This chapter addresses many of the specific issues raised by the IWG for the committee’s consideration and provides suggestions for a path forward. It concludes that, in the longer term, the development of a new damages module, satisfying the scientific criteria stated in Recommendation 2-2, in Chapter 2 (scientific basis, uncertainty characterization, and transparency), and addressing some of the challenges identified by the committee and by the IWG in its 2010 Technical Support Document, is merited. Since such a research effort is likely to consume significant resources and time, this chapter also recommends a set of improvements the IWG could undertake in the near term.
The first section below reviews the damage components of the integrated assessment models used to estimate the social cost of carbon (SC-IAMs).1 The second section discusses alternate approaches to estimating climate damages as well as some of the recent literature on damage estimation. The third section provides the committee’s recommendations for improvements in the near term. In the final section the committee offers recommendations for a new damage module that could be developed in the longer term and outlines its properties.
1 These are the three integrated assessment models widely used to produce estimates of the SC-CO2: the Dynamic Integrated Climate-Economy (DICE) model, the Framework for Uncertainty, Negotiation and Distribution (FUND) model, and the Policy Analysis of the Greenhouse Effect (PAGE) model (see Chapter 1).
Currently, the damage component of an SC-IAM translates streams of socioeconomic variables (e.g., income and population and gross domestic product [GDP]) and physical climatic variables (e.g., changes in temperature and sea level) into streams of monetized damages over time. To do this, it must represent relationships among physical variables, socioeconomic variables, and damages. To date, the SC-IAMs and related literature consists of damage representations that are either simple and global (e.g., global damages as a function of global mean temperature) or sectorally and regionally disaggregated (e.g., agricultural damages as a function of regional temperature, precipitation change, and CO2 concentrations).
The damage formulations in the SC-IAMs differ substantially in their sectoral and regional disaggregation of damages, functional forms, drivers of damages, and consideration of parametric uncertainty: see Table 5-1. All three SC-IAM damage components take global mean temperature, global mean sea level, and socioeconomic projections (global population and GDP) as inputs for computing damages. The models differ in their use of the drivers of damages with respect to other climate variables (e.g., CO2 concentrations, regional temperature), regional socioeconomic projections and sectoral detail (e.g., the agricultural share of the economy, energy efficiency of space cooling and heating), demographic detail (e.g., population density), and other factors. The models also vary in the representation of adaptation, which is implicit in the DICE parameterization, explicit in FUND and PAGE and endogenous only in FUND.
The IWG currently runs each of the SC-IAMs in a simulation mode with information passed from one module to another in a once-through fashion. Thus, the models do not optimize the social response to climate change (except for FUND’s adaptation to sea level rise). There are varying degrees of feedbacks to socioeconomic elements (e.g., through effects on GDP or capital stocks) and climate (e.g., through effects on emissions or albedo) in the IWG SC-IAMs (shown in Table 5-1).
In the SC-IAMs, all damages are represented in terms of dollars as fractions of global or regional GDP. Damages therefore scale with the size of the economy, with the rate varying across models and sometimes regions (Rose et al., 2014b). Global damages are a simple summation across sectors and regions (or just across sectors in the case of DICE). Physical units are computed first for some damages, such as mortality and morbidity effects in FUND, but not for all damages in all three models.
The current approach to damage calculations taken by the IWG, to varying degrees, considers three kinds of uncertainty—input (temperature and CO2 concentration changes, sea level rise, and socioeconomic), parametric (within model), and structural (via the differences in damage
TABLE 5-1 Structural and Implementation Characteristics of Damage Components in SC-IAMs
|Characteristic||DICE 2010||FUND 3.8||PAGE 2009|
|Regions||1 region||16 regions||8 regions|
|Damage Sectors||2 sectors: sea level rise, aggregate non-sea level rise*||14 sectors: sea level rise, agriculture, forests, heating, cooling, water resources, tropical storms, extratropical storms, biodiversity, cardiovascular and respiratory mortality, vector borne diseases, morbidity, diarrhea, migration||4 sectors: sea level rise, economic, noneconomic (i.e., not in GDP), discontinuity (e.g., abrupt change or catastrophe)|
|Sea Level Rise Damage Specification (Fraction of Income)||Quadratic function of global sea level rise||Additive functions for coastal protection costs, dryland loss, and wetland loss, based on an internal cost-benefit rule for optimal adaptation||Power function of global sea level rise|
|Drivers of Sea Level Rise Damage||Global mean sea level rise, income||Global mean sea level rise, dryland value, wetland value, topography, protection cost, population density, income density, per capita income||Global mean sea level rise, regional coast length scaling factor relative to European Union, adaptation capacity and costs, per capita income, income|
|Non-Sea Level Rise Damage Specifications (Fraction of Income)||Quadratic function of global temperature||Uniquely formulated nonlinear functions by sector (see Anthoff and Tol, 2014)||Power function of regional temperature|
|Characteristic||DICE 2010||FUND 3.8||PAGE 2009|
|Non-Sea Level Rise Damage Drivers||Global mean temperature, income||Global mean temperature, CO2 concentrations (for carbon fertilization and storms), population, income, technological change||Regional temperature, regional scaling factor relative to the European Union, adaptation capacity and costs, population, income|
|Adaptation||Implicit (damages net of adaptation)||Explicit for agriculture and sea level rise, implicit otherwise (econometric studies of net response to warming)||Two types of exogenous fixed adaptation policy that reduce impacts for a cost|
|Climate Benefits||Implicit (damages net of benefits)||Explicit outcome of certain sectoral damage functions (e.g., avoided heating demand, agriculture benefits from CO2 fertilization)||Assumes small economic benefits at low levels of warming|
|Damages Due to Abrupt Climate Change||Included in calibration of aggregate damages not from sea level rise*||No explicit representation||Unspecified “discontinuity” impact occurs with a positive probability at global average temperature changes greater than 3 °C|
|Feedbacks from Damages||Damages affect global income, which affects future global capital stocks and income levels, but projected emissions are unaffected||No economic feedback||No economic feedback|
*These damages are an aggregate based on a calibration of sectoral damages according to Nordhaus and Boyer (2000) and rescaled using external aggregate damage information. For additional details, see note a to Table 5-2.
