Keywords anthropogenic change, climate models, climate sensitivity, early agriculture, greenhouse warming, past climate, pre-industrial climate

Holocene Special Issue Comparisons of atmosphere ocean simulations of greenhouse gas-induced climate change for pre-industrial and hypothetical no-anthropogenic radiative forcing, relative to present day The Holocene 21(5) 793 801 The Author(s) 2011 Reprints and permission: sagepub.co.uk/journalspermissions.nav DOI: 10.1177/0959683611400200 hol.sagepub.com J.E. Kutzbach, 1 S.J. Vavrus, 1 W.F. Ruddiman 2 and G. Philippon-Berthier 1 Abstract We compare climate simulations for Present-Day (PD), Pre-Industrial (PI) time, and a hypothetical (inferred) state termed No-Anthropogenic (NA) based upon the low greenhouse gas (GHG) levels of the late stages of previous interglacials that are comparable in time (orbital configuration) to the present interglacial. We use a fully coupled dynamical atmosphere ocean model, the CCSM3. We find a consistent trend toward colder climate (lower surface temperature, more snow and sea-ice cover, lower ocean temperature, and modified ocean circulation) as the net change in GHG radiative forcing trends more negative from PD to PI to NA. The climatic response of these variables becomes larger relative to the changed GHG forcing for each step toward a colder climate state (PD to PI to NA). This amplification is significantly enhanced using the dynamical atmosphere ocean model compared with our previous results with an atmosphere slab ocean model, a result that conforms to earlier idealized GHG forcing experiments. However, in our case this amplification is not an idealized result, but instead helps frame important questions concerning aspects of Holocene climate change. This enhanced amplification effect leads to an increase in our estimate of the climate s response to inferred early anthropogenic CO 2 increases (NA to PI) relative to the response to industrial-era CO 2 increases (PI to PD). Although observations of the climate for the hypothetical NA (inferred from observations of previous interglacials) and for PI have significant uncertainties, our new results using CCSM3 are in better agreement with these observations than our previous results from an atmospheric model coupled to a static slab ocean. The results support more strongly inferences by Ruddiman concerning indirect effects of ocean solubility/sea-ice/deep ocean ventilation feedbacks that may have contributed to a further increase in late-holocene atmospheric CO 2 beyond that caused by early anthropogenic emissions alone. Keywords anthropogenic change, climate models, climate sensitivity, early agriculture, greenhouse warming, past climate, pre-industrial climate Introduction Great attention is being given to possible changes of climate in the next 50 100 years in response to continuing increases in the concentration of greenhouse gases (GHGs) (Intergovernmental Panel on Climate Change (IPCC), 2007). There is also interest in establishing how much the GHG increases of the industrial-era have already influenced the climate of the past 100 200 years (IPCC, 2007; Meehl et al., 2004). This topic of past changes has led to simulations of the climate of the pre-industrial (PI) period with a variety of models (Otto-Bliesner et al., 2006a; Sedlacek and Mysak, 2008a, b; Vavrus et al., 2008). PI climate simulations employ reduced net radiative forcing based on estimates of GHG concentrations prior to the time of their rapid increase. The simulated PI climate, with lowered GHGs, is typically 1 1.5K colder than present-day (PD) climate simulations. This simulated difference is larger than the estimated observed difference (0.7 1.2K colder), where the substantial range, 0.5K, reflects both measurement and sampling uncertainties especially for early periods extending back to 1850 (IPCC, 2007; Jones and Mann, 2004; Jones et al., 1999). (Our choice of 1850 for the time of PI, the time when rapid increase in industrial-era GHGs commenced, is described in the Appendix.) Interest in explaining the cause of GHG changes and estimating their role in climate change now extends back to the past several thousand years. Ruddiman (Ruddiman, 2003, 2008; Ruddiman et al., 2011, this issue) has established that the trends in GHGs in the Holocene differ significantly from the trends in GHGs in similar stages of previous interglacials. Whereas CO 2 and CH 4 trend downward in varying amounts following Northern 1 University of Wisconsin-Madison, USA 2 University of Virginia, USA Received 23 March 2010; revised manuscript accepted 11 November 2010 Corresponding author: J.E. Kutzbach, University of Wisconsin-Madison, Center for Climatic Research, Nelson Institute for Environmental Studies, 1225 W. Dayton Street, Madison WI 53706, USA. Email: jek@wisc.edu

794 The Holocene 21(5) Hemisphere (NH) summer insolation maxima in previous interglacials, the downward trend in CO 2 and CH 4 stops around 7000 5000 years ago in this interglacial (the Holocene) and thereafter the trend is upward. The estimated difference in GHG levels between the late stages of previous interglacials (roughly comparable in orbital forcing to the present) and the GHG levels of the present day is termed a state of No-Anthropogenic (NA) GHG forcing. In response to the question: Why do GHGs trend upward in the current interglacial but trend downward in similar stages of previous interglacials?, Ruddiman (2003, 2008) proposed that early agriculture and animal husbandry was the explanation, and developed estimates of the likely magnitude of these anomalous emissions in the period prior to the industrial revolution see Kaplan et al. (2010, this issue); Yu (2011, this issue), Fuller et al. (2011, this issue), and Ruddiman et al. (2011, this issue) for the most recent estimates. Alternatively, if human activities are not the cause of this atypical upward trend in GHGs in the late Holocene, then an alternative natural explanation (or process) is required to explain both the upward trends of this interglacial and the absence of such trends in previous interglacials (Ruddiman, 2008; Ruddiman et al., 2011, this issue). Our previous modeling studies using reduced GHG forcing examined the climatic response to the GHG trends observed in earlier interglacials for various model configurations including atmosphere static slab ocean (Ruddiman et al., 2005), atmosphere static slab ocean with interactive terrestrial vegetation (Vavrus et al., 2008), and dynamical atmosphere ocean (Kutzbach et al., 2010). The experiments with interactive terrestrial vegetation exhibit a positive feedback in high latitudes related to snow cover/vegetation interaction that enhances the cooling simulated with atmosphere ocean models. We have not yet included two other potentially important biogeophysical factors in our simulations. The biogeophysical impact of changes in land cover associated with human activities (such as the albedo increase from deforestation associated with the spread of agriculture) has not been included because there has been ongoing work to estimate the potentially large changes in area of land-use per capita among early agriculturalists compared with present day this larger-than-present agricultural footprint of earlier times could influence the climatic impact of early agriculture even although the population was relatively small (Kaplan et al., 2009, 2010 (this issue); Ruddiman and Ellis, 2009). We intend to include these new estimates of the agriculture footprint in future simulations. One recent modeling study simulated the combined biogeophysical and biochemical (CO 2 ) climatic effects of anthropogenic land cover change (ALCC) in the past millennium (Pongratz et al., 2010). They found that the higher surface albedo caused by the ALCC change produced a slight global cooling ( 0.03K) whereas the CO 2 rise of about 18 20 ppm led to a strong warming (0.18K); regionally, such as in Europe, China, and eastern North America, the biogeophysical change alone produced a more pronounced cooling but the combination of the increased surface albedo and the CO 2 rise caused a net warming everywhere. Finally, we have not yet experimented with models with interactive ocean biogeochemistry, a topic mentioned briefly in the Discussion and conclusions. This paper addresses only three items within this large overall topic: (1) the characteristics of the climate simulated by coupled dynamical atmosphere ocean models for lowered levels of GHG forcing associated with PI and NA (relative to PD); (2) the magnitude of the climatic response caused by the decreased GHG forcing from PD to PI relative to the corresponding response from PI to NA; and (3) some implications of these results for oceanrelated CO 2 feedbacks and for comparisons of simulations and observations. Simulations, models, and boundary conditions We compare selected results from two publications describing three climate simulations: Present Day (PD), the time around 1850 (PI), and the state of No Anthropogenic forcing (NA). The NA simulation (Kutzbach et al., 2010 where NA was called NOANTHRO), the PI simulation (Otto-Bliesner et al., 2006a) and the PD (modern control) simulation are described in greater detail in the original publications. The comparison of these three simulations will allow us to partition the relative climatic response to changed GHG forcing (from NA to PI, and from PI to PD). This partitioning will permit new insights concerning the climatic response to GHG forcing for two climate states (NA and PI) that are colder than PD. All three simulations (PD, PI, NA) are made with the NCAR Community Climate System Model version 3 (CCSM3), a fully coupled dynamical atmosphere ocean model (Collins et al., 2006a); the resolution of the atmospheric model is T42. The land cover model employs prescribed present-day vegetation (Dickinson et al., 2006). The CCSM3 has biases that are relevant to studies of the response of Northern Hemisphere temperature/snow to reduced radiative forcing (Kutzbach et al., 2010; Vavrus et al., 2008). In middle to subpolar latitudes, summer temperatures are biased 1 2K cold and winter temperatures are biased 0 1K warm (zonal averages). The summer cold bias is large in some areas, for example, as much as 2 to 6K over parts of northeastern Siberia and northwestern North America. However, with regard to snow cover, these seasonal surface temperature biases compensate at least partially such that the simulated extent of residual snow cover during summer is fairly realistic. Nevertheless, cold-biased control climates can influence the sensitivity of snow cover changes to reduced insolation (orbital) forcing as demonstrated by Vettoretti and Peltier (2003a, b). On the other hand, a model intercomparison of six coupled atmosphere ocean models for the last glacial maximum (LGM) showed that the response of CCSM3 to LGM forcing changes (ice sheet, GHG, and orbital changes) was similar to that of the other five models for large-scale measures such as global average and Northern Hemisphere average surface temperature (Braconnot et al., 2007). We will return to this topic of model bias in the section Results. The concentrations of five GHGs for the three simulations (PD, PI, NA) are listed in Table 1 along with the associated net change in GHG radiative forcing relative to PD: 2.05 W/m 2 for PI and 3.06 W/m 2 for NA. The GHG values for CO 2 and CH 4 for the hypothetical (inferred) state NA are based on the levels found at the late stages of previous interglacials that are comparable in time (orbital configuration) with the present interglacial. For CO 2, this value was initially estimated to be 240 245 ppm based on examination of three previous interglacials (Ruddiman, 2003). We picked the lower value (240 ppm) for our modeling studies, begun in 2004, and have continued to use that value for consistency over a suite of sensitivity experiments with various model configurations of increasing components and complexity. Estimates of GHG gases for this hypothetical state NA are now based on seven previous interglacials, rather than three, and now yield

