Initial Validation of Surface Ocean Properties in MITgcm Arctic Regional Model

In ice-covered regions it can be challenging to determine air-sea exchange – for heat and momentum, but also for gases like carbon dioxide and methane. The harsh environment and relative data scarcity make it difficult to characterize even the physical properties. Here, we seek a mechanistic interpretation for the rate of air-sea gas exchange (k) derived from radon-deficits. These require an estimate of the water column history extending 30 days prior to sampling. We used coarse resolution (36km) regional configuration of the MITgcm with fine near surface vertical spacing (2m) to evaluate the capability of the model to reproduce conditions prior to sampling. The model is used to estimates sea-ice velocity, concentration and mixed-layer depth experienced by the water column .We then compared the model results to existing field data including satellite, moorings and Ice-tethered profilers. We found that seaice coverage have 88 to 98% accuracy, sea-ice velocities have 78% correlation which resulted in 2 km/day error in 30 day trajectory of sea-ice. Model showed the capacity to capture mixed layer evolution trends although with a bias and water velocities showed only 29% correlation with actual data. Using the capacity of the model to produce same order of magnitude of water speed we calculated an average radius of possible origins of water parcel equal to 10.3 km.

TKE is mostly dominated by wind speed on open ocean interface (Wanninkhof 1992;Ho et al. 2006;Wanninkhof and McGillis 1999;Nightingale et al. 2000;Sweeney et al. 2007;Takahashi et al. 2009). In the polar oceans wind energy and atmospheric forcing are transferred in a more complex manner as a result of sea ice cover (Loose et al. 2009(Loose et al. , 2014Legge et al. 2015). Sea ice drift due to Ekman flow (McPhee and Martinson 1992), freezing and melting of ice leads on the surface ocean (Morison et al. 1992) and short period waves (Wadhams et al. 1986;Kohout and Meylan 2008) all constitute important sources of momentum transfer. Considering the scarcity of data on marginally covered sea-ice zones (Johnson et al. 2007;Gerdes and KöBerle 2007), especially during arctic winter time, comparing and validating these models are quite challenging.
The radon deficit method involves sampling 222Rn and 226Ra in the mixed layer to examine any difference in the concentration or (radio) activity of the two species.
Radon is a gas, radium is a cation; in absence of gas exchange 222Rn and 226Ra enter secular equilibrium meaning the amount of 222Rn produced is equal to decay rate of 226Ra. Any missing 222Rn in the mixed layer is attributed to exchange with atmosphere (Peng et al. 1979).
Since the 222Rn concentration in air is very low, less than 5% (Smethie et al. 1985) and considering that concentration is proportional to activity/decay rate A, we can use Eq.
(1) to determine gas exchange. Where k gas transfer velocity in (m d-1), AE is the activity or decay rate of 222Rn which in secular equilibrium is equal to 226Ra activity, AM is 222Rn measured decay rate in mixed layer, λ is decay constant of 222Rn (0.181 d-1) and h is the mixed layer depth , Gas transfer velocities from equation (1) reflect the memory of 222Rn for a period of two to four weeks (Bender et al. 2011), which is four to eight times the half-life of 222Rn (3.8 days).The mixed layer depth, h, is calculated from the measurements performed at the hydrographic stations during 222Rn sampling process.
While a valuable tool this method is based on two premises, a) an invariant mixed layer and b) invariant TKE forcing during the entire period of "memory" that gas in the mixed-layer has experienced. To further illustrate the discrepancy caused by utilizing the invariant assumptions, described above, consider a mixed layer that rapidly changes by a factor of 2 just prior to sampling for radon. If the mixed-layer becomes shallower (stratification) hdefined by the density profile -will be smaller by factor of 0.5 while AE/AM in the mixed layer remains the same. Based on equation 1, this causes k to be half of its true value. That is, prior to stratification TKE forcing was sufficient to ventilate the ocean to a depth greater than the apparent h (Bender et al. 2011).
Conversely, if the mixed layer deepens due to mixing, h increases and a new parcel of water with AE/AM= 1 is added to mixed layer, causing the activity ratio to come closer to unity. These two influences on equation 1 (increasing H and AE/AM approaching unity) work against each other, but the net effect is to cause k to appear larger. The change of factor of 2 in mixed layer depth in less than two weeks has been observed during several studies (Acreman and Jeffery 2007;Ohno et al. 2008;Kara 2003).
