HB-1-2

Sea-ice models are used to understand sparse data of ice thickness and drift. Yet, they are based on many assumptions and parameterizations. Many standard parameterizations have been introduced based on physical reasoning, but details of the parameterizations and parameter choices often remain unclear. For example, Tremblay and Mysak developed a granular material rheology for sea ice based on the Mohr-Coulomb yield curve and a non-normal flow rule. Wang and Wang identified this rheology to give lead patterns in better agreement with observations when compared to the more widely used elliptical yield curve and normal flow rule. More generally, the sensitivity of sea-ice models to numerical details and to parameterizations has been described. Until recently, these models have been run at relatively low resolution and the details of the linear kinematic features (LKF or leads) in the pack ice were not well resolved, so that any rheology model including compressive and shear strength lead to indistinguishably good results when compared to sea ice drift observations. In addition to the low resolution, viscous plastic sea ice models were generally not iterated to convergence adding to the problem of poorly defined LKF. As the resolution of numerical grids becomes higher, LKF are better resolved, but current models fail to reproduce the observed statistics of the ice deformation and new approaches to simulating sea ice are needed. At very high resolution the details of the rheology model become important for the interesting dynamics of leads in sea ice. The ever higher resolution also requires more accurate and efficient numerical techniques for solving the momentum equations of sea-ice. The adjoint technique allows the exact computation of gradients with respect to model variables within complex numerical models. The gradient information can be used for minimizing cost or objective functions that describe model-data misfit in order to obtain so-called state estimates (i.e., the fit of complex numerical models to many different observations). Adjoint sea-ice models have been used to analyze and assess the sensitivity of sea-ice distribution to forcing factors on seasonal time scales (Kauker et al., 2009), or of ice-transport through Lancaster Sound on multi-year time scales. State estimates with adjoint sea-ice models in regional domains (Labrador Sea) can be found in Fenty (2010). Modeling system and adjoint and inverse techniques in Heimbach et al. (2010) and Fenty (2010) were the same as the ones used in, for example, Losch and Heimbach (2007) and Losch and Wunsch (2003) in purely oceanographic contexts. Different data assimilation techniques, such as Kalman Filters or Monte-Carlo techniques, have been used only sporadically in the context of sea-ice models and observations. To this date comprehensive sensitivity studies that take into account a full range of objectives and control parameters are not known. At the same time the uncertainty of the parameterizations of the sea-ice models need to be assessed in the context of observations. State estimates are required that lead to sea-ice simulations that are dynamically consistent with icerelated observations.

These hypotheses will be tested with a general ocean circulation model with sea-ice (MITgcm) in an Arctic-North-Atlantic model domain with southern boundaries near 45°N. An adjoint model will be set up to explore sensitivities of observables such as ice export through gateways, ice extent, ice volume to initial conditions, boundary conditions (forcing), and internal model parameters. The ensuing general strategy is outlined as follows: constrain the model with available observations; reject implemented physics, if no state consistent with observations can be found, and give recommendations for revised physics; revise physics using (adjoint) sensitivities to modify rheology (including solution techniques), resolution, or parameterizations and re-iterate optimization. As a result, sensitivities of sea-ice model observables and an optimal state (i.e., a set of model parameters and surface forcing) consistent with available observations (see project HB-1) and implemented physics will be available. These results will be combined with the results of CA-1 and CA-2 and form the basis for predictions of Arctic sea-ice distribution on short and long scales.