IF past greenhouse-gas emissions from fossil-fuel combustion and other human activities have already changed our climate, what risks do we run if we continue on our current course? As discussed in chapter 1, this report attempts to help answer that question. While our focus is the economic risks of climate change, the analysis necessarily starts with an assessment of ways in which the climate may change in the years ahead.
A growing body of evidence shows conclusively that continued emission of CO2 and other greenhouse gases will cause further warming and affect all components of Earth’s climate system. While there have been significant advances in climate science in recent years, Earth’s climate system is complex, and predicting exactly how global or regional temperatures and other climate variables will change in the coming decades remains a challenge. It’s important to be honest about the uncertainty involved in forecasting our climate future if we are to provide policy makers, businesses, and households with the information they need to manage climate-related risks effectively (Heal & Millner 2014). Scientists face five major sources of uncertainty in predicting climate outcomes: (1) socioeconomic uncertainty, (2) global physical uncertainty, (3) regional physical uncertainty, (4) natural climate variability, and (5) tipping points. In this chapter we discuss each and provide an overview of how they are addressed in our analysis.
SOCIOECONOMIC UNCERTAINTY
Future levels of greenhouse-gas emissions will depend on the pace of global economic and population growth, technological developments, and policy decisions—all of which are challenging to predict over the course of a decade, let alone a century or more. As a consequence, the climate-science community has generally preferred to explore a range of plausible, long-run socioeconomic scenarios rather than rely on a single best guess (Bradfield et al. 2005; Moss et al. 2010). Each scenario includes assumptions about economic development, energy-sector evolution, and policy action—capturing potential futures that range from slow economic growth, to rapid economic growth powered primarily by fossil fuels, to vibrant economic development in a world transitioning to low-carbon energy sources. Each scenario results in an illustrative greenhouse-gas emission and atmospheric concentration pathway.
A broadly accepted set of global concentration pathways was recently developed by the Integrated Assessment Modeling Consortium (IAMC) and used in the Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC). These four pathways, termed “Representative Concentration Pathways” (RCPs), span a plausible range of future atmospheric greenhouse-gas concentrations. They are labeled based on their radiative forcing (in watts per square meter, a measure of greenhouse-gas concentrations in terms of the amount of additional solar energy the gases retain) in the year 2100 (van Vuuren et al. 2011). The pathways also include different assumptions about future changes in emissions of particulate pollution, which reflects some of the Sun’s energy to space and thus dampens regional warming. The RCPs are the basis for most global climate modeling undertaken over the past few years.
At the high end of the range, RCP 8.5 represents a modest increase in recent global emissions growth rates, with atmospheric concentrations of CO2 reaching 940 ppm by 2100 (figure 3.1) and 2,000 ppm by 2200. These are not the highest possible emissions: Rapid conventional economic growth could lead to a radiative forcing 10 percent higher than RCP 8.5 (Riahi 2013). But RCP 8.5 is a reasonable representation of a world where fossil fuels continue to power relatively robust global economic growth and is often considered closest to the most likely “business-as-usual” scenario absent new climate policy by major emitting countries.
FIGURE 3.1. Representative Concentration Pathways
Atmospheric concentration of CO2 in parts per million
Source: van Vuuren et al. (2011)
At the low end of the range, RCP 2.6 reflects a future achievable only by aggressively reducing global emissions (even achieving net negative emissions by the end of the twenty-first century) through a rapid transition to low-carbon energy sources. Atmospheric CO2 concentrations remain below 450 ppm in RCP 2.6, declining to 384 ppm by 2200. Two intermediate pathways (RCP 6.0 and RCP 4.5) are consistent with a modest slowdown in global economic growth and/or a shift away from fossil fuels more gradual than that in RCP 2.6 (Riahi 2013). In RCP 6.0, CO2 concentrations stabilize around 750 ppm in the middle of the twenty-second century. In RCP 4.5, CO2 concentrations stabilize around 550 ppm by the end of the twenty-first century.
We include all four RCPs in our analysis for two reasons. First, an individual RCP is not uniquely associated with any particular set of population, economic, technological, or policy assumptions; each could be attained through a variety of plausible combinations of assumptions. For example, a rapid emissions decline in the United States combined with continued emissions growth in the rest of the world could result in a concentration pathway similar to RCP 8.5. Likewise, if the current decline in U.S. emissions reverses course but the rest of the world makes a rapid transition to a low-carbon economy, a concentration pathway similar to RCP 4.5 is still potentially possible. Given the uncertainty surrounding emissions pathways in other countries, American policy makers must assess the risks associated with a full range of possible concentration futures. This is especially true for local officials and American businesses and households, as these local stakeholders have little control over America’s overall emission trajectory, let alone global concentration pathways.
