The full process for obtaining the (partial) sea-icefield analysis involves numerous, organized steps. At first, images need to undergo a pre-processing phase, where camera artefacts regarding vignetting effect (directly affecting intensity based methods) and lens distortions (undermining orthorectification accuracy) are removed. In the case of the vignetting effect, a method is derived for estimating such effect and removing it through a de-vignetting mask; while the lens distortion is estimated and removed by means of a powerful tool from the Matlab environment. In addition, during the pre-processing phase, images are geometrically orthorectified from their perspective view by means of two distinct algorithms: a derived geometry-based algorithm and a homography-based transformation using transformation frames and rotation matrices. In order to obtain the attitude of the camera with respect to the world, an IMU device is used. The orthorectification process provides a “birds eye view” of the icefield, where sea-ice floes display their true shape.
Even though sea-ice is an important component of the global climate system, it has been observed only rather recently. Sea-ice observations are mainly limited to sea-ice concen- tration (the relative amount of area covered by ice, compared to some reference area), to ice drift (displacement of a sea- icefield) and, recently, to total sea-ice freeboard (height of the top of snow or sea-ice above sea level, hereafter called ice freeboard) and thickness. Ice concentration and drift have been observed at large scale since the late 1970s, in both the Arctic and Antarctic regions, using passive microwave sensors on board satellites (e.g. Gloersen et al., 1992). Com- paratively, ice thickness observations are much sparser, ham- pering a proper estimate of the ice volume. Originally, they stemmed only from upward-looking sonar by submarines in the Arctic (Rothrock et al., 2008) and ship-based visual ob- servations in the Southern Ocean (Worby et al., 2008).
For both years, SIA shows overall improvement during the whole year, but this is not the case for the SIE. In 2005 the SIE good match between initial iteration and satellite data during summer months disappears after assimilation, with considerable underestimation of SIE. In 2007 there is an overall SIE improvement after the assimilation, but there are again months with a considerable SIE underestimation. Both metrics suffer from the inability to guarantee that improve- ments in this metric also lead to an overall improved match in the spatial seaice coverage, since a perfect total SIA or SIE evolution may still correspond to considerable differences to the data in their regional distribution. Chances of having SIE distribution close to observations with quite different spa- tial shape of the seaicefield are very high. This calls for changing the common practice of model evaluation by only comparing their ability to simulate present day SIE without considering the seaice spatial distribution (e.g., Dukhovskoy et al., 2015).
Already when the ITD parameterization originally was developed, two main problems were identified that are still the biggest sources of uncertainty today: (1) the redistribution of ice between different ice thickness categories by ridging processes (Thorndike et al., 1975) and (2) the assumption that the deformation energy is either lost to friction or converted to potential energy as ice floes ridge and raft (Rothrock, 1975). Both Thorndike et al. (1975) and Rothrock (1975) make assumptions about the mechanical processes that govern seaice ridge formation, but Pritchard (1981) already showed that they were missing important parts of the energy balance. At the time there were only a few observations of thickness and ridge profiles available (see e.g. Parmerter & Coon, 1972, and references therein), and dynamical mod- eling studies provided the most reliable understanding of ridging processes (Parmerter & Coon, 1973). The amount of available data has increased since. After discrete element models of the ridging process (Hopkins, 1998), labo- ratory experiments of ridging (Tuhkuri, 2002), and in-situ measurements of stresses in ice floes (Tucker & Perovich, 1992; Richter-Menge & Elder, 1998), the analysis of ridging properties is still an important field of ongo- ing research. Methods range from evaluating airborne observations (Herzfeld et al., 2015) and basin-wide process-oriented model simulations (Hopkins & Thorndike, 2006) to the analysis of conceptual models (Godlovitch et al., 2011). A common notion is that the details of the physical processes during ridging and their large-scale statistical properties, that is, the key features in shaping an ITD and determining the amount of energy necessary for de- formation, are still not sufficiently well understood.
