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

---

Empirical Estimation of Climate Sensitivity and Anthropogenic Sulfate Forcing

Michael Schlesinger

University of Illinois

Urbana, Illinois

See also: When We Don't Know the Costs or the Benefits: Adaptive Strategies for Abating Climate Change

---

Schlesinger discussed methods of estimating climate sensitivity (the change in equilibrium surface air temperature that results from a doubling of carbon dioxide) and sulfate aerosol forcing. He summed up the uncertainties in using general circulation models (GCMs) for estimating climate sensitivity as follows: "One can get any answer out of a GCM depending on how one treats the physical processes that are not explicitly resolved. Ergo, we cannot trust these models to correctly estimate this most important climatic quantity." To illustrate this point, Schlesinger points to the model from the United Kingdom Meteorological Office, the standard form of which produces a climate sensitivity of 5.2°C; but by changing the way it treats clouds (specifically, how fast water falls out of ice clouds relative to how fast it falls out of liquid water clouds), the sensitivity changes drastically, from 5.2°C to 1.9°C.

In discussing problems with trying to deduce climate sensitivity using GCMs, Schlesinger points out that the present generation of GCMs resolves only two (the planetary and synoptic scales) of the 14 orders of magnitude of physical processes that determine Earth's climate. If we wanted to increase the resolution of one of these models by one order of magnitude in scale, computing time for the simulation would increase from about 500 hours to about 50 years. But even though we can not practically resolve these processes explicitly, we can not ignore them, since many are important to determining climate. Therefore, we parameterize them, meaning that we try to determine the statistical effects of these unresolved processes on the resolved processes using information only about the resolved processes. This involves making many assumptions about how nature operates, and how we parameterize important processes (e. g., cumulus convection) largely determines the climate sensitivity produced by the models.

One can get any answer out of a GCM depending on how one treats the physical processes that are not explicitly resolved. Ergo, we cannot trust these models to correctly estimate this most important climatic quantity.

Schlesinger then discussed a "sequential decision analysis" he undertook with colleagues at the RAND Corporation. They compared two near-term policies for dealing with climate change: a moderate policy (conservation only, characterized by a 30 percent decrease in energy intensity over 20 years) and an aggressive policy (involving fuel switching at a rate at which half of all fossil fuel-using facilities are replaced within 40 years). They assumed that in ten years we would learn exactly what the climate sensitivity is and what the climate target should be (the maximum allowable warming we want the system to undergo). The climate policy for the long term could then be adapted as needed at the end of ten years, depending on what was learned. Looking at long -term costs, the key conclusion of this analysis is that it is most important to learn the climate sensitivity because the cost of abatement changes primarily based on climate sensitivity and the climate target, and virtually not at all on whether the aggressive or moderate policy is chosen.

So, if determining climate sensitivity is the most important thing we need to know, and the models are not reliable for estimating it, how should we do it? One approach is to use past climate data to infer the climate sensitivity. There are several problems with this approach. First, past climate data are derived from proxies (such as tree rings and ice cores) rather than direct observational measurements. Second, in addition to knowing the climate changes, we need to know the forcing mechanisms. And third, the climate sensitivity value we seek is not a constant, but rather depends upon the climate perturbed. So if using past climate data is not the best approach, we are left with using instrumental temperature observations to determine the climate sensitivity.

The key conclusion of this analysis is that it is most important to learn the climate sensitivity because the cost of abatement changes primarily based on climate sensitivity and the climate target, and virtually not at all on whether the aggressive or moderate policy is chosen.

Given this, Schlesinger discussed a simple modeling method for using the observational record of surface air temperature and interhemispheric temperature difference reported in the IPCC 1992 and 1995 assessments. He points out that there is a larger downward trend in the interhemispheric temperature difference from the 1995 data than from the 1992 data. Schlesinger fed these data, along with specified forcing due to greenhouse gases, tropospheric ozone, and solar output variations, into a simple energy balance climate model /upwelling-diffusion-ocean model. (The calculations were done with and without solar forcing and a large difference was found.) In this simple model, the Northern and Southern Hemispheres are forced separately, all feedbacks are included by the estimated climate sensitivity, the ocean sets the time scale for the response, upwelling occurs in non-polar regions and downwelling in polar regions, and deep water formation occurs in polar regions only.

Several estimates of climate sensitivities were made for the different temperatures, forcings, and sulfate emission rates to determine which of these contributes most to the uncertainties; probability distributions were also figured for these estimated quantities. (To get radiative forcing due to sulfate aerosols, sulfate emission rates are multiplied by a model -determined parameter.) The simple model is run with a number of different specified sulfate forcings and climate sensitivities and the best fit between the observed climate data and the model is determined to be the solution. Because sulfate forcing is negative (it partially compensates for greenhouse-gas forcing), the higher it is, the higher the climate sensitivity. The sulfate forcing is determined by using data on the interhemispheric temperature difference. The climate sensitivity is determined by the global-mean temperature.

