Human Contribution to the European Heatwave of 2003
P.A. STOTT et al / Nature v.432, 2dec04
Peter A. Stott1, D. A. Stone2,3 & M. R. Allen2
1Met Office, Hadley Centre for Climate Prediction and Research (Reading Unit), Meteorology Building, University of Reading, Reading RG6 6BB, UK 2Department of Physics, University of Oxford, Oxford OX1 3PU, UK 3Department of Zoology, University of Oxford, Oxford OX1 3PS, UK
Letter To Nature
The summer of 2003 was probably the hottest in Europe since at latest AD 15001–4, and unusually large numbers of heat-related deaths were reported in France, Germany and Italy5. It is an ill-posed question whether the 2003 heatwave was caused, in a simple deterministic sense, by a modification of the external influences on climate—for example, increasing concentrations of greenhouse gases in the atmosphere—because almost any such weather event might have occurred by chance in an unmodified climate. However, it is possible to estimate by how much human activities may have increased the risk of the occurrence of such a heatwave6–8. Here we use this conceptual framework to estimate the contribution of human-induced increases in atmospheric concentrations of green-house gases and other pollutants to the risk of the occurrence of unusually high mean summer temperatures throughout a large region of continental Europe. Using a threshold for mean summer temperature that was exceeded in 2003, but in no other year since the start of the instrumental record in 1851, we estimate it is very likely (confidence level >90%)9 that human influence has at least doubled the risk of a heatwave exceeding this threshold magnitude.
Temperatures near the Earth’s surface are rising globally10, and evidence is mounting that most of the warming observed in recent decades has been caused by increasing atmospheric concentrations of greenhouse gases9,11,12. Anthropogenic increases in annual-mean temperatures have also been detected on continental scales, in Europe, North America and other land regions13–15. We first investigate the origins of long-term changes in decadal-mean European summer (June–August) temperatures, determining the changes attributable to anthropogenic drivers of the climate system and changes attributable to natural drivers. We then estimate how the risk of mean June–August temperatures exceeding a particular extreme threshold in any individual summer has changed as a result of this anthropogenic interference in the climate system.
Over the course of the twentieth century, June–August temperatures in Europe exhibited an overall increase, and a distinctive temporal pattern of temperature change, including cooling in the 1950s and 1960s (Fig. 1). We focus on the region bounded by 108 W and 408 E and 30–50° N (Fig. 1 inset), this being one of the regions chosen in previous studies13,16 to represent climatically coherent regions sufficiently large to exhibit climate change signals above the noise of natural internal variability. We use a pre-selected region in order to minimize any bias that could result from selecting our region already knowing where the most extreme temperatures occurred. Even in such a large domain, 2003 was the warmest summer on record. The history of temperature change averaged over this region is well reproduced by simulations of the HadCM3 climate model17, even at the model’s relatively low spatial resolution (3.758 longitude by 2.58 latitude), when driven with both anthropogenic and natural drivers of climate change (Fig. 1; see red, green, blue and turquoise lines). Four simulations (denoted ALL) were made with different initial conditions18, each with the same combination of well mixed greenhouse gases, sulphate aerosols and changes in tropospheric and stratospheric ozone, as well as natural changes in solar output and explosive volcanic eruptions12. A calculation of the temperature changes due to natural drivers alone (obtained by combining a simulation with solar forcing and a simulation with volcanic forcing, and denoted NAT) shows no warming in the latter part of the century (Fig. 1, yellow line).
Figure 1 June–August temperature anomalies (relative to 1961–90 mean, in K) over the region shown in inset. Shown are observed temperatures (black line, with low-pass-filtered temperatures as heavy black line), modelled temperatures from four HadCM3 simulations including both anthropogenic and natural forcings to 2000 (red, green, blue and turquoise lines), and estimated HadCM3 response to purely natural natural forcings (yellow line). The observed 2003 temperature is shown as a star. Also shown (red, green and blue lines) are three simulations (initialized in 1989) including changes in greenhouse gas and sulphur emissions according to the SRES A2 scenario to 210022. The inset shows observed summer 2003 temperature anomalies, in K.
Figure 2 Power spectra of European mean summer temperature. Solid line, observed spectrum after removing an independent estimate of the externally forced response provided by the ensemble mean of the ALL simulations. Shaded region, 5 to 95 percentile region of the estimated range of spectra of natural internal variability estimated from segments taken from the HadCM3 control run of the same length as the observations. Model spectral densities have been inflated by a factor of 1.25 to allow for sampling uncertainty in the ALL ensemble mean.
