9.6.4 Summary of Observational Constraints for Climate Sensitivity
Any constraint of climate sensitivity obtained from observations must be interpreted in light of the underlying assumptions. These assumptions include (i) the choice of prior distribution for each of the model parameters (Section 9.6.1 and Supplementary Material, Appendix 9.B), including the parameter range explored, (ii) the treatment of other parameters that influence the estimate, such as effective ocean diffusivity, and (iii) the methods used to account for uncertainties, such as structural and forcing uncertainties, that are not represented by the prior distributions. Neglecting important sources of uncertainty in these estimates will result in overly narrow ranges that overstate the certainty with which the ECS or TCR is known. Errors in assumptions about forcing or model response will also result in unrealistic features of model simulations, which can result in erroneous modes (peak probabilities) and shapes of the PDF. On the other hand, using less than all available information will yield results that are less constrained than they could be under optimal use of available data.
While a variety of important uncertainties (e.g., radiative forcing, mixing of heat into the ocean) have been taken into account in most studies (Table 9.3), some caveats remain. Some processes and feedbacks might be poorly represented or missing, particularly in simple and many intermediate complexity models. Structural uncertainties in the models, for example, in the representation of cloud feedback processes (Chapter 8) or the physics of ocean mixing, will affect results for climate sensitivity and are very difficult to quantify. In addition, differences in efficacy between forcings are not directly represented in simple models, so they may affect the estimate (e.g., Tett et al., 2007), although this uncertainty may be folded into forcing uncertainty (e.g., Hegerl et al., 2003, 2007). The use of a single value for the ECS further assumes that it is constant in time. However, some authors (e.g., Senior and Mitchell, 2000; Boer and Yu, 2003) have shown that ECS varies in time in the climates simulated by their models. Since results from instrumental data and the last millennium are dominated primarily by decadal- to centennial-scale changes, they will therefore only represent climate sensitivity at an equilibrium that is not too far from the present climate. There is also a small uncertainty in the radiative forcing due to atmospheric CO2 doubling (<10%; see Chapter 2), which is not accounted for in most studies that derive observational constraints on climate sensitivity.
Despite these uncertainties, which are accounted for to differing degrees in the various studies, confidence is increased by the similarities between individual ECS estimates (Figure 9.20). Most studies find a lower 5% limit of between 1°C and 2.2°C, and studies that use information in a relatively complete manner generally find a most likely value between 2°C and 3°C (Figure 9.20). Constraints on the upper end of the likely range of climate sensitivities are also important, particularly for probabilistic forecasts of future climate with constant radiative forcing. The upper 95% limit for ECS ranges from 5°C to 10°C, or greater in different studies depending upon the approach taken, the number of uncertainties included and specific details of the prior distribution that was used. This wide range is largely caused by uncertainties and nonlinearities in forcings and response. For example, a high sensitivity is difficult to rule out because a high aerosol forcing could nearly cancel greenhouse gas forcing over the 20th century. This problem can be addressed, at least to some extent, if the differences in the spatial and temporal patterns of response between aerosol and greenhouse gas forcing are used for separating these two responses in observations (as, for example, in Gregory et al., 2002a; Harvey and Kaufmann, 2002; Frame et al., 2005). In addition, nonlinearities in the response to transient forcing make it more difficult to constrain the upper limit on ECS based on observed transient forcing responses (Frame et al., 2005). The TCR, which may be more relevant for near-term climate change, is easier to constrain since it relates more linearly to observables. For the pre-instrumental part of the last millennium, uncertainties in temperature and forcing reconstructions, and the nonlinear connection between ECS and the response to volcanism, prohibit tighter constraints. Estimates of climate sensitivity based on the ability of climate models to reproduce climatic conditions of the LGM broadly support the ranges found from the instrumental period, although a tight constraint is also difficult to obtain from this period alone because of uncertainties in tropical temperature changes, forcing uncertainties and the effect of structural model uncertainties. In addition, the number of studies providing estimates of PDFs from palaeoclimatic data, using independent approaches and complementary sources of proxy data, are limited.
Thus, most studies that use a simple uniform prior distribution of ECS are not able to exclude values beyond the traditional IPCC First Assessment Report range of 1.5°C to 4.5°C (IPCC, 1990). However, considering all available evidence on ECS together provides a stronger constraint than individual lines of evidence. Bayesian methods can be used to incorporate multiple lines of evidence to sharpen the posterior distribution of ECS, as in Annan and Hargreaves (2006) and Hegerl et al. (2006a). Annan and Hargreaves (2006) demonstrate that using three lines of evidence, namely 20th-century warming, the response to individual volcanic eruptions and the LGM response, results in a tighter estimate of ECS, with a probability of less than 5% that ECS exceeds 4.5°C. The authors find a similar constraint using five lines of evidence under more conservative assumptions about uncertainties (adding cooling during the Little Ice Age and studies based on varying model parameters to match climatological means, see Box 10.2). However, as discussed in Annan and Hargreaves (2006), combining multiple lines of evidence may produce overly confident estimates unless every single line of evidence is entirely independent of others, or dependence is explicitly taken into account. Hegerl et al. (2006a) argue that instrumental temperature change during the second half of the 20th century is essentially independent of the palaeoclimate record of the last millennium and of the instrumental data from the first half of the 20th century that is used to calibrate the palaeoclimate records. Hegerl et al. (2006a) therefore base their prior probability distribution for the climate sensitivity on results from the late 20th century (Frame et al., 2005), which reduces the 5 to 95% ECS range from all proxy reconstructions analysed to 1.5°C to 6.2°C compared to the previous range of 1.2°C to 8.6°C. Both results demonstrate that independent estimates, when properly combined in a Bayesian analysis, can provide a tighter constraint on climate sensitivity, even if they individually provide only weak constraints. These studies also find a 5% lower limit of 1.5°C or above, consistent with several studies based on the 20th-century climate change alone (Knutti et al., 2002; Forest et al., 2006) and estimates that greenhouse warming contributes substantially to observed temperature changes (Section 18.104.22.168).
Overall, several lines of evidence strengthen confidence in present estimates of ECS, and new results based on objective analyses make it possible to assign probabilities to ranges of climate sensitivity previously assessed from expert opinion alone. This represents a significant advance. Results from studies of observed climate change and the consistency of estimates from different time periods indicate that ECS is very likely larger than 1.5°C with a most likely value between 2°C and 3°C. The lower bound is consistent with the view that the sum of all atmospheric feedbacks affecting climate sensitivity is positive. Although upper limits can be obtained by combining multiple lines of evidence, remaining uncertainties that are not accounted for in individual estimates (such as structural model uncertainties) and possible dependencies between individual lines of evidence make the upper 95% limit of ECS uncertain at present. Nevertheless, constraints from observed climate change support the overall assessment that the ECS is likely to lie between 2°C and 4.5°C with a most likely value of approximately 3°C (Box 10.2).