Frazil ice growth and ice production during katabatic wind events in Ross Sea polynyas, Antarctica

During katabatic wind events in the Terra Nova Bay and Ross Sea polynyas, wind speeds exceeded 20 m s-1, air temperatures were below -25 °C, and the mixed layer extended as deep as 600 meters. Yet, temperature and salinity profiles were not perfectly vertical, as one would expect with vigorous convective heat loss. Instead, the profiles revealed bulges of warm and salty water starting at the ocean surface and extending to the top tens of meters. Considering both the colder air above and colder water below, we surmise that the increase in temperature and salinity reflects latent heat and salt release during unconsolidated frazil ice production throughout the upper water column. We use a simplified salt budget to analyze these anomalies to estimate in-situ frazil ice content 5.8 and 0.13 kg within the top 50 m of the water column. Estimates of vertical mixing by turbulent kinetic energy dissipation reveals rapid convection in these unstable density profiles, and mixing lifetimes from 2 to 30 minutes. The corresponding ice production rates yield an average ice thickness of 52 cm day-1, which compares well with previous empirical and model estimates. However, our individual estimates of production up to 358 cm day-1 reveal the intensity of short-term ice production in the windiest sections of the Terra Nova Bay Polynya.

Latent heat polynyas form in areas where prevailing winds or oceanic currents create divergence in the ice cover, leading to openings surrounded by extensive pack ice . The open water of polynyas is critical for airsea-heat-exchange, since ice covered waters are better insulated .     this figure depicts the release of latent heat of fusion and brine rejection as a frazil ice crystal is formed. Includes key features of frazil ice crystals including diameter, thickness, and shape.
Polynyas drive extreme oceanic heat loss which creates in-situ "supercooled" water, that is colder than the freezing point ). Two criteria for ice production in polynyas from supercooled water are large net heat loss from the water and transport of the frazil ice away from the formation region; both criteria are achieved in the polynya by katabatic winds and cold air temperatures . These conditions generate sea ice as fine disc-shaped or dendritic crystals called frazil ice. These frazil ice crystals depicted in Figure 2, measure about 1-4 millimeters in diameter and 1-100 micrometers in thickness . Katabatic winds sustain the polynya by clearing frazil ice, forming pancake ice which piles up at the polynya edge to form a consolidated ice cover (Morales . The production and sweeping away of frazil ice crystals creates an-efficient ice production mechanism whereby seawater is kept in contact with cold air, unmitigated by an insulating layer of ice . Brine rejection ) and a large amount of latent heat release accompany the continuous ice production. In the Ross Sea, these coastal polynyas produce the precursor to Antarctic Bottom Water, a water mass known as High Salinity Shelf Water (HSSW) that is created by the large volumes of brine rejection Sansivero et al, 2017;Tamura et al, 2007;Cosimo & Gordon, 1998;.
Given the importance of Antarctic Bottom Water to global circulation, polynya ice production rates have been widely studied and modeled. Gallee (1997), , , and Sansivero et al (2017) used models to calculate polynya production rates on the order of tens of centimeters per day. Schick et al (2018) and  used heat fluxes to estimate polynya ice production rates, also on the order of tens of centimeters of ice thickness per day.
However, quantitative estimation of polynya ice production is challenging due to the difficulty of obtaining in-situ measurements (Tamura et al, 2007).
During a late autumn 2017 oceanographic cruise expedition to the Ross Sea as part of the PIPERS (Polynyas, Ice Production and seasonal Evolution in the Ross Sea) project, Conductivity, Temperature, and Depth (CTD) vertical profiles acquired in the Ross Sea coastal polynyas indicated anomalous regions of saltier, warmer water near the surface. Simultaneously, visual field observations noted active frazil ice formation in these same locations. We hypothesize that the excess temperature is evidence of latent heat of fusion from frazil ice formation and that the excess salinity is evidence of brine rejection from frazil ice formation. We evaluate the reliability of these CTD measurements by comparing the shape and size of the profile anomalies with estimates of the CTD precision and stability, and by using supporting evidence of the atmospheric conditions that are thought to drive frazil ice formation (e.g. temperature and wind speed). Next, we estimate the production of frazil ice using the temperature and salinity anomalies. Finally, we attempt to put bounds on the mixing timescale of these anomalies, by asking how long they would take to mix into the background, if the formation process ceased. This, in turn provides an estimate of near instantaneous frazil ice production. Last, we discuss the implications of these results.  . The RSP is Antarctica's largest recurring polynya; it forms in the central and western Ross Sea . The average area of the RSP is 27,000 km 2 but can grow as large as 50,000 km 2, depending on environmental conditions . It is located to the east of Ross Island, adjacent to the Ross Ice Shelf, and typically extends the entire length of the Ross Ice Shelf . TNBP, located to the north of Drygalski ice tongue, and MSP, the smallest of the three polynyas, are both located in the western Ross Sea, depicted in Figure 3  . The area of TNBP, on average is 1000 km 2 , but can extend up to 5000 km 2 ; the oscillation period is 15-20 days . This paper focuses primarily on TNBP and secondarily on RSP.
During the autumn and winter season, Morales Maqueda et. al (2004) estimated TNBP cumulative ice production around 40-60 meters of ice, or approximately 10% of the annual sea ice production that occurs on the Ross Sea continental shelf. The RSP, while having a lower daily ice production rate, produces three to six times as much as TNBP annually due to its much larger size . Ice production in polynyas plays an important role in the modification of HSSW. In areas over the continental shelf, brine rejection paired with super-cooled temperatures at or below the freezing point produces especially dense shelf waters ). In the case of the Ross Sea, the cold, dense HSSW formed on the shelf eventually becomes Antarctic Bottom Water off the shelf, the densest water in global circulation. TNBP produces especially dense HSSW, driven by its higher salinity, and despite being smaller than RSP, it produces approximately 1-1.23 Sv annually .