A necessary requirement for any detection of significant warming is that HadCM3 adequately represents the natural internal variability of European summer temperatures. A quantitative test, in which an estimate of the forced response calculated from the ALL ensemble mean is subtracted from the observations and the residual compared with HadCM3 simulated internal variability, shows no significant discrepancy in variance on either interdecadal or 2- to 10-year timescales (Fig. 2). If anything, the model appears to be overestimating observed variability in this season and region, making our estimates of attributable risk relatively conservative. We also find no evidence of a secular change in variance on subdecadal timescales in either model or observations: over the twenty-first century, the standard deviation of the forced simulations under the SRES A2 scenario (Fig. 1), when a second-order polynomial trend has been removed, increases by an insignificant 0.01 K from the first 40 years (1990–2030) to the last 40 years (2060–2100). This contrasts with the conclusions of ref. 2, although their results applied to a much smaller region.
We now apply a standard optimal detection analysis to European summer temperatures, similar to those applied to global scale patterns of temperature change11,12,19. The analysis is a regression between decadal-mean seasonal-mean observations over 1920–99 and simulated temperature changes over the same period from both anthropogenic and natural forcings (ALL) and from natural forcings alone (NAT), and is in most respects identical to that of ref. 13 (see Methods). Figure 3a shows the estimated likelihood functions for the factors by which we could scale up or down the amplitudes of the model-simulated responses to anthropogenic forcing (red curve) and natural forcing (green curve) while remaining consistent with the observations. As the fifth percentile of the scaling factor on the anthropogenic response (red curve in Fig. 3a) is greater than zero, an anthropogenic influence on decadal-mean European summer temperature is detected at the 5% significance level (that is, the hypothesis that there is no positive anthropogenic influence can be rejected at the 5% level). We conclude from this investigation of decadal-mean summer temperatures that it is very likely that past anthropogenic forcing is responsible for a significant fraction of the observed European summer warming.
We now calculate the changed risk of extremely hot summers that can be attributed to past anthropogenic emissions, allowing for uncertainty in the anthropogenic warming. Averaged over the region of interest (Fig. 1 inset), summer temperatures in 2003 exceeded the 1961–90 mean by 2.3 K (Fig. 1, black star). To quantify changes in risk, we need an objective definition of the event in question. Using 2.3 K itself is problematic for three reasons: first, relying too closely on the details of what actually occurred when defining what we are looking for introduces a selection bias in our attribution procedure; second, temperature anomalies in 2003 may have been amplified by soil-moisture feedbacks2 or interactions with the North Atlantic20, both of which may be under-estimated, although we do observe similar magnitude spikes in model summer temperatures in Fig. 1; third, given the length of model-simulated variability we have available, inferring the statistics of temperature excursions over 2 K requires extrapolation of extreme value distributions, which introduces further uncertainties. 2003 was the first year to reach or exceed a threshold of 1.6 K (2001 being the second-warmest European summer, at 1.5 K). We therefore consider how the probability of exceeding this threshold has changed, by comparing this estimated late-twentieth-century probability with the estimated probability of exceeding the same absolute threshold if there had been no anthropogenic influence on climate. Increasing this threshold to any value up to 2.3 K strengthens our conclusions regarding attributable risk; hence using a threshold that only just exceeds the second warmest summer is relatively conservative.
Assuming that subdecadal continental-scale variability is stationary and adequately represented by HadCM3, we estimate possible distributions of temperatures in individual summers in the presence and absence of anthropogenic influence. We do this by adding HadCM3 control variability to reconstructions of 1990s decadal-mean temperatures both with all external factors included, and with anthropogenic factors removed, allowing for uncertainty in these decadal-mean temperatures from the detection analysis (Fig. 3b). Figure 4a shows the estimated likelihood of the risk (probability) of exceeding a 1.6 K threshold in the presence of anthropogenic climate change (red line) and in the absence of anthropogenic change (green line), expressed both as a frequency (number of occurrences per thousand years, bottom axis) and as a return period (top axis). The clear shift from the green to the red distribution implies that an appreciable fraction of the risk of such hot summers can be attributed to human influence on climate. Even in the presence of anthropogenic warming, we conclude that the estimated probability of exceeding 1.6 K appears to be low (best estimate is a 1 in 250 year event (Fig. 4a, red curve) but this risk may be increasing rapidly).
