Reconstructing Monthly Arctic Sea Ice Conditions Using Online Data Assimilation

Few spatially complete records of Arctic sea-ice concentration and thickness exist prior to satellite observations. Arctic sea-ice reconstructions are used as boundary conditions for reanalysis products and other model simulations, thus gaining better estimates of Arctic sea ice throughout the Instrumental Era (1850-present) would not only improve our understanding of Arctic sea ice but Arctic climate more broadly. Here we employ data assimilation along with a Linear Inverse Model (LIM) to reconstruct global sea-ice concentration and thickness along with other climate fields on monthly timescales from 1850 to present. Specifically, we use a Kalman Filter technique to optimally combine climate model output and temperature observations, and the LIM is used to efficiently emulate the dynamics of a global climate model, thereby incorporating memory into the reconstructions. We find that the LIM outperforms persistence forecasts of Arctic sea-ice concentration and thickness on monthly timescales and reveals the dynamical modes that contribute most to Arctic sea-ice predictability. We validate our reconstructions against satellite observations and find good performance during this time period. We also compare our results with previous work and with an offline approach (no memory between timesteps). Finally, we use reconstructions to investigate the seasonality and dynamical conditions associated with the early 20th century warming (1910-1940), which was the largest Instrumental Era warming event prior to the satellite era. In particular, we investigate the role of sea-ice thickness in this event compared to satellite era sea-ice loss.