The ocean temperature, θo, and conductivity, γo, can be monitored by a temperature and conductivity recorder (Sea-Bird MicroCAT SBE 37SIP or Valeport mini CT). θo and γo measurements are converted to a salinity, So (g kg-1), using either the GSW toolbox for representative ocean salt composition or using equations presented in for pure NaCl. A sonar (Aquascat 1000R) measures the position of the waterline at the start of the experiment and the position of the base of the sea ice throughout the experiment. We use the GSW toolbox to calculate the speed of sound for a given So and θo and scale the raw sonar output to produce an accurate distance measurement. We can deploy a Liquicel 1.7 × 5.5 MiniModule in our ocean coupled to a LGR GGA-30r-EP OA-ICOS laser spectrometer to measure the concentration of dissolved CH4 and CO2 in the ocean without removing water from the experimental system.

Temperature, θ, profiles through the ocean and sea ice are measured using chains of digital thermometers (Table ). These have a resolution of 116 ∘C and are calibrated against θ measured by the Sea-Bird 37SIP before each run while they sit in well-mixed water. The chains have 1 to 8 cm resolution and are 10 to 80 cm long. Wireharps are used to measure a liquid fraction profile, ϕ, through the sea ice using 1ϕ=γ0R0γt(θ,Sbr)Rt, where R is the resistance between a given wire pair, γ is the conductivity of the solution between the wire pairs, and the subscripts 0 and t indicate the point in time at which the sea-ice front passes a wire pair and some later point in time. We calculate the bulk sea-ice salinity, Ssi (g kg-1), using 2Ssi=ϕSbr(θ), where Sbr(θ) is a the brine salinity. Sbr(θ) is retrieved from the third-order polynomial presented in (for natural seawater composition) 3Sbr/(gkg-1)=-18.7(θ/∘C)-0.519(θ/∘C)2-0.00535(θ/∘C)3 or (for NaCl, fit to the data of ), 4Sbr/(g kg-1)=-17.6(θ/∘C)-0.389(θ/∘C)2-0.00362(θ/∘C)3.

Our version of the wireharps use two alternating current frequencies, 2 kHz (as used in ) and 16 kHz. When using the 2 kHz channel, we calculated γ0γt(θ,Sbr) following . When using the 16 kHz channel, we presume that the electronic double layers driving the conductivity changes are not present and so take γ0γt(θ,Sbr)=1. We have also deployed fibre optics and photodiode light sensors Table  in the sea ice and ocean. Fibre optics can be connected to a spectrometer outside of the cold room (Ocean Optics USB2000+). The light sensors consist of photodiodes, waterproofed in resin, measuring in the blue, green, red, and clear. The wavelength range integrated by the photodiodes is, to a good approximation, 400 to 700 nm. The response of the photodiodes is negligible to light with wavelengths shorter than 400 nm. At wavelengths above 700 nm, the response is generally less than 10 % of the maximum response. Diffusing glass plates sit atop the photodiodes to increase the angle of incidence at which light reaches the diodes. The data sheet for the photodiodes is given in the Supplement.

The temperature, wind speed, and relative humidity of our atmosphere are measured using a weather station (WS600-UMB). Two Los Gatos Research (LGR) greenhouse gas analysers measure CO2, CH4, and H2O mole fractions in the atmosphere and can also analyse the air stream of the Liquicel equilibrator (Table ). A Teledyne T200UP measures NO and NOx and a T200U measures NO and NOy. A Teledyne T400 measures ozone. There is also a zero air generator (T701 H, Teledyne) and a T700U dynamic dilution calibrator with ozone generator.

We set up instrumentation in a dry, clean tank. Sea-ice instrumentation is attached to poles that are free to rise in the vertical as sea ice grows and floats. Ocean and atmosphere instrumentation and sample lines are mounted in a fixed position. The nature of the experiment will determine the state of the tank sides and base. For optical experiments (Sect. ), the tank's inner surfaces should be covered to simplify the light field. We use mirrored sides and a matt black base to approximate infinite lateral boundary conditions and a non-reflective ocean (Fig. ).

