Journal article Open Access
The ice thickness distribution (ITD) is one of thecore constituents of modern sea ice models. The ITD ac-counts for the unresolved spatial variability of sea ice thick-ness within each model grid cell. While there is a generalconsensus on the added physical realism brought by theITD, how to discretize it remains an open question. Here,we use the ocean–sea ice general circulation model, Nu-cleus for European Modelling of the Ocean (NEMO) ver-sion 3.6 and Louvain-la-Neuve sea Ice Model (LIM) ver-sion 3 (NEMO3.6-LIM3), forced by atmospheric reanalysesto test how the ITD discretization (number of ice thicknesscategories, positions of the category boundaries) impacts thesimulated mean Arctic and Antarctic sea ice states. We findthat winter ice volumes in both hemispheres increase with thenumber of categories and attribute that increase to a net en-hancement of basal ice growth rates. The range of simulatedmean winter volumes in the various experiments amounts to∼30 % and∼10 % of the reference values (run with fivecategories) in the Arctic and Antarctic, respectively. Thissuggests that the way the ITD is discretized has a signifi-cant influence on the model mean state, all other things beingequal. We also find that the existence of a thick category withlower bounds at∼4 and∼2 m for the Arctic and Antarc-tic, respectively, is a prerequisite for allowing the storageof deformed ice and therefore for fostering thermodynamicgrowth in thinner categories. Our analysis finally suggeststhat increasing the resolution of the ITD without changingthe lower limit of the upper category results in small but notnegligible variations of ice volume and extent. Our study pro-poses for the first time a bi-polar process-based explanationof the origin of mean sea ice state changes when the ITDdiscretization is modified. The sensitivity experiments con-ducted in this study, based on one model, emphasize thatthe choice of category positions, especially of thickest cat-egories, has a primary influence on the simulated mean seaice states while the number of categories and resolution haveonly a secondary influence. It is also found that the currentdefault discretization of the NEMO3.6-LIM3 model is suffi-cient for large-scale present-day climate applications. In allcases, the role of the ITD discretization on the simulatedmean sea ice state has to be appreciated relative to other in-fluences (parameter uncertainty, forcing uncertainty, internalclimate variability).