PIPERS Expedition
We collected our data during the 63-day PIPERS expedition aboard the RVIB   In many instances, the up cast recorded a similar thermal and haline anomaly, however the 24 bottle CTD rosette package creates a large wake that disturbs the readings on the up cast. Therefore, we use only the down cast profiles for this analysis.
The SBE 911 data were post-processed with post-calibrations by Seabird, following standard protocol, and quality control parameters. Profiles were binaveraged at two size intervals: one-meter depth bins and 0.1-meter depth bins, to compare whether bin averaging influenced the heat and salt budgets. We observed no difference between the budget calculations derived from one-meter vs 0.1-meter bins; the one-meter bins are presented in this publication. All thermodynamic properties of seawater were evaluated via the Gibbs Seawater toolbox which uses the International Thermodynamic Equation of Seawater -2010 (TEOS-10).

Weather observations
During the PIPERS expedition, multiple katabatic wind events were observed within the TNBP and RSP. The NB Palmer was in TNB from May 1 through May 13; during this period the hourly wind speed and air temperature data from Weather Station Manuela, shown on Figure 3a, the automatic weather station on Inexpressible Island, was compared to NB Palmer's meteorological suite, normalized to a height of 10 meters, Figure 4. In most cases, the winds and air temperature from both locations follow the same pattern, with shipboard observations from the NB Palmer observations being lower in intensity (lower wind speed, warmer temperatures) than Station Manuela. While in the RSP May 16-18, the wind speed and air temperature from NB Palmer is compared to Station Vito, shown on Figure 3a and located on the Ross Shelf Ice Sheet. At Station Vito, the air temperature is colder, but the wind speed is less intense, most likely due to higher drag across the ice sheet.
During the CTD sampling within TNBP there were 4 periods of intense katabatic winds, with each period lasting for 24 hours or longer. During the CTD sampling within RSP there was one period of near katabatic strength winds that lasted 24 hours or longer. During each wind event, the air temperature oscillated in a similar pattern and ranged from approximately -10 ℃ to -30 ℃.

EVIDENCE OF FRAZIL ICE FORMATION
During PIPERS, the CTD profiles acquired in the RSP and TNBP defied expectations for vertical profiles in the presence of strong winds. Despite air temperatures well below freezing and strong winds, the profiles presented with anomalous regions of warmer water near the surface. The excess temperature was accompanied by anomalous regions of saltier water. Simultaneously, visual field observations noted active frazil ice formation in these same locations. We suggest that the excess temperature is evidence of latent heat of fusion from frazil ice formation and that the excess salinity is evidence of brine rejection from frazil ice formation.
We evaluate the reliability of these CTD measurements by comparing the shape and size of the profile anomalies with estimates of the CTD accuracy, and by using supporting evidence of the atmospheric conditions that are thought to drive frazil ice formation (e.g. temperature and wind speed). Next, we estimate the production of frazil ice using the temperature and salinity anomalies.