The second reason is to identify the extent to which global efforts to reduce greenhouse-gas emissions can reduce climate-related risks associated with the absence of deliberate mitigation policy (i.e., RCP 8.5 or, under a slower global economic growth scenario, RCP 6.0). This is not to recommend a particular emission-reduction pathway, but to identify climate outcomes that are potentially avoidable versus those that are already locked in.
GLOBAL PHYSICAL UNCERTAINTY
Even if we knew future emissions growth rates with absolute certainty, we would still not be able to predict their impact precisely because of the complexity of Earth’s climate system. At a global level, the largest source of physical uncertainty resides in the magnitude and timescale of the planet’s response to a given change in radiative forcing, commonly represented by equilibrium climate sensitivity and transient climate response. The former, typically reported as the response to a doubling of CO2 concentrations, reflects the long-term response of global mean temperature to a change in forcing; the latter reflects how that response plays out over time.
The effect on global temperature of the heat absorbed and emitted by CO2 alone is fairly well understood. If CO2 concentrations doubled but nothing else in the Earth system changed, global average temperature would rise by about 2°F (Hansen et al. 1981; Flato et al. 2013). Across the entire climate system, however, there are several feedback mechanisms that either amplify or diminish this effect and respond on different timescales, complicating precise estimates of the overall sensitivity of the climate system. These feedbacks include an increase in atmospheric water vapor concentrations; a decrease in the planet’s reflectivity because of reduced ice and snow coverage; changes in the rate at which land, plants, and the ocean absorb carbon dioxide; and changes in cloud characteristics. Significant uncertainties remain regarding the magnitude of the relatively fast cloud feedbacks and of longer-term or abrupt feedbacks, such as high-latitude permafrost melt or release of methane hydrates, which would amplify projected warming (see the discussion in the section “Tipping Points” later in this chapter). Such longer-term feedbacks are not included in the equilibrium climate sensitivity as conventionally defined.
Uncertainty in the equilibrium climate sensitivity is a major contributor to overall uncertainty in projections of future climate change and its potential effects. Based on observed climate change, climate models, feedback analysis, and paleoclimate evidence, scientists have high confidence that the long-term climate sensitivity (over hundreds to thousands of years) is likely 3°F to 8°F of warming per CO2 doubling, extremely likely (95 percent probability) greater than 2°F of warming per CO2 doubling, and very likely (90 percent probability) less than 11°F of warming per CO2 doubling (Collins et al. 2013). This warming is not realized instantaneously because the ocean serves as a heat sink, slowing temperature rise. A more immediate measure, the transient climate response, indicates that a doubling of CO2 over 70 years is likely to cause a warming of 2°F to 5°F over that period of time (Collins et al. 2013).
These ranges of climate sensitivity values are associated with significantly different projections of future climate change. Many past climate-impact assessments have focused only on the “best estimates” of climate sensitivity. To capture a broader range of potential outcomes, we use MAGICC, a commonly employed simple climate model (Meinshausen, Raper, & Wigley 2011) that can emulate the results of more complex models and can be run hundreds of times to capture the spread in estimates of climate sensitivity and other key climate parameters. MAGICC’s model parameters are calibrated against historical observations (Meinshausen et al. 2009; Rogelj, Meinshausen, & Knutti 2012) and the IPCC’s estimated distribution of climate sensitivity (Collins et al. 2013). A more detailed description of our approach is provided in appendix A.
REGIONAL PHYSICAL UNCERTAINTY
Because deliberate planetary-scale climate experiments are largely infeasible and would raise profound ethical questions, scientists must rely on computer models to conduct experiments on Earth’s complex climate system, including projecting how climate will change at a regional scale in response to changes in greenhouse gases. Global climate models are descended from the first numerical weather-prediction models developed after World War II (Phillips 1956; Manabe & Wetherald 1967; Edwards 2011). Over time, they have been expanded to include the dynamic effects of oceans and sea ice, atmospheric particulates, atmospheric-ocean carbon cycling, atmospheric chemistry, vegetation, and most recently land ice. Model projections of the central components of long-term, human-induced climate change have grown increasingly robust, and recent generations of increasingly complex models provide greater detail and spatial resolution than ever before.
There are dozens of global climate models, with a range of different model structures and parameter assumptions. Since the 1990s, the global climate modeling research community has engaged in structured model comparison exercises, allowing them to compare experiments run in different models to one another and to the observational record. The differences identified among the models allow estimates to be made of the uncertainties in projections of future climate change and highlight which aspects are robust and where to focus future research efforts to improve results over time. By comparing and synthesizing many models, clear trends emerge.
