of sea ice melt onset, a general hypothesis of the important energy balance terms has emerged. In most cases, it seems that surpluses in the net radiation balance provide the necessary energy for melt [Papakyriakou, 1999; Sedlar et al., 2011; Persson et al. 2012], which is consistent with observations over high-latitude terrestrial [e.g., Ohmura, 1982] and glacial [e.g., Bennartz et al., 2013] snowpacks. Turbulent heat fluxes are thought to be less important in driving melt onset, however, they are known to contribute significantly to the energy budget of lower-latitude snowpacks under certain atmospheric conditions [Granger and Male, 1978]. Conductive heat fluxes may be an important source of energy to the surface during the winter, but become negligible during the transition to melt [Papakyriakou, 1999; Persson et al., 2012].

The causes of the radiation surplus driving melt onset are debatable, and probably vary spatially and interannually [Maksimovich and Vihma, 2012]. Papakyriakou [1999] reported substantial interannual variability in albedo and melt onset dates due to the presence or absence of snowfall in the early spring period, but most studies have emphasized the importance of periods of increased net longwave radiation driven by synoptic weather events. Since net longwave radiation is positively correlated with clouds, total atmospheric column water vapor, and air temperature [Jin et al., 2006; Raddatz et al., 2013], the passage of cyclones or low-pressure centers are likely to be the key weather systems responsible for driving melt onset. This tendency for clouds to increase available energy at the surface was illustrated by Zhang et al. [1996], who used a radiative transfer model to show that the increase in downwelling longwave radiation associated with clouds generally outweighs the decrease in downwelling shortwave radiation. Correspondingly, Persson et al. [2012] found that warming of the snowpack and eventual melt onset at the SHEBA site was driven by a series of frontal systems that increased net longwave radiation. It is interesting to note that no studies have identified the seasonal increase in incoming solar radiation as the crucial variable triggering melt onset. Instead, synoptic weather events are thought to play the most important role in initiating sea ice melt, either through the role of precipitation on albedo, or via modifications of the net radiation budget.

In this paper, we examine the surface energy budget of landfast sea ice in the Canadian Arctic Archipelago during the transitions from winter to melt onset, and then to melt pond formation. Our objectives are to examine the role of the various energy balance terms, and assess the hypothesis that synoptic atmospheric systems play a key role in the initiation and progression of melt.

2 Methods

2.1 Study Location and Instrumentation

The data for this study were collected from 10 May to 20 June 2010 during the Arctic-ICE (Ice-Covered Ecosystem) campaign, near Resolute Bay, Nunavut, Canada (Figure 1). The main sampling site for the campaign was located approximately 5 km from shore, on landfast sea ice with snow depths initially ranging from 12 to 30 cm and ice thickness ranging from 1.4 to 1.5 m.

Figure 1.

Map of the study area. The red dot is the location of the field camp (74.708°N and 95.250°W).

A tall (4.5 m) tower at the site (Figure 2) supported the atmospheric sensors, including a propeller anemometer (RM Young 05103) at 4.5 m height; a temperature/relative humidity probe (Vaisailla HMP45C212), sonic anemometer (Campbell Scientific CSAT3), and open path CO2/H2O gas analyzer (LiCor LI-7500) all at 3.9 m height; and a pressure transducer (RM Young 61205V) at 1.3 m height. A shorter (1.3 m) tower supported a Kipp & Zonen CNR1 net radiometer system deployed at 1.3 m height. The CNR1 system is composed of paired pyrgeometers/pyranometers that simultaneously look upward and downward. The system was checked daily to ensure that it remained level and free of frost, condensation, and snow accumulation. During this experiment frost and snow events were rare, but some postprocessing was required to remove affected data. Also at the radiation tower was a thermistor string that measured snow temperatures at 3 cm intervals (from 3 to 21 cm above the snow-ice interface), and ice/seawater temperatures at 10 cm intervals (from 10 cm below the snow-ice interface to 220 cm depth). Thermocouple data were filtered for any obvious effects of solar heating in the near-surface levels.

Figure 2.

Photograph of the energy balance instrumentation. The tower with the eddy covariance and atmospheric monitoring equipment is shown in the foreground. In the background is the radiation tower and the thermocouple strings.