Selection of profiles of interest
We used the following selection criteria to identify profiles from the two polynyas that appeared to be under the influence of frazil ice formation: (1) a deep mixed layer extending several hundred meters, and down to 600 meters in one case meters of the profile (e.g. Figure 7). Each temperature profile was individually plotted over the entire depth range to identify the deep mixed layer, ranging from 100 to 600 meters.
Figure 5: 1000-meter Conservative Temperature profiles of all 57 out of 58 PIPERS CTD stations. One station not included due to significantly warmer temperature outside temperature range shown here. The CTD stations from TNB and RS with frazil ice anomalies and deep mixed layers are highlighted in blue and the stations without anomalies are represented in red. In addition to the large mixed layer, these profiles also represent the coldest temperatures.   Figure 7.b and 7.e have a salinity anomaly that approaches 50 meters, so the plot extends to 80 meters to best highlight it. All of the plots (a-h) have an x-axis representing 0.03 g kg -1 .

Evaluating the fidelity of the CTD measurements
To evaluate the uncertainty associated with the temperature and salinity anomalies at each of the polynya stations, we compared each anomaly to the initial accuracy of the SBE 911: ± 0.001 ℃ and ± 0.0003 S m -1 or 0.00170 g kg -1 when converted to absolute salinity. To quantify the maximum amount of the temperature anomaly, the baseline excursion, ΔT, was calculated throughout the anomaly ΔT = Tobs -Tb, where Tobs is the in-situ conservative temperature and Tb is the in-situ baseline, which is extrapolated from the far field conservative temperature within the well-mixed layer below the anomaly. Taking the single largest baseline excursion from each of the 11 anomalous CTD profiles and averaging them, we compute the average baseline excursion of 0.0064 •C. While, this is a small change in the temperature, it is still 32 times larger than the stated precision of the SBE 911 (0.0002 ℃). The same approach applied to the salinity anomalies is 0.0058 g kg -1 , which is 10 times larger than the instrument precision (0.00004 S m -1 ). Table 1 includes the maximum temperature and salinity anomalies for each CTD station.
One concern was that frazil ice crystals could interfere with the conductivity sensor. It is possible that ice crystals smaller than 5 mm can be ingested into the conductivity cell and create spikes in the raw conductance data. Frazil crystals smaller than 100 µm are theoretically small enough to fit in between the conductivity cell electrodes and thereby decrease the conductance/salinity that is reported by the instrument ). To test for frazil interference, the absolute salinity was plotted from raw conductivity data and from 1-meter binned data for the CTD Stations with anomalies, Supplemental Figure 1.The raw data shows varying levels of noise in the signal and spikes of lesser magnitude values that are likely due to frazil ice crystal interference. However, the 1-meter binned data, does not follow the spike excursions, indicating that binning minimizes or removes the effects of the noise and spikes. We conclude that there is frazil interference in the conductivity, but the lesser magnitude and 1-meter bins negates the effects.
Considering the consistency of the temperature and salinity measurements within and below the anomalies, and also considering the repeated observation of anomalies at 11 CTD stations, we infer that the observed anomalies are not an instrumental aberration and can be interpreted as valid CTD profiles.

EISCam Observations of frazil ice formation
During PIPERS a EISCam (Evaluative Imagery Support Camera) version 2 instrument was operating in time lapse mode, recording figures of the ocean surface continuously. The images of the water surface, that coincide in time with the 11 anomalous CTD profiles, reveal long streaks and large aggregations of frazil ice in every frame (Figure 8). The winds were strong enough at all times to set up wave fields or advect frazil ice and resulted in downstream frazil streaks and pancake ice in most situations. Smaller frazil streaks and a curtain of frazil ice below the frazil streak are also visible. Figure 8: Images from NB Palmer as EISCam (Evaluative Imagery Support Camera) version 2. White areas in the water are loosely consolidate frazil ice crystals being actively formed during a katabatic wind events. d.) brightened to allow for better resolution.  conducted laboratory experiments to reproduce the conditions observed in polynyas. They exposed their tank, measuring 2-m length, 0.4-m width and 0.6-m depth to air temperatures at -10 ℃ and wind speeds of 6 −1 .

