Sensitivity of Local Predictions to Initial Conditions

Abstract

A hydrostatic, anelastic mesoscale numerical model is used to study the predictability of a sea breeze type of circulation over La Plata River in South America. The experiments are designed to test the sensitivity of the predictions to uncertainties in the initial wind field. They consist in the comparison of control and perturbed forecasts which differ only in the initial specification of the wind held. In the case of small-domain prediction systematic uncertainties are probably more representative of actual initial uncertainties than are random, uncorrelated uncertainities. In order to study such cases we systematically add to or subtract from the control run at every grid point a constant value equal to the typical error in the measurement of the wind. We also perform other experiments on a larger domain covering most of South America. The results show that the predictions of the sea breeze type of circulation in the boundary layer around the La Plata River are sensitive to errors in the initial wind held. These results hold both for random uncertainties as well as systematic uncertainties, and lead to pessimistic conclusions about the deterministic predictability of this type of weak low-level flow. This conclusion reverses, however, in those cases where the local observation accuracy exceeds current observation accuracies by a factor of approxirnately 3. A more optimistic conclusion also holds in the vicinity of a very rough lower boundary, such as the Andes Mountains.