Supplemental Material for PNAS 2006

For full supplemental material, please see PNAS Web Site. Overview:

1 Model

The National Center for Atmospheric Research Climate System Model (NCAR-CSM) Version 1.4 is an updated version of the model described in a Special Issue of Journal of Climate (issue June 1998). It is a global coupled atmosphere-ocean-sea ice-land surface model without flux adjustments (Boville and Gent, 2001; Otto-Bliesner and Brady, 2001). The atmospheric model is CCM3, a spectral model with 18 levels in the vertical (Kiehl et al., 1998). For the experiments presented here, it was run at T31 resolution (an equivalent grid spacing of 3.75 by 3.75 degrees). The land surface model had specified vegetation types and a comprehensive treatment of surface processes (Bonan, 1998). The land model used the same grid spacing as the atmosphere model. Freshwater balance was maintained with the precipitation-scaling scheme described in Boville and Gent (1998). The ocean component was the NCAR CSM Ocean Model (NCOM) with 25 levels, 3.6° longitudinal grid spacing, and latitudinal spacing of 1.8° poleward of 30° smoothly decreasing to 0.9° within 10° of the equator (Gent et al., 1998). The sea ice model (Weatherly et al., 1998) included ice thermodynamics based on the three-layer model of Semtner (1976) and ice dynamics based on the cavitating fluid rheology of (Flato and Hibler, 1995). The ice grid spacing was the same as that of the ocean model. Further information can be found at the NCAR-CCSM website: the CSM 1.4 Coupled Climate Model can be found here.

2 Experiments

2.1 Spin-up and branch procedures

The experiments presented in Ammann et al. (2006) were run in transient, fully coupled mode for the climate of 850-2000 AD. A full-length control experiment was run in parallel to the long forced simulations. The spin-up procedure for the model was done in two major steps:
First, branching off a present-day control experiment performed by the CSM team (Boville and Gent, 1996), an initial control was run using pre-industrial atmospheric composition and solar irradiance (we used Lean et al., 1995). In this step, estimates of pre-industrial climatological SSTs and sea ice (output from a CSM mid-Holocene simulation) were set as monthly boundary conditions while the atmosphere and land model were run interactively for 20 years. The atmospheric and land model output of last 5 years of this simulation were then used in a loop over a nominal 20-year surface/1000-year accelerated deep ocean (using a vertical acceleration profile reaching a factor of 50 in the deep-ocean) simulation with only the ocean and sea ice models running interactively. The ocean and sea ice fields at the end of this simulation were then used as initial conditions of a subsequent non-accelerated fully coupled integration with pre-industrial atmospheric composition for 200 years.
Second, the atmospheric composition used the pre-industrial control (~1870 AD) was then modified to 850 AD conditions and run forward in un-accelerated coupled configuration using the restart of the end of the 200-year coupled experiment of step 1. The simulation was monitored over 100 years. No significant drift was recognized over this interval, which is in agreement with a small radiative forcing (a few tenths of a W/m2) resulting from the change from 1870 AD to 850 AD atmospheric composition and solar irradiance.
All long experiments were started using the year 50 restart conditions of this control from step 2. Both long integrations (medium and large scaled solar experiments) and the control integration were then run in parallel for 1150 years in transient, un-accelerated fully coupled mode.
After about 200 years, the control experiment started to exhibit a small drift that carried on for several centuries. After about ~700 years, the global trends were reduced to essentially zero. This drift reflects a consistent problem for long transient climate simulations of coupled climate models because the deep ocean climate suffers from an extremely long equilibration time scale. Similar problems have been found in other experiments (Zorita et al., 2005; Kiehl and Shields, 2005). Because the persistent drifts can only be recognized after many centuries, we did not have the computational resources to restart the transient forced simulations but applied a grid-point based correction to the model output. The control climate time series, for each month of the year, at each grid-point separately, was fit with a very smooth (1000-year low-pass) cubic spline. This smooth trend was then removed from the 1150-year long transient control and forced simulations (medium and large solar), for each month of the year, and at each grid point. Such a drift correction is clearly not ideal, but it is a necessary step.
The shorter low-scaled solar simulation (1575-2000 AD) was initialized using the end of the control (model year 1200) conditions and did therefore not have to be detrended.

