8.3.3 Sea Ice
The magnitude and spatial distribution of the high-latitude climate changes can be strongly affected by sea ice characteristics, but evaluation of sea ice in models is hampered by insufficient observations of some key variables (e.g., ice thickness) (see Section 4.4). Even when sea ice errors can be quantified, it is difficult to isolate their causes, which might arise from deficiencies in the representation of sea ice itself, but could also be due to flawed simulation of the atmospheric and oceanic fields at high latitudes that drive ice movement (see Sections 8.3.1, 8.3.2 and 11.3.8).
Although sea ice treatment in AOGCMs has become more sophisticated, including better representation of both the dynamics and thermodynamics (see Section 8.2.4), improvement in simulating sea ice in these models, as a group, is not obvious (compare Figure 8.10 with TAR Figure 8.10; or Kattsov and Källén, 2005, Figure 4.11). In some models, however, the geographic distribution and seasonality of sea ice is now better reproduced.
Figure 8.10. Baseline climate (1980–1999) sea ice distribution in the Northern Hemisphere (upper panels) and Southern Hemisphere (lower panels) simulated by 14 of the AOGCMs listed in Table 8.1 for March (left) and September (right), adapted from Arzel et al. (2006). For each 2.5° x 2.5° longitude-latitude grid cell, the figure indicates the number of models that simulate at least 15% of the area covered by sea ice. The observed 15% concentration boundaries (red line) are based on the Hadley Centre Sea Ice and Sea Surface Temperature (HadISST; Rayner et al., 2003) data set.
For the purposes of model evaluation, the most reliably measured characteristic of sea ice is its seasonally varying extent (i.e., the area enclosed by the ice edge, operationally defined as the 15% contour; see Section 4.4). Despite the wide differences among the models, the multi-model mean of sea ice extent is in reasonable agreement with observations. Based on 14 of the 15 AOGCMs available at the time of analysis (one model was excluded because of unrealistically large ice extents; Arzel et al., 2006), the mean extent of simulated sea ice exceeds that observed in the NH by up to roughly 1 x 106 km2 throughout the year, whereas in the SH the annual cycle is exaggerated, with too much sea ice in September (~2 x 106 km2) and too little in March by a lesser amount. In many models the regional distribution of sea ice is poorly simulated, even if the hemispheric areal extent is approximately correct (Arzel et al., 2006; Holland and Raphael, 2006; Zhang and Walsh, 2006). The spread of simulated sea ice extents, measured as the multi-model standard deviation from the model mean, is generally narrower in the NH than in the SH (Arzel et al., 2006). Even in the best case (NH winter), the range of simulated sea ice extent exceeds 50% of the mean, and ice thickness also varies considerably, suggesting that projected decreases in sea ice cover remain rather uncertain. The model sea ice biases may influence global climate sensitivity (see Section 8.6). There is a tendency for models with relatively large sea ice extent in the present climate to have higher sensitivity. This is apparently especially true of models with low to moderate polar amplification (Holland and Bitz, 2003).
Among the primary causes of biases in simulated sea ice (especially its distribution) are biases in the simulation of high-latitude winds (Bitz et al., 2002; Walsh et al., 2002; Chapman and Walsh, 2007), as well as vertical and horizontal mixing in the ocean (Arzel et al., 2006). Also important are surface heat flux errors, which in particular may result from inadequate parametrizations of the atmospheric boundary layer (under stable conditions commonly occurring at night and in the winter over sea ice) and generally from poor simulation of high-latitude cloudiness, which is evident from the striking inter-model scatter (e.g., Kattsov and Källén, 2005).