1. Introduction
The accumulation of snow on sea ice plays an important role in modifying the properties of sea ice in several ways. With respect to sea-ice growth and the surface heat budget, the thermal insulation effect and process of snow-ice formation have mainly been investigated for sea ice that is thick enough to accumulate snow as a dry layer on top (e.g. Holtzmark, Reference Holtzmark1955; Fichefet and Maqueda, Reference Fichefet and Maqueda1999; Wu and others, Reference Wu, Budd, Lyte and Massom1999; Fichefet and others, Reference Fichefet, Tartinville and Goosse2000; Maksym and Jeffries, Reference Maksym and Jeffries2000; Jutras and others, Reference Jutras2016; Sturm and Massom, Reference Sturm, Massom and Thomas2017). On the other hand, the effect on the freeze-up process of thin sea ice, complicated by the formation of slush containing sea water exposed to the air, is not fully understood due to the lack of observations. Snowfall on a wet, thin sea-ice surface can potentially affect the production rate and subsequent crystal alignment through the seeding effect which induces crystallization (Martin, Reference Martin1981; Gow, Reference Gow1986; Weeks and Ackley, Reference Weeks, Ackley and Untersteiner1986; Svensson and Omstedt, Reference Svensson and Omstedt1994). Besides, the addition of ice thickness due to the incorporation of a surface slush layer may reduce the heat exchange with the atmosphere considerably (Maykut, Reference Maykut1978). Although traditionally the effect of snow has been omitted when estimating the sea-ice production rate in polynyas (e.g. Pease, Reference Pease1987; Martin and others, Reference Martin, Drucker and Yamashita1998; Tamura and others, Reference Tamura, Ohshima and Nihashi2008), this decision seems to have been for simplicity rather than based on observational facts and can give rise to large uncertainty in the computation (Spreen and Kern, Reference Spreen, Kern and Thomas2017).
Observational studies on thin ice growth processes have been relatively limited due to logistical difficulties (e.g. Gow and others, Reference Gow, Meese, Perovich and Tucker1990; Perovich and Richter-Menge, Reference Perovich and Richter-Menge2000; Wettlaufer and others, Reference Wettlaufer, Worster and Huppert2000; Granskog and others, Reference Granskog2004), and focused on the relationship between the heat budget and entrapped brine or chemical properties. Although Gow and others (Reference Gow, Meese, Perovich and Tucker1990) did explore the contribution of wind-blown snow from perennial ice to snow-ice formation using a transect measurement across Arctic leads, we still need more detailed data to better understand and quantify the effect of snow. Fortunately, we had the opportunity to observe the effect of snow on the crystal textures and growth of thin sea ice from field experiments at the Saroma-ko Lagoon, Hokkaido, Japan (Fig. 1). To our knowledge, this experiment and resulting data are novel.
Map of the Saroma-ko Lagoon and the experiment site. The locations of the ALMO and the Tokoro automated meteorological station are also shown.
The stability of the field site on the Saroma-ko Lagoon makes it very suitable investigating this issue. Accordingly, there have already been several experiments targeting thin ice and snow, examining its growth rate (Hasemi, Reference Hasemi1974), the salinity evolution associated with snow-ice formation (Takizawa and Wakatsuchi, Reference Takizawa and Wakatsuchi1982; Takizawa, Reference Takizawa1983, Reference Takizawa1984), the surface heat budget on sea ice (Ishikawa and Kobayashi, Reference Ishikawa and Kobayashi1984) and the upward brine migration in sea ice (Kasai and Ono, Reference Kasai and Ono1984). However, few studies addressed the effect of snowfall on the crystal alignment and surface heat budget for thin sea ice. Although Kawamura (Reference Kawamura1982) investigated the evolution of crystallographic orientations in thin sea ice, the effect of snow was not considered. And although similar experiments were conducted at a comparatively stable site in the Baltic Sea (Granskog and others, Reference Granskog2004), their focus was on the behavior of chemical components.
The purpose of this paper is to present a case study that quantitatively estimates the effects of snowfall on ice structure and heat exchange with the atmosphere during the freeze-up process of thin sea ice. These estimates are based on results obtained from artificial pool experiments and a traditional 1-D thermodynamic sea-ice growth model. The model is based on a simple surface heat balance investigation (Maykut and Untersteiner, Reference Maykut and Untersteiner1971; Semtner, Reference Semtner1976; Maykut, Reference Maykut1978) and was used to estimate the effect of snow on ice growth and the individual heat fluxes at the ice surface. Since snow relates partly to the formation of granular ice through snow-ice formation or superimposed ice (e.g. Kawamura and others, Reference Kawamura, Ohshima, Takizawa and Ushio1997), our study seeks to advance understanding about the substructure of granular ice and its impact on ice growth.