SOURCE: Adapted from Rose et al. (2014b, Table 6-1).
formulations among the three models). Input uncertainty is considered in the form of alternative climate and socioeconomic input projections (see Chapters 3 and 4). Parametric uncertainty is considered in the damage formulations of two of the current SC-IAMs, FUND and PAGE. DICE in its standard formulation, used by the IWG, does not consider parametric uncertainty, although a variety of studies have explored some forms of parametric or structural uncertainty with versions of the DICE damage function (e.g., Nordhaus and Popp, 1997; Azar and Lindgren, 2003; Keller et al., 2004; Ackerman et al., 2010; Kopp et al., 2012; Cai et al., 2016; Lemoine and Traeger, 2016). The parametric uncertainty specifications in FUND and PAGE differ, with FUND representing larger uncertainty in annual damages through 2100, but less than PAGE after 2100, and PAGE exhibiting higher average annual damages (Rose et al., 2014b). Structural uncertainty is considered to a degree in the IWG’s framework by including the three SC-IAMs. However, as Table 5-2 shows, the most recent SC-IAM formulations for PAGE and DICE exhibit some degree of dependency on the other models (see discussion below on model dependency).
Another attribute of the SC-IAMs that underpin the current IWG estimates is that much of the research on which they are based is dated. As Table 5-2 shows, the damage formulations do not in many cases reflect recent advances in the scientific literature (e.g., some using sources not more recent than the 1990s and early 2000s).
Figure 5-1 illustrates that there are significant differences across models in global damage response to key input drivers of damages. DICE and PAGE yield higher damages for a given level of warming and income and
TABLE 5-2 Literature Sources for Current SC-IAM Damage Component Specifications
|Model (Version)||Damage Type||Study||Basis for Damage Estimate|
|DICE 2010a||Aggregate non-sea level rise||Literature surveys Intergovernmental Panel on Climate Change (2007a) and Tol (2009) used to rescale Nordhaus and Boyer (2000) sectoral damagesb||Calibration|
|SLR coastal impacts||Undocumented|
|Model (Version)||Damage Type||Study||Basis for Damage Estimate|
|FUND 3.8||Agriculture||Kane et al. (1992), Reilly et al. (1994), Morita et al. (1994), Fischer et al. (1996), Tsigas et al. (1996)||Calibration|
|Tol (2002b)||Income elasticity|
|Forestry||Perez-Garcia et al. (1995), Sohngen et al. (2001)||Calibration|
|Tol (2002b)||Income elasticity|
|Energy||Downing et al. (1995, 1996)||Calibration|
|Hodgson and Miller (1995)||Income elasticity|
|Water resources||Downing et al. (1995, 1996)||Calibration|
|Downing et al. (1995, 1996)||Income elasticity|
|Coastal impacts||Hoozemans et al. (1993), Bijlsma et al. (1996), Leatherman and Nicholls (1995), Nicholls and Leatherman (1995), Brander et al. (2006)||Calibration|
|Diarrhea||Global Burden of Disease 2000 estimatesc||Calibration|
|Global Burden of Disease 2000 estimatesc||Income elasticity|
|Vector-borne diseases||Martin and Lefebvre (1995), Martens et al. (1995, 1997), Morita et al. (1994)||Calibration|
|Link and Tol (2004)||Income elasticity|
|Cardiovascular and respiratory mortality||Martens (1998)||Calibration|
|Storms||CRED EM-DAT database,d World Meteorological Organization (2006)||Calibration|
|Toya and Skidmore (2007)||Income elasticity|
|Ecosystems||Pearce and Moran (1994), Tol (2002a)||Calibration|
|Model (Version)||Damage Type||Study||Basis for Damage Estimate|
|PAGE09||SLR||Anthoff et al. (2006)e||Calibration and income elasticity|
|Economic||Warren et al. (2006)f||Calibration|
|Noneconomic||Warren et al. (2006)||Calibration|
|Discontinuity||Lenton et al. (2008), Nichols et al. (2008), Anthoff et al. (2006), Nordhaus (1994a)g||Calibration|
|Adaptation costs||Parry et al. (2009)||Calibration|
aThe committee assembled the following information related to the calibration of DICE 2010 based on communications with William Nordhaus, Nordhaus and Boyer (2000), and Nordhaus (2010). DICE global damages have historically been calibrated to the aggregate results of another model, RICE, which has regional and sectoral damage calibrations. RICE 2000 is the last full set of regional and sectoral damage estimates that are fully documented for the DICE/RICE family of models, and DICE 2000’s global estimate was calibrated to RICE 2000. These estimates were based on Nordhaus and Boyer (2000) and calibrated at the sector level using the following sources as the main references: agriculture (Darwin et al., 1995), health (Murray et al., 1996), energy (Nordhaus and Boyer, 2000), recreation (Nordhaus and Boyer, 2000), human settlements and natural ecosystems (Nordhaus and Boyer, 2000), coastal impacts (Yohe and Schlesinger, 1998), and catastrophic damages (Nordhaus, 1994a; Nordhaus and Boyer, 2000). Updates to DICE/RICE prior to DICE 2013 have used the same sectoral breakdown of damages as RICE 2000 but changed the aggregate based on further information. According to Nordhaus (2010), Tol (2009), and Intergovernmental Panel on Climate Change (2007a) were the additional information used for DICE/RICE 2010. However, the specifics of the recalibration are not available. For information regarding DICE 2007, which was used for the Interagency Working Group on the Social Cost of Carbon (2010) SCC estimates, see Nordhaus (2007, 2008).
cSee http://www.who.int/healthinfo/global_burden_disease/estimates_regional_2000/en [November 2016].
are much more responsive to both temperature change and income than FUND, while none of the models are particularly responsive to global population size.2 Communicating and providing scientific justification for these differences is critical, as discussed below. For each model, the slope of the temperature response is indicative of the projected incremental damages resulting from a pulse of CO2.
