GROWTH AND DECAY OF ERROR IN A NUMERICAL CLOUD MODEL DUE TO SMALL INITIAL PERTURBATIONS AND PARAMETER CHANGES

The behavior of a numerical cloud model is investigated in terms of it s sensitivity to perturbations with two kinds of lateral boundary cond itions: 1) with cyclic lateral boundary conditions, the model is sensi tive to many aspects of its structure, including a very small potentia l temperature perturbation at only one grid point, changes in time ste p, and small changes in parameters such as the autoconversion rate fro m cloud water to rainwater and the latent heat of vaporization; 2) wit h prescribed lateral boundary conditions, growth and decay of perturba tions are highly dependent on the flow conditions inside the domain. I t is shown that under relatively uniform (unidirectional) advection ac ross the domain, the perturbations will decay. On the other hand, conv ergence, divergence, or, in general, flow patterns with changing direc tions support error growth. This study shows that it is the flow struc ture inside the model domain that is important in determining whether the prescribed lateral boundary conditions will result in decaying or growing perturbations. The numerical model is inherently sensitive to initial perturbations, but errors can decay due to advection of inform ation from lateral boundaries across the domain by uniform flow. This result provides one explanation to the reported results in earlier stu dies showing both error growth and decay.