Conceptualising and quantifying the nonlinear, chaotic climate: implications for climate model experimental design

Abstract

Uncertainty in climate system initial conditions (ICs) is known to limit the predictability of future atmospheric states. On weather time scales (i.e. hours to days), the separation between two atmospheric model trajectories, initially "indistinguishable" (compared to unavoidable uncertainties) from one another, diverges exponentially-on-average over time, so that the "memory" of model ICs is eventually lost. In other words, there is a theoretical limit in the lead time for skilful weather forecasts. However, the influence of perturbations to climate system model ICs - particularly in more slowly evolving climate system components (e.g., the oceans and ice sheets) - on the evolution of model "climates" on longer time scales is less well understood. Hence, in order to better understand the role of IC uncertainty in climate predictability, particularly in the context of climate change, it is necessary to develop approaches for investigating and quantifying - at various spatial and temporal scales - the nature of the influence of ICs on the evolution of climate system trajectories. To this end, this study explores different conceptualisations and competing definitions of climate and the climate system, focussing on the role of ICs. The influence of ICs on climate quantifications, using probability distributions, is subsequently investigated in a climate model experiments using a low-resolution version of the Community Climate System Model version 4 (CCSM4). The model experiment consists of 11 different 50-member ensemble simulations with constant forcing, and three 50-member ensemble simulations under a climate change scenario with transient forcing. By analysing the output at global and regional scales, at least three distinct levels of IC influence are detected: (a) microscopic influence; (b) interannual-scale influence; and (c) intercentennial-scale influence. Distinct patterns of interannual-scale IC influence appear to be attributable to aperiodic and quasi-periodic variability in the model. It is found that, over some spatial domains, significant (p < 0.01) differences in atmospheric variable "climatologies", taken from 60-year distributions of model trajectories, occur due to IC differences of a similar order to round-off error. In addition, climate distributions constructed using different approaches are found to differ significantly. There is some evidence that ensemble distributions of multidecadal temperature response to transient forcing conditions can be influenced by ICs. The implications for quantifying and conceptualising climate are considered in the context of the experimental results. It is concluded that IC ensemble experiments can play a valuable role in better understanding climate variability and change, as well as allowing for superior quantification of model climates.