Kutzbach et al. 795 Table 1. Greenhouse gas concentrations in PD, PI, and NA simulations, and the associated change in net radiative forcing relative to PD (the control). Refer to the Appendix for discussion of additional changes in radiative forcing in PI Greenhouse gases PD PI NA CO 2 (ppm) 355 280 240 CH 4 (ppb) 1714 760 450 N 2 O (ppb) 311 270 270 CFC 11 (ppt) 30 0 0 CFC 12 (ppt) 50 0 0 Radiative forcing change relative to PD (W/m 2 ) 2.05 3.06 long-term mean value). Because of the relatively small presentday orbital eccentricity and near average axial tilt, the present-day Northern Hemisphere summer insolation minimum is not as extreme as in the comparable stages of previous interglacials (the combined effects of these two orbital forcing factors are illustrated for the present interglacial and seven previous interglacials using caloric summer half year insolation in Kutzbach et al., 2010). Reduced GHG forcing (values lower than PI) has been shown to augment the effects of low summer insolation due to orbital (astronomical) forcing on summer temperature and snow cover (Vettoretti and Peltier, 2004); this combination of low Northern Hemisphere summer insolation forcing and reduced GHG forcing is present in both the PI and NA simulations. For reasons noted in the Introduction, we use the same prescribed vegetation in all three experiments. an NA estimate of ~250 ppm for CO 2 (Ruddiman et al., 2011, this issue). The NA value for CO 2 will be adjusted upward to 245 ppm in our future simulations for NA. However, a sensitivity experiment by Vavrus et al. (2008) found that using 250 ppm rather than 240 ppm for CO 2 (with no changes in other GHGs) produced only minor differences in the simulated NA climate. The PI simulation by Otto-Bliesner et al. (2006a) included other small changes in forcing but the net effect of these changes is assumed to be small (see Appendix). The length of the simulations, the simulation procedures, and the averaging intervals for each experiment are summarized in the Appendix. We use modern (PD) orbital parameters in all three simulations. (The orbital changes and associated insolation differences between PD and PI are trivially small.) The use of modern (PD) orbital parameters is consistent with the assumption that the hypothetical climate state NA has GHG levels appropriate to the late stages of previous interglacials i.e. times with a Northern Hemisphere summer insolation minimum (aphelion in northern summer). The present-day Northern Hemisphere summer insolation minimum (aphelion in northern summer) is not as extreme as previous minima because the orbital eccentricity is small Ruddiman et al. (2011, this issue) and Kutzbach et al (2010). Low obliquity (small axial tilt) has also been shown to contribute to cooler summers and lower annual-average temperature in high latitude (Phillipps and Held, 1994, Vettoretti and Peltier, 2004); the present-day obliquity is not at a minimum (but instead near its Results We focus on four indicators of the simulated climate changes associated with lowered GHG concentrations: global average and polar temperature, snow cover, sea-ice cover, and several ocean variables (Table 2 and Figures 1 and 2). Changes in these variables for PI and NA, relative to PD, are given in absolute terms or percents. The standard deviations (or coefficients of variation) for these variables for PD are listed in Table 2. The differences between the experiments are large relative to these indicators of variability within PD (and PI and NA, not shown). Temperature and snow cover The global annual average surface temperature for PD is 14.7 C. With the lowered GHG concentrations, the temperature is lowered to 13.5 C (PI) and 12.0 C (NA). Relative to these changes in global temperature, the annual temperature changes over polar and subpolar lands are larger (Table 2). With lowered temperatures, the annual-average area of NH snow cover increases by 13% (PI) and 29% (NA), and the area of year-round snow cover increases by 71% (PI) and 129% (NA). The regions of year-round snow cover are limited mainly to Greenland and Ellesmere Island in PD but they expand to northern Baffin Island, and parts of Alaska, eastern Siberia, and the Eurasian Arctic in PI; the area of year-round snow cover increases further in each of these four regions in NA (Figure 1). The Table 2. Changes in selected global and regional climate variables, annual average, for NA-PD, NA-PI, and PI-PD. Percentage changes are calculated relative to PD. Arctic and Antarctic are defined as the area poleward of 60 degrees. Control (PD) interannual standard deviations (for temperature, K) and coefficients of variation (for areas, % the ratio of the interannual standard deviation to the mean, in percent) are small relative to the differences between experiments. Global ocean temperature standard deviations (not shown) are very much smaller than the global surface temperature standard deviations Climate variables NA-PD NA-PI PI-PD PD: s, CV Net longwave radiative forcing (W/m 2 ) 3.06 1.01 2.05 Global surface temperature (K) 2.74 1.55 1.19 0.07 Arctic surface temperature (K) 7.16 4.07 3.09 0.36 Antarctic surface temperature (K) 4.08 2.77 1.31 0.42 N. H. land snow area (%) 29.1 16.5 12.6 2.5 N. H. permanent snow area (%) 129.0 58.3 70.7 14.0 N. H. sea ice area (%) 48.8 28.2 20.6 1.2 S. H. sea ice area (%) 44.0 32.3 11.7 3.3 Global ocean temperature (K) 1.25 0.88 0.37 Global ocean temperature between 0 and 1km (K) 1.39 0.79 0.60 Global ocean temperature between 1km and the bottom (K) 1.19 0.90 0.29

796 The Holocene 21(5) Figure 2. Annual average meridional overturning circulation (MOC) in Sv as a function of latitude and depth for PD, PI, and NA Figure 1. Months of snow cover for PD, PI, and NA.Year-round snow cover is shown in white. See text for discussion of fractional area of snow cover within each white grid square 129% increase in area (NA) indicates that the total area of yearround snow extends over more than twice the corresponding area of Greenland/Ellesmere Island in PD. The areas of increased year-round snow cover correspond well to areas with expanded mountain glaciers during the Little Ice Age (Kutzbach et al., 2010), an indication that the simulations produce increased yearround snow cover in regions that have seen glacial expansion in the past (see also Mysak, 2008). The area of snow cover in these simulations is less than the area of the white grid squares in Figure 1 because CCSM3 includes an algorithm that translates average snow depth for a grid square into fractional snow cover. This algorithm is based on the idea that if a grid square has varying small-scale topography and surface characteristics, then it is likely to be only partially snow-covered if the average snow depth of the square is small. In Figure 1, we show the entire grid square as white if it has permanent year-round snow of sufficient depth so that at least 5% of the area of the square would be snow covered.

