Free beta access. 50 free AI tokens each week. Optional top-ups available. See what is included
Opening Word document…

Tipping points are likely---only 1.5 degrees solves

Yongyang Cai 16, Computational and Environmental Economist, Associate Professor in the Department of Agricultural, Environmental, and Development Economics, PhD in Copmutational and Mathematical Engineering, at Stanford, Timothy M. Lenton, Thomas S. Lontzek, Risk of multiple interacting tipping points should encourage rapid CO2 emission reduction, http://www.nature.com/nclimate/journal/v6/n5/pdf/nclimate2964.pdf

Existing scientific studies suggest there are multiple climate tipping points that could be triggered this century or next if climate change continues unabated7,8 , and there are causal interactions between tipping events such that tipping one element affects the likelihoods of tipping others8 (Fig. 1). The likelihood of specific tipping events varies, but is generally expected to increase with global temperature7,8 . However, internal variability within the climate system, and relatively rapid anthropogenic forcing, mean that even if deterministic tipping points could be precisely identified, the actual systems could be tipped earlier or later9 . Thus, any assessment of their policy implications needs to represent the stochastic uncertainty surrounding when tipping points could occur10. Furthermore, the impacts of passing different tipping points are expected to vary7,11, and to unfold at different rates depending on the internal timescale of the part of the climate system being tipped7,11 Relative to this scientific understanding, most cost–benefit analyses of climate change allow for only simple and scientifically unrealistic representations of climate tipping points11. Most previous IAM studies of climate catastrophes have treated them in a deterministic fashion, sometimes giving them a probability distribution5,12–15. Some recent IAM studies have considered one stochastic climate tipping point impacting economic output10, nonmarket welfare16, climate sensitivity17, or carbon cycle feedbacks17 . This can lead to up to 200% increases in the SCC in extreme cases10, with the results clearly sensitive to the timescale over which tipping point impacts unfold, as well as the final magnitude of those impacts10. However, there has been little consideration of multiple tipping points and interactions between them, or of how an appropriate representation of risk aversion affects the optimal response to the prospect of future tipping points. A recent IAM study18 has examined three loosely defined tipping points that instantaneously alter climate sensitivity, carbon cycle feedbacks, or economic output, and interact through their effects on atmospheric CO2 , global temperature, or economic output. Here we consider five carefully defined tipping points7,8 and the direct causal interactions between them identified by scientific experts8 (Fig. 1). These interactions occur primarily through aspects of the climate system that are not resolved in simple IAMs. The impacts of our tipping points unfold at a rate appropriate for the system being tipped, in contrast with instantaneous changes17,18 in climate sensitivity and carbon cycle feedbacks, which are scientifically questionable10. Our tipping points principally affect economic output, although we also consider their feedback effects on the carbon cycle. Instead of arbitrarily specifying the likelihood of the tipping points18, we calibrate their likelihoods (and the causal interactions between them) on the basis of the results of an existing expert elicitation8 . Furthermore, in contrast to recent work18, we alter the specification of the social planner’s preferences regarding risk aversion and intergenerational equity, in a manner appropriate for the stochastic uncertainty surrounding future tipping points. Modelling tipping points We use the dynamic stochastic integration of climate and economy (DSICE) framework19 to incorporate five stochastic tipping points and causal interactions between them into the 2013 version of the well-known DICE model20 (see Methods and Supplementary Figs 1 and 2). This means solving a sixteen-dimensional stochastic model—the first time in the field of economics of climate change that an analysis on such a scale has been accomplished (our previous work10 solved a seven-dimensional system, whereas other simplified stochastic versions17 of DICE consider only four dimensions). In our stochastic version of the DICE model, we use annual time steps, and calibrate parameters in the carbon cycle and temperature modules against the emulated median response of complex climate models for the four RCP (representative concentration pathway) scenarios21 (see Supplementary Methods). In a deterministic setting within our model (without considering climate tipping points) our calibration gives a SCC in 2010 of US$15 per tCO2 (all results are in 2010 US dollars). For reference, the DICE-2013R model20 that uses five-year time steps and is calibrated against one RCP scenario also has a 2010 SCC of US$15 per tCO2 In IAMs such as DICE, greater emission control at present mitigates damages from climate change in the future but limits consumption and/or capital investment today. A ‘social planner’ is assumed to weigh these costs and benefits of emission control to maximize the expected present value of global social welfare. When faced with stochastic uncertainty about future tipping events, the social planner’s response will depend on their preferences regarding risk and smoothing consumption. DICE adopts a specification of risk aversion that is inversely tied to the decision-maker’s preferences to smooth consumption over time (that is, the intertemporal