We use several observational data sets to evaluate the sea-ice model. An extensive historical data set, named BASIS, cov- ers the winters 1960/1961 to 1978/1979. This data set con- tains the, at that time, best available information on the ice concentration, thickness, as well as dominant ice types. BA- SIS is based on hand-drawn sea-ice charts which were pro- vided by the local weather services for shipping. The sea-ice charts were derived from direct ice measurements and es- timates from voluntarily observing ships, coast guards, ice breakers, lighthouses, and harbor authorities. Additional in- formation came from over-flights by the FMI, the SMHI, and the Swedish Air Force. From the late 1960s onwards satel- lite observations were partly included. Thus, the underlying ice charts were extrapolated from the irregular (as regards space and time) observations. The associated uncertainties are unclear and are presumably largest away from the major shipping lines. Nevertheless, BASIS is the best available in- formation on historic ice conditions in the Baltic Sea. These ice charts were collected and then digitized in 1981 in a joint project of the Finnish Institute of Marine Research (today the FMI) and the SMHI. The original data were hard to access as BASIS ice was designed for storage on punchcards. Thus, Löptien and Dietze (2014) provided an easier-to-access ver- sion in the NetCDF free file format via www.baltic-ocean.org (or PANGEA https://doi.org/10.1594/PANGAEA.832353). BASIS ice thicknesses were originally indexed by numbers from 1 to 9. These numbers were assigned to thickness classes (1–2, 3–6, 7–12, 13–20, 21–30, 31–42, 43–56, 57– 72, and more than 73 cm). Thus a lower bound for the uncer- tainty when it comes to ice thickness is the precision given by these classes. Note that when we calculate integrated metrics, e.g., total ice extent and total ice volume, the BASIS data set
Nares Strait, the converging part of Kane Basin and upstream of coasts and islands, the ice builds up along narrow and ori- ented features (see Fig. 11a). The effect of numerical dif- fusion in smoothing these features and reducing the local- ization of the thickest ice increases with ice drift velocities downstream of the domain. Nevertheless, at all times the sim- ulated probability density function is strongly asymmetric, consistent with thickness distributions estimated for seaice with little history of melting in the open Arctic ocean (e.g., Haas, 2009) and at the entrance of Nares Strait (Haas et al., 2006). As in the idealized case, the strong localization of the ridged ice translates into an exponential tail for P (h) of the form of Eq. (13), with h ∗ increasing with time (see Fig. 11b). According to Eq. (5) and to the simple redistribution scheme employed here, the evolution of h is a function of the mechanical redistribution term given by Eq. (7), which itself is a function of the flux of ice concentration ∇ · (Au) = u · ∇A + A∇ · u. The spatial distribution of h in the present Maxwell-EB model therefore depends essentially on the sim- ulated velocity field. Furthermore, both simulations, ideal- ized and realistic, show a similar exponential decrease for the tail of P (h). This suggests that the localization of the thickest ice does not arise from the complexity of the domain geom- etry. The shape of the tail of P (h) is also conserved when using a lower spatial resolution, for instance, with 1x = 4 and 8 km in idealized simulations (orange, dotted and dotted- dashed curves in Fig. 10b), indicating that this property of P (h) is not resolution-dependent either. This is consistent with the fact that there is no characteristic scale for the lo- calization of damage and deformation in the model beyond the scale of the model element (see Dansereau et al., 2016, Sect. 6.1 and 6.2); i.e., at all spatial resolutions, the simu- lated deformation is highly localized. Additional simulations with different ranges of cohesion (not shown) indicate that the value of C does not impact the shape of the ice thickness distribution but rather only the rate at which the exponential tail flattens, i.e., the rate at which the ice cover thickens. In brief, the strong localization of ridged ice in the model there- fore appears to be only the consequence of its capability to reproduce the extreme localization of ice deformation and associated sharp gradients in the ice velocity field.