Kutzbach et al. 797 As mentioned in the section Simuations, models, and boundary conditions, the modern control climate of CCSM3 (PD) has biases in summer and winter land surface temperature that could influence its seasonal snow climatology, although its annual average and summertime snow cover climatology agrees reasonably well with observations (Kutzbach et al., 2010; Vavrus et al., 2008). Nevertheless, this cold bias in summer land temperature in PD could have influenced the sensitivity of the model to lowered GHG forcing. Indeed, two regions of large cold bias in PD, northwestern North America and eastern Siberia, are regions of significantly increased snow cover in PI and NA (Figure 1). Sea-ice cover, ocean temperature, and ocean circulation Annual average sea-ice cover increases in both hemispheres: in the NH by 21% (PI) and 49% (NA), and in the SH by 12% (PI) and 44% (NA). The global and vertically averaged ocean temperature changes by 0.37K (PI) and 1.25K (NA), both relative to PD. While the vertical and latitudinal distribution of the ocean temperature change is not uniform, the cooling occurs at all depths and all latitudes (Table 2, and see latitude depth ocean temperature change, Kutzbach et al., 2010: figure 6). The lowered ocean temperature and expanded sea ice cover (Table 2) are linked to changes in latitude depth distribution of ocean salinity in PI and NA (not shown, but see Kutzbach et al., 2010; Otto-Bliesner et al., 2006b). The expanded area and thickness of sea ice enhances the salt flux to the ocean in the ice formation regions, particularly in the region surrounding the Antarctic continent. The colder and more saline surface water sinks to the deep ocean and flows northward at depth. In contrast, and except for these sea ice formation zones of higher surface salinity, the upper ocean is generally less saline (less evaporation from the tropical and subtropical oceans, melting of sea-ice in ice export regions, and, in the Northern Hemisphere, reduced northward transport) (Kutzbach et al., 2010; Sedlacek and Mysak, 2008a, b). These changes in water flow are apparent in changes of the meridional overturning circulation (MOC) for the three simulations: PD, PI, and NA (Figure 2). The narrow Antarctic Cell (area shown in blue centered near 70S) intensifies and elongates ( 4.7 Sv in PD, 5.9 Sv in PI, 6.6 Sv in NA) as the cold, saline water sinks, and the large Antarctic Bottom Water (AABW) cell (area shown in blue centered at a depth of 3 4 km) intensifies ( 16.9 Sv in PD, 21.2 Sv in PI, 24.7 Sv in NA) as this water flows north at depth. In contrast with these two major changes, the Deacon Cell (area shown in red and orange centered near 45S) weakens only slightly, but shifts slightly equatorward and retracts at depth. The North Atlantic cell intensifies slightly in PI and weakens in NA (relative to PD). (The North Atlantic cell, averaged for the Atlantic basin only, is not shown; however a somewhat similar response in the North Atlantic was reported by Otto-Bliesner et al., 2006b, where the North Atlantic overturning increased slightly in PI but decreased significantly at the LGM.) These changes in the North Atlantic in our simulations are reflected to some extent in the changes in global average MOC in the NH (Figure 2) (areas shown in yellow and green north of 30N). Changes in the global average MOC from PD to NA are in the same direction as those reported by Stouffer and Manabe (2003) for simulations with CO 2 forcing of 300 ppm and 150 ppm, respectively. The northward transport of heat by the ocean decreases slightly in PI and decreases by about 10% in NA (relative to PD) especially in the NH (not shown). This decrease may contribute to NH sea ice expansion (Table 2). Carbon dioxide feedbacks The changes in temperature, sea ice, and the MOC for NA (relative to PD), summarized above, may in turn cause further changes in the atmospheric concentration of CO 2 (Kutzbach et al., 2010). First, the lowering of ocean temperature by 1.25K (Table 2) (NA, compared with PD) increases the solubility of CO 2 in the ocean and would be associated with a reduction in atmospheric concentration of CO 2 of ~12.5 ppm (using the lower-end conversion factor of 10 ppm CO 2 per 1K described in Martin et al., 2005). The ocean cooling of 0.88K between PI and NA (Table 2) is, correspondingly, equivalent to a CO 2 reduction of ~9 ppm. This change, 9 ppm, is almost twice our previous estimate (5 ppm, Kutzbach et al., 2010) and is explained as follows. The change of 9 ppm is based upon the large simulated change in global ocean temperature from NA to PI (0.88K), whereas our previous estimate of a 5 ppm change was based upon attributing only 0.38 of the global ocean temperature change to the period from NA to PI (~0.5K). The small fraction, 0.38, was the fractional change in global slab ocean temperature from NA to PI that we had calculated using the CAM3+SO model; i.e. the different results stem from the different GHG sensitivity of the slab ocean model compared with the full ocean model (section Comparison of size of climatic responses of PI and NA, relative to PD ). Second, the increased sea-ice cover and the reduced upward motion from the deep ocean near 60S (Figure 3, Table 2) could have reduced the ventilation of CO 2 from the deep ocean to the atmosphere as discussed in the Introduction. An increase in global mean salinity could decrease ocean solubility of CO 2. However the mean salinity changes in PI (~0.005%) and NA (~0.01%) were extremely small because this version of CCSM3 largely omitted the important contribution to salinity change from snow buildup on land by capping snow depth increase at 1 m water equivalent (and in any case the simulation was relatively short for allowing substantial snow/ice buildup on land see Appendix). However, even for the last glacial maximum when the observed increase in salinity was several percent (Adkins et al., 2002) the salinity-related effect on CO 2 solubility due to greatly enlarged continental ice sheets was considerably smaller in magnitude (and of opposite sign) compared with the effect of the much colder glacial age ocean (Sigman and Boyle, 2000). Therefore any salinity-related change in CO 2 solubility in PI or NA would be extremely small relative to the solubility effect of the simulated change in ocean temperature described above. We return to further discussion of CO 2 feedbacks in the Discussion and conclusions. Comparison of size of climatic responses of PI and NA, relative to PD We compare the relative forcing and relative response of the industrial era (PI-PD) and the early anthropogenic era (NA-PI) (Figure 3). The net radiative forcing associated with GHG changes of the industrial era (PI-PD) is 2.05 W/m 2, whereas the net forcing of the early anthropogenic era (NA-PI) is 1.01 W/m 2, i.e. the early anthropogenic GHG forcing change represents 33% of the total change (NA-PD) (Figure 3). In contrast to the 33% change in

