2.1. Air, water and ice observations

Air temperature was recorded near the roof of the tank at approximately 2.5 m height. Underwater pressure transducers placed at the left wall of the tank (when facing from wave maker towards beach) recorded the wave amplitude. Several thermistors placed on this wall provided temperature information at different levels across the ice/water surface.

Water electrolytic conductivity and temperature were measured (with a sampling interval of 15 s) with two MicroCat SBE37-SM conductivity-, temperature- and depth-measuring devices (CTDs) placed stationary at the centre of the tank, at 0.6 and 0.8 m depth. To convert from conductivity to NaCl salinity we use the UNESCO algorithms (Reference Fofonoff and MillardFofonoff and Millard, 1983) modified to account for a higher conductivity of an aqueous NaCl solution compared with sea water (e.g. Reference KaufmannKaufmann, 1960). We apply a standard conductivity of 45.3172 mS cm^–1 at 15^˚C and 35 g kg^–1 NaCl.

The sampling procedure of in situ frazil-ice samples was adopted from earlier studies (e.g. Reference WilkinsonWilkinson, 2005; Reference Smedsrud and SkogsethSmedsrud and Skogseth, 2006) using a transparent plastic cylinder of 89.07 mm diameter, open at both ends and with a 22.98 mm diameter stick through the middle with a bottom rubber lid. the cylinder was lowered through the frazil ice until the lower edge was well below the ice surface and the stick pulled up to entrap the sample and seal off the bottom with the rubber lid. the cylinder was then lifted to measure the undrained frazil-ice thickness (to ±0.5 cm accuracy) using a centimetre measuring scale fixed to the cylinder. Every 4 m along-tank (refer to Fig. 1) the frazil ice was collected in a hand-held sieve to drain off the water for several minutes, before melting it in bottles at room temperature. the salinity of the melted frazil/brine samples was inferred from measuring their conductivity with a hand-held WTW LP191 conductimeter (accurate to ±0.1 mScm^–1). Sample weight and volume were obtained using a standard measuring scale (accurate to ±0.01 g) and a volumetric flask (accurate to ±0.1 mL), respectively.

Surface brightness temperature, T _B, was recorded with a thermal infrared (IR) camera (VarioCAM hr 384 M, manufactured by JENOPTIK) mounted at 2.4m height above the water/ice surface. the instrument provided IR temperature images in the 7.5–14 μm band with 7.6 Hz sampling rate and a field of view (FOV) of 15^˚×12^˚. At the given height, the surface viewed was 66 cm×52cm (384×288 pixels) with 1.72 mm pixel resolution. To reduce side-lobe effects we omit rim pixels and use the inner 354×280 pixels for our image analysis. A 5 s running mean was applied to the T _B fields to reduce the wave-induced fluctuations. the manufacturer reports a temperature resolution better than 80mK and an accuracy of ±1.5 K. As described in section 3, we convert brightness temperature to physical temperature (T _B = ɛT _s) on the basis of water temperature peaks.

3.1. Surface emissivity and grease–pancake thresholding

To derive surface temperatures from the observed IR brightness temperatures (T _s = T _B/ɛ) one needs to know the emissivity, ɛ. However, literature values show a rather wide range due to variable background radiation, surface roughness and incidence angles (e.g. Reference Rees and JamesRees and James, 1992). We approach this problem by taking into account the character of the wavy ice field and assume that, due to convergence and divergence as well as brine convection, some pixels in our 60 cm ×50 cm FOV will always reflect the water temperature, T _w, monitored below the ice. We hence calculate an effective ɛ by assuming that the 1% highest T _B in each image determines ɛ = T _B/T _w. This approach yields ɛ between 0.9953 and 0.9978 for the IR images (Fig. 2a), with no apparent time dependence and a mean of 0.9968±0.0033. This standard deviation in ɛ corresponds to a temperature uncertainty of 0.09 K. That local pixel temperatures may occasionally exceed the water temperature by 0.1 K (Fig. 2b) may be interpreted as instrumental noise (within the temperature range of the IR camera resolution) and is not considered further here.

Fig. 2.