2. Observations
The experiment was conducted over two nights from 25 to 27 February in 2019 as part of the Saroma-ko Lagoon Observations for sea-ice Physico-chemistry and Ecosystems 2019 program (SLOPE2019; Nomura and others, Reference Nomura2020). The experiment site was located at 44°07′ N, 143°58′ E, ~200 m away from the eastern shore of the Saroma-ko Lagoon, Hokkaido, Japan (Fig. 1). The dimensions of the square pool made for this experiment were 1.5 m × 1.5 m (Fig. 2a). The depth and salinity of the sea water in the pool were 1.2 m and 31.2 psu, respectively, and it was surrounded by 0.32 m thick land fast ice with 0.09 m of snow. According to the formula of UNESCO (1978), the freezing temperature (T f) is estimated to be −1.71°C from the salinity. With the significant wintertime solar radiation at this latitude (Toyota and Wakatsuchi, Reference Toyota and Wakatsuchi2001), our experiment was limited to nighttime when the sea-ice growth occurred.
Photographs of our experiments at an artificial pool site to examine how meteorological conditions affect sea-ice properties at the early growth stage. (a) Pool site with a dimension of 1.5 m × 1.5 m. Ice conditions were monitored with a monitoring camera. Air and water temperature near the ice surface were monitored with a thermistor at the corner of the pool. (b) Photo showing the setup of our radiometer and AWS. (c) Photo of ice conditions at the sampling time at 09:26 on 26 February and (d) same as (c) except for at 09:38 on 27 February, taken by a monitoring camera in Figure 2a. Note that there is no snow layer on sea ice.
2.1 Meteorological and ice conditions
Air temperature, relative humidity and sea level pressure were recorded to monitor the meteorological conditions and calculate surface heat fluxes during ice growth. Measurements were performed at a height of 1.5 m at 1-min intervals by an Automatic Weather System (AWS, Kestrel manufacturer) on the Saroma-ko Lagoon (Fig. 2b), mounted 2 km away from the eastern coast of the lake (Nomura and others, Reference Nomura2020). The nominal accuracies of air temperature, relative humidity and wind speed are 0.5°C, 2% and 3% of reading, respectively. The time series of air temperature and wind speed for the experimental period are shown in Figures 3a and b. Air temperature varied from −15 to −3°C with southerly to westerly winds ranging from 0 to 5 m s−1 during our experiment. In Figure 3b it is noted that some wind data were partly missing at nighttime, presumably due to the freezing of the anemometer under the cold air (−15 to −10°C) and undersaturated humidity (85–90%) conditions. These data gaps were filled with wind speed measurements from the Tokoro automated meteorological station operated by Japan Meteorological Agency, located ~5 km away from the Saroma-ko Lagoon (Fig. 1). This was justified by the strong agreement between the two sites for daytime measurements.
Meteorological data observed on the Saroma-ko Lagoon with light gray shades showing the ice growth periods (18:00 to 06:00) and dark gray shades showing a snowfall event (19:30 to 20:30 on 26 February). (a) Air temperature with the data at Tokoro (Fig. 1a) in broken line. (b) Wind speed with the data at Tokoro (Fig. 1a) in broken line. (c) Radiative fluxes of downward shortwave (red), upward shortwave (green), downward longwave (orange), upward longwave (blue) and net (purple). Note that radiation data were taken on the snow surface.
To assess the upwelling and downwelling radiative fluxes, shortwave (305–2800 nm) and longwave (3–50 μm) radiation was monitored every minute with a radiometer (EKO Instruments Co., Ltd., MR-40) with the accuracy of ~1–2% of the measured values. The radiometer was mounted on the snow surface near the AWS to avoid impacting the freezing conditions in the pool (Fig. 2b). The time series of each radiative flux, together with net flux, are shown in Figure 3c. The coupled variation of the downwelling longwave radiation and shortwave radiation likely indicates the presence/absence of clouds. Hence, it is inferred that the weather during the experiments was fair but occasionally cloudy for the first night (25–26 February) whereas mostly cloudy for the second night (26–27 February). The net radiative flux was positive (up to 150 W m−2) from 06:00 to 15:00 whereas it became close to zero or negative from ~15:00 through to the early morning of the next day. Due to the quite different surface conditions between the radiometer site (snow surface) and pool site (thin ice), the upwelling longwave radiative flux used for the model was calculated from the Stefan–Boltzmann law (refer to Section 3.2 for details).
To monitor the freezing conditions, vertical temperature profiles of the air and water near the surface (at 0.01, 0.06 and 0.22 m above the water and at the depths of 0.03 and 0.20 m) were recorded at one corner of the pool (Fig. 2a) during the observation, using five thermo-recorders (RT-32S from ESPEC MIC Corporation), with a measurement accuracy of 0.1°C. The time series of water temperatures at depths of 0.03 and 0.20 m are shown in Figure 4. Although temperatures at both depths increased by ~0.5°C in the daytime through the absorption of solar radiation into the water, they began to decrease after 15:00 and remained close to the freezing point (−1.7°C) from 18:00 to 06:00 through the night. This is consistent with the diurnal variation of the net radiative flux in Figure 3c, where the net radiative flux became negative at ~15:00 on both days, and therefore it is likely that the solar heat absorbed into the water during the daytime was completely released back into the air by 18:00 on both days. Based on this result, we presumed that sea-ice formation began at 18:00 and the sensible heat flux from the underlying water was not taken into account in the thermodynamic model.