Figure 5-2 displays the estimated incremental damages over time produced by the three SC-CO2 models in response to an identical incremental change in projected temperature from a CO2 emissions pulse in
2 The population results in Figure 5-1 are from a sensitivity analysis that scales global population with the regional distribution fixed. Population enters each of the SC-IAMs differently. In DICE, population affects total factor productivity, income, and the capital stock. In FUND, population affects per capita income and is an explicit input variable in a number of individual damage categories (water resources, energy consumption, ecosystems, various human health damage categories, tropical storms). In PAGE, population affects per capita income, which enters into each category of damages within the model.
2020.3 Annual incremental damages differ in sign, magnitude, timing, and regional and sectoral composition. Underlying Figure 5-2 are significant differences in total global damage levels. For example, DICE and PAGE produce annual global damages in 2100 that are four times larger than those from FUND for the same reference climate and socioeconomic future used for Figure 5-2. The models also differ notably in the size and sensitivity of their responses to key uncertain inputs, as shown in Figure 5-1 (Rose et al., 2014b).
The differences in model characteristics (shown in Table 5-1) drive the differences in results, with specific characteristics playing a prominent
3 Each model’s damage component was first run with identical reference temperature, CO2 concentration, and socioeconomic projections using the IWG’s highest emissions (and corresponding socioeconomic) scenario, and then run again with identical incrementally higher temperature and CO2 concentration projections resulting from a 2020 1 billion ton carbon (3.7 Gt CO2) emissions pulse. Projected incremental damages over time are the difference between the projected damages in the two scenarios. The results in Figure 5-1 reflect differences across models in the modeling of sea level, regional temperatures, and damages, which are all driven by global average temperature change in the models (see Table 5-1). See Rose et al. (2014b) for more details and discussion.
role. For example, in DICE, damages are based on quadratic functions of temperature and sea level rise; in FUND, net benefits in the agricultural sector result at lower warming levels, adaptation addresses much of the risk from sea level rise, cooling energy demand costs are a large fraction of damages, and “catastrophic” damages are not included; and, in PAGE, regional damages are computed by scaling damages between regions, and a large fraction of damages are from those that do not directly impact GDP and an unspecified discontinuity damage.
The committee evaluated the damage components of the three SC-CO2 IAMs according to the criteria in Recommendation 2-2, in Chapter 2. Overall, the damage formulations of the three models used by the IWG, and the overall IWG damage modeling approach, differ in the extent to which they satisfy the three criteria: scientific basis, uncertainty characterization, and transparency. None of the damage components fully satisfies all the criteria. Analysis of the IWG documentation, individual model documentation, and outside research suggest a number of elements in the current damage functions, individual model results, and damage components as a whole that can be improved.
The committee notes that the Interagency Working Group on the Social Cost of Carbon (2010) identified a number of potential shortcomings and critiques of the current damage formulations, which are discussed further below. These include:
- incomplete treatment of noncatastrophic damages;
- incomplete treatment of potential catastrophic damages;
- uncertainty in extrapolation of damages to high temperatures;
- incomplete treatment of adaptation and technological change;
- omission of risk aversion with respect to high-impact damages;
- failure to incorporate intersectoral and interregional interactions; and
- imperfect substitutability of consumption for environmental amenities.
The committee defines climate impacts as the biophysical or social effects driven by climate change (e.g., changes in land productivity, mortality, morbidity, water supply, coastal flooding, or conflict) and climate damages as the monetized estimates of the social welfare effects of climate impacts (see Box 2-2 in Chapter 2). Impacts estimates are either an explicit input or implicit element of projected damages.
The damage component of a reduced-form IAM is composed of damage functions that monetize climate change effects, with functional forms
and calibrations that are in some way derived from and calibrated to more detailed climate change impacts and damage analyses, other parameters (e.g., economic elasticities), and a modeler’s judgment (see Table 5-2). SC-IAM damage functions are thus constrained by the available literature, and they typically need to extrapolate beyond the relationships characterized in the detailed supporting analyses, for instance, beyond the warming levels evaluated or locations studied.
The scientific literature has produced studies of damages and impacts using physical process models, structural economic models, and empirical models.
- Physical process models describe the dynamics of a physical process to identify a climate change–induced physical impact and evaluate its implications: for example, crop models assess the impact of temperature, precipitation, CO2 concentrations, and other drivers on plant productivity.
- Structural economic models describe the structure and dynamics of economic decisions and markets to evaluate the net economic implications of climate-induced physical changes: for example, they can assess the economic consequences of climate-related changes in the productivity of land or the labor force, as well as the demand for heating or cooling.
- Empirical models estimate statistical relationships between weather (short-run) or climatic (long-run) variables and human or ecological responses from historical data: for example, they are used to estimate dose-response functions between exposure to temperature and mortality.4
The literature includes impact and damage research that varies in scope. It includes studies of individual and multiple sectors, studies at local, national, and global geographic scales, studies using higher and lower spatial resolution, studies that model different processes and interactions, and studies that focus on market and nonmarket damages. The differences in methodologies and in scope create challenges for users trying to synthesize understanding of impacts or damages. For instance, structural and empirical methods are fundamentally different from one
4 Early empirical work largely relied on cross-sectional techniques (i.e., comparing the relationship between climate and outcomes across space, potentially capturing other factors that lead to spatial variability). The most recent empirical literature has employed methodological insights from the causal inference literature, which has resulted in numerous estimates for a number of sectors that actually reflect causal relationships between weather/climate and economically relevant variables, represent populations of interest in damage function calibration, incorporate nonlinearities, and empirically reflect historical forms of adaptation.
another and, as a consequence, they produce results that may not be directly comparable. These comparability and scope issues need to be addressed in some way when developing damage functions.
As discussed above, each of the current SC-IAM damage components has some direct or indirect link to the damages literature of the 1990s and early 2000s. The literature has, however, evolved substantially since then. This more recent literature yields economic estimates that could be integrated into SC-CO2 modeling in the near term. The research community has also initiated activities that will yield useful impacts and damages information in the longer term. These activities are important to monitor and are discussed in the context of our proposals for the longer term, in the final section of this chapter.
