Near Surface Boreal Summer Climate as Simulated by Three General Circulation Models

Abstract Ensembleintegrations of three general circulation models (COLA, NCAR and NCEP) have been performedover five different boreal summer seasons (June through September of 1986, 1987, 1988, 1993 and1994) with prescribed observed sea surface temperature to assess the predictability of seasonal climate duringthe boreal summer. Beyond some inconsistent initialization of soil wetness among the models, there isno land surface contribution to predictability that can be assesed. An evaluation of the systematic errors ofthe models shows strikingly similar patterns among them. Despite large systematic errors, aspects of the time evolution of climate during the borealsummer are well simulated, with particular attention paid in this analysisto the major monsoon,regions of Asia and North America. Potential predictability is assessed by examining in tandem the models’ skill as measured by their anomaly correlation coefficients, and the models’ signal to noise ratio (essentially interannual versus intra-ensemble variance) as a measure of confidencein the results. Co-location of skill in anomaly simulation and a robust signal is a strong indicator ofpotential predictability. Predictability of interannual climate variations is found to be low outside the deep tropics.Useful predictability of precipitationover land is found consistently only over Indonesia. There is some extratropical skill in simulating surface temperatures, particularlyover North America, but not all modelsshow potential predictability. Predictability appears strongest in June, probably due to the influence ofthe initial conditions. With only SST as a driving boundary condition, the poor performance of these modelsduring summer,may indicate that we must turn to the land surface in order to harvest potential predictability. -1-