Analysis of the range of potential climate effects on the United States for this report is based on climate projections developed as part of the Coupled Model Intercomparison Project Phase 5 (CMIP5) with a suite of 35 different global climate models (Taylor, Stouffer, & Meehl 2012). This suite of complex models has become the gold standard for use in global climate assessments (including by the IPCC in AR5) and for regional assessments (including the third U.S. National Climate Assessment, released in 2014). Major U.S.-based models participating in CMIP5 have been developed by teams led by the NASA Goddard Institute for Space Studies, the NOAA Geophysical Fluid Dynamics Laboratory, and the National Center for Atmospheric Research.
The global climate models that participated in CMIP5 typically have spatial resolutions of ~1° to 2° (about 70 to 150 miles at midlatitudes). To produce projections at a finer spatial resolution, researchers have used a variety of downscaling approaches. The projections in this report build upon one particular downscaling technique: bias-corrected spatial disaggregation (BCSD; Wood et al. 2002; Brekke et al. 2013). We use a BCSD data set generated by the Bureau of Reclamation (Brekke et al. 2013) from the CMIP5 archive. In addition to the uncertainty in the global climate models themselves, further uncertainty is introduced by the downscaling step. Alternative downscaling approaches can give rise to different localized projections, particularly of extremes (Bürger et al. 2012).
It is important to recognize that the CMIP5 model projections are not a probability distribution, but instead an “ensemble of opportunity” (Tebaldi & Knutti 2007). The models are not fully independent of one another, instead sharing overlapping lineages and a common intellectual milieu (Edwards 2011). Moreover, every modeling team that participates in the Coupled Model Intercomparison Project has striven to develop a model that captures a suite of important physical processes in the oceans and atmosphere and has tuned some of the parameters of its model to reproduce historical behavior reasonably. Attempts to interpret the CMIP5 ensemble as a probability distribution will accordingly undersample the distribution tails and oversample the best estimates.
For this reason, we use estimates from MAGICC of the probability of different temperature outcomes at the end of the century to weight the projections of more complex global climate models. For those parts of the probability distribution for global temperature not covered by the CMIP5 models, primarily in the tails, we create “model surrogates” by scaling spatial patterns of temperature and precipitation change from the CMIP5 models using temperature projections from MAGICC. In appendix A, we compare our key results to those we would estimate if we treated the CMIP5 projections as though they formed a probability distribution.
NATURAL CLIMATE VARIABILITY
As discussed earlier, natural climate variability can range in timescale from day-to-day temperature variations, to interannual patterns such as El Niño, to longer-term patterns such as the Pacific Decadal Oscillation. In addition to the trends in climate associated with climate change, global climate models simultaneously simulate natural climate variability. The magnitude of such variability renders the differences in climatic response among plausible emissions pathways essentially undetectable at a global scale until about 2025. The relative magnitude of climate variability, physical uncertainty, and scenario uncertainty differs from place to place and for different variables. For example, in the British Isles, internal variability in decadal mean surface air temperature dominates scenario uncertainty through the middle of the century (Hawkins & Sutton 2009). Internal variability, not fully captured by climate models, probably accounts for a significant fraction of the slowdown in global warming over the past decade (Trenberth & Fasullo 2013) and for the absence of net warming in parts of the southeastern United States over the past century (Kumar et al. 2013). While unprecedentedly warm years will occur with increasing frequency, climate variability means that the annual mean temperatures of cooler years in most of the United States will be in the range of historical experience until at least the middle of the century (Mora et al. 2013).
Extreme weather events like heat waves, hurricanes, and droughts are examples of natural climate variability experienced on more compressed timescales. By nature, the probability of these events occurring is low, putting them at the far “tails” of statistical weather distributions. There is increasing evidence, however, that climate change is altering the frequency or severity of many types of these events (Cubasch et al. 2013). Although most individual extreme events cannot be directly attributed to human-induced warming, there is relatively high confidence that heat waves and heavy rainfall events are generally becoming more frequent (Hartmann et al. 2013). As the climate continues to warm, certain types of storms such as hurricanes are expected to become more intense (though not necessarily more frequent), although less is known about how other types of storms (such as severe thunderstorms, hailstorms, and tornadoes) may respond (National Academy of Sciences & The Royal Society 2014).
