• GG 312: Global Climate Change and Environmental Impacts (Fall 2002)


    Chapter 6: Detection of Climate Change and Attribution of Causes

    6.1. The Meaning of Detection and Attribution
    6.2. Results from IPCC-1995
    6.3. Internal Climate Variability
    6.4. Climate Forcings and Responses
    6.5. Climate Response to Natural Forcings
    6.6. Climate Response to Anthropogenic Forcings
    6.7. Detection and Attribution Studies
    6.8. Conclusions

    6.1. The Meaning of Detection and Attribution

    The response to anthropogenic changes in climate forcing occurs against a backdrop of natural internal and externally-forced climate variability that can occur on similar temporal and spatial scales.

    Internal climate variability, by which we mean climate variability not forced by external agents, occurs on all timescales from weeks to centuries and millennia. Slow climate components, such as the ocean, have particularly important roles on decadal and century timescales because they integrate high frequency weather variability and interact with faster components. Thus the climate is capable of producing long timescale internal variations of considerable magnitude without any external influences.

    Externally-forced climate variations may be due to changes in natural forcing factors, such as solar radiation or volcanic aerosols, or to changes in anthropogenic forcing factors, such as increasing concentrations of greenhouses gases or sulphate aerosols.

    The presence of this natural climate variability means that the detection and attribution of anthropogenic climate change is a statistical "signal-in-noise" problem.

  • Detection is the process of demonstrating that an observed change is significantly different (in a statistical sense) than can be explained by natural internal variability. However, the detection of a change in climate does not necessarily imply that its causes are understood.

  • Unequivocal attribution of climate change to anthropogenic causes (i.e., the isolation of cause and effect) would require controlled experimentation with the climate system in which the hypothesised agents of change are systematically varied in order determine the climate's sensitivity to these agents. Such an approach to attribution is clearly not possible. Thus from a practical perspective, attribution of observed climate change to a given combination of human activity and natural influences requires another approach. This involves statistical analysis and the careful assessment of multiple lines of evidence to demonstrate, within a pre-specified margin of error, that the observed changes are:
    • unlikely to be due entirely to internal variability;
    • consistent with the estimated responses to the given combination of anthropogenic and natural forcing; and
    • not consistent with alternative, physically-plausible explanations of recent climate change that exclude important elements of the given combination of forcings.

    6.2. Results from IPCC-1995

    The climate change study for identifying the signal in near surface air temperature used equilibrium model responses to CO2 and sulphate aerosol forcings. The summertime (JJA) near surface air temperature changes (deg C) in AGCM experiments with forcing by CO2 only (Figure 6.1a) and by both CO2 and anthropogenic sulphate aerosols (Figure 6.1b) were investigated.

    • The temperature change is very different between CO2 only and CO2 + sulphate aerosols.
    • The latter shows warmings and coolings, which is more consistent with observations.

    This study also evaluated the significance of the pattern of change predicted by the model with that of the observations. The behavior of a pattern correlation statistic, R(t), measuring the similarity between model predicted and observed patterns of near surface temperature change are shown in Figure 6.2a and Figure 6.2b. The model predictions are from equilibrium response experiments with forcing by early 1990s CO2 levels (Figure 6.2a) and by the combined effects of CO2 and sulphur emissions (Figure 6.2b). For each experiment, a single pattern characterises the temperature change signal. The signal is then searched for in observed time-varying records of near surface temperature change, The figures show the results for the autumn (SON) season.

    • In the case of combined CO2 and sulphate forcing, there is a positive linear trend in R(t) time series over the last 50 year period (1944-1993), indicating that sub-global features of the observed temperature change patterns are becoming increasingly similar to the predicted signal pattern.
    • No such increasing similarity is found for the CO2 only signal.
    • The 50 year trend in R(t) in the experiment with combined CO2 and sulphate aerosol forcing was highly significant relative to estimates of internally generated natural variability from two extended AOGCM control runs.
    • The initial decrease in R(t) from roughly 1910 to 1945 has not been fully explained, but is likely due to the fact that the observed warming in the 1930s and 1940s had some similarity to a CO2 only signal, and was different in character from more recent changes.