2.2 Radiative forcing series

The radiative forcing series shown in Figure 1 can be accessed here

Solar forcing series were taken from Bard et al. (2000), who offer directly scaled series to irradiance background changes of 0.25% and 0.65% back to the Maunder Minimum (1645-1715 AD). The magnitudes are slightly larger than those in Crowley (2000), who scaled the absolute minimum solar irradiance to 0.25%. In the Bard et al. series, the minimum does not occur during the Maunder Minimum, but rather during the previous Spoerer Minimum (~1415-1540 AD).
Volcanic forcing is based on a compilation of individual ice cores readily available from (http://www.ncdc.noaa.gov/paleo/icecore.html). Timing and hemispheric sulfate flux were estimated and scaled to the 1815 AD eruption of Tambora. The global radiative forcing was calculated using a conversion factor of 21 between the radiative flux changes and annual mean visible optical depth of the volcanic aerosol (F=21*tau_vis), recently suggested by Hansen et al. (2002). The implementation of the volcanic forcing in the atmospheric component of CSM is described in Ammann et al. (2003).
Natural sulfate aerosol was included using an annual cycle for all runs prior to 1870 AD. Subsequently, only the "natural only" experiments continued with the same natural background aerosol, while the ‘full forcing’ experiments were driven with monthly model-simulated fields of natural plus anthropogenic aerosol based on surface emissions of various sulfur containing gases.
Pre-20th century greenhouse gas concentrations (CO2, CH4, N2O) were specified using available ice core data. CFC-11 and CFC-12 were added during the 20th century following SRES guidelines.

Not included are slow forcings from changes in orbital configuration (Milankovich parameters) and landuse changes. Milankovich forcing is globally small, but potentially of regional importance. It might affect the millennium scale trend (see also Bertrand et al, 2002). Because of problems regarding the deep-ocean drift, this issue is not the primary focus of our simulations and thus omission of the orbital forcing will have no effects on the conclusions. Landuse changes are not included because no reliable dataset exists. Given the present knowledge of landuse influence (see Bonan, 1997; Feddema et al., 2005), it can be of local and regional importance. On a global basis, however, the overall perturbation is probably small, and temporally slow during pre-industrial times. Future experiments would nevertheless benefit if these forcings could be included. The same is true for the indirect aerosol effect as well as absorbing aerosols.

