Present-day climate simulation by a 20km horizontal resolution (TL959L60) global atmospheric general circulation model (AGCM) was validated using re-analysis and precipitation datasets for the period of 1979-2000. This model was developed based on the Japan Meteorological Agency operational forecast model. The model was integrated under the observed sea surface temperatures of the HadISST (Hadley Center sea Ice and Sea Surface Temperature dataset). The result was compared with those of 60km (TL319L60), 120km (TL159L60) and 180km (TL95L60)-mesh AGCMs.

In this study, we focused on how large-scale monthly mean climatology is improved (or not) by model resolutions. Statistical scores such as the Root Mean Squared Error (RMSE) and the Taylor's skill score were compared for monthly mean fields interpolated onto uniform 2.5 degree grids. Two kinds of comparisons were made; first, except for the change of horizontal resolutions, we didn't carry out any particular tuning suited for each model resolution. Secondly, we also compared the results of the 180km and 60km models which were tuned for lower resolution models to the extent that global averaged precipitation and radiation balance were comparable with those of the 20km model. The reason was that these models showed too large bias issued from physical processes when the parameters for the 20km model were applied to the lower resolution models.

We found that climatological precipitation was improved by higher resolution models, but the global skill for 20km and 60km models were comparable. An approximate 10% RMSE reduction was seen for the 60-km model simulation as compared with the 180-km model on average. On the other hand, further improvement of the score was not seen necessarily for all kinds of the score as the 20km model was compared with the 60km model.

For atmospheric fields, we analyzed zonal mean and stationary wave components of temperature, sea-level pressure, geopotential height and winds. The scores were improved generally for the higher resolutions. The representation of zonal mean temperature and wind fields of 20km model was the best among them. The improvement of the stationary wave component was comparable between the 20km and 60km models..

The superiority of 20km model to 60km model in simulating large scale fields was not so clear. However, we could see advantages of 20km model in specific regions and seasons. Excessive precipitation bias over and around the Himalaya and Andes mountains was much improved. Deficient precipitation relating to North America summer Monsoon was improved over Arizona, New Mexico and the Sierra Madre Occidental. The increased precipitation strengthened the upper monsoon anticyclone over Mexico, which tended to be too weak in the 60km model and the lower resolution models. In Asia-Australia Monsoon regions, large reduction of the RMSE of precipitation by the 20km model was also found over lands and the surrounding oceans. Seasonal march in early stage of the Asian Monsoon was simulated best in the 20km model. Corresponding to the improvement for precipitation, low level wind error in 20km model was significantly reduced over the Asia-Australia Monsoon region. It is speculated that better representation of the effect of complex topography on water vapor transports result in better simulation of precipitation and monsoon circulations in North America Monsoon and Asia-Australia Monsoon.

These results shows that the 20km model has advantages not only in the representation of tropical cyclones and severe weather but also improve large-scale climatological mean fields in specific regions and seasons due to more accurate representation of the fine topography even as compared with the 60km model.