Comment on “Aerosol radiative forcing and climate sensitivity deduced from the Last Glacial Maximum to Holocene transition” by Petr Chylek and Ulrike Lohmann


[1] Chylek and Lohmann [2008] (hereafter referred to as CL08) recently presented a new estimate of climate sensitivity (CS) based on analysis of a single paleo-climate record (Vostok ice core). CL08 inferred a narrow range of climate sensitivities (1.3–2.3°C) which essentially lies below 90% of the CS range given in the last IPCC report (2.0–4.5°C, [Meehl et al., 2007]). Moreover, CL08 claimed a much higher confidence in the range of uncertainties than that stated in the latest IPCC report (95% as given by CL08 versus 66% in IPCC report). In addition, CL08 estimated aerosol forcing during the Last Glacial Maximum (LGM) to be 3.3 ± 0.8 W/m2, which is well above previous estimates. Here we show that both the low CS and high aerosol forcing estimated by CL08 crucially depend on the authors' choice of uncertainty ranges for several key characteristics, as well as a number of assumptions which are hard to justify. In the following we demonstrate that CL08 strongly underestimate related uncertainties and therefore the method used by CL08 does not provide meaningful constraints on the range of climate sensitivity and LGM aerosol forcing.

[2] The ratio of past global temperature change to past global radiative forcing can be taken as an approximate measure of CS, as many authors have done, focusing on different periods in climate history [Covey et al., 1996]. This method is based on the assumption that CS does not strongly depend on the type of radiative forcing and climate state. The accuracy of this assumption remains debatable [Hansen et al., 2005; Schneider von Deimling et al., 2008]. In addition, the method of CL08 requires an estimate of global radiative forcing and global temperature anomaly from a local climate change signal. These two estimates introduce considerable uncertainty. To quantify the impact of glacial dust content on global LGM forcing, CL08 considered a second time interval in the ice core record (42 KyBP). The focus on this additional time period requires two additional assumptions. First, that the ratio between global and local temperature change remains the same over different periods of time; and second, that the local (Vostok) dust record represents a quantitative proxy for the global aerosol forcing. Again, these two assumptions introduce additional uncertainties. Below we show that total uncertainty of CS obtained with the method proposed by CL08 is much higher than that claimed by the authors.

[3] In their study, CL08 consider an Antarctic LGM cooling of 10.2°C without attaching an uncertainty estimate to this value. When applying equation (3) of CL08 to infer a 95% quantile of CS, uncertainty in both variables that enter their equation should be taken into account. Given the uncertainty in transferring measured δ18O content to a temperature signal, the accuracy of the stated 10.2° cooling is at best 20–30% and the range for estimated Antarctic LGM cooling is 7–11°C [Masson-Delmotte et al., 2006]. Even more important are the uncertainties in the ratio between Antarctic and global cooling at the LGM (ΔTA/ΔTG). CL08 choose the range 2.0–2.5 for this ratio. Since the inferred range of CS is inversely proportional to the range of ΔTA/ΔTG, the choice of the latter requires serious justification, but CL08 failed to give any justification except for referring to their own paper about recent temperature change in Greenland. However the ratio between regional and global temperature change derived from the instrumental data for the past century cannot be applied to the glacial conditions because (i) glacial climate forcing has a very different spatial pattern from the one responsible for global warming, and (ii) the transient climate system response on centennial time scale is very different from the quasi-equilibrium climate changes seen in paleoclimate records. At the same time, due to the poor global coverage of paleo-proxies it is impossible to infer global LGM cooling accurately enough from paleodata alone. Hence the use of climate models remains the only option to estimate ΔTA/ΔTG for glacial climate conditions. We note that the IPCC 2007 report [Meehl et al., 2007] gives a model-based ΔTA/ΔTG estimate of 1.4 ± 0.3 for equilibrium response to an increase in CO2 concentration. Whether the same ratio can be applied to glacial climate change is rather questionable: on one hand, glacial forcing had been strongly concentrated in the Northern hemisphere through the presence of large ice sheets and, to a lesser extent, due to pronounced changes in Northern hemisphere vegetation and dust content. Therefore, the glacial temperature anomaly was strongly asymmetric between the hemispheres, which is supported by the fact that glacial cooling in Greenland was 2–3 times stronger than in Antarctica [Masson-Delmotte et al., 2006]. On the other hand, local temperature change in Antarctica is very sensitive to local factors, such as change in elevation of the ice sheet. Based on results of PMIP2 experiments for LGM conditions, Masson-Delmotte et al. [2006] reported a median value of ΔTA/ΔTG = 1.9 for a prescribed 400 m increase in elevation of the Antarctic ice sheet during the LGM. Yet such a large increase in the elevation of the East Antarctic ice sheet is not supported by paleo-data. When model results are corrected for this elevation change, the median value for ΔTA/ΔTG is only 1.2 [Masson-Delmotte et al., 2006]. We should mention that those modeling studies had neglected cooling caused by changes in glacial dust content and vegetation changes, which can additionally reduce ΔTA/ΔTG due to a strong hemispheric asymmetry of these two forcings.