Time series of water temperature at 0.03 and 0.20 m depths at the pool site with gray shading showing the ice growth periods (18:00 to 06:00). Arrows denote the timings of sample collection.
The ice conditions and snowfall events at the pool site were monitored at 1-min intervals from 07:00 to 17:00 on each day with a camera mounted near the pool (Fig. 2a). To detect snowfall events each night, we referred to records at the Abashiri Local Meteorological Observatory (ALMO), located ~25 km east of the site, and the hourly snow data from the Tokoro station (Fig. 1). According to the monitoring camera photos at the pool, light snowfall events occurred at 13:05–13:20, 14:38–15:25 and 16:04–17:00 on 26 February during the observation period. However, based on the following observational facts, we infer that there was no significant snowfall except for during 19:30–20:30 on 26 February for the second night, when snowfall accumulated in a 0.01 m thick layer at the pool:
(1) The document at ALMO reported that a significant snowfall with visibility <2 km was limited to 30 min duration between 20:40 and 21:10 and it resulted in a 0.01 m thick layer during 20:00–21:00 on 26 February.
(2) No snowfall was recorded at either Tokoro or Abashiri except for the above period.
(3) According to Figures 3c and 6b, net longwave radiation was close to zero from 19:30–20:30 on 26 February, indicating thick clouds cover over this area during this period.
(4) Snow pit measurements conducted on the Saroma-ko Lagoon on the morning of 27 February revealed a 0.01 m thick new snow layer on top of the pre-existing snow layer (Table 1).
Snow pit measurements
aIn Grain shape, RG, DH, MF and PP denote rounded grains, depth hoar, melt forms and precipitation particles, respectively. Measurements were conducted at the same site on Lake Saroma, close to a radiometer. Mean snow density was 309 ± 32(sd) kg m−3. Snow type classification is based upon The International Classification for Seasonal Snow on the Ground (Fierz and others, Reference Fierz2009). Snow temperatures at the heights of 0.00, 0.03, 0.06 and 0.09 m from the snow/ice boundary were −3.1, −2.6, −2.5 and −2.6°C, respectively, on 25 February, and −3.4, -2.9, −2.7 and −2.5°C, respectively, on 27 February.
2.2 Sea-ice sampling
The newly formed sea ice of ~2–3 cm thickness was collected at 09:26 on 26 February and at 09:38 on 27 February (Local time) by cutting out an area of 0.15 m × 0.15 m area with a portable saw (Figs 2c and d). As shown in these figures, there was no dry snow layer on the sea ice in the pool at the sampling time. The ice samples were kept in resealable bags that were immediately stored with coolant (−25°C) in an insulation box, and then sent to the Institute of Low Temperature Science (ILTS), Hokkaido University in Sapporo within the sampling day. They were kept in a cold room at a temperature of −15°C for analysis. Although the ice samples experienced large temperature changes before analysis, this is assumed to have had a negligible impact on the crystal texture and salinity of the ice samples that we were focused on.
3. Analytical method
3.1 Sample analysis
Sea-ice samples were processed in the ILTS cold room at −15°C to carry out thick (5 mm) and thin (1 mm) section analyses for examining the inclusions and crystal alignments, respectively. First, ice samples were vertically sliced to make a 7 mm thick ice section with a bandsaw, and then they were attached onto slightly warmed glass. The samples were placed on dark clothes and photographed with scattered light to observe inclusions. Next, the samples on the glass were sliced carefully with a microtome until the thickness reached 1 mm, and then photos were taken through crossed polarizers to observe crystal alignments.
The salinities of the samples were also measured. Since the two-layered structure was prominent for each sample (Fig. 5), the salinities of the individual layers were measured. To do so, rectangular columns with dimensions of 6 cm × 6 cm (see Table 2 for individual ice thicknesses) were cut out from the ice samples, and then each column was cut horizontally at the depth of the boundary between the two layers. Salinities were measured with a salinometer (WTW Cond 3110, nominal accuracy: 0.1 psu) after melting the ice samples at room temperature. The results are listed in Table 2. The averaged salinities of the granular and columnar ice layers are 10.4 and 10.2 psu, respectively. There was no significant difference between these two ice types or sampling dates. These values are comparable with the 12 psu reported by Takizawa (Reference Takizawa1984) for snow ice formed in the pool at the Saroma-ko Lagoon. The reason for the lack of salinity contrast between these two layers seems to be because of the increased permeability of sea ice when it is near the freezing point. The brine volume fraction was estimated to be 24–30% using the formula of Frankenstein and Garner (Reference Frankenstein and Garner1967) by substituting the model-derived T s (−2.4 to −1.7°C) and measured bulk salinity (10.2 psu), well above the traditional 5% threshold of permeability (Golden and others, Reference Golden, Ackley and Lytle1998). This enhanced brine drainage induced by increasingly permeable structure, irrespective of ice type, was previously highlighted by Takizawa (Reference Takizawa1984).