3 Solar Irradiance as a Climate Pace Maker

Solar irradiance as taken from Bard et al. (2000) was scaled using three different scalings: high (equivalent to a Maunder Minimum to present background trend of 0.65%), medium (0.25%) and small (no background trend). The impact of the different scaling factors on the solar ‘pacing’ of climate is shown in Figure SUP-1. As the magnitude of the solar irradiance range (and thus the forcing itself) increases, the of a climate response develops.
The large solar forcing leads to a very strong and direct influence of the global surface temperature (Panel C in Figure SUP-1). The global temperature tracks the time-evolution of the solar irradiance (black dashed line with separate y-axis, see magnitude at the bottom of Figure SUP-1) very closely. The vertical gray bars in Figure SUP-1 indicate periods of increased volcanism. Hardly any of these volcanic periods cause a significant departure of the surface temperature away from the superposed track of the solar irradiance series.
Medium scaled solar irradiance shows generally a similar picture as the large scaling, but because of the reduced amplitude of the forcing, the model simulation experiences more variations. Some departures from a close tracking to the solar irradiance (black dashed line) are clearly caused by volcanic forcing. During these periods, surface temperatures drop in accordance with the negative forcing from the volcanic aerosol. However, not all deviations are due to volcanism, because the overall trend in the solar forcing is much smaller than in the large scaled experiment, the internal variability of the coupled climate model has also more impact on the temperature evolution. While external forcing still drives the general multi-decadal structure, decadal variations are getting more significant in this experiment. This is confirmed in a parallel ensemble simulation over 700 years (not shown), which exhibits the same multi-decadal structure, but departs from its parallel case at interannual (outside of volcanic cooling) and decadal structure.
Logically, the low scaled solar forcing, representing absence of a significant trend in solar irradiance from the present to the Maunder Minimum, is the least driven by solar irradiance. Volcanic forcing seems very dominant in guiding its multi-decadal surface temperature structure. The temporal evolution over the available interval is overall quite similar to the medium scaled solar run, although the mean temperature is offset towards a warmer mean during the pre-industrial period, particularly before the large eruptions of the early 19th century. Subsequently, the experiment is very similar in response to the increase in anthropogenic influence.
There is good agreement between surface temperatures in the low solar run with both the large and medium scaled experiment from the late 19th century onward. Given internal variability in a coupled model – no clear distinction can be found between the three simulations. This observation underlines our conclusion that, independent of the magnitude of pre-industrial cooling due to a potential change in the background level of solar irradiance, the 20th century warming is mostly driven by the anthropogenic change in atmospheric composition. Thus, the recovery out of a cool Little Ice Age cannot explain the warming in the late 20th century because both a large and a very limited cool mean state of pre-industrial climate exhibit the same warming during the 20th century. Figure SUP-1 (panels A-C) also show that a purely natural extension of the transient experiment (holding anthropogenic forcings fixed at 1870 AD conditions) would extend the relatively close relationships that the global surface temperature had exhibited to solar irradiance over 1000 years. However, in order to reproduce the warming seen in instrumental data (dark gray lines in Figure SUP-1, observational data from Jones and Moberg, 2003), the anthropogenic forcing has to be included. This was also shown in more detail in Figure 3c in Ammann et al. (2006).
The change in solar ‘guidance’ of global surface temperature during pre-industrial time (850-1874 AD) between a large and medium scaled solar irradiance forcing is also shown in Figure SUP-2. The global surface temperature was decomposed using a non-decimated discrete wavelet technique (Oh et al., 2003). The decomposition level of surface temperature representing a roughly 200-year band of the long CSM experiments is compared with the same level of decomposition applied to the solar irradiance series. The solar series (black line) was taken from the medium scaled irradiance series, thus the actual scale of the y-axis on the right hand side is valid for the comparison with the medium scaled experiment (red series), although its temporal structure is identical for the large solar comparison. As can be expected, there is no consistent correspondence between the solar irradiance series and the control run evolution, where solar forcing remained constant (light-blue line). A very strong connection can be assumed between the solar series and the large solar experiment (dark blue line). Both the phasing and the relative amplitudes of individual intervals are very similar. The medium-scaled experiment generally shows a good correspondence to the solar forcing, but both amplitude and phase are less well reflected. As the magnitude of the solar forcing variations decreases, the relative role of other forcings, namely volcanic eruptions, is increasing. While there is still a clear indication of solar irradiance guidance on centennial temperature evolution in the medium scaled experiment (while dominating in the large solar run), the volcanic influence modified the exact timing of low-frequency variations, so that the phase is less exact. The volcanic influence is also present in the large solar experiment, causing somewhat of an early peak in temperatures around 1600 AD compared to the solar forcing (see also main article), but this influence is significantly stronger (larger phase shift) in the medium scaled experiment.
These results are consistent with Goosse et al. (2005), who have commented that global and hemispheric results are generally closely tied to the underlying forcings, but that at regional scales the internal variability is often dominant.
One additional observation can be gained from Figure SUP-2. The magnitude of variability in the coupled climate model simulations of control, medium and large scaled solar irradiance can be compared directly. The unforced control simulation does exhibit variability at century timescales. Its magnitude is about +/- 0.075ºC. The medium scaled solar irradiance forcing enhances this variability to +/- 0.15ºC while clearly imposing a phase in accordance with the solar forcing. The large solar experiment nearly doubles the magnitude of ~200-year variability (almost +/- 0.3ºC) compared to the medium scaled run. Note, this window of ~200 years does not represent the full difference between the forcing series as lower-frequency variations at millennial time scales are omitted. But it focuses on a band that contains little influence from other factors.
In summary, the solar forcing is clearly recognizable in both large and medium solar experiments. While the large solar run was very strongly affected by the solar irradiance, the medium scaled simulation also exhibits influence of volcanism from the interannual to the centennial time scales.