In practice the tie points are defined from analysis of satellite data and given as brightness temperatures. A retrieval of, for example, 110% is understood as a mixture of 110% seaice of tie point radiative properties and 10% open water. Wherever possible, we have applied the most recent tie points provided by the authors of a given algorithm; Table 1 provides the pertinent references. Many of these tie points have been tuned to daily average brightness temperature data and our use of swath data may therefore introduce a slight inconsistency that may minimally affect the error standard deviations, but not the correlations, in subsequent comparisons. This possible in- consistency must be weighed against the possible large error from an increased time offset between SSM/I and reference observation. In addition, the diurnal variation in tie point emissivity during winter is minimal and the comparison statistics will only be slightly affected by a minor change in tie points. It is common to limit the range of the algorithms to the physically meaningful range in a postprocessing step, which makes good sense for most users of seaice concen- trations. However, ‘‘saturated’’ ice concentration estimates will have reduced sensitivity to real openings in the ice and this practice additionally complicates comparisons of algo- rithm statistics as part of the true bias and variance are hidden in the cut off portion of the retrievals. In the present study we avoid all such postprocessing with the exception of NT2 and N90. For N90, the problem is that the smooth interpolation between the ice and water points is only strictly valid inside the 0 to 100% concentration interval. In fact, outside the interval and depending on the tie points, the concentration is not generally monotonously increasing with decreasing polarization [Spreen, 2004]. Our choice of tie points results in a well-behaved characteristic in the domain of interest in this study. However, to avoid errors, pixels where the polarization falls below the ice tie point and the retrieved concentration is below 100% are dis- carded. This error is only found in very few cases. For the NT2 algorithm tie points are integrated in tables of simulated brightness temperatures that are used in a mini- mization scheme to find the combination of seaice and atmospheric contributions matching the satellite observa- tions. Since only solutions for ice concentrations up to 100% are allowed, the variability of the NT2 algorithm is not directly comparable to the remaining algorithms. We have therefore extended these tables to enable solutions to be found in the range between 0 and 120%; we name this version of the algorithm ‘‘unconstrained’’ and refer to it as NT2U henceforth. Owing to the dynamic scheme used to
In order to identify the origin of the (Arctic) drift and to counter- act it, we have run an additional experiment (hereafter referred to as HI_PRESERVE, see also Table 1 ), which is compara- ble to MULTI but for which the category ice volume vicen is no longer added to the state vector. In each category, vicen is diagnosed from the update of the category ice concentration (aicen), so that the forecast ice thicknesses in all categories are preserved. HI_PRESERVE mitigates the drift in the bias and is now indiscernible from FREE during the last two years (Fig. 2 c,e). The reduction of the bias leads to a reduction of the rmse of the same amplitude. This suggests that the drift observed in MULTI is mainly the consequence of the ice volume postprocessing and not of the ice concentration postprocessing (which are the same in MULTI and in HI_PRESERVE). More surprisingly, it is found that HI_PRESERVE achieves similar performance as MULTI for all variables (see Fig. 4 ) and even for the overall ice thickness hi, which is a mean of the category thicknesses (which remain unchanged during the assimilation steps) weighted by concentration area of the different category. Thus the benefits found in hi and in the ice volumes in MULTI are a direct consequence of the improved category ice concentrations aicen and not from the updates of the vicen. This merely implies that adding the category volume does not add skill but causes a drift in the performance of the system. It would be interesting to study whether adding vicen to the assimilation state vector could add skill if a higher frequency of assimilation or a larger ensem- ble size would be used. It would also be interesting to assess whether we could further improve the impact of assimilation by
effects on other experiments (e.g., Else et al., 2015). Ice sam- ples were collected using ceramic knives or a Kovacs Mark II coring system depending on the ice thickness. Sampling was performed from a movable bridge to avoid walking on the ice surface and to ensure only undisturbed sites were sampled. Ice cores were collected from one end of the pool (half me- ter away from the edge of the pool) and at least 20 cm away from previous cored sites. Ice cores were packed in clean plastic bags and kept frozen during the 20 min transport to a cold laboratory and processed within a few hours. Seawater was sampled for total alkalinity (TA) and total dissolved in- organic carbon (TCO 2 ) with a peristaltic pump (Cole Parmer
Our study investigates the short-term response of seaice and upper ocean to prescribed changes of the surface fresh- water input, both in magnitude and spatial distribution. We do consider these modifications as a perturbation of the current-climate environment. We specifically include the dy- namic response of the sea-ice–ocean system in our analysis. Generally, previous studies use very crude renderings of the spatial distribution of the freshwater addition. Here, we focus on the differences between a widely used uniform runoff dis- tribution around Antarctica and more complex spatially vary- ing distributions. In our study, we employ an eddy-permitting ocean–sea-ice model. Four experiments with differing spatial distribution and magnitude of the Antarctic surface freshwa- ter flux are analyzed with respect to the response of seaice concentration, thickness, and velocity in space and extent, volume, and production over time. In addition, the develop- ment of the on-shelf water column and the dense shelf water at the main sites of dense shelf water formation are presented. While these changes do not directly correspond to the trends observed around Antarctica in recent decades, they tell us how the seaice reacts dynamically and thermodynamically to spatially limited changes in the freshwater input. Thus, they provide a measure of what to expect as a seaice response to observed changes in the runoff and also offer explanations for observed changes in seaice and water properties. Three ad- ditional experiments with enhanced freshwater amounts ex- ceeding the range of current estimates (70–290 Gt yr −1 ; Rig- not et al., 2008; Joughin and Alley, 2011; Shepherd et al., 2012; Vaughan et al., 2013; Wouters et al., 2013; Rignot et al., 2013; Velicogna et al., 2014) allow us to gain insight into possible future seaice changes.