Parallels between the PIPERS profiles and lab experiments
They observed supercooling in the range of 0.1 to 0.2 ℃ at the water surface and found that after 20 minutes the rate of super-cooling slowed due to release of latent heat, coinciding with visually observed frazil ice formation. Simultaneously with formation of frazil ice crystals, they observed an increase in salinity from the rejection of brine. After ten minutes of ice formation, the temperature of the frazil ice layer was 0.07 ℃ warmer and the layer was 0.5 to 1.0% saltier .
In this study, we found the frazil ice layer to be on average 0.0064 ℃ warmer than the underlying water. Similarly, the salinity anomaly was on average 0.0058 g kg -1 saltier, which equates to 0.017% saltier than the water below. While our anomalies were significantly smaller than those observed in this experiment, the same trend of super-cooling, followed by onset of frazil ice formation and the appearance of a salinity anomaly, was observed during PIPERS as by .
The forcing conditions and dimension constraints of the tank experiment can explain the discrepancies in the size of temperature and salinity anomalies formed.

Similarities to Platelet Ice formation
In the polynya, katabatic winds and sub-freezing air temperatures create supercooled water near the surface, which in turn drives frazil ice formation. While the mechanism for supercooling differs, Robinson et al (2017)

The anomalous profiles from TNBP an RSP appear to trace active frazil ice formation
Throughout sections 2 and 3, we have documented that the anomalous profiles from TNBP and RSP appear to trace frazil ice formation. In §3.1 and §3.2, we showed that the CTD profiles in both temperature and salinity are reproducible and large enough to be distinguished from the instrumental noise. In §2.4, the strong winds and sub-zero air temperatures supported both ice formation and advection. The coincident EISCam measurements reveal significant accumulation of frazil ice crystals on the ocean surface, while the NB Palmer was in TNBP and RSP. In §3.4 and §3.5, we note the commonalities between the PIPERS polynya profiles and frazil ice formation during platelet ice formation and during laboratory experiments of frazil ice formation.
Given the correlation of strong winds, cold air temperatures, water temperature around the freezing point, we find no simpler explanation for the apparent warmer, saltier water near the surface of these 11 CTD profiles. Considering the similarity in conditions found during lab experiments, platelet ice formation, we concluded these profiles reflect measurable frazil ice formation.

ESTIMATION OF FRAZIL ICE CONCENTRATION USING CTD PROFILES
Having selected the CTD profiles that reveal frazil ice formation, we next ask "how much frazil ice formation is inferred by these T and S profiles?". The inventories of heat and salt from each profile can provide independent estimates of frazil ice mass, that should be comparable. To simplify the inventory computations, we neglected the horizontal advection and diffusion of heat and salt; this is akin to assuming that lateral variations are not important because the neighboring water parcels are also experiencing the same intense vertical gradients in heat and salt. We first describe the computation using temperature in § 4.1 and the computation using salinity in § 4.2.

Estimation of frazil ice concentration using temperature anomalies
We used the temperature profiles to compute the "excess" heat inside the anomaly. Utilizing the latent heat of fusion as a proxy for frazil ice production we estimated the amount of frazil ice that would be formed in order to create such an anomaly. For each station, we first estimated the enthalpy inside the temperature anomaly  as follows. Within each CTD bin, we estimated the excess temperature ΔT = Tobs -Tb, where Tobs is the in-situ conservative temperature and Tb is the in-situ baseline or far field conservative temperature. The excess over the baseline is graphically represented in Figure 9a. Because we lacked multiple profiles at the same location, we were not able to observe the time evolution of these anomalies. Consequently, Tb represents our best inference of the temperature of the water column prior to the onset of ice formation; it is highlighted in Figure 9a with the dashed line. We established Tb by looking for a near constant value of temperature in the profile directly below the temperature bulge. In most cases the temperature trend over depth was very linear, monotonic and close to the freezing point, however it did have slight variations. After selecting the starting location, the conservative temperature was averaged over 10 meters to remove slight variations in the conservative temperature and minimizing selection bias.
The Mass of ice derived represents the total mass of ice, in kg, in the volume of water, = * . A more detailed explanation of equations 1 and 2 is contained in Supplemental 1. The mass of ice derived from the temperature anomaly for each station is listed in Table 1.