The "memory" of gas exchange forcing that radon experiences is further complicated by the presence of sea ice. Consider two alternate water parcel drift paths that lead to the 222Rn sampling station in sea-ice zone (Fig. 1). Path B demonstrates a history which water column spends most its back trajectory under sea-ice. Path A shows a water column which experiences stratification and shallowing of mixed layer depth equal to δh when drifting through water that is completely uncovered by ice. During most of Path B gas transfer happens in form of diffusion through sea-ice and it will have a very low k (Crabeck et al. 2014;Loose et al. 2011). In contrast Path A will have a greater radon deficit, but a smaller h because of stratification. In either case, it is critical to take into account the time history of gas exchange forcing, including changes in the mixed-layer and ice cover, which has led to the apparent radon deficit at the time of measurement.
This observation about drift paths in the sea ice zone strongly implies that we must consider both time and space in estimating the forcing conditions that are recorded in the radon deficit. In other words, we require a Lagrangian back trajectory of water parcels to track the evolution of mixed layer and its relative velocity 4 weeks prior to sampling.
Although satellite data, Ice tethered drifters ) and moorings Proshutinsky et al. 2009) have provided valuable seasonal and spatial information about the sea ice zone, they do not track individual water parcels and tend to convolve space and time variations. The spatial limitation of these data pose a challenge to producing a a back trajectory of the water parcel hence we used a 3D numerical model to simulate these data. This is the principle goal of our studywe want to know if a coarse resolution model that can be run on a desktop multi-core processor can provide us with spatial and temporal information to better model the back trajectory of a radon-labelled water parcel. The null hypothesis might therefore be, that simple assumptions such as an invariant mixed-layer, Ekman ice drift and area-averaging of sea ice cover are as good or better at reproducing the radon-based estimates of k.
We set up a numerical simulation that captures near surface phenomena as a means to access the history of a water parcel sampled for radon, and the forcing conditions that led to the observed radon deficit. For this study we are focusing on near surface phenomena in the Arctic using the Arctic Regional configuration of the MITgcm with the domain of ECCO2 (Marshall et al. 1997;Menemenlis et al. 2008;Losch et al. 2010;Heimbach et al. 2010;Nguyen et al. 2011). A number of Arctic ocean-ice models which have been compared as part of Arctic Ocean Models Intercomparison Project (AOMIP) (Proshutinsky et al. 2001;Lindsay and Rothrock 1995;Proshutinsky et al. 2008) in regard to their capability to represent main ice-ocean dynamics. This model has shown to have better correlation with remotely sensed sea ice data (Johnson et al. 2012) and high near surface vertical resolution.
It is worth noting that model studies using higher resolution evaluate model skill are already in print. For example, Holloway et al., (2011) use a 4 (km) version of the MITgcm regional model has a high skill in producing near surface velocities and capturing eddies. So, the choice to use the 36 (km) resolution model may seem anachronistic. However, we approach this problem as geochemists who are searching for a data interpretation tool. We seek higher resolution in the top 100 m of the water column, as compared to these prior studies. Furthermore, the 36 (km) model can be run on a desktop multi-core processor, and the simulation is completed in less than a week.
The remainder of the article is organized as follows: In section 2 we introduce the changes we made to increase the near surface resolution. Section 3.1 and 3.2 contain the comparison of the new results and original model outputs including sea-ice concentration, velocity and trajectory to observed data from satellite and Ice tethered profiler. Sections 3.3, 3.4 and 3.5 are used to study the model output salinity and temperature structure resulting upper ocean density structure and mixed layer. CHAPTER 2

METHOD
The model has a horizontal resolution of 36km, and 50 vertical layers employing the z* coordinate system (Adcroft and Campin 2004)  Satellite estimation of sea-ice cover at 25km horizontal resolution (Comiso 2000) is being interpolated on a grid system and then compared with A0 and A1. Then sea-ice drift gathered by 28 Ice Tethered Profilers (ITP)  which has more than 2 month of data in Beaufort Sea between 2006 and 2013 has been used to do the ice velocity and trajectory comparison. We selected only ITPs with 2 or more months of drift data in the Beaufort Sea.
We compared near surface water velocity data from ITP-V (

Sea ice concentration
Both A0 and A1 model scenarios are used to study sea ice extent (area of cells with more than 15% ice cover) from 1992-2013. At the Arctic Basin scale, the model depicts a decrease in the September minimum sea ice extent of 0.82 million square kilometres per decade or 10% of the starting value at 1992 compared to satellite data which display a 1.2 million square kilometres per decade, the model show an over estimation of sea-ice extent (Fig. 2). For further analysis we introduced a grid system covering the Beaufort Gyre and interpolated the data from satellite (Comiso 2008) and A0 and A1 on to the grid.