To fill the tank, we add 100 kg of salt and mix this with tap or 18.2 MΩ water. The salt can be pure NaCl (used for Sect.  and ) or a natural sea salt composition (Tropic Marin sea salt classic, used for Sect. ). The tank takes around 18 h to fill with de-ionised water or a few hours from the tap, and 100 kg of salt dissolves in a few hours under vigorous circulation by the pumps. Alternatively, real sea water can be delivered to the tank. Filling the tank to 1 m gives an ocean volume of around 3.2 m3. Once the tank is full, we begin to cool the water. The tank surface can be covered at this stage to reduce evaporation. With an atmospheric temperature of -20 ∘C, and an uncovered tank, the temperature of 3.2 m3 of ocean drops by around 1 ∘C every 4 h. If the experiment requires an isolated cold room atmosphere, the cold room is sealed at this stage. Sea-ice growth can be initialised by continuing to cool under constant temperature and vigorous circulation. Alternatively, once the water is within a few tenths of a degree of its freezing point, turning the pumps to a low setting, or off, initialises sea-ice growth within a few minutes. As a rule of thumb, it takes one full week to go from a dry tank to first sea ice.

The water circulation rate and the atmospheric temperature determine the nature of the sea ice . Nilas will grow in quiescent water. In circulating water, a layer of grease ice will form that subsequently consolidates. When the growth temperature is less than -25 ∘C, frost flowers form on the sea-ice surface (Fig. ). By running the heating pads between the main tank and insulation, we are able to maintain free-floating sea ice to 20 cm thickness with no visible water gap at the tank sides. We have verified that the sea ice is free-floating in several experimental runs by gently pressing the corner of the sea ice and noting that it bobs. With no, or insufficient, side heating, the sea ice attaches to the tank sides and the surface floods e.g., resulting in a shiny, liquid surface layer. Sea ice, fast to the tank walls, has been grown up to 25 cm thickness and could potentially be grown thicker. With insufficient side heating and insulation, the ocean tends to supercool (Fig. ). When the ocean supercools, ice grows on the tank sides and instruments, causing problems with measurements and potentially damaging equipment. Growth phases typically last for a few days to a few weeks.

Underlying water may be sampled throughout, providing the volume is replaced. We sample sea ice either by taking cores using a 7.5 cm diameter Kovacs ice corer or using the procedures outlined by to extract sea-ice “slabs”. Sea-ice cores have a known bias, particularly problematic in young sea ice, as brine is lost from the permeable sea ice near the ocean interface. The method of seeks to minimise this bias by collecting sea-ice samples that are still floating and then freezing the sample and surrounding ocean at -40 ∘C to immobilise the brine before processing the sample. The method of becomes difficult and time consuming when sea ice is thicker than around 25 cm due to limitations of the power tools used to extract the slabs and the weight of the slabs. We cut cores and slabs into discrete vertical profiles using a bandsaw with a precleaned blade at -25 ∘C.

For most experiments, it is desirable to have the sea-ice–atmosphere interface exposed to the bulk air within the cold room. Leaving the atmosphere of the tank uncapped ensures that the temperature of the atmosphere overlying the sea ice is responsive to the cold room atmosphere. When the atmosphere is contained by some headspace, the temperature tends to be much warmer than the cold room e.g.. Here, we quantify the degree to which the cold room can be sealed from the outside by deriving the air exchange rate coefficient, k, and time constant, τl, of CH4 exchanging between the cold room to the outside (Fig. ).