The fraction attributable risk (FAR) is estimated in Fig. 4b (see Methods). In certain circumstances, the figure of relevance in establishing possible liability for compensation has been FAR ¼ 0.5, corresponding to a doubling of risk over natural conditions21 (meaning that one event in two would have happened naturally). According to our calculation, there is a greater than 90% chance that over half the risk of European summer temperatures exceeding a threshold of 1.6 K is attributable to human influence on climate. Although there is a large spread, reflecting the remaining uncertainties in the effects of climate change on this spatial scale, the anthropogenic FAR could be substantially greater than 0.5. Also marked on Fig. 4b is a vertical line representing an overall ‘best estimate’ of the human contribution to the increased risk of these very hot European summers7, given the information that we have available at present. On this basis, human influence is to blame for 75% of the increased risk of such a heatwave.
Our analysis shows that European summers are warming owing to anthropogenic climate change. Under unmitigated emissions scenarios, summers like 2003 are likely to be experienced more frequently in future; HadCM3 projections (Fig. 1) indicate that the probability of European mean summer temperatures exceeding those of 2003 increases rapidly under the SRES A2 scenario22, with more than half of years warmer than 2003 by the 2040s. By the end of this century, Fig. 1 shows that 2003 would be classed as an anomalously cold summer relative to the new climate, for the scenario and model under consideration.
Figure 3 Estimated likelihood functions for anthropogenic and natural contributions to European summer temperature changes. The curves show estimated distributions of anthropogenic (red) and natural (green) scaling factors on model-simulated responses (a). 1990s summer temperatures (relative to pre-industrial climate) including all external drivers of climate change (red) and with anthropogenic drivers removed (green) (b). A scaling factor of zero (horizontal solid line in a) implies no contribution to observed 1990s temperatures from this driver, while unity (horizontal dashed line in a) implies no systematic under- or over-estimate by the model of the observed response to this driver. The width of these distributions reflects the uncertainties for these probabilities.
Figure 4 Change in risk of mean European summer temperatures exceeding the 1.6 K threshold. a, Histograms of instantaneous return periods under late-twentieth-century conditions in the absence of anthropogenic climate change (green line) and with anthropogenic climate change (red line). b, Fraction attributable risk (FAR). Also shown, as the vertical line, is the ‘best estimate’ FAR, the mean risk attributable to anthropogenic factors averaged over the distribution.
We may have underestimated the FAR of a heatwave in 2003 by using 1990s figures for attributable background warming. Conversely, the FAR may have been overestimated by selecting an area that has recently been subject to extreme temperatures, although this effect should be alleviated by our use of an independently specified region. Modelling and forcing uncertainties, and any errors in the variability characteristics of the model, could mean that our assessment of the likelihood of the FAR exceeding 0.5 is in error. Including a different combination of the available natural HadCM3 simulations changes the results quoted by less than 5%, but a systematic exploration of both modelling and forcing uncertainty would require a very large multi-model ensemble, varying model parameters across their range of possible values and exploring the range of potential forcing estimates23,24.
Anthropogenic warming trends in Europe imply an increased probability of very hot summers. Given the effects of the 2003 heatwave, this suggests a greater risk of associated adverse impacts, although to make a quantitative attribution assessment of any specific social, economic or ecological impact will require detailed modelling of both local meteorological conditions and their relationships with the impact in question. Nevertheless, it seems likely that past human influence has more than doubled the risk of European mean summer temperatures as hot as 2003, and with the likelihood of such events projected to increase 100-fold over the next four decades, it is difficult to avoid the conclusion that potentially dangerous anthropogenic interference in the climate system is already underway. The FAR provides a potential measure for redistributing the associated costs of such extreme events21.