The analysis grid extends from 70° to 80° north and 130° to 170° west, covering most of Beaufort Gyre (Fig. 3). Grid points can be divided into two main geographic zones that are marked out based on sea ice cover. The first zone contains grid points where the annual average sea ice cover is greater than 80%. These sets of points are fully covered by sea-ice most of the year. The second zone can be described as "marginally ice covered" wherein the ocean surface is free of ice for some fraction of the year. We where the water is predominantly covered by 100% or 0% ice (P1 and P3), the model captures the seasonal advance and retreat and the percent ice cover itself is accurate.
However, in the transition regions that are characterized by marginal ice for much of the year (P2), the model has more difficulty reproducing accurate sea ice cover as well as the timing of the advance and retreat. These results are the same for both A0 and A1 experiments, and this behavior is consistent with the description that has been explained by Johnson et al. (2007), that models have a higher accuracy predicting sea ice concentration in central arctic and less accuracy near periphery and lower latitudes.
The spatial sensitivity of the model can be observed using root mean square (RMS) error Eq.
(2), calculated over the 1992-2013 period (Fig. 5). The area with most error coincides with area between the 80% and 60% contour lines (Fig. 3) and is concentrated primarily in the Western Beaufort. The RMSE error of 0.2 is the maximum value away from land, this same level of error can also be found near land which is caused by fast-ice generation. Fast Ice in the model is replaced with pack of drifting sea ice; this error is common between numerical models and has been studied during AOMIP (Johnson et al. 2012, p. 20). (2) If we compare the monthly climatology for sea ice cover over the 1992-2013 period, the RMS error between model and satellite data is least during the early winter months (e.g. Jan-Mar) when sea ice is close to its maximum extent. Comparing A0 and A1 Fig. 5 depicts an increase in RMSE during July, August, September and October and a minor decrease in May and November. The RMSE appears to be greater during the summer months of ice retreat, and slightly less during the autumn months of ice advance. Overall, the periods of transition (melt and freeze) coincide with the greatest RMSE.
The 2m revised grid (A1), with smaller vertical intervals near the surface has produced a greater RMSE than the optimized model (A0). We are still exploring why this took place. It is possible that the convective parameterization in the ice model is somehow negatively affected by short i.e. 2 m vertical layers. To reproduce changes in the mixed layer will require the greater resolution reflected in the A1 model run. To generate a statistically large sample size, this calculation is preformed such that every day is treated as a starting point for simulation and the trajectory of that parcel is traced for a 30 day period. Figure 7 shows the result of this calculation for ITP 53.

Vertical Salinity Temperature profiles
We chose 4 profiles in Beaufort sea to represent the simulated vertical salinity and temperature. The first two sets of profiles are from ITP-1 winter (Fig. 11) and summer 2006 (Fig. 11). The third set is from ITP-43 during winter 2010 (Fig. 12) and the fourth is from ITP-13 during summer 2008 (Fig. 13).
During winter time, a well mixed layer reaches below 15 meters in the model T and S profiles (Fig. 11), followed by a very large gradient. The mixed-layer temperature is close to the local freezing point in a condition called "ice bath" (Shaw et al. 2009).
The ITP profiles are similar; however the ITP mixed layer depth is deeper by nearly 10 meters, indicating more ice formation and convective heat loss over this water column, as compared to the model water column. In summer (Fig. 11)

Density profiles
We compare the MITgcm density to the time series of density profiles from ITP-62 ( Fig. 15-16 have both temporal and spatial changes in them (Fig. 14).
We are able to discern some broad similarities in the model and ITP density profiles.
From September to mid-November, the density profile above 50 (m) tended to increase, consistent with the period of cooling and ice formation. From December through March, both ITP and model density profiles remain relatively constant.
Between March and April, ITP-62 appears to drift through a unique water parcel, with lower density above 70 (m). The same feature can be observed in the MITgcm density.
However, on a smaller scale, there is significantly more variation in the ITP data than what the model represents.
For exploring the reason behind the density signals we are going to use the simulated fraction of sea ice cover and ice thickness (Fig. 17). The dominating effect appears to result from sea ice fraction and when there is continuously covered area. The changes from sea ice thickness can be observed in volume of fresh water in the water column.
A peak in near surface density can be seen late in March when a decrease in ice fraction from 100% to 90% exposed the surface water to cold atmosphere, which generated newly formed sea ice and inserted brine into the water column. This signal will be further discussed on mixed-layer section.