2.2 Definition and Calculation of Energy Balance Terms

Defining a surface energy budget for sea ice is somewhat challenging, because while most energy exchanges do take place at a finite surface (i.e., the interface with the atmosphere), a proportion of solar energy penetrates beyond the surface. Here we follow Persson et al. [2012] (and many other authors who have taken a similar approach) and define the surface energy balance for sea ice including any overlying snow or melt ponds, as:

display math(1)

where Fnet is the net energy flux at the surface, Q* is the net radiative flux, Hs is the turbulent sensible heat flux, Hl is the turbulent latent heat flux, and C is the conductive heat flux. The net radiative flux is composed of the net shortwave (K*) and longwave (L*) components, further composed of incoming (↓), outgoing (↑), and transmitted (T) components:

display math(2a)
display math(2b)

Here we define fluxes that direct energy toward the surface as positive regardless of whether the flux is upward or downward. The exception is the turbulent fluxes, which follow micrometeorological conventions where upward fluxes are defined as positive. This is admittedly a bit confusing since upward turbulent fluxes represent energy flow away from the surface, however, it allows for direct comparisons with previous studies that typically use the same convention.

Of these terms, the incoming and outgoing radiative fluxes were measured directly using the net radiometer system, and the turbulent fluxes were measured directly using eddy covariance. The eddy covariance technique makes use of high-frequency (in this case, 20 Hz) wind velocity and temperature measurements from the sonic anemometer, combined with water vapor measurements from the open path gas analyzer. The fluxes are calculated over an averaging period (here we used 20 min) as:

display math(3)
display math(4)

In these equations, instantaneous measurements are denoted with the prime symbol (′), and the overbars denote mean values over the averaging period. The required inputs are air density (ρ), specific heat capacity of air (cp), vertical wind velocity (w), air temperature (T), molecular mass of water vapor (mv), latent heat of sublimation (Ls, used prior to melt pond formation), latent heat of vaporization (Lv, used after to melt pond formation), mixing ratio of water vapor in dry air (χv), and water vapor concentration (cv). We calculated high-frequency fluctuations in T from sonic temperature using a boot-strapping method that incorporates the water vapor measurements from the open path gas analyzer [e.g., Kaimal and Gaynor, 1991]. The Hs term included in equation (4) accounts for the dilution effect of temperature fluctuations on the open path gas analyzer [Webb et al., 1980; Leuning, 2004]. Prior to solving the equations the three-dimensional wind vector was rotated into a stream-wise coordinate system that sets math formula [e.g., Lee et al., 2004], and “spikes” in the vector and scalar quantities were removed using a simple detection and interpolation technique.

The remaining terms in the energy balance equations (C and KT) could not be measured directly with our instrumentation, and thus required estimation. The conductive heat flux was estimated as:

display math(5)

where k is the thermal conductivity of snow or ice, and ΔTz is the temperature gradient across the snow or ice volume. When snow was present, we estimated ΔTz by linear fit to the snow thermocouples, and used k = 0.3 W m−2 K−1 [Sturm et al., 2002]. Once the snow cover had melted completely, we estimated ΔTz by linear fit to the ice thermocouples, and used k = 2.0 W m−2 K−1 [Persson et al., 2002]. Transmission of solar irradiance through the snow pack and ice was estimated following Persson et al. [2012] as:

display math(6)

where α is the albedo (calculated as K/K), fv and fn are the fractions of shortwave radiation in the visible and near-infrared portions of the spectrum, kxs is the solar extinction coefficient in snow, I0v and I0n are the ice surface transmission parameters for the visible and near-infrared portions, kxiv and kxin are the extinction coefficients in ice for the visible and near-infrared portions, Ds is the snow depth, and Di is the ice depth. We used the values suggested by Persson et al. [2012]: fv = 0.6, fn = 0.4, I0v = 0.95, I0n = 0.37, kxs = 10 m−1, kxiv = 0.72 m−1, kxin = 4.6 m−1. By evaluation of these constants in equation (6), we can see that the fraction of K transmitted to the ocean is very small (typically <1% when snow is present). Therefore, although this transmitted flux term is calculated for the entire snow/ice column, it is still useful conceptually think of equation (1) as a surface energy balance, at least during the snow covered period. Calculated hourly averages of the energy balance components (equation 1) along with ancillary atmospheric data and metadata can be accessed from the Polar Data Catalogue (http://www.polardata.ca). Search for CCIN reference number 11876.