Vertical structures by thick (5 mm) (upper) and thin (1 mm) (lower) section analysis for sea-ice samples collected at (a) 09:26 on 26 February near the margin of the pool, (b) 09:38 on 27 February near the center of the pool, and (c) 09:38 on 27 February near the margin of the pool. Total ice thicknesses for (a)–(c) are 23, 30 and 35 mm, respectively. In each figure, top of the sample corresponds to the ice surface.
Salinities of sea-ice samples
aSSL and Cl denote surface solidification layer and congelation ice, respectively.
3.2 1-D thermodynamic model
To examine the snow effect quantitatively, we calculated the ice thickness evolution with a thermodynamic ice-growth model used in Toyota and Wakatsuchi (Reference Toyota and Wakatsuchi2001). This model is originally based on Maykut (Reference Maykut1978) and calculates ice growth from a simple surface heat balance equation expressed by Eqns (1) and (2). Sensible heat flux (FSH) is obtained from ρ ac pC su(T a − T s) with the bulk method, where T a is the observed air temperature, T s is the surface temperature of the ice given by solving Eqn (1), ρ a is the air density (1.3 kg m−3), c p is the specific heat of the air (1004 J kg−1 K−1), C s is the transfer coefficient for sensible heat and u is the observed wind speed. Latent heat flux (FLH) is obtained from 0.622 ρL vC eu(re sa − e ss)/p with the bulk method, where L v is the latent heat of sublimation (2.84 × 106 J kg−1; Yen, Reference Yen1981), C e is the transfer coefficient for latent heat, r is the observed relative humidity, p is the observed surface pressure and e sa and e ss are the saturation vapor pressure in the atmosphere and at the ice surface, respectively. The dependence of e s on air temperature is expressed as a fourth-order polynomial developed by Maykut (Reference Maykut1978). C s and C e are taken to be 1.0 × 10−3 after Aota and others (Reference Aota, Shirasawa and Takatsuka1989).
The downward longwave radiation (FLW↓) was given by the radiometer mounted near the AWS, whereas the outgoing longwave radiation (FLW↑) was calculated by $\varepsilon \;\sigma T_{\rm s}^4$
, where ɛ is the emissivity and taken to be 0.97 (Maykut, Reference Maykut and Untersteiner1986), and σ is the Stefan–Boltzmann constant because the surface conditions at the pool site were significantly different from those at the radiometer site. Since the penetration depth of longwave radiation is less than a few millimeters (e.g. Bae and others, Reference Bae, Nam, Sing and Kim2010), there seems no need to consider the penetration fraction of FLW↓. The conductive heat flux in sea ice (FCI) can be written as k i(T B − T s)/H i on the assumption of the ice being snow-free, where k i and H i are the thermal conductivity and thickness of sea ice, respectively. This assumption is justified from the photos at the sampling time (Figs 2c and d). k i is set to 2.0 W K−1 m−1 as the representative value for sea ice at ~−2°C. T B is the temperature at the ice bottom and is taken to be T f (= −1.71°C). T s is solved using Eqn (1) with the Newton–Raphson method, assuming that all heat fluxes are balanced at the surface thin ice layer:
Thereafter the individual fluxes are calculated by substituting the obtained T s. The ice growth rate was calculated from the following heat-balance equation at the ice bottom on the assumption of no ocean heat flux
where ρ i is the ice density (900 kg m−3). L f is the latent heat of fusion and taken to be 2.27 × 105 J kg−1 by substituting the sea-ice salinity (10.2 psu) for congelation ice (Table 2) and the freezing temperature (−1.71°C) into the Yen (Reference Yen1981) formula. The initial ice thicknesses (0.10 × 10−3 m and 0.11 × 10−3 m for the first and second nights, respectively) were calculated from the heat loss per second at the water surface at the freezing point at 18:00 on each day, assuming that sea ice began to grow at 18:00. The subsequent ice thickness evolution was obtained by integrating Eqn (2) with a time step of 1 min. The assumption of no ocean heat flux is justified by the fact that the water temperature remained at the freezing point at depths of both 0.03 and 0.20 m in Figure 4.
