In order to assess the performance of two versions of the IAP/LASG Flexible Global Ocean-Atmosphere- Land System (FGOALS) model, simulated changes in surface air temperature (SAT), from natural and anthropogenic forcings, were compared to observations for the period 1850-2005 at global, hemispheric, continental and regional scales. The global and hemispheric averages of SAT and their land and ocean components during 1850-2005 were well reproduced by FGOALS-g2, as evidenced by significant correlation coefficients and small RMSEs. The significant positive correlations were firstly determined by the warming trends, and secondly by interdecadal fluctuations. The abilities of the models to reproduce interdecadal SAT variations were demonstrated by both wavelet analysis and significant positive correlations for detrended data. The observed land-sea thermal contrast change was poorly simulated. The major weakness of FGOALS-s2 was an exaggerated warming response to anthropogenic forcing, with the simulation showing results that were far removed from observations prior to the 1950s. The observations featured warming trends (1906-2005) of 0.71, 0.68 and 0.79 C (100 yr) 1 for global, Northern and Southern Hemispheric averages, which were overestimated by FGOALS-s2 [1.42, 1.52 and 1.13 C (100yr)-1 ] but underestimated by FGOALS-g2 [0.69, 0.68 and 0.73 C (100 yr)-1 ]. The polar amplification of the warming trend was exaggerated in FGOALS-s2 but weakly reproduced in FGOALS-g2. The stronger response of FGOALS-s2 to anthropogenic forcing was caused by strong sea-ice albedo feedback and water vapor feedback. Examination of model results in 15 selected subcontinental-scale regions showed reasonable performance for FGOALS-g2 over most regions. However, the observed warming trends were overestimated by FGOALS-s2 in most regions. Over East Asia, the meridional gradient of the warming trend simulated by FGOALS-s2 (FGOALS-g2) was stronger (weaker) than observed.
