Models and climate forcing

Numerous simulations devoted to the study of the past millennium climate have been performed using Energy Balance Models (EBMs; e.g. Free and Robock 1999; Crowley 2000; Hegerl et al. 2003), general circulation atmospheric models coupled to a simplified ocean model (e.g. Shindell et al. 2001; Waple et al. 2002), Earth Models of Intermediate Complexity (EMIC; e.g. Bertrand et al. 2002; Bauer et al. 2003; Gerber et al. 2003, Goosse et al. 2005a), as well as comprehensive

Figure 7.2 Typical history of past radiative changes, as used in Goosse et al. (2005a). The evolution of solar irradiance follows the reconstruction of Lean et al. (1995) extended back in time by Bard et al. (2000). The effect of volcanism is derived from Crowley (2000). The forcing due to change in land-use is based on Ramankutty and Foley (1999), using a linear interpolation before 1700 as in Brovkin et al. (2006). Only the direct effect of sulfate aerosols due to anthropogenic activity is taken into account here so the green curve is a lower bound for this forcing (e.g. Andreae et al. 2005). A 25-year running mean has been applied to all the time series.

Atmosphere-Ocean General Circulation Models (AOGCMs; e.g. Cubasch et al. 1997; Gonzalez-Rouco et al. 2003, 2006; Widmann and Tett 2003). Such types of model have also been used to study other periods of the Holocene (cf. Crucifix, this volume; Claussen, this volume).

To reproduce the observed changes, climate models have to be driven by reconstructions of past variations in external forcings. These forcings can be grouped into two categories: natural and anthropogenic. The main radiative perturbations of purely natural origin over the past millennium are related to changes in solar irradiance (cf. Beer and van Geel, this volume) and the release of aerosols into the atmosphere during explosive volcanic eruptions. Orbital forcing, which is dominant on longer time-scales (cf. Crucifix, this volume), plays a weaker role during the past millennium. The largest radiative perturbation induced by human activity is related to the increase in greenhouse gas concentrations in the atmosphere. The increase in the atmospheric aerosol load is associated with a significant radiative forcing, particularly close to the main industrialized regions (e.g. Boucher and Pham 2002). These two forcings are mainly restricted to the past 250 years, with a strong amplification during the past 50 years. The rate of land-use change has also increased during the latter period, although in some regions its impact is significant throughout the past millennium (Ramankutty and Foley 1999) and earlier (cf. Oldfield, this volume).

Figure 7.2 shows typical reconstructions of external forcings during the past millennium. The uncertainties in these forcings are relatively large, except for the one associated with the increase in greenhouse gas concentrations. The majority of the forcings have to be reconstructed from indirect sources such as the amount of sulfates in ice cores for volcanic activities or of cosmogenic isotopes for solar irradiance (cf. Beer and van Geel, this volume). In addition, the magnitude of the negative aerosol forcing depends on numerous complex phenomena and the estimates of the present-day aerosol forcing range from nearly zero to more than 2 W m-2 (e.g. Andreae et al. 2005). This illustrates that the time evolution of the past radiative forcing, as well as its magnitude, is not well constrained.

Some model simulations include only a small number of the forcings mentioned above (generally one), the goal being to understand precisely the mechanism behind the response to a particular forcing (e.g. Shindell et al. 2001, 2004; Waple

Figure 7.3 Annual mean temperature anomaly averaged over the Northern Hemisphere in simulations performed with two AOGCMs (CCSM, Ammann et al., cited in Jones and Mann (2004) in green, and ECHO-G, Gonzalez-Rouco et al. 2003, 2006) and in one three-dimensional Earth model of intermediate complexity (ECBILT-CLIO-VECODE, Goosse et al. (2005a) in blue). The two simulations performed in ECHO-G (ERIK and ERIK2, dashed red and red respectively) differ only in their initial conditions, ERIK using one that is probably too warm. The reconstruction of Jones and Mann (2004) is in magenta, with the reconstruction plus and minus two standard deviations in grey. The time series are grouped in 10-year averages. The reference period is 1500-1850. This reference period was chosen to eliminate possible problems related to initial conditions and to the strong differences of specified anthropogenic forcing in the 20th century.

900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Time (years)

900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Time (years)


et al. 2002; Goosse and Renssen 2004; Mann et al. 2005a). Other simulations that try to reproduce the observed climate evolution tend to include all the important forcings. They all take into account at least the solar, volcanic, and greenhouse gas forcings, and some simulations include a more exhaustive list of forcings, accounting for the effect of land-use change, anthropogenic aerosols, and orbital forcing.

Figure 7.3 displays the annual mean temperature averaged over the Northern Hemisphere in such simulations using various three-dimensional climate models. The model results are in the range of the uncertainty associated with the reconstructions, showing generally colder temperatures during the 16—19th centuries and a larger warming during the 20th century. Nevertheless, differences between the models can be noticed, attributable to different factors. Firstly, the simulations use different reconstructions of past forcing changes. For solar radiations, despite the relatively large range in the reconstructions, recent model simulations tend to use similar reconstructions. The selection of the reconstruction of past solar irradi-ance has thus a weak impact on the differences shown in Figure 7.3 (Goosse et al. 2005b; Osborn et al. 2006). Models that do not include the anthropogenic aerosol forcing (such as the simulations using the ECHO-G model) simulate a much larger warming during the 20th century than the models that take into account this important forcing.

