An in depth discussion of the physical effects of all of the 25 parameters which varied in the EnKF experiments is beyond the scope of this paper (and perhaps of little interest to those using different models). Here we briefly describe the statistical behaviour of the most significant parameters along with their characteristics to the extent that they illustrate some the results described above.
Table 4 shows the correlations of the temperature differences between the experimental climate states and the control climate for 9 of the 25 parameters which were allowed to vary independently in the EnKF experiments. These 9 parameters (defined in Table 3) are the only ones which individually showed even a marginally (at the previously mentioned 1% level) significant correlation for any of the three experimental climates (considering T2 changes with respect to the CTRL climate) at the global or tropical scale.
At the global and tropical scale for the 2×CO2 climate, the clearly dominant parameter is “prctau”, but interestingly, this parameter is less dominant for the LGM and LGMGHG climate states. This parameter is the time scale for ice precipitation in the clouds. Therefore a higher value should correspond to a higher total amount of ice in the clouds. The relationship of ice, clouds and sensitivity is complex, but Tsushima et al. (2006) find less cloud ice to be linked to in a larger pole-ward shift in cloud water and therefore a reduced cloud albedo effect, amplifying the overall warming. However, at least for the two versions of MIROC considered in that work (among other GCMs) and an intermediate unpublished version, their overall sensitivity to LGM boundary conditions is rather similar, indicating that this change in model formulation has little effect under strong cooling conditions. The other 3 parameters which are significant for the 2×CO2 temperature changes in the tropics are not very significant on the global scale. Zonal analysis (not shown) shows that this is caused by a sharp decrease in the correlation (or even opposite correlation in some cases) in the southern sea ice region. While the LGM temperature changes also show significant correlation with these parameters particularly in the tropics, the LGM picture is further complicated by effects from four other parameters. This result is consistent with the results in Table 2 and Fig. 3, which shows considerable scatter in the relationship between LGM and 2×CO2 temperature changes. Of these 4 additional parameters, 2 (alp, snfrs) are not significant for LGMGHG. It is perhaps unsurprising that alp (gravity wave drag) and snfrs (related to albedo) are more strongly related to the temperature changes over the ice sheet. The overall similarity in the parameters that are significant for both LGMGHG and LGM climate changes is consistent with the general similarity of the results for these two climates shown in Sub-plot A of Fig. 4.
Columns 8 and 9 of Table 4 show the correlations of the ratios of the T2 temperature changes for LGM/2CO2 and LGMGHG/2CO2 respectively. These ratios should indicate which parameters are linked to the scatter around the red line in Fig. 3 and the spread of the histogram in Fig. 5. “prctau” is important for the LGM/2CO2 ratio, but “vice0” is more important for the LGMGHG/2CO2 ratio. “vice0” is a scaling factor on the ice fall speed so has a similar effect to “prctau” in that its value affects the total amount of ice in the clouds. It seems plausible that some of the asymmetrical effect between warming and cooling is, therefore, linked in this model to the distribution of ice in clouds. The other parameter important in the temperature ratios in columns 8 and 9 of Table 4 is “dffmin”, the minimum vertical diffusion coefficient. Higher values of this parameter are expected to lead to warmer winter conditions at high altitudes, so the result here, which indicates that the parameter is more important for cooler than warmer climates, is reasonable.