    Let us consider modeled and observed temperature trends. Figure 6.1a depicts the JJA mean near surface air temperature changes (deg C) from equilibrium response experiments with an atmospheric GCM, forced with early 1990s CO2 levels. The combined CO2 and sulphate aerosol forcing results are shown in Figure 6.1b. The observed temperature changes from 1955-74 to 1975-94 shown in Figure 6.3 are qualitatively more similar to the changes in the combined forcing experiment than in the CO2 only experiment.

    Based on this and many other pieces of evidence, IPCC-1995 concluded that The body of statistical evidence when examined in the context of our physical understanding of the climate system now points to a discernable human influence on global climate.

    6.3. Internal Climate Variability

    Detection and attribution of climate change is a statistical "signal-in-noise" problem, it requires an accurate knowledge of the properties of the "noise". Ideally, internal climate variability would be estimated from instrumental observations, but a number of problems make this difficult.

  • The instrumental record is short relative to the 30-50 year time scales that are of interest for detection and attribution of climate change, particularly for variables in the free atmosphere. The longest records that are available are those for surface air temperature and sea surface temperature. Relatively long records are also available for precipitation and surface pressure, but coverage is incomplete and varies in time.

  • The instrumental record also contains the influences of external anthropogenic and natural forcing. A record of natural internal variability can be reconstructed by removing estimates of the response to external forcing. However, the accuracy of this record is limited by incomplete knowledge of the forcings and the climate model used to estimate the response.

    Stouffer et al. assessed variability simulated in three 1000-year control simulations (Figure 6.4). The models are found to simulate reasonably well the spatial distribution of variability and the spatial correlation between regional and global mean variability although there is more disagreement between models at long timescales (>50 years) than at short timescales. None of the long model simulations produces a secular trend which is comparable to that observed.

    The power spectrum of global mean temperatures simulated by the most recent coupled climate models (Figure 6.5) compares reasonably well with that of detrended observations (solid black line) on interannual to decadal timescales. However, uncertainty of the spectral estimates is large and some models are clearly underestimating variability (indicated by the asterisks). Detailed comparison on interdecadal timescales assessment is difficult because observations are likely to contain a response to externalforcings that will not be entirely removed by a simple linear trend.

    Figure 6.4: Global mean surface air temperature anomalies from 1000-year control simulations with three different climate models, HadCM2, GFDL R15 and ECHAM3/LSG (labelled HAM3L), compared to the recent instrumental record. No model control simulation shows a trend in surface air temperature as large as the observed trend. If internal variability is correct in these models, the recent warming is likely not due to variability produced within the climate system alone.

    Figure 6.5: Coloured lines: Power spectra of global mean temperatures in the unforced control integrations that are used to provide estimates of internal climate variability. Solid black line: spectrum of observed global mean temperatures over the period 1861-1998 after removing a best-fit linear trend. Dotted black line: spectrum of observed global mean temperatures after removing an independent estimate of the externally-forced response provided by the ensemble mean of a coupled model simulation. Asterisks indicate models whose variability is significantly less than observed variability on 10-60-year timescales after removing either a best-fit linear trend or an independent estimate of the forced response from the observed series.

    Summary: There is still considerable uncertainty in the magnitude of internal climate variability. Various approaches are used in detection and attribution studies to account for this uncertainty.

  • Some studies use data from a number of coupled climate model control simulations and choose the most conservative result.
  • In other studies, the estimate of internal variance is inflated to assess the sensitivity of detection and attribution results to the level of internal variance.
  • Some authors also augment model-derived estimates of natural variability with estimates from observations.
  • A method for checking the consistency between the residual variability in the observations after removal of externally forced signals and the natural internal variability estimated from control simulations is also available.

    6.4. Climate Forcings and Responses

    There are several reasons why one should not expect a simple relationship between the patterns of radiative forcing and temperature response. First, strong feedbacks such as those due to water vapour and sea-ice tend to reduce the difference in the temperature response due to different forcings. This is illustrated graphically by the response to the simplified aerosol forcing used in early studies. The magnitude of the model response is largest over the Arctic in winter even though the forcing is small, largely due to ice-albedo feedback. The large-scale patterns of change and their temporal variations are similar, but of opposite sign, to that obtained in greenhouse gas experiments (Figure 6.6).