4. Results
4.1 Sea-ice structure
The vertical thick and thin sections of ice samples are shown in Figure 5. Since the sea-ice samples were collected at the margin and center of the pool site on 27 February, two sections are shown for this day. The thick section structure in Figure 5 clearly shows a layered structure for each sample and the contrast between the two experiments is noticeable. For the 26 February sample, the whitish layer in the top 5 mm is followed by a thin dark layer and then a relatively coarse translucent layer (Fig. 5a), whereas the 27 February sample shows two fine-structured translucent layers separated by a distinct boundary (Figs 5b and c). Correspondingly, the contrast in the crystal alignments for the thin sections from the two experiments is also noticeable for the full depth of the samples. The 26 February sample was composed of granular ice (~5 mm thick), a thin dark layer with a vertical c-axis, and columnar ice with a horizontal c-axis and large grain size (~18 mm thick in Fig. 5a), whereas the 27 February sample had fine grained granular ice (~10 mm thick in Fig. 5b and 15 mm thick in Fig. 5c) and then much thinner-structured columnar ice. The fine-grained granular ice indicates that this layer formed within a short time. The granular ice in Figure 5c is further divided into the upper 5 mm thick layer and the lower 10 mm thick layer, where the upper layer has somewhat larger grain sizes than the lower layer. The difference in ice thickness between the two samples results from the absence of this upper layer in Figure 5b. The formation process of this layer will be discussed in Section 5.2.
The crystal structure in the thin section of Figure 5a is quite similar to that of the ice grown in the tank experiment under calm conditions by Toyota and others (Reference Toyota2013), except for the granular ice at the top (compare with their Fig. 2). This suggests that the sea ice formed under the relatively calm conditions on the first night (25–26 February), was free from any disturbance. On the other hand, the meteorological conditions on the second night likely affected the sea-ice structure during the growth period. Given that the air temperature and wind speed conditions were relatively similar for these two nights (Fig. 3), the key difference was the snowfall event that occurred at ~20:30 on 26 February, as described in Section 2.1. Therefore, it is most likely that the 10 mm thick granular ice layer overlying the columnar ice in Figures 5b and c is attributed to snow-related processes. It is interesting to note that the thickness of this layer is comparable with the estimated snowfall amount.
4.2 Ice growth and surface heat budget
Here the results from the 1-D thermodynamic model are presented. The ‘no snow’ ice growth calculated each night from 18:00 to 06:00 is shown with a red solid line for the first night and a black solid line for the second night in Figure 6a. We stopped the calculation at 06:00 on the next day because our focus is ice growth at night and the effect of solar radiation cannot be neglected (Fig. 3c), i.e. Eqn (1) does not apply, after 06:00. The ice thickness at 06:00 amounted to 16 mm for 26 February and 17 mm for 27 February. These values are close to the observed thicknesses of columnar ice in Figure 5 (18 ± 1 mm for 26 February and 20 ± 1 mm for 27 February). We confirmed that the difference in the predicted ice thickness caused by the measurement errors of the meteorological observations is within only ±0.5 mm. Considering the possibility of slight growth between 06:00 and the sampling time (~09:30), our calculation closely reproduces the observed ice thickness. The time series of the sensible, latent and net longwave radiative heat fluxes used for calculating the ice growth rate during the first and second nights are shown in Figures 6b and c, respectively, and the averages of the individual fluxes for each night are depicted in Figure 7. These figures indicate that FLW is the dominant heat flux in determining the sea-ice growth rate. The large variation of FLW in Figures 6b and c was caused by the variation of downward longwave radiation, which is controlled mainly by the amount of cloud cover. Thus, cloud cover plays an important role in determining the ice growth rate. The values in Figure 7 are comparable with those of past observations conducted over thin ice at the pool on the Saroma-ko Lagoon by Ishikawa and Kobayashi (Reference Ishikawa and Kobayashi1984). These results all indicate that our model reproduced the real ice growth and heat fluxes successfully.
Sea-ice growth during the nights calculated with a 1-D thermodynamic ice growth model. (a) Temporal evolution of ice thickness. (Red line: without snow for the first night, black solid line: without snow for the second night and black broken line: for the second night, including the surface solidification layer (SSL).) Two bars ~06:00 denote the real columnar ice thickness observed in the ice samples. (Red bar: 26 February and black bar: 27 February.) (b) Temporal variation of individual heat fluxes during the first night (25–26 February). (Green: net longwave radiation, red: sensible heat flux and blue: latent heat flux.) Note that net longwave radiation was obtained by summing up the downward flux observed by the radiometer and the upward flux calculated by the model. (c) Same as (b) except for during the second night (26–27 February).
Surface heat fluxes during the night, averaged from 18:00 to 06:00. FSH: sensible heat flux, FLH: latent heat flux and FLW: net longwave radiation; SSL stands for surface solidification layer.