2.3 Error Estimates for Energy Budget Terms

Since the magnitude of Fnet can be quite small, it is important to assess whether it can be reliably resolved over the measurement uncertainty associated with the individual energy balance terms. Ideally, measurement uncertainty can be estimated by the use of replicate instruments at the same site [e.g., Persson et al., 2002; van den Broeke et al., 2004], or by comparing periods of similar atmospheric conditions observed on different dates [e.g., Hollinger and Richardson, 2005]. Unfortunately, limited instrumentation and constantly changing conditions rendered both of these approaches unsuitable for our site. Therefore, we used a literature-based approach for estimating the error associated with the various instruments and techniques that we have employed in this study.

For measurement uncertainty of the CNR-1 system, we adopted the ±2% uncertainty reported for the four radiation components in a similar cold, high-latitude environment by van den Broeke et al. [2004]. This should be a conservative estimate, given that the instruments used by van den Broeke et al. [2004] were left unattended, while our instruments were checked daily for tilt and frost accumulation.

For turbulent fluxes measured by eddy covariance, errors typically increase with flux magnitude, topographical complexity, and surface heterogeneity [Hollinger and Richardson, 2005; Vickers et al., 2010]. In our case, flux magnitudes were quite small, the surface was very flat, and, at least until melt pond formation, highly homogeneous. We therefore expect turbulent flux errors to be small in our data set, justifying the use of the ±20% uncertainty reported by Vickers et al. [2010] for simple surfaces.

Conductive heat fluxes are typically subject to higher uncertainty, due to thermocouple measurement errors, and uncertainties in the thermal conductivity coefficient. Accounting for these sources of error, Papakyriakou [1999] estimated conductive flux uncertainty to be 20–40% through the snow, and 60–70% through ice. Since our conductive fluxes were calculated using a similar approach, we adopted similar uncertainty values.

The most difficult error term to estimate is the transmitted irradiance (KT), since it is a parameterization composed largely of empirical observations. Fortunately, the KT term is quite small (∼1 W m−2 daily average) during the snow-covered period, which is the main focus of this study. Therefore, even large errors in this term would not be reflected as large errors in Fnet, and as such we have not included an uncertainty estimate for this term in our error analysis.

To calculate uncertainty in Fnet, we calculated daily error terms for each of the energy balance components (in W m−2), and then calculated the square root of the sum of squares. Using this approach, daily uncertainty in Fnet was typically around ±10 W m−2.

3 Results

3.1 Atmospheric Conditions

Atmospheric conditions throughout the study period are displayed in Figure 3. Near the start of the experiment on 10 May, daily air temperatures ranged from −15 to −10°C (Figure 3a). A fairly linear increase in temperature followed, with air and surface temperatures first exceeding 0°C briefly on 1 June. Temperatures then returned to slightly below freezing until 6 June, after which they fluctuated diurnally around 0°C, with midday highs consistently exceeding the freezing point. Although snow depths were not measured systematically, the first in situ observations of snowmelt were made on 6 June [Mundy et al., 2014], which we adopt as the date of melt onset. After 6 June, the snowpack rapidly deteriorated, and melt ponds were first observed on 14 June.

Figure 3.

Hourly mean atmospheric conditions observed during the study period: (a) air temperature, and surface temperature derived from the upwelling longwave radiation; (b) wind speed, and wind direction by sector; (c) atmospheric pressure; (d) vapor pressure (ev), and saturation vapor pressure (ev sat) over ice. Red vertical lines are located at the first observations of positive surface temperature (1 June), snowmelt onset (6 June), and melt pond onset (14 June). These lines divide the period into four subsections that are discussed in the text.