Mixed layer depth
There are many different methods in the literature for calculating mixed layer depth (Brainerd and Gregg 1995;Wijesekera and Gregg 1996;Thomson and Fine 2003;de Boyer Montégut et al. 2004;Lorbacher et al. 2006;Shaw et al. 2009). The methods can be divided into two main types (Dong et al. 2008 (Brainerd and Gregg 1995;Lorbacher et al. 2006). A more sophisticated approach to type 1 of this criteria is to utilize a differential between (ρ 100m -ρ surface ) as the cut of point (instead of using a fixed δρ) to account for the effects of surface ρ changes during winter and summer (Shaw et al. 2009). Here, we have implemented two of these methods M1 and M2, with M1 using δρ equal to 0.2 of (ρ 100m -ρ surface ) (Shaw et al. 2009) and M2 with a gradient (∂ρ/∂z) cut off point equal to 0.02 (kg m -4 ) which matches innate model parametrization of MLD (Nguyen et al. 2009).
We compare these 2 methods by applying them to the profiles from Fig. (10 -11-12-13) which result in (Fig. 18). In case (a) and (b) M1 produces a mixed layer depth that is 8 to 12 meters deeper, compared to the other method. A visual examination of profiles appears to indicate that the M1 criteria may be too flexible of a criteria. The results from M2 appear to be intermittently "realistic", whereas M1 can be difficult to implement on high resolution data with greater small-scale variability. In practice, we find M2 is the most straight-forward to implement.
It should be mentioned that it is difficult to consistently compare performance of the M1(δρ) and M2 methods on ITP and model data, because the model data extends to the free surface, but the ITP data stops at 7 (m) depth and it has been shown that summer mixed layer in the Canada basin can be less than 12 meters ). To account for this effect, we apply an additional restriction wherein any profile whose mixed-layer depth is less than 2 (m) below the shallowest ITP measurement is discarded. This restriction effectively removes any ML depths shallower than 10 meters due to ITP sampler not resolving the upper 8 meters of water column. In some cases, a remnant mixed-layer from the previous winter may exist in the water column. In this case, the methods incorrectly identify the remnant ML as actual ML depth.
To compare the methods over a longer time period, we calculated the mixed-layer depth from model data and ITP-62 data along the ITP-62 drift track. We used both M1 and M2 to determine the ML depth for the model data and for ITP-62 data . M1 and M2 show almost similar results. The model shows a shallower ML compared to the ITP data; the most prominent feature in early March corresponds to a sudden change in density found in (Fig. 15). During the months of June and July, the model predicts zero mixed layers with stratification almost to the surface. The ITP data, beginning at 8 m cannot reflect this stratification, but we know this model result to be plausible based upon our comparisons with shipboard CTD profiles (data not shown here) and what is known about ice melting and stratification.
For further exploring the forcing that drives the mixed layer, we use the Japanese reanalysis (JRA-25) (Onogi et al. 2007) wind speed, air temperature data and sea-ice fraction from the model and interpolated them on the path of ITP-62. The prominent feature during March to April can be explained by the reduction in ice cover to 95% , which resulted in water column exposure to low or moderate wind from 5 (ms -1 ) to 10 (ms -1 ) and low air temperatures, which in turn increased mixing and deepening of ML by 100% (Fig. 17). The wind appears to have led to a divergence in ice cover, which in-turn exposed the ocean to the cold atmosphere, leading to a loss of buoyancy and an increase in the mixed-layer depth. In this regard, the wind appears to play a facilitating role, leading to sea ice divergence, but there is no evidence to support that a strong wind event caused the mixed layer to deepen.
Further exploring the dependency of ML evolution to wind and temperature we calculated cross correlation between ML-wind and ML-temperature (Fig. 22). This analysis shows responses with respect to air temperature happen with no significant lag and wind with 10 days of delay which would result in a maximum 0.36 correlation for wind and 0.39 correlation for temperature. Without accounting for the effect of sea ice cover, which can somewhat modulate momentum inputs by the wind and strongly modulate heat and moisture fluxes, it appears that neither temperature nor wind has a dominant effect on MLD in the Arctic, but rather both play a role.

Circulation
The time-averaged upper ocean circulation in the Arctic has been described by oceanographers based upon the origin of the water masses that enter and exit the Arctic Ocean. These are primarily the Atlantic water (AW) and Pacific Water (PW).
