Reproducible and Irreproducible Components in Ensemble Simulations with a Regional Climate Model

Abstract

High-resolution limited-area models (LAMs) have been widely employed to downscale coarse-resolution climate simulations or objective analyses. The growing evidence that LAM climate statistics can be sensitive to initial conditions suggests that a deterministic verification of LAM solutions in terms of finescale atmospheric features might be misguided. In this study a 20-member ensemble of LAM integrations with perturbed initial conditions, driven by NCEP?NCAR reanalyses, is conducted for a summer season over a midlatitude domain. Ensemble simulations allow for the separation of the downscaled information in two parts: a unique, reproducible component associated with lateral-boundary and surface forcing, and an irreproducible component associated with internal variability. The partition in the reproducible and irreproducible components and their seasonal statistics is examined as a function of horizontal length scale, geographical position within the domain, height, and weather episodes during the season. The scale analysis of time-dependent model variables shows that, at scales smaller than a few hundred kilometers, the irreproducible component dominates, on average, the model solution, implying that the downscaled information at these scales is mainly in stochastic form. The constraint exerted by the surface forcing on the internal variability is weak. For seasonal averages, the reproducible component dominates at all scales, although for precipitation the reproducible and irreproducible components are of the same order of magnitude at scales smaller than 150 km. These results imply a need for a probabilistic approach to LAM climate simulations and their verification, especially for shorter integration times, from months to seasons.