Here, we examine the effect of the solidification of the surface slush layer on the subsequent ice growth and turbulent heat flux with our model, by artificially adding 10 mm to the ice thickness at 21:00 on 26 February on the assumption that this layer formed immediately after the main snowfall event. We continued the ice growth calculation after 21:00 with the same atmospheric forcing for the second night. In this calculation, we used the same k i for the thermal conductivity of the solidification layer based on the near match in salinity values for the granular ice and columnar ice (Table 2). The temporal evolution of ice thickness is depicted with a broken line in Figure 6a, and the averaged heat fluxes are shown with dotted bars in Figure 7. Figure 6a shows a slight decrease in the sea-ice growth rate after the addition of the solidification layer, compared with the result without it. This is explained by the decrease in T s (predicted from −2.5°C without snow to −2.9°C with solidification layer at 06:00 on 27 February), which results in the reduction in FSH, FLH and FLW, and the reduction in the vertical temperature gradient in ice (from −46.4 K m−1 without snow to −45.2 K m−1 with solidification layer at 06:00 on 27 February), which are both associated with an addition of ice thickness. Accordingly, the turbulent heat flux (FSH + FLH) released from the ice to the atmosphere is somewhat reduced (Fig. 7). However, the decrease in ice thickness at 06:00 and the reduction in turbulent heat flux (FSH + FLH) is estimated to be <1 mm and 2.5 W m−2, respectively. These results indicate that the addition of a 10 mm thick solidification layer affected neither ice growth nor surface heat flux significantly.
To confirm our result, we further examined the effect of the solidification of the surface slush layer by changing the layer thickness in our 1-D thermodynamic sea-ice model from 10 to 30 mm, assuming that the formation of this layer would be limited to less than a few centimeters because of large amount of heat flux needed to form it (Section 5.1). Figure 8a shows the temporal evolution of the additional ice thickness from 21:00 on 26 February until 06:00 on 27 February for each solidification layer thickness. The corresponding averaged heat fluxes of FSH, FLH and FLW are depicted in Figure 8b. These figures show that although both ice thickness growth and turbulent heat flux decrease in association with the increase in the thickness of the solidification, the decrease is relatively low. Setting the solidification layer at 10, 20 and 30 mm only resulted in a 0.7, 1.3 and 1.8 mm decrease in ice thickness (<11% of the total thickness), respectively, and only 3, 5 and 7 W m−2 decrease for turbulent heat flux (<15% of total flux). This indicates that the solidification of the surface slush layer during the freeze-up process does not have a significant effect on the subsequent ice growth and turbulent heat flux for snowfall amounts less than a few centimeters. This result may justify the omission of snow effects, as far as snowfall amount is less than a few centimeters, in estimating the ice production rates or heat fluxes in the polynyas by the past studies (e.g. Pease, Reference Pease1987; Martin and others, Reference Martin, Drucker and Yamashita1998; Tamura and others, Reference Tamura, Ohshima and Nihashi2008).
The effects of snow-ice thickness (10–30 mm) on (a) ice growth after 21:00 on 26 February until 06:00 on 27 February, and (b) individual heat fluxes averaged from 21:00 on 26 February to 06:00 on 27 February. SSL stands for surface solidification layer.
5. Discussion
In the previous section, we showed some evidence of the solidification of a surface slush layer resulting from accumulated snow and examined its effect on the ice growth and surface heat fluxes. Here we discuss the processes involved in more detail and its application to lake ice, based on observations.
5.1 Possible factors responsible for efficient surface slush layer solidification
In Section 4.1, we suggested that the surface slush layer solidified efficiently during the freeze-up process, judging from the additional ice thickness being comparable with the snowfall amount. It is likely that the incorporation of considerable snow particles contributed to the solidification of the surface slush layer quite effectively, presumably as a seeding effect. According to the estimated ice growth (Fig. 6a), the ice thickness before the snowfall event was only 2 mm. This thin ice layer might have acted as a collector of snow particles, which otherwise might have scattered and diluted in the sea water. Besides, very fine grain sizes (≪1 mm) of granular ice in the thin section of Figures 5b and c indicate that this snow-ice layer formed within a short time. Clearly, this fine-grained granular ice with randomly oriented c-axes resulted in the significantly thin-structured columnar ice through the geometric selection (Weeks and Ackley, Reference Weeks, Ackley and Untersteiner1986). Here we discuss why surface slush layer could solidify so effectively.
As suggested above, the seeding effect of snow particles is one of the likely factors. However, for this effect to work continuously, the latent heat generated by the freezing of sea water should be released into the air efficiently. According to Figure 6c, net longwave radiation (FLW) decreased by ~40 W m−2 soon after the main snowfall events at 20:30 on 26 February, presumably associated with the disappearance of snow clouds. Although this might have enhanced the amount of heat released from the surface slush layer which was produced by the snowfalls, it is not enough to explain the efficient solidification. This is because the heat flux needed to be released to produce a 0.01 m thick solidification layer within 1 h is estimated to as much as 450 W m−2 (calculated from ρ iL f × 0.01 m/3600 s, assuming that the density of new snow is 200 kg m−3). Here we would like to point out another possible contributing mechanism occurring at the microscale.
