Internal and forced climate variability during the last millennium: a model-data comparison using ensemble simulations

Abstract

A three-dimensional climate model was used to perform 25 simulations over the last millennium, which are driven by the main natural and anthropogenic forcing. The results are compared to available reconstructions in order to evaluate the relative contribution of internal and forced variability during this period. At hemispheric and nearly hemispheric scale, the impact of the forcing is clear in all the simulations and knowing the forced response provides already a large amount of information about the behaviour of the climate system. Besides, at regional and local scales, the forcing has only a weak contribution to the simulated variability compared to internal variability. This result could be used to refine our conception of Medieval Warm Period and Little Ice Age (MWP and LIA). They were hemispheric-scale phenomena, since the temperature averaged over the Northern Hemisphere was, respectively generally higher/lower during those periods because of a stronger/weaker external forcing at that time. Nevertheless, at local-scale, the sign of the internal temperature variations determines to what extent the forced response will be actually visible or even masked by internal noise. Because of this role of internal variability, synchronous peak temperatures during the MWP or LIA between different locations are unlikely.

Introduction

Reconstructions of surface temperature averaged over the Northern Hemisphere display relatively warm conditions at the beginning of the 2nd millennium AD, albeit cooler than in the late 20th century (e.g., Jones et al. 1998; Mann et al., 1999; Crowley and Lowery, 2000; Briffa et al., 2001; Briffa and Osborn, 2002; Esper et al., 2002). The amplitude of the variations of this Medieval Warm Period (MWP) as well as the timing of the temperature maximum differs in these various reconstructions but they agree that warm conditions ended before 1300 AD (e.g.; Bradley, 2000; Crowley and Lowery 2000; Grove 2001). The relatively cool conditions were followed by a slow temperature recovery at the end of the 14th century, after which a gradual cooling set in, leading to particularly cold conditions during the 17th and the early 19th centuries, which are the coldest period of the Little-Ice Age (LIA). This long-term cooling trend was interrupted by relatively short, warm periods, as for instance in the middle of the 18th century. The cold period ended in the 19th century and was followed by a warming in the 20th century which leveled off during the 1940s. The latest, relatively fast, increase of Northern Hemispheric temperature since the 1950s can be mainly attributed to anthropogenic greenhouse warming (e.g., Hegerl and North, 1997; Crowley, 2000; Bertrand et al., 2002; Stott et al., 2000; Meehl et al., 2002).

Despite historical accounts and multi-proxy evidence, there is presently no accepted definition of the MWP and the LIA. Because of that, some people suggest avoiding the use of these terms. Although on hemispheric scales all temperature reconstructions exhibit a general cooling tendency from the beginning of the 2nd millennium AD until the 19th century, it is still unclear whether the MWP and LIA were truly global phenomena or whether current temperature reconstructions reflect more regional conditions. Analysing a large number of records, Hughes and Diaz (1994) argued that, in some areas of the world, the temperatures were relatively high during the MWP, in particular in summer. However, in other regions such as the Southeastern United States or Southern Europe along the Mediterrranean Sea, the climate during that time was not different from the periods that preceded or followed. Furthermore, the timing of the various warm episodes was not synchronous between different regions, as also noticed by Crowley and Lowery (2000) and Bradley et al. (2003). The cooling during the LIA appears more coherent over the mid-latitudes of the Northern Hemisphere. However, not all regional records show a consistent and synchronous cooling (e.g. Bradley and Jones(Eds.), 1992, Bradley and Jones, 1993; Jones et al., 1998, Jones et al., 2001; Bradley, 2000; Mann et al., 2000).

Climate model simulations could be used to study the characteristics of the MWP and LIA as well as to gain insight into the mechanisms that caused these climate variations. When driven by estimates of solar and volcanic forcing, physical and statistical models reproduce quite well the low frequency evolution of the northern hemisphere surface temperature as deduced from proxy records over the period AD 1000–1850 (e.g., Crowley, 2000; Bertrand et al., 2002; Bauer et al., 2003; Gerber et al., 2003). This suggests that external forcing plays a key role as pacemaker of low-frequency variations. In order to simulate the observed warming of the last century, it has been necessary to include also anthropogenic forcings (e.g., Crowley, 2000; Stott et al., 2000; Bertrand et al., 2002; Meehl et al., 2002).

