Predictability in limited area and global models

Selected small domain LAM forecasts modulated by highly corrugated underlying topography, and driven by different state-of-science outer models suggest that uncertain outer model guidance for LAMs produces large, domain averaged sensitivity. A further literature survey indicates that many LAM forecasts are relatively insensitive to details of the local initial state, and that mesoscales show slight error growth, in contradiction to classical predictability theory. A series of global predictability experiments is presented in order to reconcile the contradiction. The experiments imply that, even in baroclinically unstable atmospheres, the most common sources of local error growth are associated with small uncertainties of the larger spatial scales rather than small uncertainties of the smaller spatial scales. Variable resolution, real-data experiments of barotropic versions of the global model display substantial mesoscale error growth, due principally to the effect of larger scales. The uncertainties processing largest spatial scale appears as boundary uncertainties in LAMs, and explain the strong boundary sensitivity and weak local initial data sensitivity observed in many LAMs. We infer that accurate depiction of the largest spatial scales is a first order priority for accurate local prediction, and that for the advective portion of the dynamics, errors of the outer model that provides lateral boundary conditions may impose the largest current practical limitation for many LAM predictions.