EVALUATION OF TROPOSPHERIC WATER-VAPOR SIMULATIONS FROM THE ATMOSPHERIC MODEL INTERCOMPARISON PROJECT

Simulations of humidity from 28 general circulation models for the per iod 1979-88 from the Atmospheric Model Intercomparison Project are com pared with observations from radiosondes over North America and the gl obe and with satellite microwave observations over the Pacific basin. The simulations of decadal mean values of precipitable water (W) integ rated over each of these regions tend to be less moist than the real a tmosphere in all three cases; the median model values are approximatel y 5% less than the observed values. The spread among the simulations i s larger over regions of high terrain, which suggests that differences in methods of resolving topographic features are important. The mean elevation of the North American continent is substantially higher in t he models than is observed, which may contribute to the overall dry bi as of the models over that area. The authors do not find a clear assoc iation between the mean topography of a model and its mean W simulatio n, however, which suggests that the bias over land is not purely a mat ter of orography. The seasonal cycle of W is reasonably well simulated by the models, although over North America they have a tendency to be come moister more quickly in the spring than is observed. The interann ual component of the variability of W is nor well captured by the mode ls over North America. Globally, the simulated W values show a signal correlated with the Southern Oscillation index but the observations do not. This discrepancy may be related to deficiencies in the radiosond e network, which does not sample the tropical ocean regions well. Over all, the interannual variability of W as well as its climatology and m ean seasonal cycle, are better described by the median of the 28 simul ations than by individual members of the ensemble. Tests to learn whet her simulated precipitable water, evaporation, and precipitation value s may be related to aspects of model formulation yield few clear signa ls, although the authors find, for example, a tendency for the few mod els that predict boundary layer depth to have large values of evaporat ion and precipitation. Controlled experiments, in which aspects of mod el architecture are systematically varied within individual models, ma y be necessary to elucidate whether and how model characteristics infl uence simulations.