Comparison of model noise in AGCM independent ensemble runs and continuous simulation

Abstract

Atmospheric General Circulation models (AGCMs) forced by prescribed sea surface temperature (SST) climatological seasonal cycle simulate interannual variabilities that have cyclic characteristics. Such cyclic characteristics generate relationships between one year and the next in the model output. We document these relationships by computing lag-1-year autocorrelation in hundreds of years of CAM3 and ECHAM5 simulations. The autocorrelation is found to be generally less than 0.2, but contain robust structures. In case of zonal averaged zonal wind and air temperature the winter hemisphere is characterized by negative autocorrelation and the summer hemisphere characterized by positive autocorrelation. The presence of autocorrelation means that an average over a 10∼25 year AGCM simulation in an effort to reduce the influence of interannual variability on externally-driven climate change might not be very effective. In view of this, we investigate if ensemble runs instead of a continuous simulation is more effective in reducing such influences. The reduction gain by using N 1-year long ensemble runs over N years of continuous run is generally less than 30% and mainly limited to the areas where the autocorrelation is positive. We thus conclude that each year in a continuous simulation can generally be treated as largely independent of the next year in an AGCM run with fixed SST forcing.