This study produced four sets of results: one for the 1995 IPCC data without solar forcing, one for the 1992 IPCC data without solar forcing, and one for each of these data sets with solar forcing. As far as the interhemispheric difference in warming, the model reveals the same trend as that observed. Because sulfate aerosols are more concentrated in the Northern Hemisphere, it will warm less rapidly than the Southern Hemisphere. This becomes more true as time goes on and emissions increase. The instrument observations used in this study begin in 1856.

Results using the 1995 data without solar forcing show a climate sensitivity of over 5°C, while results using the 1992 data without solar forcing show a sensitivity of about 3°C.

Results using the 1995 data without solar forcing show a climate sensitivity of over 5°C, while results using the 1992 data without solar forcing show a sensitivity of about 3°C. The significant change in results can be explained by two factors. First, substantial corrections were made to 19th century measurements since 1992 (the effect of three additional years of measurements was negligible). Secondly, the 1995 data show a much bigger downward trend line, roughly by a factor of three, in the interhemispheric temperature difference and to explain that, we need more sulfate forcing, and thus a larger climate sensitivity, to explain the global mean.

The results also make clear that including the sun in the calculation makes an even larger difference. Solar irradiance variations have been in phase with the temperature record during the 20th century, so solar forcing is positive during most of the instrumental temperature record, and adding a positive forcing brings the sensitivity down. When we add this, sensitivity comes down by factor of two (adding a factor of two uncertainty to our estimate of climate sensitivity). We don't know what the sun's role in climate change is; this is a problem that is not going to go away, and it makes a significant difference in climate sensitivity.

All of the variations in the climatic record cannot be explained wholly by the forcings we are aware of, which brings us to a discussion of natural variability or "noise." Schlesinger and colleagues performed a singular spectrum analysis on the temperature record which reveals an oscillation about half the length of the record. They tried to reproduce this and found that they could mimic one hemisphere at a time but not both simultaneously. They then developed a "bootstrap" resampling technique to generate 900 alternate (surrogate) instrumental observation records. For each of these simulated realizations, they re -did the estimation problem. The bootstrap technique does well at reproducing the power spectrum.

Schlesinger also showed marginal and cumulative probability distributions for climate sensitivity and sulfate aerosol forcing. The 50th percentile for climate sensitivity is about 3.0°C. The 50th percentile for sulfate forcing is about &endash;0.76 watt per square meter.

We don't know what the sun's role in climate change is; this is a problem that is not going to go away, and it makes a significant difference in climate sensitivity.

Finally, Schlesinger returned to further discussion of the analysis performed with colleagues at the RAND Corporation detailed in a paper titled, "When We Don't Know the Costs or the Benefits: Adaptive Strategies for Climate Change" (Lempert et al ., 1996). While most quantitative studies of climate-change policy attempt to predict the greenhouse-gas-reduction plan that will have the optimum balance of long-term costs and benefits, Schlesinger and colleagues believe that the large uncertainties associated with the climate change problem can make the policy prescriptions of this approach unreliable. In this study, they construct a large uncertainty space that includes the possibility of large and/or abrupt climate changes and/or technological breakthroughs that radically reduce abatement costs. They perform computational experiments on a linked system of climate and economic models to compare the performance of an adaptive strategy (one that makes midcourse corrections based on observations of the climate and economic systems) and two commonly advocated "best-estimate" policies (described earlier in this summary) based on different expectations about the long-term consequences of climate change.

They find that the best-estimate policies perform well in the respective regions of the uncertainty space where their estimates are valid, but can fail severely in those regions where their estimates are wrong. In contrast, the adaptive strategy can make midcourse corrections and thereby avoid significant errors. The adaptive strategy doesn't do as well as either prescriptive policy IF that prescriptive policy was right, but the penalty for being wrong is much, much smaller. Schlesinger explains that if we knew the climate sensitivity, the damage that could be expected to result from it, and the rate of technological innovation, we could choose the optimum policy. But without knowing these things, we must create a strategy that allows adaptation based on new learning. Such a policy is analogous, he says, to sending a probe to Jupiter with the ability to make course corrections en route rather than just sending it off and saying, "bon voyage, call us when you get there."

In this study, they construct a large uncertainty space that includes the possibility of large and/or abrupt climate changes and/or technological breakthroughs that radically reduce abatement costs.

The authors of this study believe that its results suggest that society might usefully recast its view of the climate-change-policy problem. Currently the debate focuses on targets and timetables for the optimum level of near-term GHG reductions that should or should not be set. The research community views its task as improving the accuracy of the predictions of the future which will provide policymakers with better estimates of the optimum level of emissions reductions. These authors suggest that climate change be viewed instead as a problem of preparing for unpredictable contingencies. Society may or may not need to implement massive reductions in GHG emissions in the next few decades. The problems for the present include developing better options for massive reductions than those currently available and determining what observations ought to trigger their implementation.

Reference

Lempert, R. J., Schlesinger, M. E., and Bankes, S. C.: 1996, When we don't know the costs or the benefits: Adaptive strategies for abating climate change, Climatic Change 33: 235-274.

The best-estimate policies perform well in the respective regions of the uncertainty space where their estimates are valid, but can fail severely in those regions where their estimates are wrong. In contrast, the adaptive strategy can avoid significant errors.