The Heat Is Online

Causes of 20th Century Temperature Change Near Earth's Surface

Nature 399, 569-572 10 June 1999


* Hadley Centre for Climate Prediction and Research, Meteorological Office, London Road, Bracknell, Berkshire RG12 2SY, UK
† Space Science Department, Rutherford Appleton Laboratory, Chilton, OX11 0QX, UK and Department of Physics, Univeristy of Oxford

Observations of the Earth's near-surface temperature show a global-mean temperature increase of approximately 0.6 K since 1900 (ref. 1), occurring from 1910 to 1940 and from 1970 to the present. The temperature change over the past 30-50 years is unlikely to be entirely due to internal climate variability2,3,4 and has been attributed to changes in the concentrations of greenhouse gases and sulphate aerosols5 due to human activity. Attribution of the warming early in the century has proved more elusive. Here we present a quantification of the possible contributions throughout the century from the four components most likely to be responsible for the large-scale temperature changes, of which two vary naturally (solar irradiance and stratospheric volcanic aerosols) and two have changed decisively due to anthropogenic influence (greenhouse gases and sulphate aerosols). The patterns of time/space changes in near-surface temperature due to the separate forcing components are simulated with a coupled atmosphere-ocean general circulation model, and a linear combination of these is fitted to observations. Thus our analysis is insensitive to errors in the simulated amplitude of these responses. We find that solar forcing may have contributed to the temperature changes early in the century, but anthropogenic causes combined with natural variability would also present a possible explanation. For the warming from 1946 to 1996 regardless of any possible amplification of solar or volcanic influence, we exclude purely natural forcing, and attribute it largely to the anthropogenic components.

The coupled model we use is HadCM2 (refs 6, 7), which has a horizontal resolution of 2.5° in latitude by 3.75° in longitude, 19 atmospheric and 20 oceanic levels, and a flux correction. Its climate sensitivity to a doubling of the atmospheric concentration of CO2 is estimated to be 3.3 K (C. A. Senior, personal communication).

The main radiative forcings of climate since 1850 are likely to be anthropogenic changes in well-mixed greenhouse gases and tropospheric aerosols (mainly sulphate), and natural changes in solar irradiance and in stratospheric aerosol due to volcanic activity8. We compare observations1 of 10-year mean near-surface temperature changes over five 50-year periods (1906-56, 1916-66, ..., 1946-96) with simulations of HadCM2 forced by the following factors (see also section 1 of Supplementary Information):

'G'. Changes in well-mixed greenhouse gases from 1860 to 19968,9,10 expressed as equivalent CO2.

'GS'. As G but also with changes in surface albedo11 representing the effects of anthropogenic sulphate aerosols from 1860 to 19966,9,10 derived from an atmospheric chemistry model12. We assume that this albedo represents both the direct and indirect11,13,14 effects of sulphate aerosols. When we consider both G and GS we define a further signal S, the pure sulphate signal, as GS - G (see section 9 of Supplementary Information).

'Vol'. Changes in stratospheric volcanic aerosols15 from 1850 to 1996.

'Sol'. Changes in total solar irradiance from 1700 to 1996 based on proxy data16 for 1700-1991 and extended to 1996 using satellite observations17.

The climate response to each of the above factors was computed from the ensemble average of four simulations started from different initial conditions. Internal climate variability is estimated from a 1,700-year control simulation (Control). 50-year segments (of 10-year mean near-surface temperatures) from the responses of the four ensembles (signals), Control and observations were processed identically to allow for the effect of changing observational coverage, to filter out variability on scales less than 5,000 km (ref. 18) and to remove the 50-year time-mean (see section 2 of Supplementary Information).

We assume that the observations (y) may be represented as the sum of simulated responses or signals (xi) and internal climatic variability (u), which we assume to be normally distributed: This assumption of linearity has been shown to be a good approximation in at least one general circulation model19,20. The amplitude, represents the amount by which we have to scale the ith signal to give the best fit to the observations. We use an "optimal fingerprinting" algorithm21,22,23,24, a form of multivariate regression, to estimate these amplitudes and uncertainty ranges (see section 3 of Supplementary Information). Here we use 5-95% uncertainty ranges and if the uncertainty range for includes unity, then the amplitude of the simulated signal (xi) could be correct. The first millennium of Control is used to define a weighting function which minimizes the influence of patterns of high internal variability on the estimated signal amplitudes (giving "optimal fingerprints"). Observations and all simulated data are then further filtered by projection on to the leading ten modes of spatio-temporal variability from this part of Control (see section 7 of Supplementary Information). The latter seven centuries, which are statistically independent, are used to estimate the uncertainty ranges.