Wind velocity (Figure 3b) and atmospheric pressure (Figure 3c) observations show evidence of several distinct synoptic-scale weather events. The start of the experiment was dominated by high pressure, and mostly light, northerly winds. The most intense period of high pressure was 25–30 May, which corresponded to very light winds. This was followed by a series of shallow low-pressure disturbances, which caused periods of stronger winds between 6 and 9 June, culminating in a period of strong southerly winds from 13 to 18 June. Water vapor content (Figure 3d) fluctuated throughout the experiment, but was mostly below saturation. According to observations at the Environment Canada station in Resolute, the only significant precipitation during the study period occurred during a light snowfall event (4 cm) on 7 June.

3.2 Radiative Conditions

Key terms in the radiation budget (equation (2)), and the calculated surface albedo are shown in Figure 4. Incoming shortwave radiation (Figure 4a) followed the expected diurnal pattern, with a gradual linear increase in mean daily irradiance. By the end of the experiment, mean daily K had increased by approximately 50 W m−2. Surface albedo (Figure 4b) varied around a mean of 0.85 for the first part of the experiment, before beginning a gradual decline after air temperatures initially exceeded 0°C on 1 June. Following snow melt onset on 6 June, the albedo decrease became more pronounced, reaching a value of around 0.66 prior to melt pond onset on 14 June. After melt ponds formed, the albedo decreased rapidly, reaching a minimum value of 0.25 at the end of the experiment. Changes in albedo appeared to be the dominant control on K*, which was stable through the early period despite increasing K. After snowmelt onset (and the subsequent decline in albedo), K* increased rapidly, with mean daily values eventually exceeding the premelt values by ∼170 W m−2.

Figure 4.

Hourly mean radiative conditions observed during the study period: (a) incoming shortwave radiation (K); (b) albedo; (c) net shortwave radiation (K*); and (d) net longwave radiation (L*). Open circles are daily means, and any black lines are linear trends significant at >95% confidence interval. Red vertical lines are located at the first observations of positive surface temperature (1 June), snowmelt onset (6 June), and melt pond onset (14 June). These lines divide the period into four subsections that are discussed in the text.

To characterize the longwave radiation conditions, we display only L* (Figure 4d), which essentially mirrors L; L is a function of surface temperature, and only increased slightly (from 300 to 320 W m−2) over the entire study period. Net longwave radiation showed considerable variation throughout the study period, with no discernable trend in daily mean values. During the initial warming period, several 1–3 day episodes of particularly strong longwave loss occurred, indicative of clear-sky conditions. Immediately prior to snowmelt onset (1–6 June), there occurred an extended period of low longwave loss (typically around 20 W m−2). After snowmelt onset there was another period of low longwave loss (10–13 June), followed by a week of generally high longwave loss. For completeness Figure 5a shows net radiation (Q*), which reflects many of the trends observed in K* and L*.

Figure 5.

(a) Net radiation (Q*), (b) Sensible heat flux (Hs), (c) latent heat flux (Hl), and (d) conductive heat flux (C). Open circles in Figure a are daily means, and the line is hourly averages. The turbulent fluxes (Hs and Hl) are displayed as 3 h averages, with standard deviations denoted by the error bars. Conductive flux is displayed hourly. Red vertical lines are located at the first observations of positive surface temperature (1 June), snowmelt onset (6 June), and melt pond onset (14 June). These lines divide the period into four subsections that are discussed in the text.

One deficiency in our radiation measurements is that they were made at a single point, and therefore did not capture any spatial variability in albedo. This is probably not a problem prior to melt pond onset; in fact, the albedo evolution we observed matches almost precisely the findings of Perovich and Polashenski [2012], who compiled a large data set of distributed albedo measurements over landfast ice. However, once melt ponds form, the spatial variability of sea ice albedo increases dramatically. In our case, a melt pond formed directly under the radiation tower, occupying most of the field of view of the downward looking radiation sensors and biasing the measured albedo lower than the mean value for the surface (we measured a minimum albedo of 0.25, while Perovich and Polashenski [2012] suggest that the mean albedo should be 0.32 for recently ponded seasonal ice). Thus, our radiation measurements (and indeed our entire energy budget) are probably only representative of the complete surface prior to melt pond onset.