4 Model comparison with proxy-reconstructions

Figure SUP-3 compares the medium scaled solar experiment with four selected proxy based climate reconstructions of the Northern Hemisphere. The overall agreement is generally good for the most recent 400 years, but even the medium scaled CSM simulation is cooler than most reconstructions in the centuries before 1600 AD. Only in comparison with Esper et al. (2002) is the overall mean temperature about the same. The mean pre-industrial temperature would also agree well with borehole estimates (see Figure 3b, Huang et al., 1998). In contrast to the overall mean temperature, the temporal structure of the 50-year Gaussian weighted series are very similar between the model simulation and Mann and Jones (2003), Crowley et al. (2003), and Esper et al. (2002) for most of the millennium. This correspondence is getting weaker in earlier centuries. The reason for this deterioration could either be due to decrease in quality of the forcings used in the simulation (forcings are also based on proxy indicators), and/or, the quality of the proxy-reconstructions is decreasing. The latter can result if proxy networks are sparser in the early part compared to later (Jones et al., 1998), or low-frequency retention of some proxies (particularly trees) could be influencing the reconstructions.
The medium scaled simulation, when compared to the four proxy-based climate reconstructions, is not at odds with the real world estimates. There are many indications that the model simulation is able to capture significant components of natural (forced) climate variability. Somewhat problematic is still the millennial scale structure both in the simulations and in the proxy reconstructions (for the latter, see for example Esper et al., 2005). The model results are affected by the need for a low-frequency detrending. This step could potentially introduce some inconsistency with an experiment that would not require this correction, particularly at the millennial time scale. Therefore, the focus of this paper was on shorter time scales unaffected by this issue. Nevertheless, the agreement with proxy records at multi-decadal time scales (Figure SUP-3) and the models ability to reproduce the instrumental record are clear indications that the natural forcing is implemented in a useful way, and despite uncertainty about the exact magnitude of external forcing, namely a possible solar background trend, the simulations indicate that the 20th century climate is essentially insensitive to the background trend differences in solar irradiance, and that the recent warming can only be explained if anthropogenic factors are included.

5 Spatial response patterns

Figure SUP-4 shows regression plots of annual surface temperature to the corresponding solar irradiance series used in the high, medium and low scales solar experiments. Only areas with a significant temperature response to solar forcing using a 95% confidence limit are shown (2.5% significance in a two-tailed test, taking into account the high serial correlation of the underlying time series). Both high and medium scaled experiments show a clear global wide positive relationship between the solar irradiance and surface temperature. The mean temperature response is somewhat larger in the large solar experiment (0.071 ºC for each W/m2 solar irradiance change) than the medium run (0.062 ºC per W/m2). Both experiments show a clear contrast between ocean and land, as well as an increase in response at higher latitudes (polar amplification). The small scaled solar forcing run exhibits some areas with significant correlations, but they should not be over interpreted because there is no significant relationship at the global level. The areas that do pass the significance test in this very short experiment (only 275 years were used between 1575 and 1849), could also result from a superposition of volcanic forcing with solar variations at the century time scale.

Caspar Ammann

Last modified: Fri Jan 20 16:18:49 MST 2006