Projections of rising greenhouse gas concentrations in the atmosphere, their climatic consequences, and their social and environmental effects are becoming increasingly sophisticated1. Nevertheless, these studies invariably highlight the need for further improvements in the data and models, and in their analysis. Such is the case with the paper by Hulme and colleagues on page 688 of this issue2, which nonetheless constitutes a considerable advance.
The authors have combined results from simulations of changing climate and its environmental effects to look at Europe in the year 2050. They ask whether the impacts of human-induced climate change, through greenhouse gas emissions, can be distinguished from those due to natural climate variability. The two indicators they use are river flow and wheat yield. Both are affected by temperature and rainfall.
Climate change modelling has come a long way since the early days. In the 1970s, most climate change scenarios and impact analyses were based on estimated global average warmings and patterns derived from palaeoclimatic and other analogues3. Then came studies based on atmospheric general circulation models (GCMs) of coarse horizontal resolution, bounded at the surface by a simple ‘slab-layer’ ocean. These were run for changed atmospheric composition, usually for equilibrium doubled CO2 concentrations. Changes at times earlier than that of doubled CO2 were obtained by linear scaling. Due to computing limitations, the equilibrated control (present climate) and doubled CO2 climate simulations were usually run only over a period of one or two decades4.
During the late 1980s, with the aid of more powerful computers, climate modellers progressed to simulations with fully coupled ocean-atmosphere GCMs which explicitly model ocean dynamics and oceanic heat uptake5. These enable simulations with continuous variations of greenhouse gas concentrations from the present or earlier (preindustrial) conditions, following some emission scenario path to 2100 or beyond. Even so, many impact studies have relied on analyses of only a few decades of output from the longer control and enhanced greenhouse simulations. A severe problem with these studies is that of ‘natural’ multi-decadal variations in the simulated control climates, and indeed in the simulated changing climates. Such variations may dominate the analyses.
Hulme et al.2 help to put this problem in perspective by attempting to quantify the impacts of human-induced climate change relative to those of natural climate variability. They use a multi-century simulation of the control climate, run by the Hadley Centre in Britain, to extract seven more-or-less independent ‘30-year climates’, and thus to provide an estimate of the simulated natural variability in climate. They also carried out multiple simulations, or ‘ensemble runs’, for changed climate under projections of 0.5% and 1% annual increases in equivalent CO2 concentrations (for comparison, the observed rate of increase over the past decade, while fluctuating, is nearer to 0.5%).
Hulme et al. used their estimates of monthly climate change from ensemble means from four simulations, interpolated to a 0.5° latitude/longitude grid over Europe, to estimate changes in river flow and wheat yield using two climate-impact models. They then compared the simulated impacts due to human-induced climate change with the maximum simulated impacts due to the estimated natural variability. For around 2050, they find that climate change scenarios show systematic increases in river flow in northern Europe, and decreases in southern Europe, both being statistically significant relative to natural variability. But central and western Europe show smaller and non-significant changes. The spatial patterns in the climate change cases are far more consistent than in the climate variability cases.
Wheat yield in Europe is highly sensitive to natural variability in temperature and rainfall. In these simulations, climate change alone by 2050 results in significant increases in yield under both scenarios only in three northern European countries. But adding estimated effects of higher CO2 concentrations on plant growth increases the yields considerably, with yield increases now being significant in all of the ten countries included in the study. Whether such increases due to higher CO2 concentration would actually be realized is uncertain, because of problems in extrapolating from idealized laboratory and modelling results to field conditions6.
The clear message of this work is that greater efforts are needed to take account of the ‘noise’ of natural climate variability when considering the ‘signal’ of climate change. In particular, care must be taken in interpreting the results of impact assessments, especially those which use climate simulations over short periods only.
Much more remains to be done. For example, the use of simulated monthly climate data by Hulme et al.2, even when interpolated to daily data by statistical methods, leaves open the possibility of systematic changes in day-to-day variability and extremes due to climate change7,8. These could well dominate the effects of climate change on both river flow and wheat yield. Taking them into account will require the analysis of daily output from the climate models, especially at fine spatial resolution.
More importantly, in understanding the message from Hulme et al., we need to distinguish between confidence levels used for the detectability of simulated or real impacts due to climate change, and those needed to decide if the information about possible future impacts is useful for decision-makers. Detectability, used as evidence of climate change, arguably requires a high level of confidence, say at the 95% or 99% level. On the other hand, if we take it as a given that climate change is occurring, and we are seeking useful advice as to what the consequences might be, a lower level of confidence may suffice.
Studies of the length of a data set necessary to detect a real change in the mean against a background of variability, for example for detection of trends in total ozone9 or river flow10, indicate that even if the change is real, it may need to have happened (and thus will have had impacts) for decades before it can be shown statistically to have occurred. In cases such as estimation of flood frequency and of the occurrence of moisture stress in crops, what is needed to identify future ‘dangerous’ levels of climate change, or to plan adaptation measures, is an estimate of changes in the probability of such effects before they happen. This is consistent with the ‘precautionary principle’ embodied in the UN Framework Convention for Climate Change. Given the continuing uncertainties in estimating future climate changes and impacts, a risk assessment approach is essential11.
Pittock, A. B. in Modelling Change in Environmental Systems (eds Jakeman, A. J., Beck, M. B. & McAleer, M. J.) Ch. 20 (Wiley, Chichester, 1993).
Hulme, M. et al. Nature 397, 688–691 (1999).
Pittock, A. B. & Salinger, M. J. Clim. Change 4, 23–40 (1982).
Mitchell, J. F. B., Manabe, S., Melieshko, V. & Tokioka, T. in Climate Change: The IPCC Scientific Assessment (eds Houghton, J. T., Jenkins, G. J. & Ephraums, J. J.) 131-172 (Cambridge Univ. Press, 1990).
Kattenberg, A. et al. in Climate Change 1995: The Science of Climate Change (eds Houghton, J. T. et al.) 285-357 (Cambridge Univ. Press, 1996).
Pittock, A. B. Environment 37 (9), 25-30(1995)
Hennessy, K. J., Gregory, J. M. & Mitchell, J. F. B. Clim. Dyn. 13, 667– 680 (1997).
Hennessy, K. J. & Pittock, A. B. Int. J. Climatol. 15, 591–612 ( 1995).
Pittock, A. B. Pure Appl. Geophys. 118, 643–661 (1980).
Chiew, F. H. S. & McMahon, T. A. Int. J. Climatol. 13, 643–653 ( 1993).
Pittock, A. B. & Jones, R. N. Environ. Monitor. Assess. J. (in the press).
About this article
Projected changes in drought occurrence under future global warming from multi-model, multi-scenario, IPCC AR4 simulations
Climate Dynamics (2008)
Climatic Change (2005)
Biological Conservation (2004)
Global and Planetary Change (2003)