Estimation of frazil ice concentration using Sea Bird CTD Salinity profiles
The mass of salt within the salinity anomaly was used to estimate ice formation. Assuming that frazil ice crystals do not retain any brine and assuming there is no evaporation, the salinity anomaly is directly proportional to the ice formed. By above the baseline salinity ( ) is = − , and is shown in Figure 9b. The initial value of salinity ( ) was established by observing the trend in the salinity profile directly below the haline bulge; in most cases the salinity trend was very linear and monotonic beneath the bulge, however in general the salinity profiles were less homogeneous than the temperature profiles. After selecting the starting location, the absolute salinity was averaged over 10 meters.
To find the total mass of frazil ice ( ) in the water column, the integral of each component of the salt ratio is taken over the depth range of the anomaly. This integral is multiplied by the total Mass of Water ( ) initially in the depth range of the anomaly. The resulting estimates of mass ice produced are listed in Table   1.
A more detailed explanation of equations 3 and 4 is contained in Supplemental 3.

Summary of the Mass of Ice derived from Temperature and Salinity
An appreciable volume of frazil ice growth in supercooled water gave rise to salt rejection near the ocean surface, as depicted in the salinity profiles (Figure 7). The derived masses of ice are listed in Table 1. We estimate between 5.8 and 0.13kg of frazil ice were formed, depending on whether temperature ( § 4.1) or salinity ( § 4.2) anomalies are used for the budget.
It is noteworthy that the salt inventories estimate between 2 and 6 times more frazil ice than temperature inventories. The smaller amount of ice derived from the heat inventory calculation is likely caused by atmospheric heat loss. Whereas, the salt rejected by frazil ice can only mix into the ocean, the heat produced by frazil ice can quickly escape to the very cold atmosphere, which is driving much of the supercooling in the first place. Additionally, the salinity calculation assumed no evaporation. Evaporation would contribute to excess salinity; however it would also decrease the temperature. Given the positive temperature anomaly and high relative humidity (on average 78.3%), the effects of evaporation on salinity were neglected.
The effects of evaporation would reduce the mass of ice derived from the salinity anomaly, however,  found that evaporation was secondary to ice production and contributed a mere 4% to salt flux. Because the heat budget has an extra loss term that we are not able to easily quantify, we suggest that ice mass from the heat inventory significantly underestimates frazil growth as compared to the salt inventory. *Station 26 did not have a measurable salinity anomaly but was included due to the clarity of the temperature anomaly. Conversely, **Station 33 did not have a measurable temperature anomaly but was included due to the clarity of the salinity anomaly.

INTERPRETING THE LIFETIME OF THE ANOMALIES
One question that arises while trying to understand these T and S anomalies is how to interpret their persistence or lifetime: are they short-lived or do they represent an accumulation over some longer ice formation period? One interpretation is that the anomalies begin to form at the onset of the katabatic wind event, implying that the time required to accumulate the observed heat and salt anomalies is similar to that of a katabatic wind event (e.g. 12-48 hours). This, in turn would suggest that the estimated frazil ice production occurred over the lifetime of the katabatic wind event. Another interpretation is that the observed anomalies reflect the near-instantaneous production of frazil ice. In this scenario, heat and salt are simultaneously produced and actively mixed away into the far field. In this case, the observed temperature and salinity anomalies reflect the net difference between production and mixing. One way to address the question of lifetime is to ask, "if ice production stopped, how long would it take for the heat and salt anomalies to dissipate?" The answer depends on how vigorously the water column is mixing, therefore in this section we examine the mixing rate. We can first get some indication of the lifetime by simply examining the density profiles.