3.3 Turbulent and Conductive Fluxes

Latent heat fluxes constituted a loss of energy from the surface throughout the entire study period (Figure 5b), consistent with snowpack sublimation prior to melt onset, and evaporation afterward. Prior to snowmelt onset, Hl exhibited a pronounced diurnal cycle, with the strongest energy loss occurring shortly after local noon. During this period of strong diurnal cycles, the mean daily energy loss driven by Hl ranged from 2 to 9 W m−2. This diurnal signal eroded in the week prior to melt onset, and in the following week the magnitude of Hl became quite small. Latent heat fluxes then became quite strong (mean daily values up to 20 W m−2, and hourly values up to 40 W m−2) shortly before pond onset, due to a synoptic-scale event, which we discuss in more detail in section 4.2.

Sensible heat fluxes (Figure 5c) mirrored many of the patterns of Hl, with pronounced diurnal cycles prior to melt onset. However, the flux was not consistently unidirectional during this time period, with energy lost from the surface during the day, and gained by the surface at night. This pattern is consistent with solar warming of the surface during the day resulting in instability, and radiative cooling at night causing a temperature inversion. Mean daily values for Hs during this time period ranged from −7 to 5 W m−2, with an overall daily mean very close to 0 W m−2. As with Hl, the diurnal signal in Hs deteriorated prior to snowmelt onset, and then became dominated by synoptic-scale events. This is consistent with the observations of Persson et al. [2002], who showed that as the snowpack becomes isothermal, energy surpluses and deficits contribute to phase changes instead of temperature changes. The period between 13 and 17 June was particularly remarkable, with sustained sensible heat fluxes directed toward the surface at rates in excess of 25 W m−2 for nearly 48 h.

Conductive fluxes (Figure 5d) also displayed a distinct diurnal cycle, which supplied energy to the surface at night, and removed it during the day. As with the Hs cycle, this pattern resulted from surface warming during the day, and cooling at night. Mean daily C during this time period varied from −7 to 6 W m−2, with an overall mean that was also close to 0 W m−2. As melt onset approached and temperature gradients in the ice and snowpack lessened, the magnitude of daily C fluctuations diminished. Eventually, the snowpack became isothermal and temperature gradients in the ice became very small, resulting in minimal conductive heat fluxes.

3.4 Surface Energy Budget

During most of the month of May the daily energy budget was nearly balanced (Figure 6), with Fnet oscillating between small positive and negative values. As described previously, this time period was marked by distinct diurnal cycles, which are summarized in Figure 7. Net shortwave radiation was the dominant energy source for the surface over the diurnal cycle, with net longwave and latent heat losses balancing most of the surplus. Conductive and sensible heat fluxes also removed energy from the surface during the high solar zenith hours (Figure 7), but contributed energy at night, helping to balance some of the longwave losses. Daily mean values of Fnet ranged from −14 to 10 W m−2 during this time period, and were generally below the overall measurement uncertainty (Figure 6).

Figure 6.

Two representations of the mean daily energy budget terms. In (b) the sign of the turbulent fluxes (Hs and Hl) are reversed from the normal convention so that all sources of energy to the surface are displayed as upward pointing bars (and vice versa). Red vertical lines are located at the first observations of positive surface temperature (1 June), snowmelt onset (6 June), and melt pond onset (14 June). These lines divide the period into four subsections that are discussed in the text.

Figure 7.

Diurnal energy balance prior to melt onset, computed as hourly bins averaged from 17 May to 1 June. Hours are relative to local noon.

Following the first positive surface temperatures (1 June), and prior to snowmelt onset (6 June), there occurred a period where Fnet was distinctly positive (Figure 6), and in excess of the measurement uncertainty. Since this period represents the crucial transition from winter to snowmelt onset, it is discussed in more detail below. For now, it is sufficient to point out that Fnet showed a mean daily surplus of 14 W m−2 (ranging from 8 to 22 W m−2) during this period, and that net longwave losses were notably low throughout.

Between snowmelt onset and melt pond onset, Fnet surpluses became even higher (mean 34 W m−2, range 15–56 W m−2), due largely to the increase in K*. However, when K* was not strong during this period net longwave losses were particularly low, which allowed Fnet to remain high. Sensible heat flux also contributed approximately 12% of the Fnet surplus (4 W m−2 daily mean), which was not the case for most of the period prior to melt onset.

