Optimal localization radius of data assimilation for Arctic sea ice initialization using CICE5/DART

Abstract

In data assimilation (DA), localization, which adjusts the influence of observations on model state vectors, is an essential process for improving initial conditions. Given that the localization radius varies depending on the model and observation, sensitivity tests were conducted to identify the optimal localization radius for assimilating satellite-derived sea ice concentration and sea ice thickness into Los Alamos Sea Ice Model version 5 (CICE5) using Data Assimilation Research Testbed (DART). In all experiments, the updated Arctic sea ice initial conditions were generally improved across Pan-Arctic regions and time periods. Based on the sensitivity tests, the optimal localization radius for univariate DA was approximately 0.05 radians which is the default value in CICE5/DART. However, for multivariate DA, it was around 0.02 radians. This implies that the optimal localization condition for assimilation may vary depending on whether univariate or multivariate observations are being assimilated.