To do so, we sealed the cold room and left the tank dry. We then diluted the chamber air by filling it with N2 gas. We performed two dilutions, the first at around 18:00 LT on 12 February 2019 and the other at around 16:00 LT on 13 February 2019. The CH4 concentration in the chamber, Ccham, was monitored using an LGR (GGA type, Table ), where measured air was returned to the cold room in a closed loop (Fig. a and d). The dilution reduced Ccham by around 0.1 ppm in both dilution periods. We stopped the dilution by stopping the N2 flow, at which point the chamber was sealed to the best of our ability. We measured the CH4 concentration in the outer laboratory, Croom, throughout using an LGR (FGGA type, Table ), and we used the difference between Ccham and Croom as the concentration gradient between the chamber and the outside (Fig. b and e). Before the experiment, the two LGR instruments measured the same air for 30 h. The offset between them was (6.4±0.2) ppb. This offset was added to Croom values to make the measurements from the two LGR instruments consistent. Data were averaged in 5 min bins. We modelled the exchange rate as a first-order process, 5dCchamdt=-k(Ccham-Croom). We approximate dCchamdt for each measurement interval using the difference in Ccham between adjacent data points over the 5 min between those data. Panels c (experiment 1) and f (experiment 2) of Fig.  show the relationship between dCchamdt and Ccham-Croom, as well as the linear fit. We derived k=(1.5±0.3) d-1 and k=(1.5±0.1) d-1 for the first and second experiments, respectively. Taking the mean k of the two experiments, the air exchange time constant is then given by 6τl=1/k=(0.66±0.07)d. The physical interpretation of this τl is that the concentration difference between the sealed chamber and the outside will reduce by a factor of e every (0.66±0.07) d.

For some experiments, it may be necessary to spike the ocean with a chemical. We may, in such cases, want to know when the water will be well mixed with respect to our spike. Similar to the air exchange rate (Sect. ), the degree to which the chemical has mixed can be quantified using a time constant, τm. To derive this time constant, we spiked our tank with saturated NaCl solution by injecting it near the tank base and observed the conductivity of our ocean over time as it mixed under the action of three pumps (Fig. ). The temperature of the ocean was stable to within 0.02 ∘C for the hour time period over which we monitored both spikes, so we expect the salinity to be linearly related to the conductivity. The conductivity of the ocean, γo, can then be described by 7γo(t)=e-tτm(γo(t0)-γo,∞)+γo,∞, where γo,∞ is the maximum ocean conductivity obtained after the spike and t0 is the time of the spike. Fitting an exponential curve to a manually defined spike period for the two runs produces τm=(4.2±0.1) min (run 1) and τm=(4.1±0.1) min (run 2). We take the mean of the individual experiments to characterise our mixing rate, such that τm=(4.2±0.1) min. The tank should therefore mix to more than 99% of the perfectly mixed concentration in less than 30 min.

Producing a homogeneous light field in a laboratory environment can be difficult, because shading and reflection can cause heterogeneities. The intensity of our lights is also temperature dependent, which is particularly important in the RvG-ASIC given the wide range of experimental temperatures. In this section, we aim to characterise our light field, both in terms of lateral heterogeneity and with changes in temperature.

To assess the spatial variability, we measured PAR (photosynthetically active radiation, λ=400 to 700 nm) across the tank area at 1 m height (Fig. ). These measurements were made using a LI-COR LI-1800 spectroradiometer integrated from 400 to 700 nm. The PAR irradiance, IPAR, at 1 m above the tank base and 25 ∘C was generally within 80 % of the maximum recorded intensity, IPAR,max, but dropped as low as 60 % in one corner of the tank.

We measured spectra for UV (UV-A lighting, 320 to 380 nm) and PAR (LED lighting, 400 to 700 nm) at -30, -20, -10, and 0 ∘C using a USB2000+ spectrometer referenced against the Metcon spectral radiometer (Fig. ). To quantify the temperature sensitivity, we integrated the UV and PAR spectral irradiance over their respective wavelength ranges (320 to 380 nm and 400 to 700 nm, respectively). The maximum irradiance was 5.6 W m-2 for UV and 26.4 W m-2 for PAR, both at 0 ∘C. The temperature sensitivity of the irradiance was (0.16±0.01) W m-2 ∘C-1 for UV and (0.028±0.003) W m-2 ∘C-1 for PAR. The UV-A lights are more strongly temperature dependent, with irradiance dropping by 83 % between 0 and -30 ∘C, while PAR irradiance dropped by just 3 %.