Figure 9a shows the time series of the air temperature at 0.01, 0.06 and 0.22 m above the water level at the pool site for the second night. Overall, the temporal variations are linked with each other for the three heights, like the temperature at the AWS in Figure 3a. At the same time, it is found that although the temperatures at 0.06 and 0.22 m heights were almost coincident, those at the 0.01 m height were always ~1°C higher. This indicates that a strong vertical temperature gradient was maintained near the surface during the night. To show this more clearly, we depicted the nightly averaged vertical temperature profiles in Figure 9b. Although the average temperature was somewhat different between the two nights, a strong vertical temperature gradient (~−22 K m−1) is commonly found between the 0.01 and 0.06 m heights for both nights. Considering that T s should be maintained at the freezing point (−1.7°C) when slush is present at the surface, a much stronger temperature gradient (~700 K m−1) was maintained at the surface during the solidification of surface slush layer. Although we cannot estimate the heat flux quantitatively due to the lack of wind data on a microscale, it is likely that such a strong, persistent temperature gradient worked efficiently to release the heat generated from the surface slush layer to the air. Since the exposure of the surface slush layer to such a strong temperature gradient in the air seems to be a unique property of thin ice, it may be said that such efficient solidification of surface slush layer occurs ubiquitously when snow falls on sea ice during the freeze-up process.
Air temperatures at 0.01, 0.06 and 0.22 m above the surface at the pool. (a) Time series of air temperatures at individual heights for the second night. (b) Vertical profiles of air temperatures averaged for the first (open circle) and second (solid circle) nights.
Without the solidification process of the surface slush layer occurring on the microscale, it would be difficult to explain the difference of ice texture between the two nights (Figs 5a and b). On the other hand, it is noted that although the traditional 1-D thermodynamic model cannot predict this microscale freeze-up process, this process does not affect the basal freezing growth rate calculated by the model when the snowfall is less than a few centimeters. This result indicates that although the traditional thermodynamic model works well in estimating the ice production rate by basal freezing, a more sophisticated model that account for the snow–ice interaction in the atmospheric boundary layer is required to reproduce realistic ice crystal texture.
The formation of surface slush layer associated with the snowfall event is nonetheless still puzzling. The presence of a wetted layer at the surface would increase T s to near the freezing point. This causes two important issues. First, the vertical temperature gradient increases above the surface, which would in turn enhance the turbulent heat flux at the surface. Second, the vertical temperature gradient within the sea ice would be reduced, which would reduce the conductive heat flux within the ice. Therefore, it might be possible to reproduce this solidification layer in the 1-D thermodynamic model, if it can produce the surface slush layer. However, future work into the micro-process of surface slush layer formation is required to gain the necessary understanding to achieve this.
5.2 Solidification of surface wetted layer by wind-blown snow particles
Here, we examine the formation process of the top layer of granular ice (5 mm thick) in the thin section of Figures 5a and c. In Figure 5c, this layer is discriminated from the lower granular ice layer (10 mm thick), based on the larger grain sizes. It is interesting to note that this layer is reduced or eliminated for the sample collected near the center of the pool (Fig. 5b). It is unfortunate that we could not obtain any photographic evidence about what caused this layer from the site camera because it occurred at nighttime. For the following reasons, however, we deduce that this layer was produced through the solidification of the surface wetted layer induced by snow particles supplied by drifting snow onto the brine that has wicked to the surface of the relatively permeable ice:
(a) Figure 5a indicates that this layer formed after congelation ice had already grown to some extent under calm conditions with no snowfall.
(b) This layer appears more prominently near the margin of the pool, relative to the inner area.
(c) Grain sizes in this layer are relatively larger than those of the precipitation particles shown in Table 1.
(d) The maximum wind speed of ~5 m s−1 observed during the experiment is enough to induce drifting snow.
Item (a) can be explained only through the solidification of wetted layer on top of the sea ice. Items (b) and (c) suggest that snow particles were supplied from the surface snow surrounding the pool site, as shown by Gow and others (Reference Gow, Meese, Perovich and Tucker1990) for the Arctic leads. Item (d) is based on the past observational fact that unconsolidated snow begins to drift at a wind speed of 5 m s−1 (Takeuchi, Reference Takeuchi1980; Andreas and Claffey, Reference Andreas and Claffey1995). It is likely that the limited snow-drift transport and horizontal saltation distance of snow particles due to the relatively weak wind speed caused an inhomogeneous snow supply that was biased to the margin of the pool in the upwind direction as shown in (b). During the experiment, the maximum wind speeds (~5 m s−1) were observed ~06:00 for the first night and at midnight for the second night (Fig. 4b). Therefore, we deduce that the main snow transport events occurred around these timings. This is consistent with the observational result, in that the uppermost solidification layer formed after the major growth of congelation ice (the first experiment) and the solidification of surface slush layer due to snowfall at 21:00 (the second experiment).
For the regions of Antarctic sea-ice cover where there are strong prevailing winds, it is suggested from observation and numerical simulation that snow loss into leads, caused by drifting snow, amounts to more than half of the total precipitation of snow on sea ice over the entire Southern Ocean (Eicken and others, Reference Eicken, Lange, Hubberten and Wadhams1994; Dery and Tremblay, Reference Dery and Tremblay2004; Leonard and Maksym, Reference Leonard and Maksym2011; Toyota and others, Reference Toyota2016). Our experiment showed some evidence that this process can occur ubiquitously in the polar regions where drifting snow prevails at wind speeds >5 m s−1.
