Arctic sea ice predictability on daily-to-weekly timescales: sensitivity to initial positional errors under different rheology formulations

Abstract. We investigate short-term (daily-to-weekly) winter Arctic sea-ice predictability using a coupled ice–ocean model, and focusing on how sensitive forecasts are to initial uncertainty in the location of sea ice features (e.g., leads, ridges, etc.). In this context, two rheologies are compared: elastic–viscous–plastic (aEVP) and brittle Bingham–Maxwell (BBM). For January–March 1997, we conduct 10 d ensemble forecasts, initialized by applying displacement perturbations to all sea-ice fields to represent initial positional errors, while keeping atmospheric forcing identical for all the ensemble members. Potential predictability is evaluated using a “perfect model” framework and probabilistic metrics for the ice-edge position errors, local state-variable errors (concentration, thickness, drift, deformation), and the spread of virtual drifters. Ice-edge forecasts are found to be largely insensitive to initial positional errors for both rheologies, indicating dominance of thermodynamic forcing rather than ice dynamics at short lead times. In contrast, BBM exhibits strong nonlinear sensitivity in pack ice: predictability is limited to 1–5 d for drift and deformation and 5–10 d for concentration. The aEVP model, on the other hand, quickly damps small-scale heterogeneities, yielding more convergent, and thus more predictable solutions. These findings have concrete implications: the BBM model produces larger regions with high probability of intense deformation and the spread of Lagrangian drifters is up to an order of magnitude greater than in the aEVP model. Our results underscore the importance of ensemble forecasting for quantifying risks in a highly nonlinear and weakly predictable sea-ice system.