Systematic biases in the SUNYA/NCAR (T31L18) AMIP experiment are analyzed by comparing simulated and observed 1980-88 climatological means. Experiments with R15 resolution are also conducted to examine the sensitivity of these biases to horizontal resolution and to cloud radiative treatment. To facilitate the analyses, the model bias is decomposed into zonal mean and eddy components of the annual average and its monthly variation. Note that, when averaged over a whole year and all the longitudes at a latitudinal band, the total (squared) biases equal to the sum of the four (squared) components.
The surface pressure bias is found to be large, dominated by the eddies of the annual average. Annual average and monthly variation of precipitation is generally realistic in the zonal mean, but large regional errors are identified especially in the tropics. Biases in the 200 hPa zonal wind show three peaks at 60°S, 30°S and the equator, and indicate a systematic northward shift of the jet with biases predominantly in the annual average. Substantial reductions in these biases occur when the resolution increases from R15 to T31. Similarly, 500 hPa height errors in the Southern Hemisphere extratropics can be attributed primarily to the annual average and these biases are also substantially reduced as resolution increases. When compared with ECMWF analysis data, a cold upper troposphere throughout the year, especially at high latitudes, exists in both resolutions. The cloud cover in T31 is about half of observations with consequent large biases in cloud radiative forcing. These are caused mainly by the errors in the annual average, which can largely be corrected by using a lower relative humidity threshold for cloud formation. Despite the fact that the global mean cloud cover is close to observations in R15, daily variability in the 200 hPa zonal wind is largely overestimated in the tropics, but it is more realistically simulated in T31. In addition, over most of the Northern Hemisphere extratropics, 500 hPa height meridional gradient biases reverse sign between R15 and T31 during winter. These results indicate that the relative importance of dynamic versus cloud-radiative forcing errors cannot be clearly identified. This is also evident from another R15 simulation with different cloud radiative treatment. Although the improved cloud radiative scheme produces more realistic seasonal variations in cloud cover and cloud radiative forcing, impacts on the geographic distribution of 200 hPa zonal wind, 500 hPa height, surface air temperature and precipitation are found to be relatively small.