    Specifically, latitude-month plot of radiative forcing and model equilibrium response for surface temperature are in Fig. 6.6. (a) Radiative forcing (W/m2 ) due to increased sulphate aerosol loading at the time of CO2 doubling (b) Change in temperature due to the increase in aerosol loading (c) Change in temperature due to CO2 doubling. Note that the patterns of radiative forcing and temperature response are quite different in (a) and (b), but that the patterns of large-scale temperature responses to different forcings are similar in (b) and (c).

    Second, atmospheric circulation tends to smooth out temperature gradients and reduce the differences in response patterns. Similarly, the thermal inertia of the climate system tends to reduce the amplitude of short-term fluctuations in forcing. Third, changes in radiative forcing are more effective if they act near the surface, where cooling to space is restricted, than at upper levels, and in high latitudes, where there are stronger positive feedbacks than at low latitudes.

    Summary: Different models may give quite different patterns of response for the same forcing, but an individual model may give a surprisingly similar response for different forcings. The first point means that attribution studies may give different results when using signals generated from different models. The second point means that it may be more difficult to distinguish between the response to different factors than one might expect given the differences in radiative forcing.

    6.5. Climate Response to Natural Forcings

    The climate response to several recent volcanic eruptions has been studied in observations and simulations with atmospheric GCMs. The stratosphere warms and the annual mean surface and tropospheric temperature decreases during the 2-3 years following a major volcanic eruption. A simulation incorporating the effects of the Mount Pinatubo eruption and observed changes in stratospheric ozone in addition to anthropogenic forcing approximately reproduces the observed stratospheric variations (Figure 6.7). It shows stratospheric warming after the volcanic eruption, superimposed on a long-term cooling trend. Variability from other sources makes assessment of the observed climate response difficult, particularly as the two most recent volcanic eruptions (Pinatubo and El Chichon) occurred in ENSO warm years.

    Figure 6.7: (a) Observed microwave sounding unit (MSU) global-mean temperature in the lower stratosphere, shown as dashed line, for channel 4 for the period 1979-97 compared to the average of several atmosphere-ocean GCM simulations starting with different atmospheric conditions in 1979 (solid line). The simulations have been forced with increasing greenhouse gases, direct and indirect forcing by sulphate aerosols and tropospheric ozone forcing, and Mount Pinatubo volcanic aerosol and stratospheric ozone variations. The model simulation does not include volcanic forcing due to El Chichon in 1982, so it does not show stratospheric warming then. (b) As for (a), except for 2LT temperature retrievals in the lower troposphere.

    Differences between the response to solar and greenhouse gas forcings would make it easier to distinguish the climate response to either forcing. However, the spatial response pattern of surface air temperature to an increase in solar forcing was found to be quite similar to that in response to increases in greenhouse gas forcing. The vertical response to solar forcing (Figure 6.8) includes warming throughout most of the troposphere. The response in the stratosphere is small and possibly locally negative, but less so than with greenhouse gas forcing, which gives tropospheric warming and strong stratospheric cooling. Hence, the conclusion that changes in solar forcing have little effect on large-scale stratospheric temperatures remains tentative. The different time-histories of the solar and anthropogenic forcing should help to distinguish between the responses. All reconstructions suggest a rise in solar forcing during the early decades of the 20th century with little change on interdecadal timescales in the second half. Such a forcing history is unlikely to explain the recent acceleration in surface warming, even if amplified by some unknown feedback mechanism.

    Figure 6.8: Response (covariance, normalised by the variance of radiance fluctuations) of zonally-averaged annual mean atmospheric temperature to solar forcing for two simulations with ECHAM3/LSG. Coloured regions indicate locally significant response to solar forcing. (b) Zonal mean of the first EOF of greenhouse-gas induced temperature change simulated with the same model. This indicates that for ECHAM3/LSG, the zonal mean temperature response to greenhouse gas and solar forcing are quite different in the stratosphere but similar in the troposphere.