However, two other sources of error can be identified. Both are linked to the definition of the boundary of the cell (usu- ally quadrangle) over which deformation is computed. Lind- say and Stern (2003) showed that unrealistic deformation is often obtained when this boundary is too irregular. Also, spu- rious openings and closings (that we will refer to as artificial noise hereafter) are caused by unfavorable orientation of the cell boundary relative to the discontinuities in the seaice mo- tion field, also called dynamic discontinuities, slip lines or linear kinematic features (LKFs). Lindsay and Stern (2003) evaluated the standard deviation of the error in area change due to the boundary definition to be about 3.2 km 2 for a 10 by 10 km square cell, which is more than twice the error from tracking mentioned above. Kwok (2006) stated that this arti- ficial noise would lead to an overestimation of the ice volume production, although no precise number was given. Lindsay et al. (2003) proposed to reduce this error by combining cells together, but this solution reduces the benefits of having high- resolution data and reduces the spatial range over which one could perform scaling analysis. The error in estimating defor- mation has also been studied by Thorndike (1978) for large spatial scales (from 100 to 500 km), but such an analysis is only valid for homogeneous and isotropic fields and does not apply to small-scale deformation as explained in Lind- say et al. (2003).
Historically, NEMO-LIM3 seaice dynamic parameters were tuned in a heuristic, ‘‘trial-and-error’’ process, with the goal to best match the ice areal export through Fram Strait (M. Vancoppenolle, personal communication, 2013). While seaice extent and thickness are realistically simulated in the Arctic in the default conﬁguration [Vancoppenolle et al., 2009], the Arctic seaice drift simulated at the daily time scale exhibits clear biases. Figure 1a is a snapshot of observed seaice motion observed during a 2 day period in April 2012. Because NEMO- LIM3 is forced by atmospheric reanalyzes, it is expected to reproduce this motion. The direction of the ﬂow is simulated overall correctly, but its intensity clearly is not (Figure 1b). In addition, seaice is nearly motionless in a large portion of Beaufort Sea and the gradients of seaice velocity are clearly too smooth compared to obser- vations. At the model’s resolution, oceanic eddies are not explicitly represented and their effect is instead par- ameterized in terms of global state variables [Madec, 2008]. The fact that the modeled drift is globally too slow may be in part attributed to this issue, but this is probably not the only reason.
There are a number of important implications which can be inferred from this anal- ysis. Although we cannot directly investigate the oceanographic impacts of a tidally altered landfast ice cover with this model, in Chapter 2 it was found that the in- troduction of landfast ice altered the location of leads of polynyas, moving them offshore. It was then suggested that this process would alter the properties of the dense water cascading at the shelf break. In this analysis it has been noted that tides had a significant effect on landfast ice location and duration, which will im- pact on the location of the offshore polynyas and leads. Tides were also found to increase the lead fraction by mechanically redistributing thin ice into thicker ice. By increasing the lead fraction the net ice production was also found to increase (Figures 3.11 and 3.25). Annually, over the Arctic the net increase in ice production caused by tides when the independent landfast ice parameterisation was employed was 0.41 cm day −1 , with strong increases in ice production observed in the HB region, Baffin Bay, around Svalbard and Greenland, in the White and Laptev Seas, in the Bering Sea and the Sea of Okhotsk. Similarly in the Antarctic the tides were found to increase the net rate of ice production by 0.37 cm day −1 , with the largest effects seen in the Ross and Weddell Seas. Climate models which do not include landfast ice and tides will misrepresent the location and rate of ice production, and the ice thickness distribution of the pack ice - two crucial characteristics of the seaice cover.
Global warming has caused a significant decrease in seaice coverage in the Arctic. This is having far reaching implications for the ecosystems, as well as dramatically changing the way that humans interact with the Arctic environment. Climate models predicted that a similar decrease in seaice would occur in the Antarctic. However, since regular observations began in 1979, the seaice extent in the Antarctic has been increasing. We review current research that identifies numerous atmospheric and oceanic factors that are influencing seaice trends. These factors have helped to explain some of the changes observed in seaice extent at a regional level, but still do not accurately predict seaice trends for the Southern Ocean as whole. A significant anomaly in seaice extent that occurred in the austral spring of 2016-17 has confounded scientists, and highlights the limitations of current science and climate models to foresee the trend in seaice in the Antarctic. Furthermore, we explore the potential implications for Antarctic ecosystems through a review of current literature. This emphasises the critical role of seaice in the life history of a vast majority of Antarctic species, making them extremely vulnerable to changes in seaice extent. Finally, we consider the implications for human activities in the Antarctic through a series of case studies. These identify the organisations and industries that will be affected by changes in seaice, and who will rely on the development of accurate models and predictions to safely plan their future activities in the Antarctic.