(a) Surface emissivity estimate for all IR T _B image field maxima using T _w as the ɛ = T _B/T _w calibration threshold. (b) Water temperature, T _w, and freezing temperature, T _f, in relation to T _s maxima, with corresponding 0.5 hour running mean, and median frazil temperatures. (c) Median frazil T _s and median and minimum pancake T _s evolution. All T _s in (b) and (c) were derived using mean ɛ = 0.9968 as a constant.

We attempted to classify the ice types by several automatic thresholding algorithms (HistThresh MATLAB Toolbox provided by A. Niemistö , 2004, http://www.cs.tut.fi/~ant/histthresh/). During the first 7 hours of the experiment, while frazil was dominant and few pancakes were visible, the algorithm from Reference Kittler and IllingworthKittler and Illingworth (1986) rendered the most stable threshold of 270.0±0.1 K. Other histogram-based algorithms produced much larger scatter of ±0.2–0.4 K. This is likely related to the fact that the frazil and pancake modes are not well separated in the histogram, and to the appearance of a third mode between the warm frazil and cold pancake peaks, associated with the rim of the pancakes. However, we found by manual threshold determination that the automatic threshold of 270.0 K overestimates the pancake fraction. A value of 269.5±0.1 K gives the most realistic pattern over the course of the experiment (see Fig. 3a and b). This revised threshold matches the intermediate mode associated with the pancake rim, which gives us confidence in our segmentation procedure.

Fig. 3.

(a) Evolution of frazil (white) and pancake (black) area cover during selected times, estimated from constant threshold set at 269.5. (b) Corresponding histogram for same times as given in (a), showing the constant threshold (dashed line) used to estimate frazil and pancake area covers. the histogram temperature range, T _range, and mean difference, T _diff, between frazil and pancake median temperatures are also given.

With the temperature threshold determined we calculated frazil and pancake area fractions, A _Fr and A _Pk, for each image by selecting the pixels above and below, respectively. Extreme temperature values were caused at times, particularly by pancake rims and by ice crystals falling from the rooftop onto the surface measured by the IR camera. We thus assign characteristic temperatures for pancake ice and for frazil ice by calculating the median temperature value of the distribution of temperatures for each ice class.

3.2. Solid-ice volume fraction

In the following we make the assumption that brine salinity (S _b) in the ice is at its freezing point, for which we use:

obtained by Reference MausMaus (2007) on the basis of data compiled for NaCl solutions. For an NaCl salinity of 32 g kg^–1 this gives T _f = –1.912^˚C.

Neglecting the air bubble content, we consider frazil and pancake ice as a mush of liquid (brine) and pure solid (ice) with densities ρ _b and ρ _s, brine volume fraction v _b and solid-ice volume fraction v _s = (1– v _b). Frazil-ice density and salinity are then given as ρ _i = ρ _b v _b + ρ _s(1 – v _b) and S _i ρ _i =v _b ρ _b S _b, respectively. These then may be combined to obtain the solid-ice volume fraction

Ice density ρ _s = 917 kgm^–3 is assumed constant while ρ _b = 1000 + 0.77S _b kgm^–3 sufficiently approximates the brine density of our NaCl solution for our purpose and property range. Equation (2) thus yields the dependence of v _s on ice and brine salinities. Below we also apply it to obtain the ice salinity S _i when v _b and S _b are known.

Our ice-sampling procedure involves the loss of some brine after which the salinity was reduced to S _im. As we will discuss further in a future paper, the frazil solid fraction, v _s, may be derived as

where V _g is the measured grease/frazil-ice volume before drainage, M _m and S _im the measured mass and salinity of the samples after drainage, and S _b the brine salinity. Although salinity S _im depends on the drainage protocol which may vary widely, we can reconstruct the true frazil-ice salinity S _i(Fr) before drainage by combining Equations (2) and (3). S _b was not measured directly, but estimated by assuming thermodynamic equilibrium with the temperature in the ice (Equation (1)). For loose grease ice, observations indicate that only a thin surface layer is below the freezing point of sea water (Reference Martin and KauffmanMartin and Kaufmann, 1981).

We use S _b corresponding to the freezing temperature T _f of the tank salt water from Equation (1), but for solid pancake ice we assume a linear temperature profile between surface temperature T _s and T _f. By integrating Equations (1) and (2) we obtain an average v _s.