Because of the tremendous damage caused by hurricanes in recent decades, there has been significant interest in understanding global and regional trends in cyclone activity and the causes of any observed changes. At a global level, the evidence for long-term changes and the influence of human-induced climate changes on hurricane activity over the past century is unclear (Knutson et al. 2010). That is not to say that human-induced warming played no role; rather, because of limitations in the quality of historical records, it is possible that such influence is simply not yet detectable or is not yet properly modeled given the uncertainty in quantifying natural variability and the effects of particulate pollution, among other factors (Knutson et al. 2010; Seneviratne et al. 2012; Christensen et al. 2013). Short-term and regional trends vary, however; hurricane activity has increased in the North Atlantic since the 1970s (Christensen et al. 2013).
Our confidence in projecting future changes in extremes (including the direction and magnitude of changes) varies with the type of extreme, based on confidence in observed changes, and is thus more robust for regions where there is sufficient and high-quality observational data (Seneviratne et al. 2012). Temperature extremes, for example, are generally well simulated by current global climate models, though models have more difficulty simulating precipitation extremes (Randall et al. 2007). The ability to project changes in storms, including hurricane activity, is more mixed. There is a growing consensus that, around the world, the strongest hurricanes (categories 4 and 5) and associated rainfall levels are likely to increase (Knutson et al. 2010; Seneviratne et al. 2012; Christensen et al. 2013). There is low confidence, however, in climate-induced changes in the origin and track of future North Atlantic hurricanes (Bender et al. 2010).
TIPPING POINTS
Many components of the Earth system exhibit critical thresholds (often referred to as tipping points) beyond which abrupt or irreversible changes to the climate or the biosphere may occur (Lenton et al. 2008; Collins et al. 2013; National Academy of Sciences 2013). Many of these tipping points are poorly represented in the current generation of climate models. Some may have direct societal or economic effects, while others may affect the global carbon cycle and amplify climate change. Such feedbacks could increase the probability of our most extreme projections (although unexpected stabilizing feedbacks could also act in the opposite direction).
Summer Arctic sea-ice cover has fallen faster than most of the previous generation of climate models had projected (Stroeve et al. 2007), although the current generation of models appears to perform better (Stroeve et al. 2012). The Arctic appears on track for nearly ice-free Septembers in the coming decades. Reduced sea-ice coverage amplifies warming in the Arctic and may also lead to slower-moving weather patterns at lower latitudes (Francis & Vavrus 2012). Slow-moving weather patterns supported the long-lived cold winter experienced by much of North America in 2013–2014, which had a significant economic effect. However, the linkage with low summer Arctic sea ice remains highly controversial (Barnes 2013).
Past mass extinctions have been tied to global climate change (Blois et al. 2013). Human activities, primarily land-use changes, have increased the global species-extinction rate by about two orders of magnitude above the background rate (Barnosky et al. 2011), and climate change is beginning to exacerbate extinction further (Barnosky et al. 2012). The economic effects of mass extinction and the associated loss of ecosystem services are difficult to estimate, but they are likely to be substantial.
Past climate change has also driven rapid ecosystem shifts (Blois et al. 2013). Some research suggests that the Amazon rain forest and northern boreal forests may be vulnerable to a climatically driven die-off, which would increase global CO2 emissions, but there is significant uncertainty about the climatic threshold for such a die-off and its likelihood (Collins et al. 2013).
Destabilization of methane trapped in ocean sediments and permafrost may have played a major role in the geologically rapid 10°F global warming of the Paleocene-Eocene Thermal Maximum, which occurred about 56 million years ago (McInerney & Wing 2011). Global warming today may trigger a similar destabilization of methane reservoirs, amplifying projected warming significantly, although such a methane release would be expected to play out over centuries (Collins et al. 2013).
Reconstructions of past sea level and physical models of ice-sheet dynamics suggest that the West Antarctic Ice Sheet can collapse and raise sea level by many feet over the course of a few centuries (Kopp et al. 2009; Pollard & DeConto 2009). Indeed, recent evidence suggests that such a collapse may be under way (Joughin, Smith, & Medley 2014; Rignot et al. 2014). The possibility of a rapid collapse is included in the projections of sea-level rise described in chapter 4, which indicate a 1-in-1,000 probability of 8 feet of global mean sea-level rise by 2100 and 31 feet of global mean sea-level rise by 2200, but its likelihood may be underestimated.
Other potential tipping points include drops in ocean oxygen content, changes to monsoons, and changes to patterns of climatic variability such as El Niño (National Academy of Sciences 2013). There may be other critical thresholds not yet considered by science. High-impact tipping points with consequences realized primarily in this century are considered unlikely, but confidence in many of these projections is low (Collins et al. 2013). As warming increases, the possibility of major abrupt changes cannot be ruled out.