798 The Holocene 21(5) Figure 3. Decomposition of the total change (NA-PD) into two components: (NA-PI), in blue, and PI-PD, in pink shown in percent, for: net GHG radiative forcing (W/m 2 ), global surface temperature (K), NH area of permanent snow cover (ru), SH area of sea ice (ru), and global ocean temperature (K). The numbers within the bars show the magnitude of the change (same as Table 2). The units for snow cover and sea ice are relative units (ru) GHG forcing (NA-PI), the partitioning of the climatic response (Figure 3) is considerably larger than 33%: the response of global surface temperature is 57%; the response of NH permanent snow cover area is 45%; the response of SH sea-ice area is 73%; and the response of global ocean temperature is 70%. This result, an amplification of the climate response relative to the GHG forcing for colder climate states, i.e. NA-PI compared with PI-PD, (Figure 3) is considerably larger for CCSM3 than for our previous experiment with an atmosphere static slab ocean model (Vavrus et.al., 2008). The atmosphere slab ocean model consisted of the Community Atmospheric Model (CAM3) the same dynamical atmospheric model contained in CCSM3 but it was coupled to a slab ocean (here called SO a variable depth mixedlayer ocean model, Collins et al., 2006b) the coupled model is here called CAM3+SO. As an example of the larger response obtained with CCSM3, the response of global surface temperature (NA-PI) is 54% of the total response (NA-PD) using CCSM3, whereas the corresponding value was 38% using CAM3+SO. Our conclusion from these comparisons with both CCSM3 and CAM3+SO is that increments of changed GHG forcing in colder climate states (NA-PI) produce a proportionately larger climate response than increments of changed GHG forcing in warmer climate states (PI-PD), and that this effect is greater for the dynamical atmosphere ocean model than for the atmosphere slab ocean model. A similar non-linear response of climate to a broad range of GHG forcing is reported by Manabe and Bryan (1985). Using CO 2 forcing of 150, 212, 300, 600, 1200, and 2400 ppm in a coupled atmosphere ocean model for an idealized landocean planet, Manabe and Bryan (1985) found a more pronounced rate of increase of surface temperature for values of CO 2 moving from 150 to 212 ppm, and from 212 to 300 ppm, than for subsequent increases in CO 2. They attributed this proportionately larger climate response to GHG forcing at colder climate states largely to temperature-albedo feedbacks. They also reported that when they repeated a similar set of experiments with an atmosphere slab ocean model, the corresponding proportional changes were smaller, thus helping to confirm their conclusion that sea ice albedo feedback processes in the fully coupled dynamical atmosphere ocean model were enhanced, relative to the slab ocean, by changes in the meridional overturning circulation and in ocean heat transport. Discussion and conclusions We compared previously published climate simulations for Present-Day (PD), Pre-Industrial (PI) times, and a hypothetical (inferred) state termed No Anthropogenic forcing (NA) based upon the GHG levels of the late stages of previous interglacials relative to the present interglacial. These simulations used a fully coupled dynamical atmosphere ocean model, the CCSM3, an improvement in model design compared with our previous examination of these three climate states using an atmospheric model coupled to a static slab ocean (Vavrus et al., 2008). We note the following points: (1) By comparing the PI simulation (Otto-Bliesner et al., 2006a) and the NA simulation (Kutzbach et al., 2010) with the control, PD, we find a consistent direction of climatic response as the net radiative forcing associated with decreased concentrations of GHGs becomes increasingly negative. The response includes lower surface temperature, more snow cover, more sea-ice cover, a colder ocean, and changes in ocean overturning as the forcing changes from PD to PI to NA.

Kutzbach et al. 799 (2) The climatic response to decreased GHG forcing is relatively larger for the colder climate states (NA-PI, compared with PI PD). This result confirms the sense of our earlier results using CAM3+SO (Vavrus et al., 2008), but the nonlinearity of this response is significantly larger for CCSM3 than for CAM3+SO. This result agrees with the results of idealized experiments by Manabe and Bryan (1985) see section Comparison of size of climatic responses of PI and NA, relative to PD. However, rather than being idealized experiments, as was the case in Manabe and Bryan (1985), in our case this non-linearity is now linked to specific climatic states (PI and NA) and therefore helps frame important questions about Holocene climate trends. (3) The relatively larger response of climate to changes in GHG forcing for colder climate states has the effect of enhancing the climate s response to the inferred early anthropogenic CO 2 increases (NA to PI) relative to the industrial-era increases (PI to PD) (Figure 3). The CCSM3 simulated a larger global ocean temperature change between NA and PI than we had previously calculated using the fractional amplification effect obtained with CAM3+SO, a temperature change of 0.88K (Table 2) rather than ~0.5K (Kutzbach et al., 2010). This larger warming of 0.88K (from NA to PI) matches very closely the temperature increase inferred from the observed late-holocene trend in marine benthic δ 18 O, relative to the trend in previous interglacials (Lisiecki and Raymo, 2005) an increase of 0.84K (Kutzbach et al., 2010). Thus there is now close agreement in the amount of global ocean warming simulated by CCSM3 (0.88K) and estimated from the observations (0.84K). (This temperature estimate based on the late Holocene δ 18 O trend obviously has some error component because the marine observations may not be representative of the entire ocean however, the observations included data from 47 ocean cores and the Holocene δ 18 O trend is larger than the standard deviation of the trends of previous interglacials (see Kutzbach et al., 2010: figure 4).) A related result of this enhanced amplification effect at the coldest climate state is that the CCSM3-simulated global surface temperature increase since the onset of the industrial era (from PI to PD), 1.2K, is smaller than the CAM3+SO-simulated increase, 1.7K, and therefore closer to observational estimates (Jones and Mann, 2004) although the accuracy of the estimate from observations is also uncertain as noted in the Introduction. (4) The potential relative contributions of direct and indirect effects to the pre-industrial CO 2 increase are different than reported in our previous studies and have changed over time. Ruddiman (2003) estimated that CO 2 increased from 240 ppm (NA) to 280 ppm (PI), an increase of ~40 ppm, because of anthropogenic factors directly related to early agriculture. Estimates of much smaller direct effects of early agriculture, 5 ppm or less, were published by Joos et al. (2004), Elsig et al. (2009), and others (see Ruddiman et al., 2011, this issue). Ruddiman (2007) likewise noted