3.3. Ice growth estimation

From the initial cooling period, prior to the onset of ice growth, we determined the heat flux in the tank from the cooling rate of the water as with T _w as the water temperature, water depth H _w = 0.85 m, water density ρ _w = 1027 (kgm^–3) and the specific heat capacity c_p = 4020.6 (J kg^–1 K^–1) assuming a salinity of 32 g kg^–1 NaCl. We converted the heat flux Q _s~82.3± 5.0Wm^–2 to an effective heat transfer coefficient for the air–water surface (Wm^–2 K^–1) as k _a = Q _s /(T _w = T _a) = 6.9±0.3, with T _a being the air temperature measured at 2.5 m height near the roof of the tank. This cooling period was too short to discriminate clearly whether cooling and freezing in the tank proceeds with constant Q _s or constant k _a. However, evaluating the ice production during day 1 clearly supported the constant Q _s model used in the present work. Such a result is expected because the thermostat on the laboratory roof maintained a constant temperature and thus heat was extracted continuously. Based on the heat flux we can compute the change in the average solid-ice fraction, v _s, as

where L _f is the latent heat of fusion (L _f = 330.7 kJ kg^–1 for an NaCl solution at –2^˚C, as in Reference MausMaus, 2007), pure ice density σ _s = 917 kgm^–3 and H _i is the mean thickness of the ice. the approach neglects changes in specific heat once freezing takes place (justified for the small temperature changes under consideration). Similar frazil- and pancake-ice thicknesses were observed over the duration of the experiment. This allows us to assume the same ice thickness for frazil and pancake ice and split up the average solid fraction into the pancake and frazil contributions as follows:

3.4. Ice salinity estimates

Our sampling procedure allows us to determine the undrained salinity S _i(Fr) of frazil ice. However, due to rapid brine drainage upon lifting pancakes, the true in situ salinity is more difficult to obtain. an alternative estimate of overall (frazil and pancake) ice salinity may be obtained from the salinity increase in the tank water, assuming that all salt released from growing ice is well mixed into the water below and recorded with the CTD instruments. the salt balance may then be written as

where H _o is the water level in the tank, ρ _w the water density and ρ _i the frazil- or pancake-ice density. Monitoring the change in water salinity, S _w, from its initial value, S _wo, at the onset of ice formation, and the ice thickness, H _i, gives the ice salinity.

This average ice salinity, S _i, may be further divided into the frazil and pancake contributions:

As in Equation (5), we made the assumption that frazil and pancake thickness are the same. We also neglect the small difference in frazil- and pancake-ice densities.

4.1. Surface area cover and temperature evolution

The clearest transition that we observed by eye during the experiment is the evolution from a frazil-ice cover into a pancake-ice cover. According to our segmentation, this transition begins after 2–3 hours (Fig. 4). the fluctuations seen in Figure 4 are associated with a limited number of pancakes travelling through the FOV of the camera and are not expected to be representative of the entire tank. Also, the slight decrease of pancake area fraction towards the end of the experiment is most likely a consequence of the limited FOV. However, the surface temperature segmentation of frazil and pancakes indicates that the transition takes approximately 10–15 hours. Images and histograms that illustrate this transition and the evolution of the temperature modes are shown in Figure 3a and b.

Fig. 4.

Evolution of pancake percentage area cover (A _Pk) derived from two methods: histogram thresholding within FOV of the camera (triangles), and salt and heat budget from Equations (2), (5), (6) and (7) (solid thick curve). Lighter curves show the respective ± standard deviation from the latter. Frazil cover may be estimated as A _(Fr)=1–A _(Pk).

Figure 2b shows that, in addition to its constant salinity and thickness, the median frazil surface temperature does not show a trend and appears to fluctuate by 0.1 K about its mean of –2.47^˚C, while pancake temperatures decrease during the experiment (Fig. 2c). Also, the 30 min averaged maximum T _s of the frazil ice remains typically 0.1 K below the freezing point (Fig. 2b). It shows, however, some low-frequency fluctuations and two distinct minima that appear to be associated with maxima in the pancake area fraction (after 8 and 16 hours), seen in Figure 4. Such a result appears reasonable as it suggests that divergence and upwelling of warm sea water decreases when pancakes are closely packed.