5.3 Application to lake ice structure
Finally, we would like to point out that our results can be applied to lake ice structure. It was noted that there are two major types of congelation ice for lake ice: (a) ice sheets composed of massive, irregularly shaped crystals with vertical c-axes and (b) ice sheets composed of vertically-elongated crystals (columnar ice) with horizontally oriented c-axes (Gow, Reference Gow1986). As for the formation process of columnar ice, the possibilities of ice nucleation (frazil ice formation) under turbulent conditions (Barns and Laudise, Reference Barns and Laudise1985) and atmospheric seeding (frozen water droplets) under quiet conditions (Gow, Reference Gow1986) were pointed out. Our results support the findings of Gow (Reference Gow1986). Considering that in the case of lake ice water temperature below the surface tends to be still slightly positive at the freeze-up stage, it is more likely that snow particles rather than frazil ice works to produce the initial granular ice which will thereafter grow into columnar ice underneath. Ohata and others (Reference Ohata, Toyota and Shiraiwa2016) also provide some evidence of this process from the δ18O profiles of lake ice at Lake Abashiri, Hokkaido, Japan.
6. Conclusions
We conducted field experiments to grow sea ice in an artificial pool on the Saroma-ko Lagoon for two nights in February 2019 to ascertain how meteorological conditions affect sea-ice properties during the freezing process. Although the weather conditions were relatively calm during both nights, a significant snowfall event occurred on the second night. This facilitated an examination of how snowfall can affect sea-ice structure, the subsequent ice growth and the turbulent heat flux. The results from our sample analysis and a 1-D thermodynamic ice growth model are summarized as follows:
(1) Snowfall during the freeze-up process of thin sea ice can affect the crystal alignments significantly through the solidification of a surface slush layer, which produces fine grained granular ice and thereafter thin-structured columnar ice.
(2) Solidification of a surface slush layer during the freeze-up process has no significant effect on the subsequent ice growth (less than a few millimeters per night) and the turbulent heat flux (only a few W m−2 on average).
(3) The rapid solidification of the surface slush layer is related presumably to the seeding effect of snow particle accumulation and the persistent strong temperature gradient near the surface.
(4) A wind speed of 5 m s−1 appears to be sufficient to induce drifting snow and carry snow particles into leads or thin ice, resulting in solidification of the wetted layer on top of thin ice that results from brine wicking upward through the porous ice.
Item (1) provides some implications about the possibility of reconstructing the freeze-up history from thin section analysis and the understanding of the formation processes of columnar ice for lake ice. Item (2) may justify the omission of snow effects, if the snowfall amount is less than a few centimeters, in estimating the ice production rates due to bottom freezing in past polynya studies. Regarding (3), it should be kept in mind that although the traditional 1-D thermodynamic model could reproduce the bottom freezing amount well, the solidification of a surface slush layer cannot yet be predicted with this model. This indicates that we need a more sophisticated model, taking the microscale process in the atmospheric boundary layer into account. Item (4) supports the past studies about snowdrift observation and presents some evidence that wind-blown snow particles contribute to the freeze-up process of thin ice as well as snowfall.
Although these results related to the properties of granular ice induced by snowfall, we believe many of them are equally applicable to the granular ice originating from frazil ice. When frazil ice is created in sea water and then swept against the thicker ice by ocean waves under the turbulent conditions, it can accumulate under quiescent conditions that are produced due to the attenuation of waves. These have previously been referred to as a ‘dead zone’ (Martin and Kauffman, Reference Martin and Kauffman1981). It is for this reason that our idealized experiments can also be applied to the harsh open water conditions of polynyas. Such surface conditions, composed of a mixture of frazil ice and sea water, would become like slush layer akin to those produced by the snowfalls in our experiments. Accordingly, the results presented in this paper could be applicable to polynyas because the solidification processes of ice crystals in sea water should be similar.
To date, the effect of snowfall on the growth of thin ice at the freeze-up process has received less attention compared with thick ice due to the logistical difficulty of field observations. Although some unique properties were revealed through our experiments, we need more field data to confirm our conclusions. Further investigation, especially the eddy flux measurement in the atmospheric boundary layer, is desirable.
Acknowledgements
This experiment was conducted as part of the Saroma-ko Lagoon Observations for sea-ice Physico-chemistry and Ecosystems 2019 program (SLOPE2019). We would like to express our sincere gratitude to Aquaculture Fishery Cooperative of Saroma Lake, Napal Kitami accommodation, and other group members of SLOPE2019 for their kind cooperation. Proof-reading by Dr Guy Darvall Williams and comments by Dr Bin Cheng, Dr Martin Vancoppenolle, and anonymous reviewers were very helpful to improve this manuscript. This work was financially supported partly by JSPS KAKENHI (#16K00511, #19K12304, #17K00534 and #18F18794).
Open access
