* Besides these points, we believe MIROC3.2 is just comparable with or even better than the other state-of-the-art climate models!

TOA radiation imbalance and deep ocean temperature drift

Due to the insufficient spinup, TOA radiation budget has a downward imbalance of ~0.5Wm-2 in high-res. and ~1. Wm-2 in med-res. The ocean temperature has a warming drift accordingly. However, this drift is limited in deep ocean; upper ocean and above surface climate do not show any serious drift.

Cold bias around tropopause

A serious cold bias up to 10K is seen around the tropopause. This is found to be mainly due to insufficient absorption of shortwave radiation by ozone. (This will be fixed in the next version of the model with revised radiation code, unfortunately not in time for the IPCC AR4)

Too shallow subtropical marine boundary layer

The subtropical marine boundary layer is too shallow and a serious dry bias is seen above it (900-700hPa). It indicates a need for improving the parameterization of cloud topped boundary layer in this model.

Too much high cloud cover

The high clouds seem too much in extent and optically too thick. The middle level clouds are then obscured by them and seem too little when looked at from the space. This may be because cloud cover diagnostics and/or cloud overlapping treatment are not very appropriate.

Thin NH sea ice in high-res./Small SH sea ice extent in med-res.

The Northern hemisphere sea ice in high-res. version is not as thick as some observational data. This seems to cause earlier decrease of sea ice extent in the NH responding to the enhancement of GHGs. This is better in the med-res. version, which has, instead, sea ice extent in the Southern hemisphere smaller than observed.

Small amplitude of ENSO

The amplitude of ENSO, in terms of the variation of NINO3 SST, for example, is smaller than reality. It seems to be at least partly attributable to a loose thermocline temperature gradient at the equator.

Problem in volcanic aerosol distribution

In the 20C3M runs, the distribution of optical thickness of volcanic aerosol is prescribed for each volcanic eruption event. It was intended that its vertical distribution has maximum just above the tropopause (diagnosed from model's temperature lapse rate) and decreases with height. However, due to an error in a model parameter, it has actually a maximum uniformly at 50hPa regardless of the temperature profile. We have confirmed that the effect of this error on overall results, including surface temperature change, is small. However, if you look at the trend of 100hPa temperature in 20C, for example, it should be problematic due to this error.

Sea level rise diagnosis (zosga)

Land ice melt is not explicitly added in the 'zosga' data but it is implicitly treated in the model. Since the mass balance of land ice is not considered in the model, snow over ice exceeding a certain critical amount is considerd to be glacier and flows into the oceans, while melted water from ice does not flow into the oceans. This treatment should be OK for equilibrium control but could be problematic for transient runs. We may revise this data with some correction of this term later.

20C historical temperature in high-res.

The 20C3M run with our high-res. version does not really reproduce the observed warming trend in early 20C and cooling trend in middle 20C. The reason for this is unclear. The 20C3M run with our med-res. version, which uses basically the same natural and anthropogenic forcing as in the high-res. run, does reproduce these trends. This might be due to natural inter-decadal variability, but we cannot confirm this because we do not have enough computer resource to make ensemble runs with the high-res. version.