Table 5-3 lists a number of studies that could be used as resources for a near-term update to individual SC-IAM damage formulations and the damages module as a whole. This table is not comprehensive, and this
TABLE 5-3 Selected List of Climate Damages Literature for a Near-Term Update
|Impacts||Regions||References and Sources of Information|
|Health, infrastructure, electricity, water resources, agriculture and forestry, ecosystems||United States||Waldhoff et al. (2015), Marten et al. (2013) www2.epa.gov/cira [December 2016]|
|Agriculture, energy, river floods, forest fires, transport infrastructure, coastal areas, tourism, human health, habitat suitability||Europe||Ciscar et al. (2011, 2014) https://ec.europa.eu/jrc/en/peseta [January 2017]|
|Agriculture, labor productivity, mortality, property and violent crime, energy demand, coastal storms and inundation||United States||Houser et al. (2015)|
|Heat extremes and health, agriculture and land use, tropical cyclones, sea level rise, drought and conflict||Global||https://chsp.ucar.edu/brace [December 2016]|
|Sea level rise, agricultural productivity, heat effects on labor productivity, human health, tourism flows and households’ energy demand||Global||Roson and Sartori (2010, 2016)|
|Impacts||Regions||References and Sources of Information|
|Sea level, agriculture, and energy demand||Global||Bosello et al. (2012)|
|Agriculture||Global||Reilly et al. (2007), Kyle et al. (2014), Nelson et al. (2014)|
|Coastal damages||Global||Diaz (2016)|
|Energy demand||Global||Isaac and van Vuuren (2009), Mima and Criqui (2009), Labriet et al. (2013), Zhou et al. (2013)|
|Energy supply||Global||Mima and Criqui (2009), Labriet et al. (2013), Kyle et al. (2014)|
|Water||Global||Hanasaki et al. (2013), Hejazi et al. (2014), Schlosser et al. (2014), Kim et al. (2016)|
|Ecosystem services||Global||http://www.naturalcapitalproject.org/invest [December 2016]|
|Empirical adaptation response||Regional, multiple sectors (agriculture, energy, mortality)||Auffhammer and Aroonruengsawat (2012), Barreca et al. (2015), Hsiang and Narita (2012); Hsiang and Jina (2014), Butler and Huybers (2013)|
section does not review and assess the literature; the time frame for this report did not allow for such an activity. This newer literature needs to be considered and, to the extent possible, incorporated in the near-term update.
Since the studies that are used to calibrate the SC-IAMs were conducted, there has been significant progress in research into both market and nonmarket damages, and in methods using both empirical and structural models. In the future, the calibration of damage functions needs to be compared to point estimates from newer literature as either validation of or justification for updates; and, where possible, assessment of the damage calibration using hindcasting and comparisons to empirical
studies will be valuable. Going forward, there are necessary and complementary roles for both empirical and structural modeling.5
In the near term, the IWG has two options for developing the damages module of an integrated SC-CO2 estimation framework: an improved damage component of a single SC-IAM (or another reduced-form IAM) or improved damage components from multiple SC-IAMs (or other reduced-form IAMs). While the committee does not recommend a specific path for the IWG, it recommends a set of steps for any damage component and module used in a near-term update of the SC-CO2 estimates.
RECOMMENDATION 5-1 In the near term, the Interagency Working Group should develop a damages module using elements from the current SC-IAM damage components and scientific literature. The damages module should meet the committee’s overall criteria for scientific basis, transparency, and uncertainty characterization (see Recommendation 2-2, in Chapter 2) and include the following four additional improvements:
- Individual sectoral damage functions should be updated as feasible.
- Damage function calibrations should be transparently and quantitatively characterized.
- If multiple damage formulations are used, they should recognize any correlations between formulations.
- A summary should be provided of disaggregated (incremental and total) damage projections underlying SC-CO2 calculations, including how they scale with temperature, income, and population.
These improvements are discussed in the four sections below.
In the near term, the IWG will need to choose which damage formulations to include in the damages module. Whether the IWG includes multiple formulations or only a single one, the damage formulations need to be consistent with the recent literature. The IWG’s choice in this matter has implications for the level of disaggregation required from
5 Information obtained through a focused literature review performed for the committee by Frances Moore (University of California, Davis) and Delavane Diaz (Electric Power Research Institute (EPRI).
the socioeconomic and climate modules in the near term. It is important to differentiate between the spatial and temporal level of aggregation in input data used in calibration of the damage formulation(s) and the level of aggregation represented in an SC-IAM. Calibration of damage formulations may be done using data at a higher resolution than represented in the IAM. The two previous chapters offer guidance on how disaggregation across regions (and sectors) could be accomplished in the near term; early coordination of disaggregation choices in the damages module with the socioeconomic and climate modules will be important for smooth implementation of the committee’s recommended modular approach. However, there is no ideal disaggregation level, as there are many factors to consider and tradeoffs with high and low resolution. (See the disaggregation section below for additional discussion.) In addition, documentation for each damage formulation—its implementation (i.e., how it is run and how uncertainty is modeled), and aggregation across formulations—needs to be provided with sufficient detail and justification for the scientific community to understand and assess the modeling.
Below, guidance is provided for a near-term revision by discussing each of the four points in Recommendation 5-1 above. In addition, Appendix G presents model-specific improvements for each of the SC-IAM damage formulations that could be pursued during a near-term update if the IWG wished to continue with some elements of one or more of the SC-IAMs. The IWG may also wish to consider additional damage formulations that have been published in the peer-reviewed literature (e.g., Roson and van der Mensbrugghe, 2012). Any alternative formulations, their implementation, and potential multi-model integration would also need to be evaluated applying the criteria in Recommendations 2-2 (in Chapter 2) and 5-1 (above).
Updating Individual Sectoral Damage Functions
As discussed above, research on climate damages has advanced beyond the studies underlying the current SC-IAM damage components. A newer and substantial body of additional empirical and structural modelling literature is now available. The literature on agriculture, mortality, coastal damages, and energy demand provide immediate opportunities to update the SC-IAMs. For example, Moore et al. (2016) provide a possible blueprint for how to achieve this for FUND. Points of departure in terms of resources that could be used for updating damage components include the studies listed in Table 5-3 (above), the empirical studies reviewed in sources, such as Dell et al. (2014) and Carleton and Hsiang (2016), and other individual peer-reviewed papers with economic damage estimates (based on either structural economic models or empirical estimates).