An apparent instability in each density profile
Initially, we expected buoyancy production from excess heat to effectively offset the buoyancy loss from excess salt within each anomaly. The result would be a stably stratified or at least neutrally buoyant water column. This seemed most likely, because the conventional interpretation is that, even though a profile may appear unstable in T or in S, an unstable density profile is swiftly destroyed by convective instability. Instead, the majority of the 11 profiles revealed that temperature did not compensate for salinity, leading to observations of an unstable water column. This suggests that dense, saline water near the surface was producing an unstable water column ( Figure 10). These density profiles are extremely unusual as any such instability will result in rapid vertical mixing and redistribution of the density anomaly, usually evading direct observation by CTD. Figure 10: Potential density anomalies (potential density minus 1000 kg m -3 ) with a reference pressure of 0 dbars for all 11 stations. The integrated excess density and assumed baseline density are depicted to highlight the instability. b) Station 26 does not present a density anomaly because it does not have a salinity anomaly. In the absence of a salinity anomaly, the temperature anomaly creates an area of less dense water, or a stable anomaly.
We hypothesize that an unstable water column that persists long enough to be profiled, must be the result of a continuously produced instability. The katabatic winds appeared to dynamically maintain these unstable profiles, through continual ice production leading to the observed heat and salt excesses at a rate that exceeds the mixing rate. If the unstable profiles reflect a process of continuous ice production, then the "inventory" of ice that we infer from our simple heat and salt budgets must reflect ice production during a relatively short period of time, defined by the time it would take to mix the anomalies away, once ice production stopped.
Similarly, Robinson et al (2017) found that brine rejection from platelet ice formation ( §3.5) also leads to dense water formation and a static instability. Frazil ice formation from continually supplied Ice Shelf Water (ISW) created a stationary instability, which was observable before being mixed by convection to the underlying homogeneous water column that extended to 200 meters. Similarly, the katabatic winds and cold air temperatures continually supply supercooled water to the polynya supporting the instability.

Relating the lifetime to turbulent eddy mixing
In the polynya the katabatic winds produce turbulent vertical eddies that continuously stir the water and disperse the excess temperature and salinity from frazil ice production into the homogeneous mixed layer found below the anomalies. The turbulence is composed of varying size and strength eddies. The largest eddies regulate the rate of dispersion (Cushman-Rosin, 2019). A characteristic timescale, t, can be approximated by relating the largest eddy size and the rate of turbulent kinetic energy dissipation (Cushman-Rosin, 2019).
Here, d is the characteristic length of the largest eddy and ε is the turbulent kinetic energy dissipation rate. In this section we discuss and select the best length scale in an environment dominated by buoyancy and wind shear. We then estimate and quantify environmental parameters critical to the length and characteristic time scale calculation. Next, we will return to and solve for the length scale and lastly we will solve for the characteristic timescale.

Estimating the length scale in a environment driven by buoyancy and wind shear
Before we can evaluate equation 5 and determine the lifetime, we needed to identify the length of the largest eddies which requires a heuristic argument in order to establish a single length scale. The largest eddies can be as large as "the domain"; in the water column, the domain might be as large as the mixed-layer (MLD), up to 600 m in some of the PIPERS profiles. However, a homogenous mixed-layer does not imply active mixing throughout the layer . On the other hand, the length of each salinity anomaly was easy to establish but does not necessarily reflect the maximum eddy size. For reference the MLDs and depth of the salinity anomalies are listed in Table 2.
Instead, the most characteristic length scale in an environment driven by both buoyancy and wind shear is the Monin-Obukhov length ( − ) ). When − is small, buoyant forces are dominant and when − is large, wind shear forces are dominant. While the − can be expressed using several different estimates of shear and buoyancy, we focus on the salt-driven buoyancy flux, because those anomalies come closest to capturing the process of frazil ice production (see §4.3 for more detail).
where * is the wind-driven friction velocity at the water surface, is gravitational acceleration, w is the water vertical velocity ΔS is the salt flux, is the coefficient of haline contraction, and is the von Karman constant. A more detailed explanation and the specific values are listed in Supplemental 4.

Estimation and Quantification of Input Environmental parameters
To solve for the length of the largest eddy using equation 6, we used the NB Palmer wind speed record, adjusted to the 10 m reference a log-wall profile . Roughness class 0 ( 0 ) was used in the calculation which is associated with water and has a roughness length of 0.0002 m.
The wind speed at 10 meters is 10 , is the NB Palmer wind speed, measured at a masthead height of = 24 . Together, these values determine the wind stress, as, where represents the density of air, with a value of 1.3406 kg m -3 calculated using averages from NB Palmer for air temperature (-18.73 ℃), air pressure (979.4 mbars), and relative humidity (78.3%). CD represents a dimensionless drag coefficient and was calculated as 1.525 x 10 −3 , using COARE 3 code, modified to incorporate wave height and speed . The average weather data from NB Palmer was paired with the wave height and wave period averaged from 04 May SWIFT to find . A more detailed explanation and the specific values are listed in Supplemental 5.
Once we found the wind stress, we could determine the aqueous friction velocity ( * ) at the air-sea interface using as follows: * = �   Table 2). The large value indicates that wind shear forces are dominant. In general, the − was longer than the salinity anomaly but smaller than the mixed layer depth. Station 35, the station with the largest salinity anomaly and highest mass of ice derived from salinity has − of 6 meters, much smaller than other stations. This indicates that at Station 35 buoyant forces are more dominant than other stations.