elasticity of substitution). Thus, a high inter-temporal elasticity of substitution is taken to imply a low risk aversion. In the baseline DICE model, risk aversion RA = 1.45, and inter-temporal elasticity of substitution IES = 1/1.45. However, empirical economic data do not support this inverse proportionality (implying timeseparable utility) and suggest instead decoupling these preferences22 . Hence, we incorporated ‘Epstein–Zin’ (EZ) preferences22 using default parameter settings23 of RA = 3.066 and IES = 1.5, which are consistent with empirical findings23 (implying non-time-separable utility). Estimates of IES > 1 have been obtained from, for example, stockholder data24, IES = 1.5 is used in a long-run risk model19,25 , and the upper bound is considered23 to be IES ∼ 2. Using IES = 1.5, equity returns data23 suggest RA = 3.066, which is in the range RA = 3–4 from a separate study of equity premiums of rare disasters26, with the upper bound considered25 to be RA ∼ 10. The five interacting, stochastic, potential climate tipping points7,8 (Fig. 1 and Table 1) represent reorganization of the Atlantic meridional overturning circulation (AMOC), disintegration of the Greenland ice sheet (GIS), collapse of the West Antarctic ice sheet (WAIS), dieback of the Amazon rainforest (AMAZ), and shift to a more persistent El Niño regime (ENSO). We used published expert elicitation results8 to derive the likelihoods (see Methods) of each of the five tipping events (Table 1), and the causal interactions between them (Fig. 1 and Supplementary Table 1). By causal interaction, we mean that the hazard rate of each tipping point depends on the state of the others. For each tipping event, we specified a transition timescale10 (Table 1, see Methods)—that is, how long it would take for the full impacts to unfold, based on current scientific understanding of the timescales of the systems being tipped7,11 (for example, ice sheets melt more slowly than the ocean circulation can reorganize). Recognizing the scientific uncertainty surrounding transition times, we explore a factor of 5 uncertainty range in either direction. We must also specify a final damage for each tipping event (Table 1, see Methods), taken to be an irreversible percentage reduction in world GDP (gross domestic product). This is the most problematic and debatable part of the parameterization, because of a gross shortage of scientific and economic estimates of tipping point damages11. We can make some scientific inferences about relative damages (for example, based on the eventual contributions of different ice sheets to sea-level rise). Past studies with DICE have loosely associated a 25–30% reduction in GDP comparable with the Great Depression with a collapse of the AMOC27,28, but when combined with other tipping points this could lead to excessively high overall damages. Our assigned damages for individual tipping points range from 5 to 15% reduction in GDP with a combined reduction in GDP if all five tipping events occur and complete their transitions of 38%. However, owing to relatively low probabilities and long transition timescales, the expected tipping point damages in our default scenario amount to only 0.53% of GDP in 2100 and 1.89% of GDP in 2200. In our sensitivity analysis, we consider a factor of 2–3 total uncertainty range in final damages for each tipping point. Finally, we include some conservative effects of tipping particular systems on the carbon cycle (Table 1, see Methods). Optimal policyThe result of including multiple interacting tipping points under the default EZ preferences (Fig. 2) is a nearly eightfold increase in the initial SCC from US$15 per tCO2 in the baseline model (grey line) to US$116 per tCO2 (black line). Across 10,000 sample paths of the model, there are cases where one or more tipping points still occur, leading to uncertainty ranges for the key variables (grey-shaded areas). The emissions control rate jumps from ∼18 to ∼56% in 2010 and rises to 100% by 2050, effectively shutting down fossil fuel CO2 emissions—whereas in the baseline model, emissions continue into the next the century. The average atmospheric carbon peaks in the 2030s at 415 ppm and then declines (owing to ongoing ocean carbon uptake)—whereas in the baseline model, atmospheric CO2 continues to rise to ∼650 ppm by 2100. Temperature rise slows down and is almost stable around 1.4 ◦C above pre-industrial by 2100—whereas in the baseline model, warming continues and approaches 3 ◦C by 2100. Following the expected path (black line) there is only an 11% probability of one or more tipping events by 2100, reduced from 46% in the baseline model, or 87% under a prescribed RCP8.5 emissions scenario (Table 2). A factor of 2.4 increase from the baseline SCC to US$36 per tCO2 is just due to the change to IES = 1.5 (dashed black line, Fig. 2), with a further factor of 3.2 increase due to the potential for multiple tipping points. With just IES = 1.5 (and no stochastic tipping points), the initial emissions control rate increases from ∼18 to ∼29% with 100% emissions control in 2100. Atmospheric carbon peaks around 550 ppm, with surface temperature stabilizing around 2.3 ◦C above pre-industrial. Tipping point interactions In the full model, there are both positive and negative causal interactions between tipping points (Fig. 1 and Supplementary Table 1), which are conservatively calibrated (see Methods). Hence their inclusion has only a modest net effect on the expected SCC, increasing it from US$109 per tCO2 to US$116 per tCO2 (see also Supplementary Fig. 3). However, a specific sample path where multiple tipping events occur before 2200 (Fig. 3, solid line) reveals that some tipping point interactions can have a strong effect on