After entering the Arctic ocean AW and PW loose heat and sink, AW to between 200 (m) and 800 (m) (Golubeva and Platov 2007) and PW to between 50 and 150m (Steele et al. 2004). During the AOMIP intercomparison study it was observed that Arctic models can produce two AW circulation in opposing directions, with observations suggesting a cyclonic circulation in Arctic basins (Lindsay and Rothrock 1995).
Potential vorticity influx from sub-Arctic oceans (Yang 2005) and unresolved eddy parametrization (Holloway et al. 2007) have been shown to diminish this discrepancy and create consistent cyclonic circulation among all models ).
We evaluated the capability of the model to reproduce the general circulation of Arctic by averaging output ocean velocities from A1 experiment during a time span of seven years from 2006 to 2012. We focused on two depth intervals 1-15 (m) depth to represent Ekman layer   (Fig. 24) and the second set is 180-250 m to model AW circulation (Fig. 23).
The flow field (Fig. 23) representing AW displays a counter-clockwise topographic boundary current in the Beaufort and Chukchi seas matching AW cyclonic circulation (Lindsay and Rothrock 1995;Proshutinsky et al. 2011). In contrast there is a clockwise boundary current in the southern Canadian basin (Fig. 24). These circulation patterns qualitatively match the hydrographic description of the time-mean circulation in the Arctic.

Velocities in the water column beneath drifting sea ice
We have very little information from direct observations that permit us to track a water parcel especially beneath sea ice. This is one area where model output could be critical as there are not obvious alternatives. To judge the consistency of the model water current field, we compared 2D model water velocity to data gathered from two sources: (1) from ADCPs mounted on moorings that were deployed starting in 2008 in Beaufort Gyre ) and (2) the ITP-V sensor equipped with MAVS (Modular Acoustic Velocity Sensors) (Williams et al. 2010), which was the only operating ITP before 2013 which had an acoustic sensor mounted on it.
We compared the magnitude of velocity without accounting for flow direction, i.e.
removing the effects of Ekman turning  in order to find a 2Dcorrelation over the duration of ITP-V working days which is from Oct 9, 2009 to Mar 31, 2010 (Fig. 25). The ITP data has been daily averaged so it matches the time interval of the model structure. Due to ITPs mechanical limitation on reporting velocities shallower than ~6 meters, the comparison starts from 7 meters.
The first result that emerges is that ITP velocity shows a lot more structure, changing speed at a much higher time frequency most probably caused by eddies which has been shown to be a characteristic feature in the top layers of Beaufort Sea (Zhao et al. 2014). Figure 26-27 depicts the simulated and observed velocity components for ITP-V, to better compare the results the data and simulation have been filtered by 4 days low pass filter corresponding to 36 (km) in distance which the ITP travels effectively low pass filtering the results.
The absolute 2D-correlation between the simulated and observed velocities for eastward direction is 20%, northward direction 22% and for the speed is 20%, these numbers represent r and not r2. We also compared the correlation versus depth and correlation of the velocity components and current speed (Fig. 28). If averaged over depth the northward velocities will have about 28% correlation with depth averaged data and also the correlation increases with depth after 40 meters. One possible reason for this lack of correlation near surface is effects of stratification in the model due to its shallower mixed layer.
Another source of ocean current data from the Arctic are ADCPs mounted on the moorings which has been deployed as part of Beaufort Gyre Observation System (BGOS), these profilers are located on the top float of subsurface moorings, measuring water velocities within the upper 30 meters of water column .
Mooring D was the first mooring outfitted with an ADCP in 2005. Figure  , , , With z the distance from the water interface and τ (z) representing the shear vector, f Coriolis parameter and C ice is the fraction of sea-ice cover. The interface friction velocity vector (u *0 ) is calculated based on the type of interface. In case of water-ice interface u *0 is based on law of the wall (McPhee 2008). Total Ekman velocity is C ice weighted summation of Ekman vectors based on ice and air (Eq. 6), we then compared the velocity components based on this method, simulation and data ( Fig. 30-31).

Sources of error in water column current velocities
There is 29% correlation between the simulation and mooring data. This is nearly identical to the poor correlation observed between the model and the ITP velocities.
These discrepancies between model and data can be caused by several reasons. The first is the model resolution is too coarse to resolve mesoscale and submesoscale eddies, which have characteristic lengths of 10 km or smaller (Boccaletti et al. 2007;Timmermans et al. 2008). Because these eddies would all hypothetically fit within one of our 36-km grid boxes, this model configuration lacks the ability to represent them.