Resolving the length and time scale of turbulent mixing
Using the − , the estimates of TKE dissipation rate ( ) can be applied to find the characteristic time scale or lifetime using equation 5. The rates of mixing raged from 2 to 29 minutes, vary by one order of magnitude, and have a 14 minute average.

RATE OF FRAZIL ICE PRODUCTION IN TERRA NOVA AND ROSS POLYNYAS
To calculate the frazil ice production rate, we focus on the mass of ice estimates that are derived from salt inventories. This is justified by the systematically smaller estimates of ice mass that are derived from the heat inventory (see §4.3). We attribute the smaller values to heat loss to the atmosphere. The frazil ice production rate is calculated using the estimates of ice mass taken from salt inventories ( ) and using the mixing lifetime (t) that was determined from TKE dissipation in §5.
Here, = 920 kg m -3 , t=lifetime, in days, and = 1 2 . The results are summarized in Table 2. A more detailed explanation and the specific values are listed in Supplemental 6.

Variability in the frazil ice production rate
The ten estimates of frazil ice production rate ranged from 7 to 358 cm day -1 .
These sea ice production rates show some spatial trends within the Terra Nova Bay polynya that correspond with conditions as we understand them, in different parts of the polynya. As shown in Figure 11, a longitudinal gradient emerges along the axis of the TNBP when looking at a subsection of stations (Station 30, 32, and 25/33).
Beginning upstream near the Nansen Ice shelf (30) and downstream along the polynya axis, to the northeast, the ice production rate decreases. The upstream production rate is 56 cm day -1 followed by midstream values of 31 cm day -1 , and lastly downstream values of 9 cm day -1 . This pattern is similar to the pattern modeled by Gallee (1997).
The production rate at Station 35, was significantly higher than all other stations production rate, but this large excess is reflected in both the heat and salt anomalies.
The excess salt value is 260% greater than the closest station in both time and quantity, Station 34.
While none of the CTD casts were in the exact same location, there were 3 pairs of stations located in close geographic proximity (see Figure 11)  *Station 26 does not have a measurable salinity anomaly and a production rate could not be calculated via this method.

How do these production rates compare to prior modeled and field estimates?
Calculated production rates from PIPERS ranged from 7 to 359 cm day -1 and are plotted in Figure 11. Station 40, the one station in RSP, represents the minimum frazil ice production rate. While there is only one data point in RSP and variability in the TNBP, TNBP was expected to outpace RSP in production. The median, 13.65 cm day -1 , very closely matches Schick et al (2018) estimated average ice production rate for the month of May 16.8 cm day -1 calculated using heat fluxes.  estimated average ice production at 30 cm day -1 for the month of May by deriving an ice production rate from heat budget analysis.
The remaining published production rates are winter averages. Our mean production rate, 52.05 cm day -1 is comparable to , who modeled a wintertime maximum rate of 48.08 cm day -1 using a sea-ice model. It is similarly comparable to Gallee (1997) modeled results for a polynya. Gallee (1997) modeled in three dimensions over four days and mapped daily ice production rates in TNBP.
Modeled ice production rates near the coast (e.g. station 35) were 50 cm day -1 , decreasing to 0 cm day -1 downstream and at the outer boundaries. Station 35 is located closest to the coast (see Figure 11 in the region where highest modeled production rates took place Gallee (1997).  modeled a wintertime maximum production rates of 26.4 cm day -1 using a coupled atmospheric-sea ice model.  applied a classic model for latent heat polynyas and modeled production rates at 85 cm day -1 for 1993 and 72 cm day -1 for 1994. We might expect our production rates to be lower than the median of prior estimates, considering that the PIPERS expedition took place in late autumn when the polynya has typically not yet reached its maximum production rate. While some of our production rates far exceed modeled results, we attribute some of that variability to the relatively short time scale of these ice production "snapshots". As our estimates integrate over minutes to hours, instead of days to months they are more likely to capture the high frequency variability in this ephemeral process. As the katabatic winds oscillate, the polynyas enter periods of slower ice production, driving average rates down. Figure 11: TNBP map of ice production rates. a). Map of TNBP ice production rates, rainbow color bar indicates the ice production rate in cm day -1 and ranges from 7-57 cm day -1 . Station 35, marked as an outlier and not included in the color bar, is displayed with a patterned white marker. b). A cross-section of TNBP stations displayed to highlight a spatial pattern of decreasing ice production rates while moving away from the Nansen Ice shelf. The prevailing wind direction is noted in a dashed blue arrow.

