A probabilistic analysis of climate variation during the period AD 1050–1800 refines available estimates of the influence of temperature change on the concentration of carbon dioxide in the atmosphere.
One impact of an increase in the atmospheric concentration of carbon dioxide and other greenhouse gases is a global warming at Earth's surface1. The climatic effect influences the carbon cycle and thus, in turn, the concentration of CO2 in the atmosphere. Elsewhere in this issue (page 527), Frank and colleagues2 describe an analysis designed to reduce the large uncertainty in estimating the scale of this feedback loop by quantifying more precisely the sensitivity of the carbon cycle to climate.
The two-way interactions between climate and atmospheric CO2 concentration lead to both negative and positive feedbacks. For instance, a warm, drying trend in regions where water availability limits plant growth will tend to reduce net primary production (the amount of new organic matter produced by photosynthesis) and thus the carbon uptake by vegetation. By contrast, a temperature increase will tend to enhance primary production and carbon uptake in cold regions. By contrast again, warming may induce acceleration in the respiration of soil organisms, so releasing more CO2 to the atmosphere. Overall, however, the net effect of these interactions is generally considered to be a positive one that amplifies variations in both CO2 and temperature3,4.
Computer models of the interplay between climate and the carbon cycle include representations of those feedbacks. But there are large differences in the results of the various models. This means that there are significant uncertainties in climate projections of the response to CO2 emissions and in estimates of the levels of emissions required to meet some target for maximum warming5. Take, for example, a comparison3 of the results of 11 models driven by anthropogenic emissions of CO2: when climate–carbon-cycle feedbacks were taken into account, the simulated CO2 concentration in the atmosphere by 2100 was higher by as little as 20 p.p.m.v. (parts per million by volume) or as much as 200 p.p.m.v. compared with idealized experiments in which the feedbacks were not taken into account.
To reduce those uncertainties, Frank et al.2 compared changes in temperature (drawn from various forms of proxy data) and in atmospheric CO2 concentration (as recorded in three ice cores) for the period AD 1050–1800. Over this period, anthropogenic emissions of carbon were relatively weak and CO2 variations had only a minor impact on temperature changes6. Assuming that the carbon cycle was mainly influenced by climate variations, it is relatively straightforward to obtain estimates of the influence of temperature changes on the atmospheric CO2 concentration. By contrast, over the past 200 years the perturbation of the carbon cycle by anthropogenic activities has been much larger, and the CO2 rise has been a major cause of global warming, at least over the past 50 years1. This leads to complex interactions and larger uncertainties for the magnitude of any individual process.
In a global approach to the system, the authors2 derived estimates of γ, the sensitivity of the carbon cycle to temperature. This is defined as the increase in atmospheric CO2 concentration associated with an increase in Northern Hemisphere mean temperature of 1 °C. Applying a formal and thorough investigation of the uncertainties, Frank and colleagues obtain a range for γ of 1.7–21.4 p.p.m.v. CO2 per °C.
This range may seem large, and most current models have values within it. However, Frank et al.2 confirm that the probability of negative values is very low. Given that about 50% of the carbon emitted by anthropogenic activities has remained in the atmosphere4, this means that, in a warmer climate, we should not expect pleasant surprises in the form of more efficient uptake of carbon by oceans and land that would reduce the fraction of the anthropogenic CO2 remaining in the atmosphere and limit the amplitude of future climate change. Frank et al. also attribute low probability to very high values for γ, at least in the range of temperatures covered by the period that they studied.
A particularly intriguing feature is the apparently different behaviour of the system when two periods, 1050–1549 and 1550–1800, are analysed separately. For 1550–1800, the correlation between CO2 and temperature and the value of γ are relatively high, whereas both values are much lower for 1050–1549. The difference might simply be related to biases and larger uncertainties in the data for the first part of this 750-year interval. Or it might also indicate that CO2 variations are a function not only of hemispheric mean temperature, but also of finer-scale geographical variation, and that additionally we should take into account the effects of precipitation, and of ocean circulation, as it affects the marine carbon cycle.
Given this broader framework, reproducing and reaching a precise understanding of the factors responsible for changes in atmospheric CO2 concentration during the whole of the past millennium is quite a challenge. Some simulations using relevant climate–carbon-cycle models are available7, however, and more are expected in the near future; not least, the past millennium is one of the key periods to be investigated in a recently launched initiative (the Paleoclimate Modelling Intercomparison Project Phase III)8.
More generally, the work of Frank et al.2 and of others9 shows that the period 1050–1800 constitutes a unique natural laboratory for studying climate variation. More accurate, high-resolution reconstructions of variations in temperature and levels of CO2, as well as additional model simulations, will help to better constrain feedbacks between climate and the carbon cycle. Such work will also help to improve understanding of natural climate variability on longer timescales than can be investigated using only the instrumental data over the past 150–200 years. It should also help in understanding the response to natural factors such as changing solar irradiance and large volcanic eruptions.
In this perspective, models and data do not necessarily need to be processed as two separate lines of information. Indeed, the first attempts have been made to directly combine observations with the constraints imposed by physical laws as represented by models10. Until now, studies using such data-assimilation techniques have not included interactive representations of the carbon cycle, but there is no formal obstacle to prevent that. Inclusion of such representations would produce estimates of the state of the coupled climate–carbon system that use all available information, and thus potentially provide stronger constraints on all of the processes involved.
Solomon, S. et al. (eds) Climate Change 2007: The Physical Basis (Cambridge Univ. Press, 2007).
Frank, D. C. et al. Nature 463, 527–530 (2010).
Friedlingstein, P. et al. J. Clim. 19, 3337–3353 (2006).
Denman, K. L. et al. in Climate Change 2007: The Physical Basis (eds Solomon, S. et al.) 499–587 (Cambridge Univ. Press, 2007).
Meehl, G. A. et al. in Climate Change 2007: The Physical Basis (eds Solomon, S. et al.) 747–845 (Cambridge Univ. Press, 2007).
Bertrand, C. et al. Tellus A 54, 221–244 (2002).
Pongratz, J. et al. Glob. Biogeochem. Cycles 23, GB4001 (2009).
Mann, M. E. et al. Science 326, 1256–1260 (2009).
Goosse, H. et al. J. Geophys. Res. doi:10.1029/2009JD012737 (in the press).
About this article
Atmospheric CO2over the last 1000 years: A high-resolution record from the West Antarctic Ice Sheet (WAIS) Divide ice core
Global Biogeochemical Cycles (2012)