A minimum requirement for a signal or signal-combination to explain temperature changes over a 50-year period is that the amplitude of the estimated residual (xi - y) should be consistent, at the 5% level, with internal climate variability24 estimated from Control. For a signal-combination to explain change over the century we require that the estimated residual be consistent in all five 50-year periods. We test this using a F-test (see sections 4 and 6 of Supplementary Information) which rules out simulated internal variability as an explanation of temperature changes during 1906-56, 1916-66 and 1946-96 (Table 1), despite being a relatively weak test in that it makes no use of the shape of the simulated signals. Furthermore, no combination of our simulated natural signals alone can explain the warming since 1946--the period when the observational coverage is best. However, every case that includes an anthropogenic signal passes the consistency test. These results hold even if the global-mean model sensitivity or the amplitude of the forcing is wrong as we place no restriction on the amplitudes of the simulated signals.

A risk in multivariate linear regression is signal degeneracy, which means that signals resemble linear combinations of one another. Applying tests for degeneracy (see pages 243-248 of ref. 25 and section 8 of Supplementary Information) to our signals and observations, we find that no more than two signal-amplitudes can be estimated simultaneously. While three-signal and four-signal combinations are possible explanations, results from them may be misleading. Thus we only discuss combinations of at most two signals.

A signal is formally detected in a signal-combination and period if that combination is consistent with observations and the uncertainty range for the signal is entirely positive (see sections 5 and 6 of Supplementary Information). This means that we have rejected the null hypothesis that the signal amplitude is zero (or that the simulated signal is of the wrong sign). In other words, it is highly likely that the signal is present in the observations.

All one-or two-signal combinations that include anthropogenic signals are, on the basis of the consistency test, possible explanations of twentieth-century temperature change, but we wish to focus on the 'best' explanation. Ideally, this explanation would include all detectable signals and no others, provided that this combination is consistent with the observations. We find that three signals (G, S and Sol) are detected during the century (Table 1). The composite signal GS (in which the relative amplitudes of the greenhouse and sulphate signals are assumed to be as in the GS experiment, giving a single pattern of response to anthropogenic forcing) can also be detected. As signal degeneracy may make results from the three-signal combination, G&S&Sol, misleading, this leaves three two-signal combinations: G&S, GS&Sol and G&Sol. Before the most recent period, none of these three two-signal combinations are an obviously better fit to the observations than any other (Table 1). But in 1946-96, combinations including the influence of sulphate aerosols (S) fit the observations better than other explanations (Table 1), which all fail to pass the F-test at the 10% level, suggesting that some sulphate influence is required to account for recent changes. Thus we consider the two combinations containing S (G&S and GS&Sol), but we explore the sensitivity of the solar detection during 1906-56 to the size of the sulphate signal included in GS.

In G&S (unlike GS), the observations are allowed to determine the relative amplitude of the sulphate and greenhouse-gas signals. This allows for error in the prescribed sulphate forcing or modelled response which varies with each period. With these two signals we detect S in 1946-96 and G in 1906-56, 1916-66 and 1946-96 (Table 1). We find that during 1906-56 the amplitude of the sulphate aerosol signal relative to the greenhouse-gas signal is significantly less than that prescribed in GS.

Some other factor may be required to explain this discrepancy with GS rather than errors in the simulated anthropogenic signals. If the relative amplitude of the anthropogenic signals is as prescribed in the GS experiment, then Sol is detected: during 1906-56, some solar influence is required to explain the temperature changes (Table 1). If we reduce the amplitude of the sulphate signal in GS by 33% or more (making GS more like G), we detect the anthropogenic signal rather than the solar signal during 1906-56. GS is detected in 1946-96 with or without this change. The reduction of the relative amplitude of sulphate aerosols to greenhouse gases is within the possible uncertainty range8; thus our detection of a solar signal should be treated with some caution.

We reconstruct global-mean temperature changes by multiplying the simulated signals, xi by the best-fit amplitudes for 1906-56 and 1946-96. During 1906-56, in G&S, greenhouse gases warm, counterbalanced by a weaker aerosol effect than was imposed in GS (Fig. 1a). The warming peak around 1940 is accounted for by a combination of internal variability and a steadily increasing temperature due to anthropogenic forcings. In GS&Sol the observed early-century warming is explained by a combination of solar irradiance changes, internal variability and an increase in greenhouse gases (partly balanced by the radiative effect of sulphate aerosols), with solar activity largely responsible for the warming peak around 1940 (Fig. 1b). In both cases, from 1946 sulphate aerosols balance the effect of greenhouse gases giving little warming until the mid-1970s when the warming due to increasing greenhouse gases dominates (Fig. 1c, d).

