Predictability and Error Growth Dynamics in Cloud-Resolving Models

Abstract

Abstract While the benefits of ensemble techniques over deterministic numerical weather predictions (NWP) are now widely recognized, the prospects of ensemble prediction systems (EPS) at high computational resolution are still largely unclear. Difficulties arise due to the poor knowledge of the mechanisms promoting rapid perturbation growth and propagation, as well as the role of nonlinearities. In this study, the dynamics associated with the growth and propagation of initial uncertainties is investigated by means of real-case high-resolution (cloud resolving) NWP integrations. The considered case is taken from the Mesoscale Alpine Programme intensive observing period 3 (MAP IOP3) and involves convection of intermediate intensity. To assess the underlying mechanisms and the degree of linearity upon the predictability of the flow, vastly different initial perturbation methodologies are compared, while all simulations use identical lateral boundary conditions to mimic a perfectly predictable synoptic-scale flow. Comparison of the perturbation methodologies indicates that the ensuing patterns of ensemble spread converge within 11 h, irrespective of the initial perturbations employed. All methodologies pinpoint the same meso-beta-scale regions of the flow as suffering from predictability limitations. This result reveals the important role of nonlinearities. Analysis also shows that hot spots of error growth can quickly (1–2 h after initialization) develop far away from the initial perturbations. This rapid radiation of the initial uncertainties throughout the computational domain is due to both sound and gravity waves, followed by the triggering and/or growth of perturbations over regions of convective instability. The growth of the uncertainties is then limited by saturation effects, which in turn are controlled by the larger-scale atmospheric environment. From a practical point of view, it is suggested that the combined effects of rapid propagation, sizeable amplification, and inherent nonlinearities may pose severe difficulties for the design of EPS or data assimilation techniques related to high-resolution quantitative precipitation forecasting.