that the direct effect may have been considerably smaller than 40 ppm (his initial estimate), and proposed that substantial CO 2 positive feedback from an anomalously warm Holocene ocean might be required to satisfy the full carbon budget required to support a 40 ppm rise in CO 2. We briefly summarize current observation-based estimates of the direct effect and the CCSM3-based estimate of the indirect (climate feedback) effect. The direct effect of early anthropogenic activities may now, according to recent studies, be considerably larger than some of the previous estimates that had been as low as 10 ppm or even lower. Indeed, the direct effect may have increased CO 2 by as much as 21 22 ppm based upon studies reported in this volume. Kaplan et al. (2010, this issue) estimates 310 GtC emissions from pre-industrial land clearance that equate to a CO 2 increase of ~22 ppm. Another estimate (300 GtC emissions and a CO 2 increase of 21 ppm) comes from a mass-balance calculation that takes into account the need for a very large release of anthropogenic terrestrial carbon to balance a similarly large burial of terrestrial carbon in boreal peats (Ruddiman et al., 2011, this issue; Yu, 2011, this issue). The indirect (positive feedback) from decreased ocean solubility may have increased CO 2 by about 9 ppm (sections Results and Comparison of size of climatic responses of PI and NA, relative to PD ). This value of 9 ppm is larger than the value we had estimated previously (5 ppm) because the simulated warming of the global ocean in CCSM3 from NA to PI (0.88K) (Table 2) is larger than the corresponding value we had estimated previously (0.5K) based upon the fractional partitioning of the slab ocean temperature response from NA to PI (0.38) the fraction we had calculated from the CAM3+SO simulation see sections Results and Comparison of size of climatice responses of PI and NA, relative to PD. As noted in (3), above, this larger temperature change (simulated) agrees closely with estimates of temperature change from marine oxygen isotope observations. Adding the direct contribution to CO 2 increase from early deforestation (perhaps up to 21 22 ppm) and the contribution from ocean-solubility feedback (up to ~ 9 ppm), the early anthropogenic-related total increase in CO 2 (direct, plus the indirect solubility feedback) could be as much as 30 31 ppm i.e. an amount closer to the value of 40 ppm originally proposed by Ruddiman (2003), and even closer if the revised value for the CO 2 difference between NA and PI is in the range 35 40 ppm, a slightly lower range that seems likely now that additional CO 2 observations from early interglacials are included in the analysis (see section Simulations, models, and boundary conditions, and Ruddiman et al., 2011, this issue). The CCSM3 does not include a biogeochemical submodel, but Southern Ocean processes that have been claimed to affect atmospheric CO 2 (such as changes in southern sea ice and southern ocean upwelling simulated in both NA and PI) are likely to have added more positive CO 2 feedback. In particular, decreased southern sea ice and enhanced wind-driven upwelling in the region of the Southern Ocean westerlies would increase ocean ventilation and add to the atmospheric CO 2 concentration (Kutzbach et al., 2010). This positive feedback mechanism is supported by inferences of enhanced upwelling as diagnosed from Southern Ocean sediments during the most recent deglaciation and associated CO 2 rise (Anderson et al., 2009). Only a climate model with interactive oceanic biogeochemical cycling can address this topic quantitatively. Finally, as described in the Introduction, this simulation did not include terrestrial vegetation processes or prescribed changes in land use associated with early agriculture. (5) There are several important caveats to be highlighted: (i) simulations by other models, and by models of higher resolution, will be needed to confirm or modify these results from CCSM3; (ii) the two simulations (PI and NA) used different computational procedures for calculating the equilibrium response to the changed forcing, and the PI simulation included additional changes in forcing that could not be calculated exactly (see Appendix); future simulations should avoid these differences an option not possible here because of computer-time limitations; (iii) quantitative assessment of the impacts of these different climate states on feedbacks affecting atmospheric CO 2 levels must await simulations with climate models that include biogeochemical feedbacks; (iv) observations show some agreement with the results for PI and NA, but more observations will be very useful; and (v) the CCSM3 has

800 The Holocene 21(5) a significant summertime cold bias that may affect model sensitivity to snow-related feedbacks with lowered GHG forcing this underscores the need for simulations with other models and improved versions of CCSM (Kutzbach et al., 2010; Vavrus et al., 2008). For example, the newly released (2010) version of CCSM, named CESM1, has a smaller summertime temperature cold bias than CCSM3. Appendix Notes on radiative forcing, simulation procedures, and the calendar time associated with Pre-Industrial (PI) Radiative forcing. The caption of Table 1 mentions that the PI simulation of Otto-Bliesner et al. (2006a) changed not only the five GHGs listed in Table 1, but other variables: namely, the concentrations of ozone, sulphate aerosols, and carbonaceous aerosols, and the solar constant and the orbital year (see table 1 of Otto-Bliesner et al. and discussion therein for details). Small decreases in ozone, sulphate aerosols, and the solar constant (relative to PD) acted to enhance the net negative forcing, whereas the increase in carbonaceous aerosols acted to diminish the net negative forcing. The slight change in orbital forcing had near zero effect. Unfortunately, the exact net effect of these additional prescribed changes in radiative forcing for PI is not available (Otto- Bliesner, personal communication, 2010). However, we have concluded that the overall net effect of these additional changes of radiative forcing was small. We base this conclusion on a comparison of results from two PI experiments using CAM3+SO (this atmosphere slab ocean model is described briefly in section Comparison of size of climatic responses of PI and NA, relative to PD ). Vavrus et al. (2008) found, using CAM3+SO, that the change in global average temperature was 1.71K in response to the reduced GHG forcing of 2.05 W/m 2 (Table 1). Otto-Bliesner found, also using CAM3+SO, that the change in global average temperature was 1.73K in response to the reduced GHG forcing of 2.05 W/m 2 plus the net effect of the other adjustments to radiative forcing described above (Otto-Bliesner, personal communication, 2010). Based upon the almost identical change in global mean temperature for both PI experiments (Vavrus et al. and Otto- Bliesner et al.) using the same model (CAM3+SO), we will assume that the net radiative forcing changes in the two PI experiments