A key challenge as noted above will be to determine how to use economic damage results from different methods that are not fully comparable. Although many studies do not follow the causal chain all the way to monetized welfare losses and are not global in extent, they still may be used for assessing the calibration of biophysical impacts and damages in particular regions. The comparisons need to be conducted with awareness of the different ways in which the studies account for adaptation. There have been significant improvements in understanding and measurement of adaptive responses for some sectors in empirical and structural modeling, which could be considered in some way in a near-term update. Table 5-3 illustrates that agriculture, energy, mortality, and coastal damages provide some of the most immediate opportunities for updates, with both empirical (last row of Table 5-3) and structural modeling analyses (various rows).
Damage Function Calibrations
The damage formulations currently used in the SC-IAMs are not clearly and adequately justified with regard to how they are parameterized and calibrated and how particular sectors and regions contribute to the overall results. This inadequacy stems from the incomplete documentation of the individual SC-IAMs. DICE and FUND do provide some documentation for the parameterization and calibration of their models, but the accounting of how sectors and regions contribute to the damage function is not transparent. It is not possible to understand with great confidence the actual damage function calibrations and the magnitude of the sectoral contributions, even after investigating different versions of the model code, documentation, and related papers. In addition, PAGE does not provide a detailed description and scientific justification of how its damage component is parameterized.
Going forward, any damage component used in the calculation of the SC-CO2 needs to provide a clear accounting of the calibration of the damage functions. Such documentation will significantly improve scientific rationale and transparency and allow for improved scientific assessment. For DICE 2010, for example, adequate documentation would mean a clear description of the calibration of the global sea level rise and non-sea-level rise damage functions, as well as details regarding any underlying calibrations at the sector and regional levels. For FUND and PAGE, adequate documentation would entail a clear description of the calibration of the region-sector damage functions. This description will likely require input from the modelers themselves. If the damage functions are updated as detailed in the preceding section, the calibration of these updated functions would need to be documented.
Combining Multiple Damage Formulations
The IWG has pooled the results of three SC-IAMs to estimate the SC-CO2. Pooling results of multiple SC-IAMs is a method to incorporate structural uncertainty, as each model provides an alternative representation of how damages depend on climate change and other factors. However, when aggregating across models, it is important to consider the degree of dependence of the estimates across models: see Box 5-1. If the models are independent, aggregation of the results provides more information than any single model, but if the models are dependent, combining results may provide little additional information. Moreover, analysts might mistakenly underestimate the degree of uncertainty about the SC-CO2 if they combine results of dependent models on the assumption that the models are independent.
If the extent of dependence among the models is known, one can estimate the extent to which the structural uncertainty that is captured is reduced, in comparison with a case in which the models are independent. Specifically, one can estimate the number of independent models that would yield an output distribution with a similar spread (Clemen and Winkler, 1985). It is difficult, however, to appropriately characterize the dependence among models. Damage components of all of the SC-IAMs draw on a common literature, yet they use very different functional forms, which contribute to the differences in damages displayed in Figure 5-2 (above). In addition, some of the damage components draw on results of the damage components of current or previous versions of the SC-IAMs (see Table 5-2). The use of a common literature is appropriate; it is desirable that models be based on the best available scientific evidence, and a model that ignored relevant parts of the literature could be improved by including those parts. The reliance on damage components of other SC-IAMs is more problematic. This reliance induces dependence among the models that affects the extent to which structural uncertainty is captured by using multiple models. This dependence needs to be recognized when aggregating the model outputs, but it is not clear how to characterize the dependence and quantify its effect on the representation of structural uncertainty.
Whether the models are independent or not does not affect the interpretation of the central value of the distribution of SC-CO2 estimates obtained by pooling results across the models. If each of the models is judged to be unbiased (in the statistical sense of not systematically overestimating or underestimating damages), then each model provides an unbiased estimate of damages. In this case, the average of their results is also unbiased. The degree of independence affects the spread of the results but not the central value.
time, regions, and sectors, as well as a characterization of the uncertainty in results. This will improve the transparency and credibility of the individual damage formulations. Given the large potential volume of data, the IWG could provide a representative, summary characterization of the disaggregated damages underlying the SC-CO2 estimates. In addition, the IWG could provide the dataset of intermediate and disaggregated results to the public. See Rose et al. (2014b) for the kind of results the committee suggests be provided in the near term. Two examples are displayed in Figures 5-1 and 5-2.
This section offers a set of desirable characteristics of a damages module that the committee believes can be developed in the longer term, given current scientific understanding. The committee believes that work on such a module could commence immediately and proceed in parallel with implementation of the committee’s near-term recommendation, discussed in the preceding Section.
RECOMMENDATION 5-2 In the longer term, the Interagency Working Group should develop a damages module that meets the overall criteria for scientific basis, transparency, and uncertainty characterization (see Recommendation 2-2, in Chapter 2) and has the following five features:
- It should disaggregate market and nonmarket climate damages by region and sector, with results that are presented in both monetary and natural units and that are consistent with empirical and structural economic studies of sectoral impacts and damages.
- It should include representation of important interactions and spillovers among regions and sectors, as well as feedbacks to other modules.
- It should explicitly recognize and consider damages that affect welfare either directly or through changes to consumption, capital stocks (physical, human, natural), or through other channels.
- It should include representation of adaptation to climate change and the costs of adaptation.
- It should include representation of nongradual damages, such as those associated with critical climatic or socioeconomic thresholds.
Developing a damages module with these characteristics would represent a major advance in understanding the monetary impacts of climate change. In the rest of this section the committee discusses in more detail each of the five features.
