An Initial State Perturbation Experiment with the GISS Model

Abstract

Monthly mean global forecasts for January 1975 have been computed with the GISS model from four slightly different sets of initial conditions?a ?control? state and three random perturbations thereof?to simulate the effects of initial state uncertainty on forecast quality. Differences among the forecasts are examined in terms of energetics, synoptic patterns and forecast statistics. The ?noise level? of the model predictions is depicted on global maps of standard deviations of sea level pressures, 500 mb heights and 850 mb temperatures for the set of four forecasts. Initial small-scale random errors do not appear to result in any major degradation of the large-scale monthly mean forecast beyond that generated by the model itself, nor do they appear to represent the major source of large-scale forecast error.