The importance of understanding the sea-ice light field is critical as thinner, fresher, and more transient sea ice becomes more common. Even thin sea ice without snow cover impacts the transmission of PAR to the ocean and has been seen to accumulate algal biomass rapidly within the sea ice . Photochemical production rates within sea ice are also dependent upon sea-ice optical properties. Irradiation through sea ice has been postulated as a source of OH radicals and as a stressor of diatoms that may lead to iodine production from algae .

The rationale for UV–Vis illumination experiments in a controlled sea-ice facility is twofold: first, to allow for experiments investigating the optical properties of sea ice (and potentially other mediums such as snow); second, to allow simple experimental simulations of photochemistry or biology occurring in the sea ice, atmosphere, or ocean. The extinction coefficient, κ, of PAR in sea ice quantifies the rate of attenuation of PAR with depth. Here, we present measurements of κ for sea ice grown in the RvG-ASIC and compare those to previously reported values.

To quantify κ in our experimental sea ice, we performed an experiment using visible lighting, measuring a vertical profile of irradiance using eight photodiodes (Sect. ). The photodiodes respond dominantly to 400 to 700 nm light, and our lighting provides irradiance predominantly in the 400 to 700 nm range, so κ as measured by the photodiodes is a good estimate of κ for PAR. The tank sides were covered with mirrored tape to approximate infinite lateral boundary conditions, and the tank base was covered with a matt black plastic sheet to approximate no reflection from the ocean. A 3 mm opal polycarbonate sheet was placed 20 cm below the light rack to create a diffuse light field, and black curtains were draped between the lights and the tank to prevent reflection off the shiny cold room walls (Fig. ). We deployed the photodiodes with 2.8 cm vertical resolution. The sensors were calibrated pre-deployment against the Metcon spectral radiometer. Before each experimental run, while sitting in open water, an additional correction was applied to the in situ sensor output to bring it in line with depth profiles measured using the spectral radiometer. We performed four experiments at growth temperatures of -10, -20 (two runs), and -30 ∘C (Table ). For each run, we present vertical light profiles taken at the end of the growth phase, as well as κ calculated by two methods (Fig.  and Table ). Method A uses 8κ=ln⁡((1-ζs)I(+1.4cm)I(z))/z. The depth, z, is chosen to be the shallowest sensor not yet frozen into the sea ice, and spectral reflectance, ζs, assumed to be 0.05. measured κ ranging from 3.1 to 4.7 m-1 in 24 to 28 cm thick sea ice (Table ). measured κ ranging from 2.9 to 4.7 m-1 in 17 to 27 cm thick sea ice. For method B, we also calculated κ by fitting a linear model to 9ln⁡(I/(Wm-2))=-κz+c, similar to . Equation () represents a light field decaying exponentially with depth. While methods A and B both provide estimates of κ, they do so in different ways, with method A using measurements external to the sea ice while method B uses only measurements within the sea ice. Previous studies, compared to our measurements in Table , have calculated κ by iteratively solving the photometric model such that κ provides the best fit to measured albedo and transmission – which we call method C.

Measured light profiles taken at the end of four experiments are shown in Fig. . The thickness varied by up to 3 cm between the four experiments. At the end of the experiments, runs 1 and 2 had six sensors frozen into the sea ice and experiments 3 and 4 had five sensors frozen into the sea ice. The light intensity at 1.4 cm above the sea ice was between 7.7 and 8.9 W m-2 and reduced to between 2.8 and 3.6 W m-2 at 18.2 cm depth. For experiments 1 and 4, the shallowest frozen sensor measured higher intensity than the sensor above the sea ice, which is a phenomenon that was predicted by and that is due to the change in refractive index across the air–ice interface.