Attribution of decadal-mean seasonal-mean changes
In a natural extension of previous work12,13,25, observed decadal-mean summer near-surface temperature changes, y, over land for 1920–99 are constructed from monthly-mean values extracted from the CRUTEM2(v) data set10 for the region 108 W-408 E, 30–508 N, requiring data for at least two out of three months (June–August) in at least half the years. They are regressed against the mean response over the observational data-mask of a four-member ensemble of HadCM3 simulations driven with both anthropogenic and natural forcings (x1, ALL, red, green, blue and turquoise lines in Fig. 1) and a second set of simulations of the response to solar and volcanic forcing alone (x2, NAT, yellow line in Fig. 1) plus noise:
y = (x1 - v1) ß1 + (x2 - v2) ß1 + v0 (1)
where 01, 02 are the unknown scaling factors to be estimated in the regression, v0 is the noise in the observations, and v 1 and v2 account for sampling uncertainty introduced by estimating model-simulated responses from a finite ensemble26. A scaling factor of 1 would imply that the response in the model simulations is identical to the observed changes, while a factor of 0 implies no correspondence between modelled and observed responses. Likelihood functions for scaling factors obtained from the regression account only for sampling variability that arises from natural internal climate variability as represented by HadCM3. They do not account for systematic errors in the shapes of the modelled patterns or uncertainties due to missing forcings (for example, land use changes or fossil-fuel black carbon emissions). If we assume a uniform prior in these scaling factors, these likelihoods can be thought of as probability distributions.
The combined effects of solar and volcanic forcings are estimated from a simulation with volcanic forcing amplified by a factor of 10 and solar forcing amplified by a factor of 5, multiplied by 0.1 and 0.2, respectively, and added. This approach was taken in order to obtain the clearest possible climate signal from a weak forcing and a limited number of climate simulations. Previous work showed that the large-scale temperature responds linearly to increasing these forcings27. Including an additional four simulations with unamplified natural forcings in the analysis makes little difference to the results.
We use the HadCM3 control run and intra-ensemble variability to estimate internal variability, v0, v 1 and v2, taking into account the enhanced signal-to-noise from averaging over the ALL ensemble and inflating the solar and volcanic forcing13. This gives scaling factors for the combined effects of all forcings and for natural forcings, from which the anthropogenic and natural scaling factors can be calculated by linear transformation27 (Fig. 3a). Where the 5 to 95 percentile uncertainty range does not include zero scaling factor, this indicates that the relevant signal has been detected at the 5% significance level, that is, there is less than 5% chance that the relevant forcing has had no effect. From the scaling factors and their uncertainties, we derive likelihood functions of 1990s temperature anomalies attributable to the combined effects of all forcings and attributable to natural factors alone (Fig. 3b).
Attribution of change in risk of exceeding a threshold
The distribution of attributed anthropogenic decadal-mean warming in the 1990s is used to calculate the FAR for an individual European summer exceeding the 1.6 K threshold as follows. We estimate the probability of exceeding the 1.6 K threshold from the generalized Pareto distribution (GPD) fitted to the HadCM3 control run temperatures28. The spectral analysis shown in Fig. 2 indicates that HadCM3 provides an adequate representation of internal variability. We calculate the probability, P0, of the 1.6 K threshold being exceeded without anthropogenic climate change (shown by the green curve in Fig. 4a, expressed as return period, top axis, and instantaneous risk: rate of occurrence per 103 yr, bottom axis) and the probability, P1, of exceeding the threshold with anthropogenic climate change (red curve in Fig. 4a). In estimating the green curve, the chance of exceeding the 1.6 K threshold is re-estimated using the GPD with the mean summer temperature adjusted to each percentile of the estimated distribution of decadal-mean temperature anomalies attributable to natural forcing and internal variability. For the red curve, the response to anthropogenic forcing is also included. A bootstrap resampling method is used to allow for uncertainty in GPD parameters. From these distributions, we calculate the distribution of attributable increase in risk (P1/P0) and the FAR ((P1 - P0)/P1)8, as shown in Fig. 4b, whose spread reflects uncertainty in the magnitude of the forced decadal-mean response and in the estimation of the extreme value distribution parameters due to limited sample size.
Received 21 May; accepted 5 October 2004; doi:10.1038/nature03089.
Luterbacher, J., Dietrich, D., Xoplaki, E., Grosjean, M. & Wanner, H. European seasonal and annual temperature variability, trends, and extremes since 1500. Science 303, 1499–1503 (2004).
Schär, C. et al. The role of increasing temperature variability in European summer heatwaves. Nature 427, 332–336 (2004).
Beniston, M. The 2003 heat wave in Europe: A shape of things to come? An analysis based on Swiss climatological data and model simulations. Geophys. Res. Lett. 31, doi: 10.1029/2003GL018857 (2004).