[4] By discussing these published results, it is hard to see why a ΔTA/ΔTG range of 2.0–2.5 has been chosen by CL08, rather than a range of 1.0–2.0, which is more in line with previous modeling studies. CL08 noted that their estimate of global LGM cooling is consistent with model results. Coupled climate models have simulated a global LGM cooling that is indeed similar to the range considered by CL08 [Masson-Delmotte et al., 2006]. Yet it should be stressed that none of these models accounted for aerosol forcing, which, according to CL08, should increase global cooling by more than 50%. Based on a large model ensemble, we have estimated that the additional global cooling through LGM dust and vegetation forcing is on the order of 1–2°C and we have inferred a total global LGM cooling of roughly 4–8°C [Schneider von Deimling et al., 2006a]. This estimate accounts for uncertainty in the magnitude of climate sensitivity and of glacial forcing, as well as for proxy-data uncertainty [Schneider von Deimling et al., 2006b]. It is based on paleo-constraints from different regions (tropical and Antarctic sites) and accounts for the forcing exerted by changes in glacial dust content and vegetation changes (glacial dust forcing fields are taken from ECHAM-5 simulations and range between 0.6 to 1.2 W/m2 globally). Hence, modeling results are in fact not consistent with the assumption of CL08 that LGM cooling had not exceeded 5°C globally. Considering a 7–11°C Antarctic LGM cooling, a range of 1–2 for ΔTA/ΔTG (as discussed above), and a global LGM radiative forcing of 8 ± 2 W/m2 [Meehl et al., 2007; Schneider von Deimling et al., 2006a], one infers a range of 1.3–6.8°C for CS – an interval which encompasses the IPCC range and which is not more narrow than most previous CS estimates based on different methods [Meehl et al., 2007]. This range is much broader than that reported by CL08.

[5] Uncertainty in the magnitude of glacial aerosol forcing does not affect our conclusion that CL08 strongly underestimate the upper bound of CS. But it is worth discussing the impact of aerosols on glacial cooling since CL08's estimate of global aerosol radiative forcing (3.3 ± 0.8 W/m2) is much higher than previous estimates based on physically based modeling of the dust cycle. Claquin et al. [2003] and Mahowald et al. [2006] computed glacial dust forcing in the tropics of the same magnitude as glacial CO2 but their estimated global dust forcing varied between 1 and 2 W/m2. To derive aerosol forcing, CL08 make use of additional information provided by paleo-climate data, namely they used (apart from LGM) another time slice (42 KyBP) when Antarctic temperature and radiative forcings were substantially different from LGM conditions. While the idea of using different time intervals of the glacial cycle to constrain CS is an interesting one, their choice for the second interval is very unfortunate. Indeed, the warm period around 42 KyBP (according to more recent age models, this interval corresponds to 47 or 48 KyBP) represents the warmest part of the so-called A2 Antarctic event, one of the strongest among numerous other similar Antarctic events [EPICA Community Members, 2006]. The main problem with using the peak of A2 warm event to constrain global aerosol radiative forcing is that it is generally believed that A2 (and similar warming events) neither represent the global nor the equilibrium climate response to the forcing considered in their paper. Indeed, warming in Antarctica during all of these events corresponds to the coldest period in Greenland (stadials) and the lowest methane concentrations. The peculiar phase relationship between these millennial scale climate anomalies in Greenland and Antarctica [EPICA Community Members, 2006] strongly support the idea that such warming events in Antarctica represent a transient Southern Hemisphere response to reorganizations of the Atlantic Ocean circulation, rather than a response to a global radiative forcing [Ganopolski and Rahmstorf, 2001]. Furthermore, glacial aerosol forcing was spatially and temporarily highly variable with a much larger negative forcing in the tropics as compared to high latitudes. Although temporal dynamics of dust depositions during glacial cycles were qualitatively similar in many regions (it is very unlikely that the global radiative forcing of dust was directly proportional to the dust deposition rate in Eastern Antarctica where during the glacial times it was related to a single source (Patagonia). Thus, deriving aerosol forcing solely from the Antarctic temperature and dust deposition records represents an ill-posed problem.

[6] If one would discard the strongest warming events in Antarctica as caused primarily by local factors and considers averaged Antarctic temperature between 40 and 45 KyBP (i.e. between A1 and A2 Antarctic events), then the difference between this temperature and the LGM would not exceed 2°C. Based on the Antarctic ice core data, we infer that changes in greenhouse gas concentrations contributed 0.7 W/m2 to that temperature change. To estimate the resulting minimum cooling, we consider a rather low CS (2°C) and a small value for ΔTA/ΔTG (1.0). This yields an Antarctic equilibrium cooling of at least 0.4°C. An additional cooling of about 0.5°C further results from the lapse rate effect associated with the sea level drop by 60 m [Lambeck et al., 2002] between 40–45 KyBP and the LGM (this effect is not considered at all by CL08). Finally, a considerable fraction of remaining Antarctic cooling is likely to result from another factor not accounted for by CL08, namely, a lower obliquity at the LGM compared to the time interval 40–45 KyBP [Jouzel et al., 2007]. Hence the temperature difference between time interval 40–45 KyBP and the LGM, which must be explained by the aerosol cooling, is uncertain and probably rather small. The fact that each individual wiggle in Antarctic temperature does not represent a change in global temperature is clearly illustrated by comparison between Holocene and previous (Eemian) interglacial (ca. 130–120 KyBP). Maximum Antarctic temperature during Eem was by 4.5°C warmer than during Holocene [Jouzel et al., 2007] which according to CL08 should imply that Eemian was at least 2°C warmer globally than Holocene. However no appreciable differences in GHG concentration, sea level and dust forcing between two interglacials exit to explain such large difference in global temperature.

[7] To summarize, here we have argued that (i) the uncertainties of estimating CS, based on the method used by CL08, are much larger than the authors claimed, and (ii) that the role of glacial aerosols on contributing to the LGM-Holocene cooling must have been much smaller than what CL08 inferred.