A range of practical activities are also possible connected to this topic, such as growing seaice in a refrigerator and examining it's structure under magnification. The connections between the seaice and the marine ecosystem could also be explored. There are also links to human exploration of polar regions and also human exploitation of these areas.
peratures are close to equalising. The ramp-down in the slow scenario does not start from the same state as the ramp-down in the fast scenario, and a hemispheric temperature “hystere- sis” is created. The temperature- seaice trajectory during the short stabilisation period (point “d” onwards) is towards the initial state, and a sufficiently long period should see the loop completed. Since the “hysteresis” is a function of the rate of ocean heat uptake and release, we would expect differences between climate models which affect transient climate sensi- tivity (Raper et al., 2002). Although the use of global temper- ature as a reference for seaice is applicable in the assessment of transient climate seaice sensitivity (Gregory et al., 2002; Ridley et al., 2007), it may not be suitable in long simulations when ocean heat uptake is significant. The inter-hemispheric temperature loop feeds through to an effect on hemispheric climate processes such as those related to seaice.
In a model study Proshutinsky & Johnson  found two modes of wind driven ice circulation in the Arctic Ocean. The two regimes are referred to as cyclonic and anti-cyclonic. Cyclonic circulation results in an enlarged Beaufort Gyre and shifting of the Transpolar Drift towards the Siberian Coast, transporting ice from the Laptev, East Siberian and Chukchi Seas. Anti-cyclonic circulation has a weakened Beaufort Gyre and the Transpolar drift is shifted into the Central Arctic. Both circulation patterns are associated with changes in oceanic circulation, the cyclonic regime hav ing fresher surface water and stronger vertical stratification than the anti-cyclonic regime. Coupled ice-ocean model results suggest that the regimes alternate and may be persistent for 5 to 7 years [Proshutinsky & Johnson, 1997]. Shifts between regimes are forced by changes in the location and intensity of the Icelandic atmospheric low and the Siberian atmospheric high. It is possible th at the Proshutinsky & Johnson model of atmospheric control of Arctic ice could explain some of the observed Arc tic ice mass variability. Decadal time scale oscillations in atmospheric low pressure systems in the Arctic and North Atlantic are well documented and linked to the NAO, first described by Sir Gilbert Walker in 1930, and the Arctic Oscillation (AO) [Thomas & Wallace, 1998]. Recently the NAO and AO has been linked to seaice extent variability [Feng & Wallace, 1994, Mysak & Veneges, 1998]. Feng and Wallace  identify that temporal variability in seaice extent is strongly coupled to the NAO, the atmosphere leading the ice extent by 2 weeks. The influence of the atmo sphere is found to be strong enough to halt climatological advance of the ice edge in some regions, enhancing it in others, for example a see-saw in seaice extent be tween the Labrador and Greenland Seas. They indicate that local wind stress and thermodynamic forcing can explain seaice variability in most regions; except in the Greenland Sea. Mysak & Veneges  performed principle component analysis on 40 years of ice concentration and sea level pressure data. They found that a standing oscillation in sea level pressure, characterised as the AO, is correlated to a cyclic os cillation in seaice extent. More recently Wang & Ikeda  analyse data from 1901 to 1995 and find that 41% of the variance in seaice extent can be explained by the AO. The second mode apparent in the seaice extent record is related to the NAO, as seaice anomalies in the Labrador and Greenland Seas are out of phase.
The potential of Na + as a marker to reconstruct past seaice changes seems to be strongly site dependent. Concentrations of Na + from an array of West Antarctic ice cores revealed at positive correlation with winter SIE at some sites but not others . In East Antarctica (Talos Dome), Severi et al. (2017)  found a good correlation between September SIE maxima and sodium flux and the same relationship is observed at Dome Fuji . In both these cases, the correlations were stronger when converting sodium concentrations to fluxes, which take into account of the amount of snowfall each year. In Dronning Maud Land, the total sea salt Na+ flux was positively correlated with winter seaice in the Weddell Sea . Besides distance from open water, another factor controlling the amount of sea spray transported over the continent is wind strength, which affects the transport efficiency towards inland sites. Thus, Na + has also been proposed as a proxy for marine air mass advection, suggesting that the positive correlations with seaice may result from strengthened atmospheric circulation at the seaice margin .