    Summary: We conclude that climate forcing by changes in solar irradiance and volcanism have likely caused fluctuations in global and hemispheric mean temperatures. Qualitative comparisons suggest that natural forcings produce too little warming to fully explain the twentieth century warming (Figure 6.9 and Figure 6.10). The indication that the trend in net solar plus volcanic forcing has been negative in recent decades makes it unlikely that natural forcing can explain the increased rate of global warming since the middle of the 20th century.

    Figure 6.9: (a) Five-year running mean Northern Hemisphere temperature anomalies since 1850 (relative to the 1880-1920 mean) from an energy-balance model forced by Dust Veil volcanic index and solar index. Two values of climate sensitivity to doubling CO2 were used; 3.0 K (thin solid line), and 1.5K (dashed line). Also shown are the instrumental record (thick solid line) and a reconstruction of temperatures from proxy records (crosses). (b) As for (a) but for simulations with volcanic, solar and anthropogenic forcing (greenhouse gases and direct and indirect effects of tropospheric aerosols).

    Figure 6.10: Global mean surface temperature anomalies relative to the 1880-1920 mean from the instrumental record compared with ensembles of four simulations with a coupled ocean-atmosphere climate model forced (a) with solar and volcanic forcing only, (b) with anthropogenic forcing including well mixed greenhouse gases, changes in stratospheric and tropospheric ozone and the direct and indirect effects of sulphate aerosols, and (c) with all forcings, both natural and anthropogenic. The thick line shows the instrumental data while the thin lines show the individual model simulations in the ensemble of four members. Note that the data are annual-mean values.

    6.6. Climate Response to Anthropogenic Forcings

    Well-mixed greenhouse gases make the largest and best-known contribution to changes in radiative forcing over the last century or so. There remains a large uncertainty in the magnitude and patterns of other factors, particularly those associated with the indirect effects of sulphate aerosol. Models run with increases in greenhouse gases alone give a warming which accelerates in the latter half of the century. When a simple representation of aerosol effects is included the rate of warming is reduced. The global mean response is similar when additional forcings due to ozone and the indirect effect of sulphates are included.

    Increases in greenhouse gases lead to a warming of the troposphere and a cooling of the stratosphere due to CO2 (IPCC 1995). Reductions in stratospheric ozone lead to a further cooling, particularly in the stratosphere. Anthropogenic sulphate aerosols cool the troposphere with little effect on the stratosphere. When these three forcings are included in a climate model albeit in a simplified way, the simulated changes show tropospheric warming and stratospheric cooling, as observed and as expected on physical principles (Figure 6.11). Note this structure is distinct from that expected from natural (internal and external) influences.

    Figure 6.11: Simulated and observed zonal mean temperature change as a function of latitude and height. The contour interval is 0.1 K. All signals are defined to be the difference between the 1986-95 decadal mean and the 20 year 1961-80 mean. (a), increases in CO2 only; (b), as (a), but with a simple representation of sulphate aerosols added; (c) , as (b), with observed changes in stratospheric ozone; (d), observed changes.

    The spatial pattern of the simulated surface temperature response to a steady increase in greenhouse gases is well documented. The warming is greater over land than ocean and generally small during the twentieth century over the southern ocean and northern North Atlantic where mixing extends to considerable depth. The warming is amplified in high latitudes in winter by the recession of sea-ice and snow, and it is close to zero over sea-ice in summer. Despite the qualitative consistency of these general features, there is considerable variation from model to model.

    6.7. Detection and Attribution Studies

    All new single pattern studies published since IPCC-1995 detect anthropogenic fingerprints in the global temperature observations, both at the surface and aloft. The signal amplitudes estimated from observations and modelled amplitudes are consistent at the surface if greenhouse gas and sulphate aerosol forcing are taken into account, and in the free atmosphere if ozone forcing is also included. Fingerprints based on smaller areas or on other variables yield more ambiguous results at present.