Eddy resolving models have better agreement with velocity data (Holloway et al. 2011). Secondly the parameterization of sea-ice in the current model utilizes "levitating" sea-ice , meaning the sea-ice always stays on top of the free surface. This assumption neglects the forcing caused by sea-ice advection whenever surface velocity is different than the sea-ice velocity .
Neglecting the actual draft of sea-ice also introduces an error in calculating the correct distance from boundary which in turn makes the direction of velocity in Ekman spiral unreliable. Omission of other physics such as tidal waves, which is a common trait between arctic models, may as well have an effect on the velocity fields.
Having established that the correlation between the model output and mooring or ITP data is generally poor (i.e. 29% as reported above), we can further ask whether, in the absence of data, we would be just as well served by assuming that the current speeds are purely Ekman in nature. Using Eq. 4-6, as described above, we generated the Ekman spiral under partial ice cover to measure the current speed using only reanalysis winds. As with the model data, we compute the vertically-averaged current speed form 0 to 20 m depth. Using this assumption of Ekman drift and reanalysis winds, it is possible to reproduce the mooring velocity magnitude with 18% correlation. Whereas the 29% correlation between model and mooring data is unacceptable in most cases, it is still possible to argue that this is an improvement over the Ekman drift assumption. This is easily explained by the fact that the model captures more of the processes driving currents, including inertial oscillations.
In spite of the poor correlation between model and data, it captures many of the processes that drive net velocity, including Ekman, geostrophic, internal wave, inertial and subinertial processes. With regard to configuration, which was necessary to capture vertical profiles in near surface ocean, our experiments showed that changing the vertical resolution of model grid cells reduced the accuracy of the sea-ice trajectory and concentration. This is likely due to the fact that the model output was tuned with a different vertical grid spacing ) and the tuning coefficients are no longer optimal, once we change that spacing. It has been shown by (Holloway et al. 2011) that optimized parametrization of the model is sensitive to horizontal resolution; vertical resolution may as well have the same effect on the model.
We observed good correlation (78%) between model ice velocity and ITP drift. The rapid accumulation of small errors in the drift speed and direction leads to significant differences in the predicted and actual drift trajectory. We found errors in drift trajectory of 2km/day. This relatively strong correlation is perhaps not surprising, considering that more than half of the ice drift variability can be explained by Ekman drift alone, and the model receives wind as an input parameter. It is also worth noting that by doubling the model resolution, error decreases by more than half, i.e. 0.8 (km/day) accuracy by using a 18-km model ).
The estimation of mixed layer depth is challenging: No algorithm performs well in all situations; CTD profiles from drifting buoys often do not include the top 7-10 m of the surface ocean where stratification can be important, and because the density structure of the ocean is affected by vertical fluxes and by geostrophy. In these model-data comparisons we found model MLD to be biased consistently shallower, but this in part depends on a surface value, which is not recorded in moored drifters such as ITPs. The evolution of the mixed layer showed that MLD correlates equally well (or poorly if constrained by sea-ice) with wind and temperature trends (36% and 39% respectively).
Despite the potential for bias in the model MLD, the time series of ITP and model MLD reveal that the model captures variability at a similar frequency. It appears 10-50% changes in MLD happen regularly over the course of days and major changes (i.e. 100-200 %) appear to be event-driven as opposed to gradual seasonal evolution.
The potential for such event-driven changes in the MLD in the time prior to sampling are clearly important to consider when evaluating surface ocean geochemical tracer fields, although it may be sufficient to look for large excursions in the wind, air temperature and ice cover from reanalysis data.
The A1 experiment showed a good visual representation of Arctic general circulation but on finer scales, individual flows did not match the data from ITPs or Mooring.
This discrepancy may be the result of two causes -the lack of eddy resolving resolution and the lack of realistic ice physics. Holloway et al. (2011) have shown that by increasing the resolution to less than 9 (km) a good agreement can be reached for 10 m vertical resolution. Our current lack of representation of velocity fields will lead to outcomes that are broadly similar, but locally incorrect. We further saw this effect on temperature profiles modeled on ITP path.
We compare each of the variables affecting gas-exchange and summarize our comparison on Table-1. Overall, we find that a coarse-resolution model can yield as much as 50% improvement over simple assumptions that are loosely constrained by data (e.g. assuming Ekman drift or using a radius to average ice properties over a given time period). However, the coarse resolution model is not up to the task of representing Lagrangian drift of a water parcel. Water parcel drift beneath sea ice remains an elusive term, difficult to estimate, and hard to do without, when interpreting geochemical tracer fields.