by Judith Curry
Some new analyses are shedding some light on deficiencies in the approach to estimate the social cost of carbon.
The debate surrounding the social cost of carbon was described in a previous post On Trial: Social Cost of Carbon. My chief concern was that I thought the US IWG used indefensible values of climate sensitivity as input into the integrated assessment models.
New report from the NAS
The IWG has requested that the U.S. National Academies assess the approaches for determine the social cost of carbon. A preliminary report was just published: The Assessment of Approaches to Updating the Social Cost of Carbon: Phase 1 Report on a Near-Term Update.
Some context from the Executive Summary:
The social cost of carbon (SCC) for a given year is an estimate, in dollars, of the present discounted value of the damage caused by a 1-metric ton increase in carbon dioxide (CO2) emissions into the atmosphere in that year or, equivalently, the benefits of reducing CO2 emissions by the same amount in that year. The SCC is intended to provide a comprehensive measure of the monetized value of the net damages from global climate change that results from an additional unit of CO2, including, but not limited to, changes in net agricultural productivity, energy use, human health effects, and property damages from increased flood risk. Federal agencies use the SCC to value the CO2 emissions impacts of various regulations, including emission and fuel economy standards for vehicles; emission standards for industrial manufacturing, power plants, and solid waste incineration; and appliance energy efficiency standards.
The Interagency Working Group on the Social Cost of Carbon (IWG) developed a methodology for estimating the SCC and applied that methodology to produce estimates that government agencies use in regulatory impact analyses. The IWG requested this Academies interim report to determine if a near-term update to the SCC is warranted, with specific questions pertaining to the representation of the equilibrium response of the climate system in the integrated assessment models used by the SCC modeling structure, as well as the presentation of uncertainty of the SCC estimates. This interim report is the first of two reports requested by the IWG: the second (Phase 2) report will examine potential approaches for a more comprehensive update to the SCC estimates.
From the section Discussions, Conclusions, and Recommendations:
CONCLUSION 2 The relationship between CO2 emissions and global mean surface temperature can be summarized by four metrics: equilibrium climate sensitivity (ECS), transient climate response, transient climate response to emissions, and the initial pulse-adjustment timescale. ECS is less relevant than the other three metrics in characterizing the climate system response on timescales of less than a century. As a long-term, equilibrium metric, ECS alone does not provide an adequate summary of the relationship between CO2 emissions and global mean surface temperature for calculating the social cost of carbon (SCC).
RECOMMENDATION 1 The committee recommends against a near-term update to the social cost of carbon based simply on a recalibration of the probability distribution of the ECS to reflect the recent consensus statement in the IPCC AR5. Consequently, the committee also recommends against a near term change in the distributional form of the ECS.
Rather than updating the ECS in the current framework, the IWG could undertake efforts to adopt or develop a common “module” that represents the relationship between CO2 emissions and global mean surface temperature change, its uncertainty, and its profile over time.
The module’s behavior should be consistent with the best available scientific understanding of the relationship between emissions and temperature change, its pattern over time, and its uncertainty. Specifically, the module should be assessed on the basis of both its response to a pulse of emissions and its response to long-term forcing trajectories (specifically, trajectories designed to assess transient climate response and transient climate response to emissions, as well as high- and low emissions baseline trajectories). Given the degree of assessment they face, including consistency with observational data, the IPCC-class Earth system models provide a reference for evaluating the central projections of a climate module.
For regulatory decision making, it is at least conceptually possible to describe the uncertainty of these inputs in SCC calculations using probability distributions. Ideally, joint probability distributions could be defined for all of the uncertain inputs to an SCC-IAM, and the impact of uncertainty on the SCC could be evaluated using Monte Carlo analysis or a related approach.
One reason for modeling uncertainty is related to nonlinearities. If the SCC calculation involves nonlinearities over the range of uncertain parameters, the average value of the SCC computed from random draws of these uncertain inputs may not be the same as the single SCC computed from the average parameter values. The implications of such nonlinearities may be difficult to know a priori, suggesting it is best to compute the SCC from random draws of uncertain inputs.
In constructing the SCC, the IWG treated some parameters of the climate system and damage functions as uncertain and random and represented these parameters using probability distributions.
RECOMMENDATION 2 When presenting the social cost of carbon (SCC) estimates, IWG should continue to make explicit the sources of uncertainty.
CONCLUSION 4 Multiple runs from three models provide a frequency distribution of the SCC estimates based on five socioeconomic-emissions scenarios, three discount rates, draws from the equilibrium climate sensitivity distribution, and other model-specific uncertain parameters. This set of estimates does not yield a probability distribution that fully characterizes uncertainty about the SCC.
The committee notes that none of the three SCC-IAMs are sufficiently comprehensive to include all of the uncertainties in the inputs that are likely to be important in calculating the SCC. Moreover, explicit distributions for some important inputs (e.g., emission scenarios, economic growth, and population) have not been developed by the IWG for use in estimating the SCC. Factors omitted or not adequately captured by the analysis need to be better characterized. In addition, a single unifying discussion of captured and omitted uncertainty is needed.