Using method A, our calculated κ ranges from 3.7 (experiment 4) to 6.1 m-1 (experiment 2). Using method B, our calculated κ ranges from 4.4 (experiment 2) to 6.2 m-1 (experiment 4). For both methods, the range of κ observed in our tank overlaps with the range of κ observed previously for thin sea ice. These results build confidence that the RvG-ASIC is a useful tool for future biological or photochemical experiments in young sea ice and validate this particular experimental setup for experiments involving visible light.

Bulk salinity is a sea-ice state variable that, when measured alongside temperature, can allow for estimation of the sea-ice liquid fraction. Growing sea ice desalinates rapidly by gravity drainage . The bulk salinity of natural, young sea ice has been measured using cores and wireharps . sampled young sea ice in North Star Bay, Greenland, in growing sea ice between 5 and 23 cm thick. Salinities ranged from around 5 to 20 g kg-1 vertically with mean salinities decreasing from 20 g kg-1 early in the growth to around 10 g kg-1 after 10 d. cut a hole in Arctic sea ice and measured the salinity profile, in situ, as it refroze up to 17 cm thickness (the depth range of their instrument). Over two experiments, they measured bulk salinities ranging from around 35 g kg-1 at the sea-ice–ocean interface to around 4 g kg-1 in the interior. Mean salinities decreased from around 35 towards 10 g kg-1 (experiment 1) and around 15 g kg-1 (experiment 2). Here, we present bulk salinities for sea ice grown in the RvG-ASIC, comparing it to the range of salinities observed in nature. Quantifying the vertical bulk salinity profile in sea-ice cores – the most common sea-ice sampling methodology – is difficult due to the known bias during coring of brine loss and bulk salinity underestimation. Several other methodologies have been proposed. present a destructive sampling methodology that attempts to prevent brine loss upon sampling (Sect. ). present instrumentation, called a wireharp, to quantify the in situ bulk salinity profile. Several groups have presented gravity drainage parameterisations that quantify the bulk salinity profile during sea-ice growth , noting that gravity drainage is the dominant process redistributing salt in growing sea ice .

We performed two sea-ice growth experiments in the RvG-ASIC and estimated the sea-ice salinity by (1) constructing a salt and mass budget, (2) taking sea-ice cores, (3) taking sea-ice slabs , (4) using a wireharp , and (5) using the gravity drainage parameterisation. Our artificial ocean was composed of Tropic Marin sea salt and deionised water such that So=(31.3±0.1) g kg-1. We grew free-floating sea ice to (11.0±0.9) cm (run 1) and (16.4 ± 2.7) cm (run 2) thickness (mean and standard deviation of measured ice core thickness). Measurements of sea-ice temperature, thickness, and the initial ocean salinity were used to force the model, which has been used previously to model experiments in the RvG-ASIC . We used the gravity drainage parameterisation, because it performed well in previous studies and has tuning parameters shown to perform well for Arctic field data . The model has two tuning parameters: the critical Rayleigh number, Rac, and the desalination strength, α. We estimate an uncertainty on the modelling by running the model for each of the three sets of tuning parameters presented in (Table ). The model is forced by measured temperature profiles and sea-ice thickness in lieu of modelled thermodynamics (Fig. ). Cores and slabs were extracted according to Sect.  and sectioned vertically into 1 to 2 cm layers. Ice samples were melted and placed in a 20 ∘C thermostatic bath. Conductivity was measured using a conductivity probe (Orion) and calibrated with certified reference material (Thermo Scientific Eutech Handheld Meters Calibration Solution). Bulk salinity was derived using the GSW toolbox . We discarded three discrete samples (one from the cores and two from the slabs for run 1) because of punctured sample bags. A single wireharp profile was taken in the middle of the tank and used to derive ϕ and Ssi (Eqs.  and ).

