An alternative multi-model ensemble mean approach for near-term projection

An ‘alternative multi-model ensemble mean’ (AMME) method was developed for the near-term projection of regional climate change by taking into account the capacity of currently available climate models in simulating specific timescale components. These components included a climatological mean (Mean), an amplitude–frequency modulated annual cycle (MAC), multi-decadal variability (MDV), a secular trend (ST), and short-term variability (SV). The latter four components were extracted adaptively by the ensemble empirical mode decomposition filter from the climate series. For each component, a reconstructed simulation was determined from ensemble of a limited number of model simulations that could reproduce the component in the observation relatively well. An AMME simulation was obtained by combining the five components. The new method was illustrated to construct an AMME simulation of the monthly near-surface temperature series for the training period 1902–1990 in eastern China and was applied to the validation period 1991–2004. For the eastern China average, the best performance arose from MPI-ESM-MR for Mean, IPSL-CM5A-LR for MAC, ACCESS1.3 for MDV, GFDL-ESM2M for ST, and GISS-E2-H-CC for SV. Serving as a novel tool for producing reasonable near-term future climate change scenarios by utilizing currently available model simulations, the AMME exhibited a better performance in reproducing both past and near-term ‘future’ climate than conventional multi-model ensemble means and weighted average schemes.