Skill improvement from increased ensemble size and model diversity

Abstract

This paper proposes an objective procedure for deciding if the skill of a combination of forecasts is significantly larger than that of a single forecast, and for deciding if the observed improvement is dominated by reduction of noise associated with ensemble averaging, or by addition of new predictable signals. Information theory provides an attractive framework for addressing these questions. The procedure is applied to El Niño-Southern Oscillation hindcasts from the North American Multimodel Ensemble (NMME) and reveals that the observed skill advantage of the NMME compared to individual models is substantially greater than that expected from increased ensemble size alone and is more consistent with the addition of new signals. Key PointsProcedure for deciding if skill of multimodel combination exceeds single modelSeasonal skill improved from model diversity rather than from larger ensembleResults strongly support the use of different models in seasonal forecasting