RECOMMENDATION 3 The IWG should expand its discussion of the sources of uncertainty in inputs used to estimate the SCC, when presenting uncertainty in the SCC estimates. The IWG should include a section entitled “Treatment of Uncertainty” in each technical support document updating the SCC. This section should discuss various types of uncertainty and how they were handled in estimating the SCC, as well as sources of uncertainty that are not captured in current SCC estimates.
The uncertainties discussed in this section would include the uncertain parameters unique to each of the models, uncertainty about climate change impacts and their valuation, and the risk of potential catastrophic outcomes. The section would also discuss the implicit, equal weight placed on the three IAMs and five socioeconomic scenarios in computing an average SCC, the possible alternatives of unequal weights or alternative models and scenarios, and the motivation for the chosen approach.
CONCLUSION 5 It is important to continue to separate the impact of the discount rate on the social cost of carbon from the impact of other sources of variability. A balanced presentation of uncertainty includes both low and high values conditioned on each discount rate.
The report makes several good recommendations, related to a more complete assessment of uncertainty, and de-emphasizing the ECS in favor of other metrics such as TCR. I am of course in favor of more complete assessment of uncertainty. And I agree that TCR is a better metric to use for this purpose than ECS, since the TCR is a better reflection of what we can expect in the 21st century.
However, I am gobsmacked that they think the current IWG values of ECS are fine (which were based on the AR4). There is a raging debate on the discrepancy between climate model estimates of ECS (which the NAS report supports) versus the values derived from historical observations using simple energy balance models. And it is far more difficult to criticize the values of TCR determined from the energy balance models.
For further info, read the full report (Section 3). Here is the summary paragraph:
In summary, the change in the ECS distribution between AR4 and AR5 is small relative to the remaining uncertainties in this and other parameters that determine the SCC. This change arose primarily from assumptions about the multicentury adjustment of the climate system to a constant forcing that remain contested in the literature since the AR5. Neglected processes primarily affect the upper bound on ECS, continuing to support a positively skewed distributional form for this parameter such as that used by Roe and Baker (2007). The AR4 did not give a likely range for TCR that is directly comparable to that in the AR5, but the AR5 did reduce the probability of TCR values greater than 3°C from 10 to 5 percent, reflecting greater confidence and consensus on the upper bound for this parameter.
Well, given the huge uncertainties in other aspects of the SCC determination, perhaps discrepancies in the ECS determination are ‘in the noise.’ But they are nevertheless non trivial in terms of the policy implications, since climate sensitivity is the main driver of all this.
But the key issue is this. There are two credible sources of information on ECS/TCR: global climate models and historical observations (paleo estimates aren’t relevant for SCC). These two sources produce different PDFs, and the large tail values in the AR5 (10% likelihood of ECS > 6C) arise from paleo estimates, or historical estimates using flawed methodologies. The plausible upper bounds from climate models and historical estimates does not exceed 4.5C. It is very difficult to defend a fat tail out to ECS > 7C (used by the IWG), and the fat tail values drive high SCC values.
There is a highly relevant new paper by Dayaratna, Mckittrick and Kreutzer entitled Empirically-constrained climate sensitivity and the social cost of carbon.
Abstract. Integrated Assessment Models (IAMs) require parameterization of both economic and climatic processes. The latter include Ocean Heat Uptake (OHU) efficiency, which represents the rate of heat exchange between the atmosphere and the deep ocean, and Equilibrium Climate Sensitivity (ECS), or the surface temperature response to doubling of CO2 levels after adjustment of the deep ocean. Due to a lack of adequate data, OHU and ECS parameter distributions in IAMs have been based on simulations from climate models. In recent years, new and sufficiently long observational data sets have emerged to support a growing body of empirical ECS estimates, but the results have not been applied in IAMs. We incorporate a recent observational estimate of the ECS distribution conditioned on observed OHU efficiency into two widely-used IAMs. The resulting Social Cost of Carbon (SCC) estimates are much smaller than those from models based on simulated parameters. In the DICE model the average SCC falls by 30-50% depending on the discount rate, while in the FUND model the average SCC falls by over 80%. The span of estimates across discount rates also shrinks considerably, implying less sensitivity to this parameter choice.
They use the Lewis and Curry (2015) estimates of ECS [link]. Note that Nic has recently updated the values to account for more recent data and lower aerosol forcing [link]. These new results are not yet published. Of particular note is that the lower aerosol forcing substantially reduces the ‘fat tail’.
Well, it is certainly a good thing that the IWG’s calculation of the social cost of carbon is being evaluated. The NAS Committee has made some good recommendations.
The main failing thus far is a serious re-assessment of the values of ECS and TCR to be used in these calculations. My recommendation is that two separate distributions be used: one from climate models, and the other from the most reliable of the estimates using historical observations. The uncertainties associated with aerosol forcing need to be accounted for. Reductions of the SCC by more than a factor of two, simply by using LC sensitivity distribution, is a significant finding in terms of how we interpret the social cost of carbon.