Black, E., Blackburn, M., Harrison, G. & Methven, J. Factors contributing to the summer 2003 European heatwave. Weather 59, 217–223 (2004).
Institut de Veille Sanitaire. Impact sanitaire de la vague chalaire d’aout 2003 en France. Bilan et perspectives. http://www.invs.sante.fr/publications/2003/bilan-chaleur-1103l (2003).
Palmer, T. N. & Räisänen, J. Quantifying the risk of extreme seasonal precipitation events in a changing climate. Nature 415, 512–514 (2002).
Allen, M. R. Liability for climate change. Nature 421, 891–892 (2003).
Stone, D. A. & Allen, M. R. The end-to-end attribution problem: From emissions to impacts. Clim. Change (in the press).
IPCC Climate Change 2001: The Scientific Basis: Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change (eds Houghton, J. T. et al.) (Cambridge Univ. Press, Cambridge, UK, 2001).
- Jones, P. D. & Moberg, A. Hemispheric and large-scale surface air temperature variations: An extensive revision and an update to 2001. J. Clim. 16, 206–223 (2003).
- Stott, P. A. et al. Attribution of twentieth century temperature change to natural and anthropogenic causes. Clim. Dyn. 17, 1–21 (2001).
- Tett, S. F. B. et al. Estimation of natural and anthropogenic contributions to 20th century temperature change. J. Geophys. Res. 107, doi: 10.1029/2000JD000028 (2002).
- Stott, P. A. Attribution of regional-scale temperature changes to anthropogenic and natural causes. Geophys. Res. Lett. 30, doi: 10.1029/2003GL017324 (2003).
- Karoly, D. J. et al. Detection of a human influence on North American climate. Science 302, 1200–1203 (2003).
- Zwiers, F. W. & Zhang, X. Towards regional scale climate change detection. J. Clim. 16, 793–797 (2003).
- Giorgi, F. Variability and trends of sub-continental scale surface climate in the 20th century. Part I: observations. Clim. Dyn. 18, 675–691 (2002).
- Johns, T. C. et al. Anthropogenic climate change for 1860 to 2100 simulated with the HadCM3 model under updated emissions scenarios. Clim. Dyn. 20, 583–612 (2003).
- Stott, P. A. et al. External control of 20th century temperature by natural and anthropogenic forcings. Science 290, 2133–2137 (2000).
- Tett, S. F. B., Stott, P. A., Allen, M. R., Ingram, W. J. & Mitchell, J. F. B. Causes of twentieth century temperature change near the Earth’s surface. Nature 399, 569–572 (1999).
- Colman, A. & Davey, M. Prediction of summer temperature, rainfall and pressure in Europe from preceding winter North Atlantic ocean temperature. Int. J. Climatol. 19, 513–536 (1999).
- Grossman, D. A. Warming up to a not-so-radical idea: Tort-based climate change litigation. Colombia J. Environ. Law 28, 1–61 (2003).
- Nakicenovic, N. & Swart, R. Special Report on Emission Scenarios (Cambridge Univ. Press, Cambridge, UK, 2000).
- Allen, M. R. & Stainforth, D. A. Towards objective probabilistic climate forecasting. Nature 419, 228 (2002).
- Murphy, J. M. et al. Quantification of modelling uncertainties in a large ensemble of climate change simulations. Nature 430, 768–772 (2004).
- Allen, M. R. & Tett, S. F. B. Checking for model consistency in optimal fingerprinting. Clim. Dyn. 15, 419–434 (1999).
- Allen, M. R. & Stott, P. A. Estimating signal amplitudes in optimal fingerprinting, Part I: theory. Clim. Dyn. 21, 477–491 (2003).
- Stott, P. A., Jones, G. S. & Mitchell, J. F. B. Do models underestimate the solar contribution to recent climate change? J. Clim. 16, 4079–4093 (2003).
- Coles, S. An Introduction to Statistical Modeling of Extreme Values (Springer, London, 2001).
Acknowledgements P.A.S. was supported by the Department for Environment, Food and Rural Affairs, D.A.S. by a Wellcome Trust Showcase Award, and M.R.A. received partial support from the NOAA/DoE International Detection and Attribution Group.
Competing interests statement The authors declare that they have no competing financial interests.
Correspondence and requests for materials should be addressed to P.A.S. (firstname.lastname@example.org).