The climate is changing, and so are aspects of the world's physical and biological systems. It is no easy matter to link cause and effect — the latest attack on the problem brings the power of meta-analysis to bear.
The article by Rosenzweig and colleagues1 that appears on page 353 of this issue is the first to formally link observed global changes in physical and biological systems to human-induced climate change, predominantly from increasing greenhouse gases. By surveying a huge literature, Rosenzweig et al. demonstrate that changes in physical and biological systems are pervasive; that these impacts lie mainly in directions consistent with warming of the climate system; and that, at least partly, they are likely to be the result of climate change caused by increasing concentrations of greenhouse gases.
The authors make the case using what is known as the 'joint attribution' approach2. They first show that the observed correspondence between impacts and warming would be very unlikely to occur if patterns of temperature change were the result of natural climate variability. They then argue that human influence has a role because observed large-scale climate change can be attributed to human influence on the climate system3.
These points emerge from Rosenzweig and colleagues' meta-analysis of the large literature on impacts, which involved synthesizing the results from studies of diverse types of system. This is probably the only available way of broadly linking impacts to climate change at global scales. Nevertheless, its very flexibility and comprehensiveness also impose limitations. For example, it would be difficult to quantify the climate–impact link with such an analysis, for two reasons.
The first is that the approach involves aggregation of results from vastly different types of system, both biological and physical (Fig. 1). The second is that only binary indicators of impacts are considered — that is, whether a given impact was consistent, or inconsistent, with warming. The link between these binary indicators and climate change is assessed by means of a technique involving a 'spatial pattern congruence' statistic, which assumes that the effects of local climate change occur locally. This measure will not fully capture connections where biological impacts result from remote climate changes, seasonal changes or changes in temperature extremes4. So the pattern of impacts does not correlate perfectly with that of annual mean warming (see Fig. 2 of the paper1 on page 355).
One of the challenges in this kind of research arises from the limitations of available data. Uncertainties result from limited and irregular sampling in space (impacts have been studied much more intensively in some regions, particularly Europe, than in others), and from the short time span of many data sets. The records mined by Rosenzweig et al. were primarily for the period 1970 to 2004, with the rule that at least 20 years of data were available. This is much shorter than the 50-year and sometimes 100-year data sets typically considered in studies detecting and attributing change in basic climate variables such as temperature3, surface pressure3 and precipitation5,6. Such long data sets are of course few and far between, particularly for effects of climate change. But the shortness of the records used in this research1 lessens the authors' ability to put changes in the context of previously observed variations, reducing the confidence in the results compared with those based on observations made over periods of a century or longer7,8.
That said, Rosenzweig and colleagues' analysis largely overcomes sampling limitations because of the sheer number of changes reported. Their synthesis is innovative and pushes detection-and-attribution research into a much broader domain than it has previously occupied. It is also a step along the road towards understanding how, and by how much, anthropogenic factors cause the observed impacts.
To estimate the size of the anthropogenic contribution, it will ultimately be necessary to undertake direct attribution of causes of change in affected systems, rather than using two-step joint attribution, in which some aspect of change in the climate system is first attributed to an external influence, and alteration in a physical or biological system is subsequently attributed to climate change. Direct attribution would require an 'end-to-end'9 modelling system that includes explicit representations of all of the main processes (climatic and non-climatic) that contribute to the variability of the system under study, and can simulate the response to greenhouse-gas increases as well as other factors that can cause changes in the observed impact. Such a tool can then be used to estimate how different external influences contribute to observed changes in systems relative to each other, much as models of the climate system are used to study the relative contributions of greenhouse gases, aerosols and natural climate variability to observed changes in surface air temperature and other basic climate variables3. Few such modelling systems are available, in part due to the difficulty of representing the relevant processes within climate models.
Likewise, only a few end-to-end attribution studies have been carried out1,2. They are limited to cases where the affected system and its interaction with climate are either relatively well understood10 or reasonably described empirically11. We need more of them. End-to-end studies will help in interpreting the results of less direct approaches to attribution. Moreover, they can also be used to evaluate and adjust projections of future impacts based on changes that have already been observed; that will be essential in formulating strategies to adapt to the consequences of climate change, and to assess their uncertainties, much as has been done with projections of future temperature change12. The ultimate goal is to provide probabilistic projections of future effects — that is, estimates of the probability that some outcome will or will not occur — and so allow decisions about adaptive measures to be based on a firmer footing.
Rosenzweig, C. et al. Nature 453, 353–357 (2008).
Rosenzweig, C. et al. in Climate Change 2007: Impacts, Adaptation and Vulnerability (eds Parry, M. L. et al.) 79–131 (Cambridge Univ. Press, 2007).
Hegerl, G. C. et al. in Climate Change 2007: The Physical Science Basis (eds Solomon, S. D. et al.) 663–745 (Cambridge Univ. Press, 2007).
Easterling, D. R. et al. Science, 289, 2068–2074 (2000).
Zhang, X. et al. Nature 448, 461–465 (2007).
Min, S.-K., Zhang, X. & Zwiers, F. Science 320, 518–520 (2008).
Parker, D. E., Legg, T. P. & Folland, C. K. Int. J. Clim. 12, 317–342 (1992).
Rutishauser, T., Luterbacher, J., Jeanneret, F., Pfister, C. & Wanner, H. J. Geophys. Res. 112, G04016 (2007).
Stone, D. A. & Allen, M. R. Climatic Change 71, 303–318 (2005).
Barnett, T. P. et al. Science 319, 1080–1083 (2008).
Gillett, N. P., Weaver, A. J., Zwiers, F. W. & Flannigan, M. D. Geophys. Res. Lett. 31, L18211, doi:10.1029/2004GL020876 (2004).
Meehl, G. A. et al. in Climate Change 2007: The Physical Science Basis (eds Solomon, S. D. et al.) 747–845 (Cambridge Univ. Press, 2007).
About this article
Journal of Time Series Analysis (2017)
Variation of Main Phenophases in Phenological Calendar in East China and Their Response to Climate Change
Advances in Meteorology (2016)
SSRN Electronic Journal (2015)
Simulating changes in the leaf unfolding time of 20 plant species in China over the twenty-first century
International Journal of Biometeorology (2014)
Global sensitivity analysis of an end-to-end marine ecosystem model of the North Sea: Factors affecting the biomass of fish and benthos
Ecological Modelling (2014)