were very similar; i.e. that the additional changes in radiative forcing made by Otto-Bliesner et al. (their table 1 and discussion) were largely offsetting and had relatively little effect on the total net forcing. We therefore conclude (sections Results and Comparison of size of climatice responses of PI and NA, relative to PD ) that we can compare results of PI with NA, assuming that the net change in radiative forcing is well represented by the forcing associated with the five GHGs listed in our Table 1. Applying this conclusion based upon the results with CAM3+SO to the results with CCSM3, is supported by the finding that the sensitivity of global average temperature to doubling of CO 2 is nearly identical for CAM3+SO and CCSM3 (Kiehl et al., 2006). Averaging intervals, initialization, and length of runs. The previously published experiments by Kutzbach et al. and Otto-Bliesner et al., conducted independently, had chosen somewhat different periods within NCAR s long 1000-year CCSM3 control simulation to define a present-day average (PD); here we use years 981 1000 for the PD control. In both studies, CCSM3 was run additional years after introducing the changes in net radiative forcing (Table 1) in order to establish new equilibriums for PI and NA. The model atmosphere, land surface, and upper ocean typically reach new equilibriums quickly, but the deep ocean and the slowly changing sea-ice require more time. Otto-Bliesner et al. ran CCSM3 an additional 400 years for the PI experiment (starting from year 700 of the long control run) after introducing the changed radiative forcing associated with PI; the PI experiment is the average of years 350 399 of the additional 400 year run. The total length of the simulation was 1100 years (700 plus 400). Kutzbach et al. initialized the CCSM3 ocean for the NA experiment (starting from year 1000 of the long control run) using the larger reduction in net radiative forcing (Table 1), but they first adjusted the initial ocean temperatures to be colder than at the end of the PI simulation (Otto-Bliesner et al.); this initialization caused the model to reach quasi-equilibrium after only 100 years of additional simulation. The model was then run an additional 33 years; the NA experiment is the average of years 1114 1133. The total length of simulation was 1133 years (1000 plus 133) (Kutzbach et al.). We mention the similarity in length of simulations for PI and NA (1100 years and 1133 years, respectively) because the temperature of the deep ocean in CCSM3 exhibits a small downward drift of about 0.04K per century, so at least this drift component of the deep ocean temperature change is similar in all experiments. In sections Results and Comparison of size of climatic responses of PI and NA, relative to PD, we compare the responses of global ocean temperature (including the deep ocean) for PI and NA. Because these responses of global ocean temperature are relatively large compared to the small rate of drift of the deep ocean temperature, we assume that any differences in deep ocean temperature between PI and NA are caused primarily by the changed radiative forcing. We cannot estimate any additional effect on our results caused by our method of initializing NA with ocean temperatures colder than those at the end of the PI simulation, but we note that all simulations had reached quasiequilibrium except that the deep ocean in all three simulations maintained approximately the same small drift of deep ocean temperature. The averaging intervals for PD, PI and NA are all relatively short (20, 50 and 20 years, respectively). However, the interannual variability within these intervals was small (see Table 2). Because CCSM3 exhibits considerable decade-scale variability, longer averaging intervals on the order of a century would have been desirable but would have required longer simulations, particularly for NA. (Brandefelt and Otto-Bliesner, 2009, provide illustrations of long time series showing considerable decade-scale variability from a CCSM3 simulation of the LGM.) We have focused our analysis on large spatial averages and on annual averages that are least likely to be affected by the relatively short averaging periods; the only exception is the mapping of months of snow cover (Figure 1) which might be expected to change somewhat depending upon the exact choice of averaging interval. Calendar time associated with the Pre-Industrial era. Various calendar dates have been associated with the start of the Industrial Era. A number of versions of coal-fired steam engines were in development by 1750, and James Watts improved engine was introduced in 1775. While these dates may influence a choice of ~ 1750 for the start of the industrial revolution, the evidence suggests that the revolution took some decades to gather full momentum hence our choice of 1850 for dating the revolution as fully underway. Several kinds of observations and emissions estimates support this date of 1850. The ice core records from Greenland and Antarctica indicate that CO 2 concentrations exceeded 280 ppm (the canonical PI level) at times during the Medieval

Kutzbach et al. 801 period and climbed consistently above the highest pre-industrial levels only around 1850 (IPCC, 2007; Ruddiman, 2007, 2008). Sulphate aerosol concentrations in ice cores exhibit a detectable rise around 1800 but this upward trend strengthens considerably by 1850 (IPCC, 2007: figure 6.15 of Chapter 6 of Fourth Assessment Report, Working Group 1). Estimates of human-related carbon emissions (millions of metric tons of C per year) are 3 in 1750, 8 in 1800, 55 in 1850, and about 7000 in 2000 (Andres et al., 2000); i.e. carbon emissions before 1850 were very small. Acknowledgements This work has been supported by National Science Foundation grants ATM-0902982 to the University of Virginia and ATM-0602270 and ATM-0902802 to the University of Wisconsin. Computational support for this research was provided by NCAR s Climate Simulation Laboratory, which is supported by the National Science Foundation. We thank Bette Otto-Bliesner, NCAR, for permitting us to use the pre-industrial simulation described herein in our analysis, and for facilitating access to the history tapes. We thank the anonymous reviewers for their criticisms, comments, and suggestions that have helped us modify, elaborate and clarify our findings. References Adkins J, McIntyre K and Schrag DP (2002) The salinity, temperature, and d 18 O of the glacial deep ocean. Science 298: 1769 1773. Anderson RF, Ali S, Bradtmiller LI, Nielsen SHH, Fleisher MQ, Anderson BE et al. (2009) Wind-driven upwelling in the Southern Ocean and the deglacial rise in atmospheric CO 2. Science 323: 1443 1448. Andres RJ, Marland G, Boden T and Bischof S (2000) Carbon dioxide emissions from fossil fuel consumption and cement manufacture, 1751 1991, and an estimate of their isotopic composition and latitudinal distribution. In: Wigley TML and Schimel DS (eds) The Carbon Cycle. Cambridge: Cambridge University Press, 53 62. Braconnot P, Otto-Bliesner