Disaggregation of Climate Damages by Region and Sector
Regional and sectoral damage resolution is needed for transparency and to connect estimates to the literature on impacts and damages. However, a priori, there is no ideal disaggregation level. There are a number of factors to consider in determining an appropriate level of disaggregation, including the timescale over which damages are projected and whether the disaggregation is needed for the implementation or calibration of a damages module. In many cases, the level of disaggregation will be determined by the findings available from the literature on impacts and damages and the resolution of economic statistics, computational constraints, and the possible tradeoffs between capturing heterogeneity in climate risks (due to differences in markets, technology, policies, cultures, and physical systems) and feedbacks between affected groups and locations. In addition, the SC-CO2 context matters. For instance, SC-CO2 modeling does not need to have the same spatial and temporal resolution as desired for adaptation planning by a local (e.g., city) decision maker as it would for a national-level decision maker.
Damages could be incorporated in an IAM in one of three ways: (1) using a global reduced-form damages module that is calibrated to spatially and sectorally disaggregated damage formulations, (2) using a damages module that includes spatially and sectorally explicit modeling of relevant processes, or (3) using a directly calibrated and estimated global damages module. DICE 2007 and earlier versions took the first approach, attempting to calibrate a global damage function based on regional and sectoral damage functions that were calibrated to sectoral studies and a reinterpretation of expert elicitation results regarding the possibility of climate-linked economic “catastrophes” (Nordhaus, 1994a; Nordhaus and Boyer, 2000). FUND takes the second approach, with individual reduced-form damage functions for a range of sectors and impact types: agriculture, forestry, water resources and energy consumption, costs of protection against sea level rise, willingness to pay to avoid ecosystem loss, diarrhea, vector-borne diseases, cardiovascular disease, and tropical and extratropical storm damage (Anthoff and Tol, 2014). Though some more complex IAMs incorporate detailed representations of specific damage pathways (e.g., for energy demand), no IAM attempts to be both detailed and comprehensive (Nordhaus, 2014). The social cost of carbon (SCC) took the third approach: it attempted to estimate a total
global damage function directly, without a disaggregated calibration. It was based on an interpretation of a meta-analysis of past global damage estimates (Tol, 2009).
A total-damage approach might also be taken based on structured expert elicitation (Nordhaus, 1994a; Pindyck, 2015; Howard and Sylvan, 2016). However, the committee does not recommend an approach based on top-down estimation of a total global damage function because it lacks traceability to damage pathways, may not have a strong scientific rationale, or may not address nonmarket damages (e.g., Dell et al., 2012; Burke et al., 2016). More specific peer-reviewed structured expert elicitation studies that address hard-to-quantify damage categories may be useful in helping to calibrate a damage function to quantitative studies that examine specific impacts.
Structural economic and empirical models, such as those listed in Table 5-3, provide the main resource for calibrating damage formulations. Due to the detailed representation of the weather and climate links to impacts, using either structural economic or empirical models to project future changes requires a high level of spatial and temporal detail in climate and, possibly, in socioeconomic projections, comparable to the level of detail in the past observations with which they are being compared. This level of detail need not necessarily be provided by the climate module of a SC-IAM; however, results from detailed structural economic or empirical models could be used to calibrate relatively simple reduced-form models that require only relatively coarse spatial and temporal detail (as is the case in the current SC-IAMs).
Climate damages do not arise directly from physical climate variables, such as temperature or precipitation. They arise through biophysical or social pathways: agricultural damages arise because temperature and precipitation influence crop yields; labor productivity damages arise because temperatures and humidity affect the quantity and quality of work; and health and longevity are lost because of changes in heat stress and disease. Some physical climate impacts are of potentially great socioeconomic importance, but challenging to translate into dollars: for example, changes in the risk of civil conflict, human migration, or global biodiversity.
Climate damages can occur through a variety of pathways, some quantifiable, some identifiable but hard to quantify, and some unknown. In principle, the SC-CO2 estimates are intended to represent total economic damages, and thus they are the aggregate over all three types of pathways. However, these types of pathways are successively more difficult to estimate.
In order to provide a satisfactory degree of transparency, it is desirable for the damages module to report impacts in physical units when possible, such as crop yield changes, mortality, or species effects. These
natural-unit measures are more straightforward to compare to the impact literature and require fewer intermediary assumptions to estimate than their monetized counterparts. Moreover, reporting physical units for impacts that cannot be monetized allows for their inclusion in regulatory impact analyses, which is consistent with regulatory guidance.6
Representation of Important Interactions and Spillovers among Regions and Sectors
Most of the structural and empirical studies that can be used to calibrate a damage function focus on a single type of impact or on the direct effect of climate change on regions in isolation. There is an emerging literature that also incorporates interactions among regions and impacts (e.g., Reilly et al., 2007; Warren, 2011; Diffenbaugh et al., 2012; Taheripour et al., 2013; Baldos and Hertel, 2014; Grogan et al., 2015; Harrison et al., 2016; Zaveri et al., 2016). For example, given global markets, migration, and other factors, effects of a crop failure in India will also have impacts in other countries, and reductions in water availability in one region will have impacts across many regions and sectors.
One set of interactions occurs through market mechanisms, such as trade. For example, the economic impacts of climate change on crop yield in one region will depend in part on the changes in crop yields in other regions. These interactions can be captured by multisectoral, multiregional economic computable general equilibrium (CGE) models. Models of global agriculture and forestry impacts have been developed over more than two decades (e.g., Reilly et al., 1994; Sohngen et al., 2001; Reilly et al., 2007; Roson and van der Mensbrugghe, 2012; Nelson et al., 2014).
Impacts can also interact with each other, and with mitigation policy, through their effects on competition for resources, such as water and land. The relationship between temperature exposure and crop yields depends strongly on whether crops are irrigated (Schlenker and Roberts, 2009; Houser et al., 2015); the ability to irrigate will in turn depend on impacts on water resources.
Some impacts may partially represent adaptations to other impacts; care needs to be taken to avoid double counting. For example, increased demand for space cooling is the major driver of the increased energy costs associated with higher temperatures (e.g., Auffhammer and Mansur, 2014). Yet the widespread adoption of air conditioning significantly reduces the effect of temperature on mortality (Barreca et al., 2013) this paper makes
6 For example, OMB Circular A-4 notes that “Even when a benefit or cost cannot be expressed in monetary units, [an agency] should still try to measure it in terms of its physical units.”
two primary discoveries. The mortality effect of an extremely hot day declined by about 80 percent between 1900-1959 and 1960-2004. As a consequence, days with temperatures exceeding 90 °F were responsible for about 600 premature fatalities annually in the 1960-2004 period, compared to the approximately 3,600 premature fatalities that would have occurred if the temperature-mortality relationship from before 1960 still prevailed. Similarly, the sensitivity of labor supply to temperature depends to a large extent on whether workers are protected from outdoor temperatures (Graff Zivin and Neidell, 2014). Thus, increases in the impact of energy demand impact may be offset by decreases in other impacts.