CONCLUSIONS
The goal of PIPERS was to study polynyas, ice production, and seasonal evolution in the Ross Sea. During the late autumn cruise and katabatic wind events in the Terra Nova Bay polynya and the Ross Sea polynya, unexpected temperature and salinity anomalies provided an in-situ method of quantifying ice production rate.
Polynyas have been regarded as ice production factories with a wide range of model estimated production rates. Traditionally it has been hard to quantitatively estimate ice production due to the challenges of obtaining in-situ ice measurements (Tamura et al, 2017). In-situ salinity and temperature anomalies observed at 11 CTD stations were correlated to frazil ice formation and used to estimate polynya ice production. Sea ice production rates vary from 7 to 360 cm day -1 , a wide range. We suggest this is because we are capturing production on short timescales (minutes). The method demonstrated in this study provides an in-situ process for estimating sea ice production more accurate production rates can be obtained via our method by temporal spread of CTD casts in the same spatial location.
The Ross Sea polynyas have high production rates and are significant contributors to Antarctic Bottom Water formation. As shown in our production rates per area, TNBP has higher production rates than RSP. However, the significantly larger size of the RSP leads it to have the highest overall ice production rate of any Antarctic polynya (Tamura et al, 2017). Since 2015, the overall sea ice extent around Antarctica has decreased, with 2017 being an abnormally low year (Supplemental   = Assumed baseline/initial density, calculated using      The wind stress, , was calculated for each CTD station based on the extrapolated wind speed at 10 meters, 10 , average air density, and average drag coefficient: =density of air=1.34 kg m -3 calculated using averages from NB Palmer summarized above. Using wind stress, we derived the friction velocity ( * ) at the air-sea interface using the wind stress and water density, . * = � (S5.4) * = friction velocity = density of water    Figure 1: Absolute Salinity plotted from raw conductivity data and from 1-meter binned data for the CTD Stations with anomalies. The x-axis for a, c, d-f, h-k are all 0.03 g kg -1 ; b and g 0.06 g kg -1 . The raw data, plotted in purple, shows varying levels of noise in the signal and spikes of lesser magnitude values. This noise and the spikes in the data likely due to frazil ice crystal interference. Values of spikes extending off the plot: f: 34.670 g kg -1 ;g: 34.800g kg -1 ;i: 34.740g kg -1 . Plots b, c, i, j display more noise than the other plots. The 1-meter bin data, plotted in green, does not follow the spike excursions, indicating that binning the minimizes or removes the effects of the noise and spikes.
Supplemental Figure 2: Timeline of TNBP and RSP CTD casts and SWIFT deployments. A timeline of CTD and SWIFT deployments while in TNBP and RSP.
To the left of the date, the geographic region is noted. This indicates when NB Palmer entered that portion of each polynya. The NB Palmer was in TNBP from May 1 to May 13. The NB Palmer was in the RSP from May 16 to May 18. To the right of the date the CTD stations with anomalies and SWIFT deployments are shown. All of the SWIFT deployments where in TNBP.
Supplemental Figure 3: Comparison of Ice production rates. This box and whisker plot shows the production rates calculated in this study. Station 35, marked as an outlier is not shown, but was included in the mean and median calculations.
Supplemental Figure 4: Antarctic Sea ice extent. This plot shows the daily sea ice extent for Antarctica plotted over the entire year from 1978 to 2018. In 2015, the sea ice extent started to decline, with 2017 representing an unusually low sea ice extent.