Statistical Significance of Small Ensembles of Simulations and Detection of the Internal Climate Variability: An Excitable Ocean System Case Study

Abstract

The correct mathematical approach to climate change requires the knowledge of the time-dependent system’s pullback/snapshot attractor (PBA). Once the governing equations and external forcing are known, the PBA can be estimated by performing an ensemble simulation (ES) of many forward time integrations differing only by their respective initialization; the resulting ensemble mean and spread are usually considered as representative of the forced and internal variability (FV and IV), respectively. In this paper the PBA of an excitable conceptual ocean model subjected to an idealized decadal time-scale aperiodic forcing is determined and is then used to show that the system’s relaxation oscillations contribute substantially to the ensemble mean, despite their intrinsic nature: as a consequence, a clear separation between the FV and IV is impossible in this case study. This provides an example of dynamical behaviour which may be typical of climate ESs under fluctuating aperiodic forcing. The impact of the number of ensemble members N on the statistical significance of the ES is then investigated. The complexity of realistic climate modelling currently imposes N = O(100): how significant is the statistical information derived from such small ESs? To answer this question for the present case study, the knowledge of the PBA is exploited to carry out a systematic comparison between the latter and small ESs with N = 50, also by using novel quantifiers specifically conceived for this purpose. The results reveal a remarkable significance of such ESs beyond the predictability time and may provide useful information for the design of future realistic ESs.