In the SC-IAMs, damages to ecosystems are most often valued using contingent valuation estimates of existence value or direct ecosystem services (e.g., Nordhaus and Boyer, 2000; Anthoff and Tol, 2014). It is important, however, to note that damages to ecosystems may amplify other impacts. For instance, vegetation affects hydrology (e.g., Davie et al., 2013). As another example, about one-third of global agricultural production depends on animal pollination (Klein et al., 2007), so the loss of diverse animal pollinators as a result of climate-driven ecosystem stress could aggravate impacts of climate change on agriculture. Similarly, reductions in biodiversity can promote the spread of vector-borne diseases (LoGiudice et al., 2003), which is also influenced by climate (e.g., Altizer et al., 2013; Caminade et al., 2014). For the damages module in general, hindcasting and empirical calibration of models will be important tools for assessing the future representation of interactions and feedbacks.
Recognition and Consideration of Damages that Directly or Indirectly Affect Welfare
The individual sectoral impact functions available for inclusion in a damages module are estimated using a range of methods, as discussed above (see, especially, Table 5-3). Many are based on structural economic models of a sector or specific climate effect. A growing number of them derive empirical estimates by applying econometric methods to historical data, and some are processed through economic or integrated assessment models that may include various interactions among sectors or regions. There are differences in the information produced by these methods. In addition, there are important differences in the assumptions required to quantify different categories of climate change impacts. As a consequence, clarity regarding the underpinnings of the damage estimate requires transparency about the components of the estimate.
One important distinction is among damages that affect human consumption, those that affect capital stocks, and those that affect welfare in ways that are not mediated through markets. One output generated by
many of the procedures underlying damage functions is an estimate of the net change in aggregate macroeconomic consumption of goods and services that are priced in markets. This measure of welfare is clear and flows directly into the discounting procedure and the SC-CO2 estimate (see Figure 2-1, in Chapter 2). However, climate change does not always affect consumption directly, and may affect the level or productivity of capital stocks (physical, human, and environmental). Consumption effects are a downstream consequence of changes in input and output markets.
Impacts that harm capital stocks, the most well studied of which are the impacts of increased coastal flooding that affects durable infrastructure, will increase the demand for new investment. In the case of coastal flooding for example, this demand may divert investment from high-productivity activities to post-flood reconstruction and replacement of lost infrastructure. Using a CGE model, Bosello and colleagues (2007) found that the indirect costs of sea level rise, mediated by land loss or the capital market effects of protective investments, are comparable in scale to the direct effects. Using a CGE model, Houser and colleagues (2015) found that the long-term growth impacts of capital destruction caused by coastal storms on the United States as a whole were several times larger than the initial cost.
Effects on a particular type of capital stock will affect production input choices and markets, as well as output. For instance, Reilly and colleagues (2007) find that the macroeconomic effects of climate change are significantly smaller than the climate productivity shocks to land due to adaptation through markets, with changes in inputs, production, and international trade. Some of the effects of impacts on capital stocks may be captured in the functions estimating monetized consumption, but not necessarily all of them. There will be feedbacks and interactions among sectors that the available research does not yet capture. Therefore, to the extent possible, it will be important to take account of these capital stock effects as input to improved estimates of consumption and for possible consideration of feedbacks in sectoral interactions. In the longer term, incorporating these feedbacks to the socioeconomic module, discussed in Chapter 3, is of key importance.
Another potentially important welfare consequence of climate change is the loss of goods and services that are not traded in markets and so cannot be valued using market prices: examples include loss of cultural heritage, historical monuments, and favored landscapes; loss of charismatic and other species; violence; and forced migration. If kept in natural units, the distinction between estimates of these effects and those based on market prices will be transparent. However, some impacts may be treated as substitutable for consumption of market goods, and these effects may be converted into monetary terms using willingness to pay or
other simulated market concepts. These nonmarket effects are an important consequence of climate change and need to be quantified in monetary terms to the extent possible. Because the assumptions underlying these estimates are fundamentally different from the assumptions that underlie procedures based on market prices, their role in any damage total needs to be made transparent.
Representing Adaptation to Climate Change and the Costs of Adaptation
Households, communities, and societies will each take action autonomously to reduce the welfare losses of a changing climate, and policy makers will also direct investment to adaptation. Understanding the effectiveness of such measures, and their cost, is part of understanding the SC-CO2. For example, estimates of the costs of morbidity and mortality from extreme heat events will be overstated if they ignore greater use of air conditioning, but the overall damages must also include the cost of the greater use of air conditioning. In principle, the loss from the effect of a change in climate on some activity is the cost of adaptation measures plus the residual loss with the adaptation in place. In practice, such calculations can be analytically difficult.
The SC-IAM damage functions, and those in many other climate effects studies, represent climate damage as a function of global and regional mean temperature. However, climate change damages are often the effect of extreme events (e.g., a heat wave, storm, drought, or flood) involving other regional climate variables. Understanding of socioeconomic and ecological responses to these extreme events is limited, particularly at the relevant spatial scale, as is understanding of the relation of the change in these extremes to a projected change in global average temperature. This complexity not only creates difficulty for constructing estimates of climate damage, but also is a problem for the individuals and firms whose adaptation response is being modeled. Moreover, decision makers at all levels may have difficulty distinguishing between climate change and unforced weather variability, and their understanding may be further challenged by their own experiences and highly uncertain or conflicting projections from experts. As a result, they may take actions that are suboptimally early or late.
