Antarctic sea ice kinematics: satellite observation, model representation and its links with ice thickness

Antarctic sea ice plays crucial roles in the global climate and polar ecosystems, including: 1) regulating moisture and heat flux by forming an insulating layer between ocean and atmosphere; 2) modulating surface energy budget through ice-albedo modification (i.e., sea ice has higher albedo so reflects more solar energy than the ocean); 3) modifying near-surface ocean physical properties via ice production and melting processes; and 4) providing a habitat for polar primary production and higher level fauna (e.g., penguins and seals). To advance our fundamental understanding of Antarctic sea ice, baseline knowledge of Antarctic sea ice physical properties (i.e., ice concentration and thickness) and ice kinematic processes (i.e., ice motion and deformation) is essential. Observations of sea ice motion from satellites and buoy arrays have been widely and repeatedly used for the past four decades to investigate ice properties and kinematics in both Arctic and Antarctic, however there exist many gaps in sea ice research: 1) compared with the Arctic, more research on Antarctic sea ice is needed, due to far fewer validation studies for Antarctic ice motion observations; 2) sea ice concentration has been accurately estimated by remote sensing for over four decades, but the estimation of sea ice thickness, particularly in Antarctica, remains challenging; 3) previous work has shown that ice deformation processes modify the sea ice thickness distribution, and ice thickness also impacts the ice kinematics (i.e., thick ice is more difficult to deform than thin ice), but the investigation of the ice thickness retrieval from sea ice kinematics observations has yet to be explored; and 4) improvements in coupled ocean-sea ice modelling now provide high spatio-temporal resolution simulations, but limitations in accurately representing in the continuum ice rheology preclude realistic simulated ice deformation. Granular models can simulate ice deformation more accurately, but are costly in computation and have not been widely used in sea ice models. To address these gaps, the research within this thesis investigates Antarctic sea ice kinematics using remotely sensed observations of ice motion and coupled ocean-ice model simulations, and explores the links between sea ice kinematics and sea ice physical properties.
The first scientific chapter presents the verification and ultimate correction of a widely-used satellite-derived, low spatial resolution (60 km ⇥ 60 km) daily Antarctic sea ice motion product. The validation based on three in situ drifting buoys in the Weddell Sea and Ross Sea identified two issues with this ice motion product: 1) two large triangular areas of erroneously low velocity relating to an error in the satellite data composite used to derive this product; and 2) invalid assumptions for the average sensing time of each pixel. Upon rectification of these, the performance of the daily composite sea ice motion product is found to be a function of latitude, relating to the number of satellite swaths incorporated. This result indicates that ice motion products derived using the “daily-maps” (DM) approach, which merges all available satellite swaths within a given time period (e.g., 24 h) to produce composite images, increase bias and reduce accuracy. The heterogeneity of the underlying satellite signal is shown to be an important determinant of the performance of ice motion products estimated from passive microwave (PM) brightness temperature DMs. Moreover, this analysis reveals that kinematic parameters derived from ice motion datasets with a shorter timescale are higher in magnitude than that with a longer timescale, indicating a high degree of sensitivity to the observation timescale.
To further investigate how the timescale of satellite observations impacts ice kinematic retrieval, the analysis presented in the first scientific chapter is extended in the second scientific chapter using the new generation of “swath-to-swath” (S2S) PM-derived sea ice motion. This S2S product calculates ice motion from individual satellite swath pairs, across a breadth of timescales from 1.5 to 72 h, and has been favourably evaluated in the Arctic. This study investigates the Antarctic sea ice differential kinematic parameters (DKPs) of divergence and maximum shear rate computed from the S2S ice motion product, and compares it with DM-derived equivalents. This analysis shows that S2S-generated DKPs are higher in magnitude than the DM-derived equivalent DKPs at short timescales, in agreement with the previous work. Moreover, the DKP magnitude derived by the S2S product is highly sensitive to timescales, with an exponential decay observed for both divergence and maximum shear rate. The parameters controlling this decay are defined here as the DKP curve. Using unsupervised clustering in conjunction with the recent satellite-derived estimates of ice thickness, it is shown that the DKP curve parameters correspond well to regions of different ice thickness, opening the possibility for the development of a new proxy for estimating sea ice thickness based on ice kinematics. This new proxy measurement has the potential to give new estimates of Antarctic sea ice thickness over the last three decades, i.e., before reliable altimetric estimates were available.
Based on the robust relationship between ice thickness and the resulting DKPs derived from satellite observations in the second scientific chapter, it suggests that an analysis of sea ice model-simulated DKPs can be used to assess the ability of the model to represent ice thickness redistribution accurately. In the third scientific chapter, a relatively high spatial resolution (0.1!) version of the ocean-ice-coupled model simulation of sea ice motion from the Australian Community Climate and Earth System Simulator (ACCESS-OM2-01) is assessed for this purpose, as well as a general validation of ice motion-derived DKP magnitude in the model. The comparison of modelled sea ice motion and that derived from satellite DMs show that the ACCESS-OM2-01 can simulate Antarctic ice motion accurately, with mean values of correlation coefficient and regression slope of 0.73 and 1.0, respectively. Furthermore, the DKP magnitude derived by ACCESS-OM2-01 ice motion is comparable with that generated from PM-derived satellite observations. Finally, the relationship between DKP magnitude and ice thickness simulated by ACCESS-OM2-01 is investigated using the supervised machine learning-based regression models (Random Forest and eXtreme Gradient Boosting). Results of the machine learning-based regression analysis indicate that model-derived DKP curve parameters cannot predict the ice thickness. This finding contradicts the results in the second scientific chapter of this thesis and the previous work on buoy arrays, which points to limitations in the modelled sea ice rheology. The sea ice component of ACCESS-OM2-01 uses the elastic-viscous-plastic (EVP) ice rheology, which due to the elastic component is known to have difficulty in producing realistic short timescale ice deformation, and this deficiency is highlighted in this analysis. In the future, the development of rheological models with accurate simulation of short timescale ice deformation is essential to improve model representation of sea ice dynamic thickness redistribution. This work considerably advances our knowledge of the impacts of temporal scale on satellite-based estimation of Antarctic sea ice kinematics. The investigation of satellite-derived timescale-dependent ice kinematics firmly establishes a link between ice kinematics and ice thickness, and shows how this link might be exploited to retrieve estimates of ice thickness back to the early 1990s. It also provides suggestions for the development of future satellite missions. Moreover, the relationship between ice kinematics and thickness enables a new way of assessing the ice kinematic accuracy of coupled sea ice-ocean model simulations. It highlights a deficiency in ice rheology, and suggests the future improvement of ice model simulation in ice dynamic thickness redistribution.