Intercomparison of downward shortwave radiation and surface air temperatures derived from dynamical and statistical downscaling models

Downward shortwave radiation (Sd) and surface air temperature for the period 1985-2004, downscaled using four dynamical models (NHRCM, NRAMS, TRAMS, and TWRF) and one statistical model (CDFDM) were compared, taking Japan as an example. Dynamical models overestimated Sd for the annual mean and seasonal variation. Seasonal variation of mean temperature (Tm), the daily maximum temperature (Tx), and the daily minimum temperature (Tn) were well simulated, except the comparatively large cold bias for NRAMS. The relations between Sd error and surface air temperature errors in annual means were compared. For NRAMS and CDFDM, the Sd error caused both Tm error and temperature diurnal range, Tr, error. A similar tendency was shown by TWRF and TRAMS, but its sensitivity was weak. Although NHRCM had a similar tendency to that shown by CDFDM for Tr, that tendency was weak for Tm. We defined the total error of Sd, Tm and Tr as the moving radius length, and its dependency as arguments in a polar coordinate system. The shortest moving radius was CDFDM. Those of NHRCM and NRAMS were also shorter than those of the other dynamical models. The total error of NHRCM caused by only Sd and Tr errors, although that of NRAMS and CDFDM resulted from all components, suggesting that the latter models 'correctly' failed to simulate Tm and Tr under given Sd conditions.