the time evolution of the SCC. Considering a no-interactions sample path (Fig. 3, dashed line) shows that, in general, passing a tipping point reduces the incentive to mitigate and therefore lowers the SCC, because it can no longer be avoided. However, with interactions, tipping of the GIS significantly increases the likelihood of AMOC tipping (which is assumed to be the most damaging event); hence, this causes a large increase in the SCC to try to avoid AMOC tipping. (This is consistent with previous suggestions29,30 that tipping points can create multiple optima—here for the SCC and corresponding emissions30.) Subsequent tipping of AMOC greatly reduces the SCC. Tipping of ENSO causes a small increase in the SCC because it increases the likelihood of tipping the Amazon. Subsequent tipping of the Amazon halves the SCC because there is now an unavoidable extra source of carbon to the atmosphere and only WAIS left to tip. There are other sample paths where the first tipping event does not increase the likelihood of others, so the SCC drops—for example, when the Amazon rainforest tips first (Supplementary Fig. 4). The SCC therefore depends on whether tipping events occur and in which order. This can also be seen by looking at the sample paths for the earliest and sole tipping before 2100 of each element (Supplementary Fig. 5). If the GIS tips first, this leads to the highest SCC path and the most stringent emission control, reaching 100% before 2040, because of the increased risk of AMOC collapse. If the AMOC tips first, this gives the lowest SCC path because it has the greatest damages, which can no longer be avoided—yet emission control remains above 60% and the SCC remains above US$110 per tCO2 . If the Amazon tips first, this also lowers SCC and emission control, but it leads to the highest atmospheric carbon and temperature trajectory because of an accompanying carbon source. If ENSO tips first, this slightly increases emission control because the likelihood of the AMAZ tipping is increased. If the WAIS tips first, there is little effect on emission control because it only slightly increases the likelihood of tipping the AMOC and GIS. CO2 emissions trajectories (Supplementary Fig. 6) therefore depend on the contemporaneous state of tipping elements. Sensitivity analysis The high SCC is robust to sensitivity analyses (see Methods). Combined variations in assumed transition times and final damages of the tipping points give a full range in initial SCC of US$50–166 per tCO2 (Supplementary Table 2). With pessimistic settings for the expert assessment of interactions between tipping elements (Supplementary Table 3), the SCC increases from US$116 per tCO2 to US$121 per tCO2 . Including an endogenous transition time for the GIS gives only a slight reduction in SCC to US$114 per tCO2 because its damages tend to be discounted away anyway. Allowing all tipping elements to have an endogenous transition time reduces SCC to US$94 per tCO2 Retaining an inter-temporal elasticity of substitution IES = 1.5 but increasing risk aversion to RA = 10 increases the SCC from US$116 per tCO2 to US$146 per tCO2 .With the original RA = 3.066 and an upper limit of IES = 2, the SCC increases to US$151 per tCO2 . Using the default DICE settings of IES = 1/1.45 and RA = 1.45 gives an SCC of US$28 per tCO2 , a factor 1.9 increase from the default US$15 per tCO2 due to the five interacting tipping points. Thus, EZ preferences magnify the effect of including potential future tipping points, causing a factor 3.2 (rather than 1.9) increase in the SCC. To disentangle the effect of IES and RA, we also investigate a case with IES = 1.5 and RA = 1/1.5, which gives an SCC of US$104 per tCO2 . That is, when we incorporate the climate tipping risks, using time-separable preferences as in DICE, an increase from IES = 1/1.45 (and RA = 1.45) to IES = 1.5 (and RA = 1/1.5 ) leads to a factor 3.7 increase in the SCC, and the additional change to our default time non-separable EZ preferences (IES = 1.5, RA = 3.066) leads to an extra SCC of US$12 per tCO2 . Discussion and conclusion Putting our results in scientific context, there is already evidence that major ice sheets are losing mass at an accelerating rate31,32 . GIS mass loss is estimated to be contributing ∼0.7 mm yr−1 to sea-level rise33, with a corresponding increase in freshwater flux to the North Atlantic34 since 1990 of ∼0.01 Sv. Although modest at present, this and other contributors to increasing freshwater input to the North Atlantic35 are thought8 to increase the likelihood of AMOC tipping, and our results suggest that this should be increasing the incentive to control CO2 emissions. WAIS mass loss is contributing ∼0.35 mm yr−1 to sea-level rise32, and there is evidence that parts of the West Antarctic ice sheet are already in irreversible retreat36–38. If the WAIS has already passed a tipping point, then mitigation cannot avoid it, but our results suggest that this should not significantly reduce the incentive to mitigate to try to avoid other tipping events. Our results and policy recommendations differ considerably from another recent study considering multiple tipping points18 , which recommends at most a doubling of the SCC that allows CO2 emissions to continue to grow past mid-century, with temperature ultimately peaking at just under 3 ◦C. In contrast, our results recommend a nearly eightfold increase in the SCC to drive a cessation of CO2 emissions by mid-century, which limits warming to <1.5 ◦C. This very different outcome is a result of our different specification of tipping points together with our change in decision-maker preferences to something more appropriate for such stochastic climate risks.

Sign in to recut, update, or save this card.