Climate models used for an assessment of externally forced millennial-scale variability have very low resolution and the majority of them are two-dimensional (i.e. latitude and altitude/depth are varying). As a consequence, their results could not be used to gain information on a regional basis, not even at a continental-scale. Potentially, such information could be obtained from the comprehensive, three-dimensional general circulation models (GCMs) but, because of the high computer-time requirements, only a few transient simulations spanning the last 500 years have been published up to now (e.g., Cubasch et al., 1997; Rind et al., 1999; Waple et al., 2002; Gonzalez-Rouco et al., 2003; Widmann and Tett, 2003). GCM simulations were used to compare simulated response patterns to external forcing with low-frequency patterns deduced from observations. Three-dimensional general circulation models were also used to study the physical processes which amplify the climate response to strong external forcing, for instance during the minimum in solar irradiance of the late 17th century (Shindell et al., 2001). In addition, unforced GCM simulations were used in order to compare patterns of internal low-frequency variability with those obtained from reconstructions (Jones et al., 1998; Collins et al., 2002).

To overcome the computing time constraints of GCMs, it is possible to use low resolution three-dimensional models including a simplified, but reasonable, representation of the most important physical processes. Such an approach has been taken by van der Schrier et al. (2002) who analysed sea level variations during the last millennium. As in the majority of studies using GCMs on time-scale longer than 150 years, van der Schrier et al. (2002) performed only one simulation for each experimental design. This may be enough to look at large-scale, low-frequency variations, to make process studies or to look at the patterns associated with a particular forcing. However, such an approach is problematic in the sense that the observed trajectory of the climate system is just one realization within an ensemble of possible (partly externally forced) trajectories. The same holds for a single model simulation. In both cases internally generated internal variability could mask the forced signal.

As a consequence, in the present study, our goal is to estimate the contributions of forced and internal variability in the Northern Hemispheric climate evolution during the last millennium. To do so, an ensemble of 25 simulations is performed with a three-dimensional model of intermediate complexity. The model includes an improved version of the atmospheric model used by van der Schrier et al. (2002) coupled to a coarse-resolution ocean-sea-ice general circulation model. First, the results of the ensemble simulation are compared to observations to test the skill of the model. In a second step, the model results are used to help in the interpretation of the reconstructed temperature time series. We address the question as to whether the MWP and the LIA are robust features which were forced by solar and volcanic activity or whether they are representations of internal climate noise. To structure our analysis, we discuss the comparisons of simulated and observed temperature time-series on three different spatial scales (hemispheric, continental and regional) and separately consider the evolution of spatial patterns. We only use reconstructed temperature time-series with high temporal resolution because a reasonable number of these types of reconstruction are available now, for various regions of the world, and because the comparison with model variables is relatively straightforward. Our analysis will cover the period 1000–1980 AD as most proxy data are not available for the last 20 years.

Another interesting perspective on the issue of externally generated climate variability is that of predictability. Lorenz (1975) introduced two kinds of climate predictability. Predictability of the first kind describes the loss of information during a forecast due to initial condition uncertainties. Predictability of the second kind on the other hand is associated with the influence of non-stationary boundary conditions on the system's evolution. Climate predictability of the first kind has been investigated by Griffies and Bryan (1997) and Grötzner et al. (1999) using coupled GCMs and ensemble simulations. Their analysis focused on the potential predictability on decadal and interdecadal timescales. The idea of these ensemble studies was to explore the possibility that simulated preferred interdecadal climate modes associated with variations of the strength of the thermohaline circulation (Delworth et al., 1993; Timmermann et al., 1998) are predictable up to a decade or so in advance, given well-defined oceanic initial conditions. For non-stationary boundary conditions (solar, orbital, volcanic forcing) this problem turns into a predictability problem of the first and second kind. The goal of our study is to explore this type of “mixed” predictability for the climate of the last millennium.

Our paper is organized as follows: In Section 2, a description of the model and of the experimental design of our simulations is given as well as a brief description of the techniques used and the limitations of the study. Section 3 analyses the results of the simulation. We first study the large-scale evolution of annual mean temperatures during the last millennium. Then, we discuss the regional climate changes driven by the external forcing and finally the evolution of large-scale patterns. Section 4 includes a discussion of the results and Section 5 summarizes our main findings and puts them into a broader perspective.