Medium-range numerical prediction of Antarctic sea ice

This research investigates the use of a multiple-thickness sea-ice model for opera- tional Antarctic sea-ice forecasts. Deterministic predictions are generated with a finite difference sea-ice model for the Southern Ocean. The forecast model includes a new method for ridging sea ice on semi-diurnal timescales and includes a modified Coulombic rheology. This approach differs from formalistic techniques used in existing thickness distribution models, and is introduced to simulate oriented fractures with multiple thickness- categories. Behaviour of the dynamic component of the model is assessed on a test grid to illustrate its properties and numerical limitations. The dynamic component is then combined with a constant salinity, enthalpy-conserving thermodynamic model on a 50 km resolution polar stereographic grid using 11 thickness categories to simulate circum-Antarctic sea ice. Sea-ice assimilations are generated for observation intensive years (1992 and 1996) to compare model output with drafts from Upward Looking Sonar, ice velocities from drifting buoys and concentration from satellite measurements. The assimilations are forced with atmospheric fields from the European Centre for Medium Range Weather Forecasting (ECMWF) and National Oceanic and Atmospheric Administration Sea Surface Temperatures. Climatological-mean ocean currents are used as a background geostrophic field beneath the ice. The sea-ice analyses are then filtered with Special Sensor Microwave/Imager (SSM/I) derived sea-ice concentrations, adjusting the inno- vation sequence (sequence of observation minus forecast at each timestep) for the unique thermodynamic lag of each ice-thickness category. This method provides an estimate of oceanic heat flux and explicitly adjusts vertical ice temperature profiles. It improves approximations of sea-ice extent and mean thickness, suggesting a significant source of error in the unassimilated model results from incorrect heat flux specification between ice, ocean and atmosphere. However the scheme was less successful at assimilating individual thickness categories of the thickness distribution, and possible future im- provements to the scheme are discussed. A series of G-day forecast case studies designed to test short-term predictability of the model are provided in the final part of this thesis. Pack forecasts are generated with both ECMWF and National Centre for Environment Prediction (NCEP) atmospheric reanalyses. They are verified against assimilations and SSM/I-derived sea-ice concen- tration. The results demonstrate that as long as the initial sea-ice state used to initialise forecasts is physically consistent, the medium-range forecast will itself be quite skilled. Information obtained from two sets of forecasts, one forced with ECMWF and the other with NCEP, shows both sets of forcings generate regionally different but simi- larly skilled forecasts. This suggests that ensembles methods may improve operational forecast guidance. Overall, the work paves the way for regular operational Antarc- tic sea-ice forecasts and the potential of determining Antarctic sea-ice thickness from remotely-sensed observations.