    Example Study: This study estimated the magnitude of modelled 20th century greenhouse gas, aerosol, solar and volcanic signals in decadal mean data. Signals are fitted by general linear regression to moving fifty year intervals beginning with 1906-56 and ending 1946-96. The signals are obtained from four ensembles of transient change simulations, each using a different historical forcing scenario. Greenhouse gas, greenhouse gas plus direct sulphate aerosol, low frequency solar, and volcanic forcing scenarios were used. Each ensemble contains four independent simulations with the same transient forcing. Two estimates of natural variability, one used for optimisation and the other for the estimation of confidence intervals, are obtained from separate segments of a long control simulation.

    Signal amplitudes estimated with multiple regression become uncertain when the signals are strongly correlated ("degenerate"). Despite the problem of degeneracy, positive and significant greenhouse gas and sulphate aerosol signals are consistently detected in the most recent fifty year period (Figure 6.12 ) regardless of which or how many other signals are included in the analysis. The residual variation that remains after removal of the signals is consistent with the model's internal variability. In contrast, recent decadal temperature changes are not consistent with the model's internal climate variability alone, nor with any combination of internal variability and naturally-forced signals, even allowing for the possibility of unknown processes amplifying the response to natural forcing.

    Figure 6.12: Best-estimate contributions to global-mean temperature change. Reconstruction of temperature variations for 1906-1956 (a and b) and 1946-1995 (c and d) for G and S (a and c) and GS and SOL (b and d). (G denotes the estimated greenhouse gas signal, S the estimated sulphate aerosol signal, GS the greenhouse gas / aerosol signal obtained from simulations with combined forcing, SOL the solar signal). Observed (thick black), best fit (dark grey dashed), and the uncertainty range due to internal variability (grey shading) are shown in all plots. (b) and (d) show contributions from GS (orange) and SOL (blue). (a) and (c) show contributions from G (red) and S (green). All time series were reconstructed with data in which the 50-year mean had first been removed.

    Summary: Results from optimal fingerprint methods indicate a discernible human influence on climate in temperature observations at the surface and aloft and over a range of applications. All recent studies reject natural forcing and internal variability alone as a possible explanation of recent climate change. Analyses based on a single anthropogenic signal focussing on continental and global scales indicate that:

    • Changes over the past 30-50 years are very unlikely to be due to internal variability as simulated by current models.
    • The combined response to greenhouse and sulphate forcing is more consistent with the observed record than the response to greenhouse gases alone.
    • Inclusion of the simulated response to stratospheric ozone depletion improves the simulation of the vertical structure of the response.
    Analyses based on multiple anthropogenic and natural signals indicate that:
    • The combination of natural external forcing (solar and volcanic) and internal variability is unlikely to account for the spatio-temporal pattern of change over the past 30-50 years, even allowing for possible amplification of the amplitude of natural responses by unknown feedback processes.
    • Anthropogenic greenhouse gases are likely to have made a significant and substantial contribution to the warming observed over the second half of the 20th century, possibly larger than the total observed warming.
    • The contribution from anthropogenic sulphate aerosols is less clear, but appears to lie in a range broadly consistent with the spread of current model simulations. A high sulphate aerosol forcing is consistently associated with a stronger response to greenhouse forcing.
    • Natural external forcing may have contributed to the warming that occurred in the early twentieth century.
    Results based on variables other than continental and global scale temperature are more ambiguous.

    6.8. Conclusions

  • Twentieth century climate was unusual.
  • The observed warming is inconsistent with model estimates of natural internal climate variability.
  • The observed change in patterns of atmospheric temperature in the vertical is inconsistent with natural forcing.
  • Anthropogenic factors do provide an explanation of 20th century temperature change.
  • The effect of anthropogenic greenhouse gases is detected, despite uncertainties in aerosol forcing and response.
  • It is unlikely that detection studies have mistaken a natural signal for an anthropogenic signal.
  • The detection methods used should not be sensitive to errors in the amplitude of the global mean forcing or response.
  • Studies of the changes in the vertical patterns of temperature also indicate that there has been an anthropogenic influence on climate over the last 35 years.
  • Natural factors may have contributed to the early century warming.

    From CLIVAR
    From NOAA
    From Leeds U.
    Are Human Activities Contributing to Climate Change?
    Detecting Man-Made Climate Change
    Detection of Climate Change
    Common Questions about Climate Change