After the transition to melt ponds the energy balance became less representative of the entire surface (as discussed in section 3.2), but some features are still worth reporting. Obviously, the main feature was the rapid increase in K* driven by declining albedo. However, the turbulent fluxes also played an important role over this period, with Hs providing a mean daily input of 12 W m−2. Those exchanges were partially offset by Hl at a mean daily rate of 7 W m−2. At times, Hs contributed nearly 20% of the incoming energy, which was substantially higher than any other points throughout the study period.

4 Discussion

4.1 Snowmelt and Melt Pond Onset Transitions

Based on the above results, we have divided the sampling period into four subsections: a period that is representative of “warming” conditions (11–31 May); a “premelt” period (1–6 June), covering several days of sustained low longwave loss and terminating at the first observations of snowmelt; a “snowmelt” period (6–14 June), which terminates at the first observations of melt ponds; and a “melt pond” period (14–20 June) that extends until the end of the study.

Although energy fluxes were nearly balanced during the warming period, the slightly positive Fnet (mean daily value of 2 W m−2) translates to a total energy surplus for the surface of 3.7 MJ m−2 (Table 1). Over this time period, we observed an increase in mean snowpack temperature of ∼4°C and mean ice temperature of ∼1°C. The required energy for these temperature increases can be approximated via:

display math(7)
display math(8)

where ci is the specific heat capacity of ice (2110 J kg−1 K−1 for fresh ice, and slightly higher for saline ice as per Ono [1966]), ρs and ρi are the densities of snow and sea ice (we assumed 350 [e.g., Langlois et al., 2007] and 910 kg m−3 [Timco and Frederking, 1996], respectively), Ds and Di are snow depth and ice thickness (0.15 and 1.45 m at the sampling site, respectively), and ΔTs and ΔTi are the above noted changes in mean snowpack and ice temperatures. This estimate yields a total energy change (i.e., ΔEsnow + ΔEice) of 3.2 MJ m−2, in good agreement with the observed energy surplus. This good agreement is interesting given that daily Fnet values were typically below our estimate of measurement uncertainty for this time period (Figure 6). This may be the result of errors cancelling out over a long averaging period, or perhaps overestimation of the error terms. If this result is accepted, then we can conclude that the role of excess energy during this period was primarily to warm the snow and ice volumes.

Table 1. Daily Mean Values for Energy Budget Terms Over the Four Periods Defined in the Texta
 Warming (11–31 May)Premelt (1–6 Jun)Snowmelt (6–14 Jun)Melt Pond (14–20 Jun)
  1. a

    Total energy is Fnet integrated with respect to time over the period of interest.

Fnet (W m−2)2.114.336.298.1
Total Energy (MJ m−2)3.77.425.059.3
K* (W m−2)40.448.672.4155.5
K (W m−2)277.0282.8301.0310.1
Albedo0.850.820.740.37
L* (W m−2)−35.0−23.4−33.2−60.6
Q* (W m−2)5.425.239.294.9
Hs (W m−2)−0.40.1−3.2−11.8
Hl (W m−2)4.66.63.76.9
C (W m−2)0.8−4.2−2.5−1.7

The energy balance terms responsible for the warming over this period can be inferred from careful inspection of Figure 6. Interestingly, Fnet was consistently negative on days where K* was high, and positive during days with low K*. In fact, K* was significantly (confidence >95%) negatively correlated with Fnet during this period (R = −0.71). So the excess energy was clearly not derived from the shortwave balance. Instead, it appears to have been driven by L*, which was significantly positively correlated with Fnet (R = 0.94). This implies that warming of the snow and sea ice volume was driven by periods of increased net radiation permitted by low longwave loss, likely associated with cloud cover and/or atmospheric water vapor. The strong correlation of Fnet with L* and anticorrelation with K* has been observed for other high albedo surfaces [Ambach, 1974; Bintanja and van den Broeke, 1996], and supports the conclusion of Persson [2012] that the longwave radiation balance is not only key for sea ice melt transitions [Zhang et al., 1996] and melt maintenance [Sedlar et al., 2011], but also for the period leading up to melt onset. Latent heat losses were also strongly correlated with Fnet over the warming period (R = 0.65), but in this case the causality is probably reversed; during periods of strong Fnet, more energy must have been available for sublimation of the snowpack.