In GS&Sol, solar influence is detectable during the 1906-56 period (Fig. 2), but contributes little to the linear trend in global-mean temperature (a best estimate of 0.025 K per decade). Conversely, G&S shows that greenhouse gases could have contributed substantially to the warming during both the 1906-56 and 1916-66 periods (Fig. 2). In all other periods, the best-estimate solar contribution is negligible. For both cases the best estimate of the total anthropogenic warming is 0.075 K per decade during 1946-96, within an approximate uncertainty range of 0.03-0.11 K per decade.

To test the robustness of our results, we carried out six sensitivity studies in which we changed our analysis procedure (see section 10 of Supplementary Information). Detection of anthropogenic signals during 1906-56 (in G&S) and 1946-96 (in both cases) was unaltered by those changes. Detection of the 1946-96 sulphate signal is significant only at the 90% level if optimization is not used. Detection of the 1906-56 solar signal is significant at the 90% level only if greater weight is given to smaller spatial scales, but remains significant at the 95% level in all other studies. We also used a different solar irradiance reconstruction26 from 1750, to produce an alternative Sol signal, and we reduced stratospheric ozone from 19744 to modify the GS signal. Our principal results are robust to both these changes. Furthermore, if we inflate the variance of Control by a factor of 4 we still detect an anthropogenic signal in the 1946-96 period in both of the two-signal cases considered.

Our simulations did not explicitly represent the effects of sulphate aerosols on cloud albedo13 or lifetime14 (though on the scales considered these are likely to be represented by the albedo changes imposed on the model), nor did we consider the climate effects of other aerosols, changes in the spectral distribution of the solar irradiance and possible associated changes in ozone27,28. Although our results are unaffected by error in the amplitude of the forcing, they could be sensitive to error in the patterns of response or other errors in the forcing, though our conclusions were unaltered by use of an alternative solar irradiance reconstruction. The test that we use to evaluate consistency between the observations and simulations could be misled by mutually compensating errors in model estimates of natural variability and the response of the model to external factors. We do not consider possible observational errors (likely to be small relative to internal climate variability) nor has our uncertainty analysis included any uncertainty due to the finite length of data used to derive the optimisation. For consistency with earlier work21,22,23,24, we have used standard estimates of pattern amplitudes based on linear regression which are biased towards zero25 when, as here, there is uncertainty in the signals.

Bearing in mind these caveats, we interpret our results as showing the following: first, the temperature changes over the twentieth century cannot be explained by any combination of natural internal variability and the response to natural forcings alone. Second, the recent warming, 0.25 K, can be explained by the response of the climate to anthropogenic changes in greenhouse-gas concentrations partly offset by cooling due to anthropogenic sulphate aerosols, resulting in little net temperature change from 1946 to the mid-1970s. Last, the warming early this century can also be explained by anthropogenic causes and internal variability. However, solar irradiance changes could have made a significant contribution, 0.125 K, if we assume little error in the relative amplitude of the forcing of sulphates and greenhouse gases prescribed in our model.

Supplementary information is available on Nature's World-Wide Web site ( or as paper copy from the London editorial office of Nature. Received 5 October 1998;accepted 25 March 1999.