Secondly, the choice of initial conditions can play a role. This is illustrated in Figure 7.3 by the two simulations performed with the ECHO-G model that use the same forcing but in one case start from warm conditions (simulation ERIK) in year 1000 ad and in the other case from colder conditions (simulation ERIK2). In

ERIK, because of the selected warm initial state, the temperature drifts during several centuries towards colder temperatures until it becomes close to ERIK2, and the other simulations, around 1500. It is not possible to define precisely from observations the conditions that should be used to start a model in 1000 ad, but longer simulations starting in 850 ad or covering the past two millennia do not show warm conditions similar to the ones used in ERIK. The initial conditions used in ERIK2 and the subsequent evolution of temperature appear more reasonable than the ones of ERIK, although the lack of tropospheric aerosols in ERIK2 is still problematic for the realism of the simulation during the 20th century or the evaluation of the net changes in temperature from the pre-industrial to modern intervals.

Finally, different models use different grids and methods for the numerical resolution of the equations as well as different representations of some physical processes governing the evolution of the climate system, with a potential impact on the response of the model to a radiative perturbation. The behavior of models is often summarized by a few important characteristics, but this is a strong simplification of the complex behavior of three-dimensional models. Model sensitivity is classically defined by the change in global mean temperature between a controlled experiment using present-day conditions and an equilibrium experiment in which the atmospheric CO2 concentration is doubled. The efficiency of the heat uptake by the ocean also plays an important role in the model response. In simple models, this could be related to bulk parameters such as an effective heat diffusion in the ocean (e.g. Hegerl et al. 2006; Osborn et al. 2006), whereas in more sophisticated models, the link with model parameters is much more complex. In Figure 7.3, such differences in model characteristics are likely responsible for the weaker temperature changes in ECBILT-CLIO-VECODE than in ECHO-G and CCSM, with the former model having relatively weak climate sensitivity.

At the hemispheric scale, precise model-data comparisons are difficult because of the uncertainties associated with the reconstructions, the forcing, and model formulation. As a consequence, an agreement between model and data is usually not a stringent test of the model results. At the regional scale, an additional difficulty is related to the large role played by the internal variability of the climate system, whereas at the hemispheric scale, the evolution of the system is largely imposed by the changes in the external forcing. As discussed later, the characteristics of some well-known modes of climate variability such as ENSO or NAO could be influenced by the forcing. Nevertheless, a large fraction of the variability of these modes is purely chaotic, related to internal dynamics, and cannot be related to any change in external conditions.

At regional scales, the internal variability of the system could easily mask the forced response. This is probably the main reason why the so-called Medieval Warm Period (covering roughly the period ad 900-1200) is not a synchronous phenomenon in all regions of the globe (e.g. Hughes and Diaz 1994; Bradley et al. 2003; Goosse et al. 2005a). The net effect of all forcings probably tended to induce warmer conditions during the period, but, depending on the sign of the anomalies

Figure 7.4 Anomaly of summer mean temperature in Fennoscandia averaged over 25 simulations that differ only in their initial conditions (black). The mean over the ensemble of 25 simulations plus and minus two standard deviations of the ensemble are in grey. The reconstruction of Briffa et al. (1992) is in green. As the reconstruction is within the range of the various simulations performed with the model, we can consider that there is no disagreement between model results and the reconstruction. The reference period is 1500-1850. The time series are grouped in 10-year averages.

associated with internal variability, the maximum temperatures occurred at different locations at different times.

Even if perfect models and perfect forcing time-series were available, each model experiment would simulate a possible evolution of the internal variability but not necessarily the one followed in the real world. Very precise initial conditions would only allow having an agreement between model and observations during a few weeks at most, a classic problem in weather forecasting (Lorenz 1963). In complex models (such as AOGCMs and some EMICs) that include a representation of the internal variability, it is possible to compare the statistics of one simulation with the observations or to try to extract the response to a forcing in the simulation and in available reconstructions. A precise comparison between temperature changes simulated in a model with a reconstruction in a particular region, for a particular year (or decade), is usually useless because it is generally impossible to state if any difference is due to uncertainties in the reconstruction, model, or forcing deficiencies, or simply to a different realization associated with the internal variability of the system.

In this framework, ensembles of simulations are particularly useful. To perform such an ensemble, slightly different initial conditions are selected. After a few simulated days, the internal variability evolves in a totally different way in the various members. If the ensemble is large enough, a reasonable sampling of the model variability could be obtained. In order to have an agreement between model and a reconstruction, the reconstruction must correspond to a reasonable member of the ensemble (Figure 7.4). As the members of the ensemble provide independent samples of the internal variability, averaging over all the members of the ensemble tends to filter out the internal variability, leaving only the forced response of the system. This provides a clear advantage compared with individual simulations of complex models that always provide a mixture between the forced response and the internal variability.

1100 1200 1300 1400 1500 1600 1700 1800 1900 Time (years)

1100 1200 1300 1400 1500 1600 1700 1800 1900 Time (years)

Was this article helpful?

0 0

Post a comment