During the 5 day premelt period, the available energy at the surface was 7.4 MJ m−2 – nearly double the energy available in the previous 20 day period. Following equation (7), about half of this energy was necessary to raise the mean snowpack temperature from −4 to 0°C. The remaining energy could have melted approximately 2 cm of snow, which is consistent with observations of melt in the upper layers of a snowpack prior to isothermal conditions [Kuhn, 1987]. Snow grain metamorphosis due to surface melt [Colbeck, 1982] also explains the gradual decrease in albedo that we observed over this period (Figure 4b).

The dramatic increase in available energy during this premelt period is very important. If daily Fnet had remained at the 2 W m−2 level experienced previously, it would have taken another 20 days to achieve melt conditions. Instead, melt conditions were reached in less than a third of that time. Figure 6 and Table 1 once again show that cloud-radiative impacts must have played a key role. Over this period longwave losses were very low (see also Figure 4d), indicating the sustained presence of clouds and atmospheric water vapor. However, K* was notably higher than during periods of low longwave loss in the warming period (Figure 6). This increase in K* was primarily driven by an increase in K; mean K during the premelt period was ∼23 W m−2 higher than on comparable days (i.e., Fnet > 0 W m−2) in the warming period, which can easily be explained by the gradual seasonal increase in K (Figure 4a). Therefore, we can conclude that melt onset at this site was driven by an extended period of low longwave loss, assisted by the seasonal increase in K.

During the snowmelt period, 24 MJ m−2 of energy were available to the surface (Table 1) which is enough energy to melt ∼20 cm of snow. This is medial to the range of snow depth measurements made in the area (12–30 cm), and shows that sufficient energy was available to significantly ablate the snowpack and trigger the onset of melt ponds. The energy budget term most responsible for this energy excess was clearly K* (Figure 6 and Table 1), due primarily to the decrease in albedo (from ∼0.80 to 0.69, Figure 4b) associated with accelerated wet snow grain metamorphosis [Colbeck, 1982; Perovich and Polashenski, 2012]. Nevertheless, cloud-radiative impacts were still important during this period, as the 2 days with the highest Fnet values occurred during a period of low longwave loss (Figure 6). Thus, melt pond onset appears to have been driven primarily by the albedo decline prompted by snowmelt, but was assisted by a weather system that limited longwave losses for several days.

4.2 The Role of Turbulent Fluxes

As Figure 8 shows, from 12 to 17 June a relatively deep (mean sea level pressure <100 kPa) cyclone transited across the Arctic Archipelago from west to east. The low-pressure center missed our study site to the southwest where it stalled and dissipated, but the effects of its passage were strongly evident in our weather and energy balance observations. As the cyclone approached our site in the early hours of 13 June, we observed a dip in pressure, associated with increasing southeasterly winds, rising air temperature, and a widening vapor pressure deficit (Figure 3). The stalling of the low-pressure center to the southwest maintained the strong southeasterly winds until 17 June, with air temperature consistently above 0°C, and vapor pressure consistently undersaturated by ∼0.2 kPa. These conditions created intense sensible heat fluxes (sustained below −25 W m−2 for most of the 4 day period), and weaker offsetting latent heat fluxes (typically above 15 W m−2) (Figure 6).

Figure 8.

Mean sea level pressure maps during the mid-June cyclone event. Values are 3 hourly means centered on local noon. The study site location is identified by the red dot. Data are from the NCEP-NARR reanalysis [Mesinger et al., 2006], plotted using the NOAA/ESRL Physical Science Division (Boulder, Colorado) web site.

Combining the latent and sensible heat fluxes over this period yields a mean flux of energy to the surface of 10 W m−2. Therefore, turbulent fluxes contributed 3.3 MJ m−2 of energy to the system over the 4 day event. The importance of low-pressure systems in influencing turbulent heat fluxes over sea ice has been previously identified [Andreas, 1985], but not to our knowledge in connection with melt onset. Although it appears that turbulent fluxes did not play a controlling role in the transitions to snowmelt and pond onset during our study, an event like this could have significantly hastened either transition had it occurred earlier. The energy associated with the cyclone passage was essentially equivalent to all of the energy input to the system during the warming period, and would have been enough to trigger melt onset by itself during the premelt period. There is some evidence to suggest that cyclones like this are occurring more frequently and with more intensity in the Arctic [Zhang et al., 2004; Sepp and Jaagus, 2011]. If this is the case, then the turbulent heat fluxes associated with such events may be partly responsible for the observed trends toward earlier melt onset dates [Howell et al., 2009; Markus et al., 2009; Wang et al., 2013].

