Seasonal and regional analysis of Arctic sea-ice evolution

Abstract

<p>We examine the seasonal and regional evolution of sea-ice coverage in the Arctic. Using satellite and reanalysis data in comparison with model simulations of the 6<sup>th</sup> model intercomparison project (CMIP6), we build on studies (e.g. Niederdrenk and Notz, 2018; SIMIP Community, 2020) that have found a strong linear relationship between September sea-ice area of the northern hemisphere and global atmospheric air temperature (TAS) as well as anthropogenic <span>CO<sub>2</sub></span> emissions. Instead of focusing on the whole Arctic and September sea ice, we perform sensitivity analyses on higher-resolved regional and seasonal scales, aiming to identify the atmospheric and oceanic drivers that govern the evolution of sea-ice coverage on these scales and to derive simple empirical relationships that describe the impact of these processes. We find clear linkages also on these higher-resolved scales, with different regions and different seasons showing diverse sensitivities of sea-ice area evolution with respect to TAS and anthropogenic <span>CO<sub>2</sub></span>. Furthermore, we make use of the Mahalanobis distance as a multivariate metric to quantify the "quality" of a single simulation matching the observations, thereby considering the different sensitivities of all months of the year. Building the combined covariance matrix of the observations and simulations as a measure of the joint uncertainties, we can determine how "close" to the observations each single member of the simulations is. This allows us to separate models whose sensitivities are in overall good agreement with the observations from those that are apparently not capable of properly simulating the response of the sea ice to the forcing throughout all months.<br>Based on our findings we can infer the dominant drivers that force Arctic sea-ice evolution on a regional and seasonal scale and also derive projections for the future evolution of Arctic sea ice for different climate scenarios based on simple empirical relationships that can directly be estimated from observational records.<br><br>