Estimation of future surface temperature changes constrained using the future-present correlated modes in inter-model variability of CMIP3 multimodel simulations

This study utilized a new method for future surface temperature changes constrained by model performance in the late 20th century (20C) climate. We applied the singular value decomposition method to investigate covariance between future and present climates in a multimodel ensemble. We established a transfer function between the expansion coefficients of the present and future climate modes. By projecting the observations onto the present modes and using the transfer function, we obtained the best estimates of the future projection. In this study, we extracted the first two significant leading modes. The first mode showed inter-model variation in spatial patterns of temperature change, with Arctic amplification in the future, and the second mode showed variability in temperature change occurring in the southern marginal regions of sea ice in the Arctic for the 20C simulation. Our evaluation suggests that the future warming of the ensemble mean projection underestimated, particularly in the Arctic region. The estimated temperature change depends mainly on the 20C state of the sea ice and surface temperature in the northern North Atlantic Ocean, which are strongly expressed in the first mode of the 20C climate simulation. The leave-one-out cross-validation indicated that our estimation can improve multimodel mean estimates of temperature change in the higher-latitudes of the Northern Hemisphere.