The mass balance was constructed by conservation of mass and salt from the start of each run (t0) to the final sampling (tend). The conservation of mass gives 10msys=mo+msi, where m is mass, and the subscripts sys, o, and si indicate the experimental system, ocean, and sea-ice compartments, respectively. Noting that msys=mo(t0), the conservation of salt at the end of the experiment (tend) gives 11msysSo(t0)=moSo(tend)+msiS‾si. Our desired variable is the vertically integrated bulk sea-ice salinity, S‾si, which we recover by substituting Eq. () into Eq. (), giving 12S‾si=rΔS+So(tend), where 13ΔS=So(t0)-So(tend) and r gives the ratio of the mass of the experimental system to the mass of the sea ice at final sampling, such that 14r=ho(t0)ρo(t0)hsiρsi. The error on Eq. () was calculated by Gaussian propagation, such that 15u(S‾si)=r2ΔS2((u(r)r)2+(u(ΔS)ΔS)2)+u(So)2, where u gives the uncertainty of the individual terms. The dominant uncertainty, accounting for >95 % of u(S‾si), propagates from u(r) and derives chiefly from the variability in measured hsi. The errors on So(t0) and So(tend) are highly correlated given the stability of the salinity sensor over the short duration of the experiment, and as such u(ΔS)≪0.01 g kg-1 and has a negligible impact on u(S‾si). The sensitivity of S‾si to our choice of ρsi is around 0.02 (g kg-1) / (kg m-3), such that a 20 kg m-3 increase in ρsi causes a 0.4 g kg-1 increase in our estimated S‾si, which is well within our uncertainty bounds.

We first turn to the vertical profiles of Ssi (Fig. ). For both runs, the wireharp profile at low frequency including the electrical double-layer correction proposed by is very similar to the high-frequency wireharp profile without such correction. This confirms that at higher frequencies the impact of the double layer becomes less important, and no more correction to the profile is necessary. For simplicity, and because the high-frequency channel has not yet registered ice growth in the deepest run 1 sensor, we proceed by discussing only the low-frequency channel. For both runs, and for all profiles, there is a salinity maximum near the sea-ice–ocean interface (z=hsi). The model and the wireharps converge towards So as z approaches hsi. The cores and the slabs produce the lowest Ssi near this lower interface, with the slabs giving higher Ssi relative to the cores in this region in run 2. For run 1, the wireharps generally measure the highest Ssi, the cores and slabs are similar and measure the lowest Ssi, and the model is intermediate. For run 2, the cores, slabs, wireharps, and model are generally consistent between z/hsi= 0.6 to 0.1. Near the upper interface, the model predicts higher Ssi than the measurements, and the wireharps give the lowest Ssi. For run 1, the cores and slabs are consistent throughout, but the wireharps, model, and discrete samples generally disagree. Towards the sea-ice–atmosphere interface (z=0), the wireharps produce minima in Ssi. The cores and slabs show an ≈1 g kg-1 increase in Ssi for the shallowest layer relative to the layer below, forming a C-shaped profile. Modelled Ssi at z=0 is around 10 g kg-1 greater than in the interior, forming a C-shaped profile that is more pronounced than in the discrete samples at both interfaces. The difference in θ between the wireharps and the forcing is at most 0.2 ∘C which, depending on Sbr, translates to a 2 % to 3 % difference in Ssi.

The mass balance produces S‾si=(11.0±2.0) g kg-1 (run 1) and S‾si=(11.0±4.1) g kg-1 (run 2) (mean and standard deviation, Table ). These values represent the vertically integrated bulk sea-ice salinity, averaged over the tank footprint. We compare the measurements and model to the mass balance by calculating S‾si for the profile produced by each method at the end of each run. For the cores and slabs, we first averaged the individual layer measurements for each sample and then took the average of these samples to be S‾si. For the wireharps, we linearly interpolated measured Ssi to the midpoint between each measurement, and took the mean of these interpolated Ssi to produce S‾si. In this way, we more closely approximate the treatment of the cores and slabs, for which the edge of the bottommost layer is at z=hsi. Model S‾si is the mean of the Ssi of individual model layers. The cores and slabs perform similarly, underestimating the mass balance for each run. The wireharps overestimate the mass balance in run 1 and capture the run 2 mass balance to within 1 standard deviation. The best estimate model S‾si is consistent with the mass balance for both runs.