In spite of these complexities, it is important when constructing a new climate damage module to favor those damage estimates that take account of both adaptation (in order to avoid SC-CO2 estimates that overstate potential future economic loss) and the costs of adaption. Calculation of these effects in some sectors is straightforward (e.g., changes in heating or cooling), yet they may be more complex as the adaptation
response spills over into other sectors (e.g., the simultaneous effects of changes on heating or cooling on both health and energy consumption). Some structural economic models of climate impacts are well suited to consider the adaptation response. Based on historical experience, empirical models are likely to capture the adaptation that has occurred in the sector or location studied, but they will have a harder time extracting the adaptation response and its costs that are relevant to future, long-term changes that are not present in historical datasets. Advances in methods to consider adaptation responses may allow quantification of the costs of adaptation for a number of important sectors (e.g., agriculture, mortality) (e.g., Auffhammer and Aroonruengsawat, 2012; Butler and Huybers, 2013).
In contrast with process models, structural economic models can endogenously model future adaptation possibilities and their costs through changing markets (e.g., Reilly et al., 2007). Managed (e.g., policy-driven) and autonomous (e.g., market-driven) adaptation responses can be assessed in such a framework. Evaluation of the adequacy of damage estimates in capturing changes in vulnerability and success in adaptation would be a separate task for each damage function. Evaluation of overall performance would be limited to a rough assessment of the fraction of estimated damage for which explicit consideration of adaptation has been possible.
Representation of Nongradual Damages
The Earth system has the capacity to exhibit “abrupt,” nonlinear shifts between states. Various terms are used to describe these discontinuous system dynamics: abrupt changes, critical thresholds, regime shifts, tipping points, surprises, discontinuities, and catastrophic events. This imprecise and inconsistent terminology complicates discussions of how these complex phenomena can be incorporated in damage estimates.
Potential “climatic tipping elements” that could exhibit such discontinuous dynamics include the Atlantic meridional overturning circulation (AMOC), monsoonal circulation patterns, sea ice, polar ice sheets, permafrost carbon, marine methane hydrates, and the Amazon rainforest (Alley et al., 2003; Lenton et al., 2008; National Research Council, 2013; Kopp et al., 2016b). Gradual changes in the physical climate may drive these tipping elements over a threshold, producing a new equilibrium state—such as one in which an ice sheet is dramatically smaller than today or the Amazon rainforest is a savannah, for example.
Outcomes with high consequences, even if they are unlikely, have the potential to dominate expected welfare changes (e.g., Weitzman, 2011); their omission could affect estimates of the SC-CO2. While difficult to
estimate, the value of reducing the probability of high consequence events due to climate change could be quite large.
Many researchers point out that the SC-IAM damage functions fail to capture the risk of uncertain Earth system dynamics in an explicit or credible manner (Hitz and Smith, 2004; Warren et al., 2006; Kopp and Mignone, 2013; Deschenes, 2014; Howard, 2014; Li et al., 2014; Revesz et al., 2014; Sussman et al., 2014). Although the existence of these risks is supported by the geologic record (e.g., National Research Council, 2013) and in some cases by Earth system models (e.g., Drijfhout et al., 2015), the governing dynamics and thresholds are generally not well understood or quantified due to insufficient data and the limitations of process models. In addition, nongradual damages may arise from critical thresholds in socioeconomic systems as well as in natural systems. For example, by increasing the probability of civil conflict (Hsiang et al., 2013), gradual climate change could tip countries into a conflict-development trap, that is, is a self-reinforcing cycle in which civil conflict leads to slow or negative economic growth, and low economic development increases the risk of civil conflict (Collier et al., 2003).
The IWG needs to evaluate the state of knowledge and understanding of critical thresholds in climatic and climatically influenced socioeconomic tipping elements, as well as their likelihoods and consequences. It also needs to consider approaches for incorporating critical thresholds that can be appropriately quantified into the damage module. For example, Kopp and colleagues (2016b) propose an approach that includes using critical threshold scenarios in physical and empirical models to assess the potential impacts of crossing critical thresholds, together with structured expert elicitation to assess the probability of crossing those thresholds. A research program on critical thresholds, as well as on physical and economic modeling frameworks that incorporate them, would improve the capacity to integrate them into the SC-CO2 estimation framework. Such a program is particularly needed because it is currently unknown whether there are critical thresholds whose crossing would lead to significant damages, including potential effects on economic growth that could also affect SC-CO2 discounting (see Chapter 6).
CONCLUSION 5-1 An expansion of research on climate damage estimation is needed and would improve the reliability of estimates of the SC-CO2.
In the near term, initial steps that could be undertaken include:
- a comprehensive review of the literature on climate impacts and damage estimation, the evaluation of adaptation responses, and regional and sectoral interactions,
as well as feedbacks among the damage, socioeconomic, and climate modules; and
- a comparison of methods for estimating damages, including characterizations of their differences, synergies, uncertainties, and treatment of adaptation.
In the medium to long term, several research priorities could yield particular benefits for SC-CO2 estimation:
- physical, structural economic, and empirical estimation of climate impact relationships for regions and sectors not currently covered in the peer-reviewed literature;
- structural and empirical studies of the efficacy and costs of adaptation;
- calibration of damage functions using empirical and structural models operating at sufficiently high temporal and spatial resolution to capture relevant dynamics;
- the development of systematic frameworks for translating estimates of impacts into welfare costs; and
- empirical observation-based and structural modeling studies of interregional and intersectoral interactions of impacts, as well as of feedbacks among damages, socioeconomic factors, and emissions.
In the long term, research priorities that could yield particular benefits for SC-CO2 estimation would include omitted critical thresholds in natural and socioeconomic systems:
- development of simple Earth system model or full complexity Earth system model scenarios in which potential critical thresholds of tipping elements (e.g., Atlantic meridional overturning circulation, monsoonal circulation patterns, sea ice, polar ice sheets) are crossed, and the use of the physical changes in these scenarios to drive models that assess impacts and damages;
- empirical observation-based and structural modeling studies of the potential for climate change to drive the crossing of critical thresholds in socioeconomic systems and of their ensuing damages; and
- expert elicitation studies of the likelihood of different tipping element scenarios, in order to allow tipping elements and their critical thresholds to be represented probabilistically in the SC-CO2 framework.