It is also worth noting that prior to melt onset, the diurnal amplitude and daily mean values of Hl that we observed were higher than previous observations in Arctic Ocean pack ice [Nilsson et al., 2001, and references therein; Persson et al., 2002; Sedlar et al., 2011]. Furthermore, Papakyriakou [1999] found even stronger latent heat fluxes at a landfast site close to ours. This difference could be a result of the influence of dry continental air masses, as opposed to the perpetually saturated air masses that persist over pack ice [Andreas et al., 2002]. Drier air would create a stronger vapor pressure gradient, and in turn drive stronger latent heat exchange. As an extreme example of this effect, Granskog et al. [2006] estimated latent heat fluxes up to ∼70 W m−2 over landfast sea ice in the Baltic Sea associated with Foehn winds. The cyclone event that we observed is also a good example of this effect, where the position of the low-pressure center (which would typically be associated with high water vapor) channeled dry continental air over our study site (Figures 8 and 3d). Thus, we can hypothesize that turbulent latent heat exchange may play a more important role in balancing excess radiation over landfast ice than pack ice.

5 Conclusions

Our results support the prevailing hypothesis that melt transitions over sea ice are primarily driven by periods of low longwave loss associated with synoptic weather events. We can also use our results to highlight a few important nuances to this theory, which we will review here.

First, as previously suggested by Persson [2012] and Maksimovich and Vihma [2012], our findings show that the longwave trapping effect of clouds and associated moisture is not only a controlling factor in melt onset, but also in the long warm-up period prior to snowmelt. Thus, we might expect melt onset dates to vary considerably year-to-year depending on the frequency and duration of certain synoptic conditions in the early spring. This conclusion supports the theory that years of extreme sea ice loss are better correlated with cloudy spring conditions [Nussbaumer and Pinker, 2012] than with sunny summer conditions [Kay et al., 2008; Zhang et al., 2008].

Second, our findings show that latent heat fluxes may be a more important energy sink over landfast ice than pack ice. This is probably because air masses advected over landfast ice do not immediately reach saturation, as is thought to occur in mobile pack ice [Andreas et al., 2002]. The sublimation driven by this energy exchange may be significant for snowpack mass balance, which in turn exerts important ecosystem controls via the transmission of photosynthetically available radiation [Iacozza and Barber, 2001; Mundy et al., 2005, 2014].

Finally, we can propose a hypothesis that turbulent fluxes associated with cyclones may have an important impact on melt progression in certain years. During our study, we observed a cyclone that delivered enough energy to the surface to considerably shorten the warm-up period, or potentially trigger melt onset if the snowpack was already warmed. The cyclone we observed occurred after melt had progressed significantly, but had it happened earlier in the season, melt may have been initiated much earlier. Spatial and temporal variability of the frequency, duration, and impact of these events likely plays an important role in melt onset variability over landfast sea ice.

Acknowledgments

The authors wish to thank Bruce Johnson, and the many Arctic-ICE participants who helped with logistics and maintenance of the micrometeorology equipment. Field support for the Arctic-ICE project was provided by the Polar Continental Shelf Program of Natural Resources Canada, and through Individual and Northern Research Supplement Discovery grants from the Natural Sciences and Engineering Research Council of Canada to C. J. Mundy, D. G. Barber, and T. N. Papakyriakou Additional support was provided by Fisheries and Oceans Canada, the Canadian International Polar Year federal program, the Canada Foundation for Innovation, the Canada Excellence Research Chair in Arctic Geomicrobiology and Climate Change (S.R.), and the Centre for Earth Observation Science at the University of Manitoba. This paper is a contribution to the ArcticNet and Arctic Science Partnership programs. We dedicate this work to Klaus Hochheim, who recently lost his life along with two of our Canadian Coast Guard colleagues while conducting sea ice research in the Canadian Arctic.