The underestimation of S‾si by the cores (Table ) highlights a known bias in sea-ice core bulk salinity measurement, which is that of brine loss upon sampling. This bias is apparent in the vertical profiles (Fig. ), where cores give the lowest estimates of Ssi in the lower third of the profile, which we expect to be most affected by brine loss. The slabs perform similarly to the cores, potentially due to brine drainage during the shock freezing process. The slabs retain more brine than the cores in run 2, as shown by their higher Ssi in the lower portion and their higher S‾si. Previous work has tuned gravity drainage parameterisations to slab Ssi measurements , and these tuning parameters may therefore be biased to predict too much desalination. The wireharp performance is more variable, overestimating S‾si in run 1 and capturing the mass balance in run 2. The performance of the model, forced with measured θ and hsi, solely reflects the gravity drainage parameterisation and its tuning , incorporating minimal error from thermodynamics. Only the model captures the mass balance for both runs. The mass balance is, for both runs, within the range of young sea-ice salinities observed in nature .

The growth rate of sea ice depends on the balance of fluxes at the sea-ice–ocean interface, the thermal conductivity of the sea ice (K), sea-ice thickness hsi, and the absolute sea-ice surface temperature (Ts). Growth rates in young sea ice have been observed to range from 2.7 to 12 cm d-1 . Following , the change in sea-ice thickness is 16dhsidt=-FC/(ρiL), where L gives the latent heat of fusion of water, and FC gives the vertical conductive heat flux through the sea ice. FC, was calculated using 17FC=K(Ts-Tf)/hsi. The vertically averaged thermal conductivity is K=Ki(1-ϕ)+Kbrϕ, where the subscripts i and br indicate pure ice and brine, respectively. Tf is given by the salinity-dependent freezing point of the ocean, and Ts is found by iteratively solving the surface energy balance of the sea ice 18FL↓-FL↑+FS+FC=0, where sensible and latent heat fluxes have been neglected because the wind speed in our experiments was less than 0.1 m s-1. We parameterised the downwelling longwave flux, FL↓, following , as 19FL↓=σ(0.746+0.0066(ψ/mbar))Ta4, where σ is the Stefan–Boltzmann constant, ψ is the water vapour pressure, and Ta is the absolute temperature of the atmosphere. The upwelling longwave flux, FL↑, is taken to be 20FL↑=σϵTs4, where the emissivity of sea ice is ϵ=0.99. We compare the growth rate of our experimental sea ice to natural sea ice by modelling and measuring the growth rate of the four experimental runs presented in Sect.  and Table  (Fig. ). When the temperature of a sensor at depth z dipped by 0.1oC below the water temperature for 30 min, we took hsi=z. For each run, we forced the sea-ice growth and desalination model (Sect. , ) with the mean Ta and ψ during the freezing period, as measured by the weather station (Table ). Incoming shortwave flux, FS, was taken to be 8 W m-2 based on measurements of PAR just above the sea-ice surface (Sect. ). In this case, Eqs. () to () were used to calculate Ts, hsi, and the internal sea-ice temperature profile. Tf was calculated from the salinity-dependent freezing point of the model ocean.

Over the four experiments, measured growth rates range from 2 cm d-1 (experiment 3, highest air temperature) to 4 cm d-1 (experiment 2, lowest air temperature). These growth rates are within the range of those reported by . Our growth rates are at the low end of this range, because the low wind speed in our facility effectively removes the latent and sensible heat fluxes present in the field. The change in hsi with time is not linear; rather, as has been observed before , the rate of increase in hsi decreases with increasing hsi. Modelled thickness captures the non-linearity in the measured growth rates and the order of the growth rates, with coldest temperatures producing fastest growth. Modelled thickness deviates from the measurements by up to 3 cm.

Modelling thickness in this way is useful for planning experiments but – considering temperature profiles are measured during each experiment – measuring temperature and thickness gives better precision. Growth rates in the RvG-ASIC are within the range of those measured in the field and are in agreement with thermodynamic modelling .