Search

A large majority of the direct observational record for glacier changes falls within the industrial period, from the 19th century onward, associated with global glacier retreat. Given this availability of data and the significant focus in contemporary glacier modelling falling on recent retreat, glacier models are typically calibrated using – and validated with – only observations of glaciers that are considerably out of equilibrium. In order to develop a broader picture of the skill of one glacier model – the Open Global Glacier Model (OGGM) – we model glaciers for extended historical timescales of 850–2004 CE using a selection of six general circulation model (GCM) outputs. We select glaciers for which long-term length observations are available in order to compare these observations with the model results, and we find glaciers with such observations in almost all glacierised regions globally. In many regions, the mean modelled glacier changes are consistent with observations, with recent observed retreat in these regions typically at the steeper end of the range of modelled retreats. However, on the scale of individual glaciers, performance of the model is worse, with overall correlation between observed and modelled retreat weak for all of the GCM datasets used to force the model. We also model the same set of glaciers using modified climate time series from each of the six GCMs that keep temperature or precipitation constant, testing the impact of each individually. Temperature typically explains considerably more variance in glacier lengths than precipitation, but results suggest that the interaction between the two is also significant within OGGM and neither can be seen as a simple proxy for glacier length changes. OGGM proves capable of reproducing recent observational trends on at least a qualitative level in many regions, with a modelling period over a considerably larger timescale than it is calibrated for. Prospects are good for more widespread use of OGGM for timescales extending to the pre-industrial period, where glaciers were typically larger and experience less rapid (and less globally consistent) geometry changes, but additional calibration will be required in order to have confidence in the magnitude of modelled changes, particularly on the scale of individual glaciers.

The large natural variability of the Antarctic sea ice is a key characteristic of the system that might be responsible for the small positive trend in sea ice extent observed since 1979. In order to gain insight in the processes responsible for this variability, we have analysed in a control simulation performed with a coupled climate model a strong positive ice-ocean feedback that amplifies sea ice variations. When sea ice concentration increases in a region, in particular close to the ice edge, the mixed layer depth tends to decrease. This can be caused by a net inflow of ice and thus of freshwater that stabilizes the water column. Another stabilizing mechanism at interannual time scales that appears more widespread in our simulation is associated with the downward salt transport due to the seasonal cycle of ice formation: brine is released in winter when ice is formed and mixed over a deep layer while the freshwater flux caused by ice melting is included in a shallow layer, resulting in a net vertical transport of salt. Because of this stronger stratification due to the presence of sea ice, more heat is stored at depth in the ocean and the vertical oceanic heat flux is reduced, which contributes to maintain a higher ice extent. This positive feedback is not associated with a particular spatial pattern. Consequently, the spatial distribution of the trend in ice concentration is largely imposed by the wind changes that can provide the initial perturbation. A positive freshwater flux could alternatively be the initial trigger but the amplitude of the final response of the sea ice extent is finally set up by the amplification related to ice-ocean feedback. Initial conditions have also an influence as the chance to have a large increase in ice extent is higher if starting from a state characterized by a low value.

Two hindcast (1983–2007) simulations are performed with the global, ocean-sea ice models NEMO-LIM2 and NEMO-LIM3 driven by atmospheric reanalyses and climatologies. The two simulations differ only in their sea ice component, while all other elements of experimental design (resolution, initial conditions, atmospheric forcing) are kept identical. The main differences in the sea ice models lie in the formulation of the subgrid-scale ice thickness distribution, of the thermodynamic processes, of the sea ice salinity and of the sea ice rheology. To assess the differences in model skill over the period of investigation, we develop a set of metrics for both hemispheres, comparing the main sea ice variables (concentration, thickness and drift) to available observations and focusing on both mean state and seasonal to interannual variability. Based upon these metrics, we discuss the physical processes potentially responsible for the differences in model skill. In particular, we suggest that (i) a detailed representation of the ice thickness distribution increases the seasonal to interannual variability of ice extent, with spectacular improvement for the simulation of the recent observed summer Arctic sea ice retreats, (ii) the elastic-viscous-plastic rheology enhances the response of ice to wind stress, compared to the classical viscous-plastic approach, (iii) the grid formulation and the air-sea ice drag coefficient affect the simulated ice export through Fram Strait and the ice accumulation along the Canadian Archipelago, and (iv) both models show less skill in the Southern Ocean, probably due to the low quality of the reanalyses in this region and to the absence of important small-scale oceanic processes at the models' resolution (~1°).