ROUGH DRAFT authorea.com/119523
Main Data History
Export
Show Index Toggle 0 comments
  •  Quick Edit
  • Initial Conditions and Quantifying Model Climates: Does it Matter Where We Come From?

    Abstract

    There does not exist an agreed-upon procedure to quantify model climates from their output. Here, we explore differences in characterisations of regional variable climatologies arising from the application of three climate quantification approaches to the output of an initial condition ensemble climate system model experiment. We find that that ensemble member trajectories, distinguished in set-up only by the round-off order differences in initial atmospheric temperature, can—over certain regional domains and for particular initial system states—produce significantly (\(p\ll 0.01\)) different variable probability distributions. In addition, using different quantification approaches to capture what might be presumed to be the same “climatic state”—which itself may be influenced by the initial climate system state—can yield significantly different distributions. We conclude that a multivariate distribution, sampled over both time and multiple ensemble members, together with measures of autocorrelation, may serve as a useful quantification approach for model climates.

    Introduction

    Many definitions have been proposed for climate in the literature (e.g., Lorenz, 1995; Werndl, 2015; Lovejoy, 2013; IPCC, 2013; WMO, 2010). For many of the central concepts in climate science, there do not appear to be any widely accepted definitions (Todorov, 1986; Daron, 2012; Werndl, 2015). One might argue that definitions used are broadly similar, that there appears to be consensus on the intuitive idea of climate (Leith, 1985) and that the descriptions used are generally “good enough” for the specific contexts in which they are applied. However, Lorenz (1995) notes that “certain questions regarding climate may be answered either affirmatively or negatively, according to the precise [definition of climate used]”. Furthermore, Lorenz (1995) suggests that in different contexts—across which the nature of the available data varies—the definition that would lead to the most meaningful characterisation of a given climatic state, may differ (see also Schneider et al., 1974). In particular, definitions of climate which are applicable in observational studies, are not necessarily the most useful in theoretical or modelling studies (Lorenz, 1995; Schneider et al., 1974; Leith, 1978).

    Of particular interest in this work are various climate quantifications—definitions which, when applied, provide a quantitative characterisation of a “model climate” (also referred to as a “model climatic state”). In an observational context, one must consider quantifications applicable to individual climate variable trajectories sampled over time; however, in ensemble modelling studies collections of such trajectories can either be sampled at a particular “instant” or over a period of time. In this work, ensembles considered are initial condition (IC) ensembles (the value of which is argued for in, e.g., Hawkins et al. (2015)), but definitions involving perturbed physics and multi-model ensemble output could also be proposed; in particular, it could be argued that the IPCC (2013) apply a multi-model ensemble definition in quantifying projected future climates.

    Following Werndl (2015), we focus on probability distributions as a means of capturing the mean state, variability and extremes of a local, regional or global climatic state. For a particular variable, over a particular domain, three such quantifications are considered here:

    Climate Quantifications

    \label{ClimQuants}
    Temporal (TCQs):

    the distribution sampled from a series of consecutive “points in time” of a single model trajectory. In this study, annual averages are considered as “points in time”, thus avoiding complexities involved with the diurnal and seasonal cycles. Note, furthermore, that in an observational setting, rather than a model setting, only TCQs are applicable, as there is only one realisation of the planetary climate that occurs over a given period.

    Ensemble (ECQs):

    the distribution sampled at a particular “point in time”, from all members of an IC ensemble at the \(n^{\textrm{th}}\) year of the ensemble duration.

    Ensemble-temporal (ETCQs):

    the distribution sampled from all members of an IC ensemble over a series of consecutive points in time (in this case, thus, several years).

    Understanding of the differences in climatic characterisations that could be produced by different approaches to defining climate, should be explored and considered in experimental design and interpretation of results. This study aims to contribute to the discussion on preferable future approaches to climate model experimental design, following.

    Methods

    Results discussed here are obtained from the output of a large IC ensemble experiment documented in detail in Conradie (2015), ch. 3. For this experiment the Community Climate System Model, version 4 (CCSM4; Gent et al., 2011) was run in a “fully coupled” configuration. This includes the following components: atmosphere, land, ocean, sea ice and river run-off. In order to minimise computational expense, the lowest resolution (f45gx3; implying that the atmospheric component is run with a \(4\times 5^{\circ}{}{}\) finite-volume dynamic core) on which the fully coupled model can be run, is used.

    For simplicity, only four of the ensembles run for that study are considered here; their relation to one another is illustrated in Figure \ref{Figure:ExperimentalDesign}. These ensembles were all run at the South African Centre for High Performance Computing (CHPC), where a 1600-year present-day control run (PDC) was performed, to serve as a basis for ensembles to be “branched” off from. Each new ensemble uses as ICs output produced at a particular model year by a previous run, for all model components. Atmospheric temperature is perturbed to distinguish individual ensemble members from one another. Individual ensembles are named according to corresponding control run model year (i.e. the number of years for which ocean circulation has been active) at their initialisation.