  1. Parker, D. E., Jones, P. D., Folland, C. K. & Bevan, A. Interdecadal changes of surface temperature since the late nineteenth century. J. Geophys. Res. 99, 14373-14399 (1994). Links Save Citation
  2. Stouffer, R. J., Manabe, S. & Vinnikov, K. Y. Model assessment of the role of natural variability in recent global warming. Nature 367, 634-636 (1994). Links Save Citation
  3. Santer, B. D. et al. A search for human influences on the thermal structure of the atmosphere. Nature 382, 39-45 (1996). Links Save Citation
  4. Tett, S. F. B., Mitchell, J. F. B., Parker, D. E. & Allen, M. R. Human influence on the atmospheric vertical temperature structure: Detection and observations. Science 247, 1170-1173 (1996). Links Save Citation
  5. Hegerl, G. C. et al. Multi-fingerprint detection and attribution analysis of greenhouse gas, greenhouse gas-plus-aerosol and solar forced climate change. Clim. Dyn. 13, 613-634 (1997). Links Save Citation
  6. Johns, T. C. et al. The second Hadley Centre coupled ocean-atmosphere GCM: Model description, spinup and validation. Clim. Dyn. 13, 103-134 (1997). Links Save Citation
  7. Tett, S. F. B., Johns, T. C. & Mitchell, J. F. B. Global and regional variability in a coupled AOGCM. Clim. Dyn. 13, 303-323 (1997). Links Save Citation
  8. Shine, K. P., Fouquart, Y., Ramaswamy, V., Solomon, S. & Srinivasan, J. in Climate Change 1995: the Science of Climate Change (eds Houghton, J. T. et al.) 108-118 (Cambridge Univ. Press, 1996). Links Save Citation
  9. Mitchell, J. F. B., Johns, T. C., Gregory, J. M. & Tett, S. F. B. Climate response to increasing levels of greenhouse gases and sulphate aerosols. Nature 376, 501-504 (1995). Links Save Citation
  10. Mitchell, J. F. B. & Johns, T. C. On modification of global warming by sulfate aerosols. J. Clim. 10, 245-267 (1997). Links Save Citation
  11. Mitchell, J. F. B., Davis, R. A., Ingram, W. J. & Senior, C. A. On surface temperature, greenhouse gases, and aerosols: Models and observations. J. Clim. 10, 2364-2386 (1995). Links Save Citation
  12. Langner, J. & Rodhe, H. A global three-dimensional model of the tropospheric sulfur cycle. J. Atmos. Chem. 13, 225-263 (1991). Links Save Citation
  13. Twomey, S. A. Pollution and the planetary albedo. Atmos. Environ. 8, 1251-1256 (1974). Links Save Citation
  14. Albrecht, B. A. Aerosols, cloud microphysics and fractional cloudiness. Science 245, 1227-1230 (1989). Links Save Citation
  15. Sato, M., Hansen, J. E., McCormick, M. P. & Pollack, J. B. Stratospheric aerosol optical depths (1850-1990). J. Geophys. Res. 98, 22987-22994 (1993). Links Save Citation
  16. Hoyt, D. V. & Schatten, K. H. A discussion of plausible solar irradiance variations, 1700-1992. J. Geophys. Res. 98, 18895-18906 (1993). Links Save Citation
  17. Willson, R. C. Total solar irradiance trend during solar cycles 21 and 22. Science 277, 1963-1965 (1997). Links Save Citation
  18. Stott, P. A. & Tett, S. F. B. Scale-dependent detection of climate change. J. Clim. 11, 3282-3294 (1998). Links Save Citation
  19. Haywood, J., Stouffer, R., Wetherald, R., Manabe, S. & Ramaswamy, V. Transient response of a coupled model to estimated changes in greenhouse gas and sulfate concentrations. Geophys. Res. Lett. 24, 1335-1338 (1997). Links Save Citation
  20. Ramaswamy, V. & Chen, C.-T. Linear additivity of climate response for combined albedo and greenhouse perturbations. Geophys. Res. Lett. 24, 567-570 (1997). Links Save Citation
  21. Hasselmann, K. Optimal fingerprints for the detection of time-dependent climate change. J. Clim. 6, 1957-1971 (1993). Links Save Citation
  22. Hasselmann, K. Multi-pattern fingerprint method for detection and attribution of climate change. Clim. Dyn. 13, 601-611 (1997). Links Save Citation
  23. North, G. R., Kim, K.-K., Shen, S. S. P. & Hardin, J. W. Detection of forced climate signals. Part I: filter theory. J. Clim. 8, 401-408 (1995). Links Save Citation
  24. Allen, M. R. & Tett, S. F. B. Checking for model consistency in optimal fingerprinting. Clim. Dyn. (in the press). Links Save Citation
  25. Mardia, K. V., Kent, J. T. & Bibby, J. M. Multivariate Analysis (Academic, London, 1979). Links Save Citation
  26. Lean, J., Beer, J. & Bradley, R. Reconstruction of solar irradiance since 1610: implications for climate change. Geophys. Res. Lett. 22, 3195-3198 (1995). Links Save Citation
  27. Haigh, J. D. Impact of solar variability on climate. Science 272, 981-984 (1996). Links Save Citation
  28. Haigh, J. D. The role of stratospheric ozone in modulating the solar radiative forcing of climate. Nature 370, 544-546 (1994). Links Save Citation

Acknowledgements. S.F.B.T., P.A.S. and computer time were funded by the Department of the Environment, Transport and the Regions. W.J.I. and J.F.B.M. were supported by the UK Public Meteorological Service Research and Development programme. M.R.A. was supported by a research fellowship from the UK Natural Environment Research Council. Supplementary support was provided by the European Commission.

Correspondence and requests for materials should be addressed to S.B.F.T. (e-mail:

Nature © Macmillan Publishers Ltd 1999 Registered No. 785998 England.