AGCI Session II: Characterizing and Communicating Scientific Uncertainty
Session Chairs: Dr. Richard H. Moss and Dr. Stephen H. Schneider
July 31 to August 8, 1996
Climate Change Detection and Attribution Assessment in the IPCC Proceedings
Benjamin D. Santer
Program for Climate Model Diagnosis and Intercomparison
Lawrence Livermore National Laboratory,
Livermore, California
Santer defines climate change detection as the process of demonstrating that an observed change is highly unusual in a statistical sense, and offered the analogy of detecting a fever by measuring body temperature. Attribution of climate change is the process of establishing cause and effect, like diagnosing the cause of the fever through a complete suite of medical tests. We are now in the process of using complex diagnostics to diagnose the causes and effects of climate change.
Major advances have been made since 1990 in defining a human -induced climate change signal, defining the noise of natural climate variability, and increasing application of pattern-based methods with greater relevance for attribution. In regard to the first of these advances, identifying the human-induced climate signal, early models showed spatially coherent warming with more warming toward the poles. Then, full ocean dynamics were added, simulating the Earth's redistribution of heat. The latest advance is the addition of the direct effects of sulfate aerosols in the models. In this ongoing evolution, new anthropogenic components will be incorporated including indirect aerosol effects, tropospheric and stratospheric ozone, and aerosols from biomass burning.
Progress in defining "noise," or the natural variability of the climate system comes from new instrumental data as well as reconstructions of past climate. Progress in paleoclimate data comes mainly from recent efforts to meld information from diverse proxy sources rather than simply looking at tree rings, ice cores, corals, or borehole temperatures individually. A 1992 study by Bradley and Jones is an example of such melding. Further development of statistical models as well as physically -based climate models with full ocean general circulation are proving helpful as well. Comparing modeled estimates of the pattern and amplitude of natural variability with observations from paleoclimatic data show that model-based estimates of climate 100 years in the future are generally less than those derived from paleoclimatic data, but progress is being made. Knowledge of changes in solar forcing is now also being incorporated into general circulation models (GCMs).
Progress in
pattern-based methods involves "fingerprints" that can help
discriminate between different possible causes of climate change.
Different forcing mechanisms have different patterns of effects.
Progress in pattern-based methods involves "fingerprints" that can help discriminate between different possible causes of climate change. Different forcing mechanisms have different patterns of effects. For example, tropospheric warming resulting from increases in carbon dioxide and other greenhouse gases would be accompanied by stratospheric cooling, while warming resulting from an increase in solar output would not be accompanied by stratospheric cooling. Since data show that the stratosphere has indeed cooled, this is part of the fingerprint that helps identify the cause. Another pattern-based indicator is that models which incorporate both greenhouse gases and aerosols yield a better correlation with observations, showing both the cooling and warming pattern observed. The combined CO2 and aerosol signal exceeds the statistical significance threshold in the last two decades, when it would be expected to, and the regional pattern is also as expected (better correlation over industrialized regions).
A methodological advance in understanding comes from new multivariable representations of the climate change signal, such Tom Karl's climate index, rather than a focus on global mean temperature alone. Karl et al., 1995, included in their climate index: asymmetric increases in maximum and minimum temperature, cold season precipitation, severe summertime drought, extreme one-day precipitation events, and day-to-day temperature variability all of which yields a more comprehensive indicator of climate change than does global mean temperature.
Santer reviewed an example of a detection and attribution study showing "coarse" (global mean retained) and "fine" (global mean subtracted) structure of an aerosol signal from Taylor and Penner, 1994. Both observations and modeled changes for increased CO 2 only (without aerosols or ozone) show both stratospheric cooling and tropospheric warming. The observed patterns of vertical temperature changes are seasonally robust. Even better agreement between the model and observation is achieved when models include combined effects of CO2 , aerosols, and ozone. Further confidence in model results comes from the fact that there are similarities between CO 2 and aerosol signals simulated by different climate models.
A methodological
advance in understanding comes from new multivariable representations
of the climate change signal, such Tom Karl's climate index, rather
than a focus on global mean temperature alone.
Conclusions for detection are that the Earth is warming and that its mean temperature is warmer than it has been in many centuries. For attribution, conclusions are that observed geographical patterns of temperature change at the Earth's surface are similar to model predictions that incorporate combined greenhouse gas (GHG) and sulfate aerosol effects. Observed patterns of atmospheric distribution of temperature change are also similar to model predictions that incorporate the effects of GHGs, sulfate aerosols, and stratospheric ozone depletion. These pattern correspondences tend to increase with time, and model patterns are different from those due to natural variability. Observations and model predictions generally agree in overall magnitude and timing of change as well.
Major scientific uncertainties still exist with regard to estimates of the magnitude, pattern and evolution of different forcings. GHGs, direct sulfate aerosol effects, and stratospheric ozone effects are included in current analyses, but uncertainties remain about these estimates. Indirect effects of sulfate aerosols, other anthropogenic aerosols, tropospheric ozone, and solar and volcanic influences have thus far been left out of the analysis, posing additional uncertainty. In regard to model predictions of response to forcing, key uncertainties involve the parameterization of clouds and ocean vertical mixing. Estimates of the natural variability of climate on time scales of decades to centuries is another uncertainty, as is the quality of observed data.
Confidence in
the emerging identification of a human-induced effect on global
climate comes largely from comparisons of modeled and observed
patterns of change. The results of such studies rest mainly on
pattern similarities at very large spatial scales.
If climate models have errors, how can we have confidence in their results? Confidence in the emerging identification of a human-induced effect on global climate comes largely from comparisons of modeled and observed patterns of change. The results of such studies rest mainly on pattern similarities at very large spatial scales, e. g., the temperature contrast between the hemispheres and the temperature differences between the stratosphere and the troposphere. It is at these large spatial scales that we have most confidence in the reliability of model results.
Additional work is needed to give us a more complete description of human-induced climate forcing through inclusion of indirect sulfate aerosols, other anthropogenic aerosols such as those from biomass burning, and a quantification of tropospheric ozone changes. More realistic estimates of climate response can be achieved through better representation of clouds and precipitation and the incorporation of land surface processes in models. Application of more sophisticated signal processing techniques in detection and attribution studies should involve multi-variable fingerprints (such as Tom Karl's work mentioned above) and optimal filtering techniques to enhance the signal to noise ratio. We can also achieve a better understanding of the systematic error in models through model intercomparison projects.
Major scientific
uncertainties still exist with regard to estimates of the magnitude,
pattern and evolution of different forcings.