4.2. Ice salinity and solid fraction

We proceed to estimate the average ice salinity, S _i(Fr+Pk), from the increasing water salinity according to Equation (6). However, uncertainties in ice thickness and the water salinity after the previous experiment (initial S _wo at onset of freezing) likely correspond to a 20% uncertainty in the CTD-derived averaged ice salinity. We thus calibrate S _wo by requiring Equation (6) to match the frazil salinity S _i(Fr) = 25.2 after 2 hours, consistent with the onset of pancake formation. the resulting S _i(Fr+Pk) is shown as the dashed curve in Figure 5. We then combined Equations (5) and (7) to estimate the pancake salinity evolution. the result, smoothed with a 0.5 hour running mean, is shown as the thick dashed line in Figure 5. As with the pancake area fraction, we do not expect the fluctuations to present true changes in basin-averaged pancake salinity. However, the range of 18–22 g kg^–1 for the period when a substantial pancake-ice cover is established (after 7 hours) may be interpreted as typical salinities of young pancakes, obtained from salt budget and surface histogram thresholding. Thus, during the transition stage, pancakes are on average 3–7 g kg^–1 less saline than the frazil ice from which they form.

Fig. 5.

Frazil-ice salinity (stars) from reconstructed undrained in situ samples S _i(Fr) using Equations (2) and (3). the bars correspond to ±1 standard deviation. Total S _i(Fr+Pk) estimate (thick dashed line) from water S _w (Equation (6)), pancake-ice salinity S _i(Pk) (thin dashed curve) estimated from histogram thresholding (Equation (7)) and S _i(Pk) (thick solid curve) from salt and heat budget model (Equations (2), (5), (6) and (7)). the thin dashed curves above and below this last curve represent the error bounds when using Q _s±5Wm^–2. All curves are shown as 0.5 hour running means.

Combining the CTD-derived pancake- and frazil-ice salinities with ice surface temperatures for frazil and pancake, we computed, using Equations (1) and (2), the solid volume fraction, v _s. As mentioned previously, for grease ice we assume the mush to be at the freezing point of the water below. For pancake ice we assumed a linear temperature profile between the ice surface and the freezing point at the ice–water interface. These results are shown in Figure 6. the mean solid fraction obtained from the observed frazil-ice salinity is 0.266±0.026. Owing to the small salinity variations, this value remains almost constant throughout the experiment. For the pancake ice, solid fractions rapidly increase at the beginning of pancake formation (3–7 hours), but this may in turn just be due to the frazil–pancake transition. Considering again the pancake results after measurement hour 7 as the most reliable, we find that the initial solid-ice fractions in pancakes lie within the range 0.5–0.6, increasing towards the end of the experiment. Largest values of 0.63 are associated with the lowest pancake temperatures.

Fig. 6.

Frazil solid-ice volume fraction, v _s (stars), derived from undrained samples S _i(Fr) (Equation (2)), total v _s (thick dashed line) from atmospheric heat budget (Equation (4)), pancake v _s (dotted curve) derived from salinity S _i(Pk), area covers and T _i profile estimated using median pancake T _s, and pancake v _s (thick solid curve) derived from salt and heat budget (Equations (2), (5), (6) and (7)). the thin dashed curves above and below this last curve represent the error bounds when using Q _s±5Wm^–2. All curves are shown as 0.5 hour running means.

So far we have only combined the salt budget (Equation (6)) with the area fractions from surface temperature segmentation. Our next step is to include the heat flux and corresponding solid fraction budget (Equations (4) and (5)). To be consistent with the salt budget we match the solid fraction 0.266 estimated for the frazil samples at the onset of pancake formation, after 2 hours. We then solve Equations (1), (2) and (4–7) iteratively to obtain the average salinity, solid fraction and area fraction of pancakes. This approach matches salt and heat budget and thus is not dependent on the limited FOV. the results are shown as solid curves in Figures 4–6, with the sensitivity to a 5 Wm^–2 heat flux uncertainty indicated. Pancake salinities are lower than from the surface segmentation approach and decrease from 18 to 15 g kg^–1, while solid fractions are higher, increasing from 0.6 to 0.7. the corresponding prediction of the pancake area coverage is much less than from surface segmentation.