B, Harrison S, Joussaume S, Peterschmitt J-Y, Abe-Ouchi A et al. (2007) Results of PMIP2 coupled simulations of the Mid-Holocene and the Last Glacial Maximum-Part I: Experiments and large-scale features. Climates of the Past 3: 261 277. Brandefelt J and Otto-Bliesner BL (2009) Equilibrium and variability in Last Glacial Maximum simulation with CCSM3. Geophysical Research Letters 36: L19712, doi:10.1029/2009gl040364. Collins WD, Bitz CM, Blackmon ML, Bonan GB, Bretherton CS, Carton JA et al. (2006a) The Community Climate System Model: CCSM3. Journal of Climate 19: 2122 2143. Collins WD, Rasch PJ, Boville BA, Hack JJ, McCaa JR, Williamson DL et al. (2006b) The formulation and atmospheric simulation of the Community Atmospheric Model Version 3 (CAM3). Journal of Climate 19: 2144 2161. Dickinson RE, Oleson KW, Bonan G, Hoffman F, Thornton P, Vertenstein M et al. (2006) The Community Land Model and its climate statistics as a component of the Community Climate System Model. Journal of Climate 19: 2302 2324. Elsig J, Schmitt J, Leuenberger D, Schneider R, Eyer M, Leuenberger F et al. (2009) Stable isotope constraints on Holocene carbon cycle changes from an Antarctic ice core. Nature 461: doi:10.1038/nature.08393. Fuller D, Van Etten J, Manning K, Castillo C, Kingwell-Banham E, Weisskopf A et al. (2011) The contribution of rice agriculture and livestock pastoralism to prehistoric methane levels: An archaeological assessment. The Holocene 21 (this issue). doi: 10.1177/0959683611398052. Intergovernmental Panel on Climate Change (2007) Climate Change 2007: The Physical Science Basis. Cambridge: Cambridge University Press. Jones PD and Mann ME (2004) Climate over past millennia. Reviews of Geophysics 42: RG2002, doi:10.1029/2003rg000143. Jones PD, New M, Parker DE, Martin S and Rigor IG (1999) Surface air temperature and its changes over the past 150 years. Reviews of Geophysics 37: 173 199. Joos F, Gerber S, Prentice IC, Otto-Bliesner BL and Valdes P (2004) Transient simulations of Holocene atmospheric carbon dioxide and terrestrial carbon since the last glacial maximum. Global Biogeochemical Cycles 18: GB2002 10.1029/2003GB002156. Kaplan J, Krumhardt KM, Ellis EC, Ruddiman WF, Lemmen C and Goldewijk KK (2010) Holocene carbon emissions as a result of anthropogenic landcover change. The Holocene 21 (this issue). doi: 10.1177/0959683610386983. Kaplan JQ, Krumhardt KM and Zimmerman N (2009) The prehistoric and preindustrial deforestation of Europe. Quaternary Science Reviews Doi:10.1016/j.quascirev.2009.09.028. Kiehl JT, Shields CA, Hack JJ and Collins WD (2006) The climate sensitivity of the Community Climate System Model: CCSM3. Journal of Climate 19: 2584 2596. Kutzbach JE, Ruddiman WF, Vavrus SJ and Philippon G (2010) Climate model simulation of anthropogenic influence on greenhouse-induced climate change (early agriculture to modern): The role of ocean feedbacks. Climatic Change 99: 351 381 (published previously online 03 September 2009), DOI 10.1007/s10584-009-9684-1. Lisiecki LE and Raymo ME (2005) A Plio-Pleistocene stack of 57 globally distributed benthic δ 18 O records. Paleoceanography 20: PA1003,doi:10. 1029/2004PA001071. Manabe S and Bryan K Jr (1985) CO 2 -induced change in a coupled ocean atmosphere model and its paleoclimatic implications. Journal of Geophysical Research 90: 11 689 11 707. Martin P, Archer D and Lea DW (2005) Role of deep-sea temperatures in the carbon cycle during the last glacial. Paleoceanography 20: PA2015, doi: 10.1029/2003PA000914. Meehl GA, Washington WM, Ammann C, Arblaster JM, Wigley TML and Tebaldi C (2004) Combinations of natural and anthropogenic forcings and 20th century climate. Journal of Climate 17: 3721 3727. Mysak L (2008) Glacial inceptions: Past and future. Atmosphere-Ocean 46: 317 341, DOI:10.3137/ao.460303. Otto-Bliesner BL, Brady EC, Clauzet G, Tomas R, Levis S and Kothavala Z (2006b) Last Glacial Maximum and Holocene climate in CCSM3. Journal of Climate 19: 2527 2544. Otto-Bliesner BL, Tomas R, Brady EC, Ammann C and Kothavala Z (2006a) Climate sensitivity of moderate- and low-resolution versions of CCSM3 to preindustrial forcings. Journal of Climate 19: 2567 2583. Phillipps P and Held I (1994) The response to orbital perturbations in an atmospheric model coupled to a slab ocean. Journal of Climate 7(5): 767 782. Pongratz J, Reick CH, Raddatz T and Claussen M (2010) Biogeophysical versus biogeochemical climate response to historical anthropogenic land cover change. Geophysical Research Letters 37: L088702, doi:10.1029/2010gl043010. Ruddiman WF (2003) The anthropogenic greenhouse era began thousands of years ago. Climatic Change 61: 261 293. Ruddiman WF (2007) The early anthropogenic hypothesis: Challenges and responses. Reviews in Geophysics 45: RG4001 10.1029/2006RG000207. Ruddiman WF (2008) The challenge of modeling interglacial interglacial CO 2 and CH 4 trends. Quaternary Science Reviews 27: 445 448. Ruddiman WF and Ellis EC (2009) Effect of per-capita land-use changes on Holocene forest clearance and CO 2 emissions. Quaternary Science Reviews 28: doi:10.1016/j.quascirev.2009.05.022. Ruddiman WF, Kutzbach JE and Vavrus SJ (2011) Can natural or anthropogenic explanations of late Holocene CO 2 and CH 4 increases be falsified? The Holocene 21 (this issue). doi: 10.1177/0959683610387172. Ruddiman WF, Vavrus SJ and Kutzbach JE (2005) Test of the overdueglaciation hypothesis. Quaternary Science Reviews 25: 1 10. Sedlacek J and Mysak LA (2008a) A model study of the Little Ice Age and beyond: Changes in ocean heat content, hydrography, and circulation. Climate Dynamics (published online 19 December 2008) DOI:10.1007/ s00382-008-0503-6. Sedlacek J and Mysak LA (2008b) Sensitivity of sea ice to wind stress and radiative forcing since 1500: A model study of the Little Ice Age and beyond. Climate Dynamics (published online in April 2008) DOI:10.10007/ s00382-008-0406-6. Sigman DM and Boyle EA (2000) Glacial/interglacial variations in atmospheric carbon dioxide. Nature 407: 859 869. Stouffer RJ and Manabe S (2003) Equilibrium response of thermohaline circulation to large changes in atmospheric CO 2 concentration. Climate Dynamics 20: 759 773. Vavrus SJ, Ruddiman WF and Kutzbach JE (2008) Climate model tests of the anthropogenic influence on greenhouse-induced climate change: The role of early human agriculture, industrialization, and vegetation feedbacks. Quaternary Science Reviews 27: 1410 1425. Vettoretti G and Peltier WR (2003a) Post-Eemian glacial inception. Part I: The impact of summer seasonal temperature bias. Journal of Climate 16: 889 911. Vettoretti G and Peltier WR (2003b) Post-Eemian glacial inception. Part II: Elements of a cryosphere moisture pump. Journal of Climate 16: 912 927. Vettoretti G and Peltier WR (2004) Sensitivity of glacial inception to orbital and greenhouse gas climate forcing. Quaternary Science Reviews 23: 499 519. Yu Z (2011) Holocene carbon flux histories of the world s peatlands: Global carbon-cycle implications. The Holocene 21 (this issue). doi: 10.1177/0959683610386982.