Abstract
Although achieving net-zero emissions is very likely to stabilize the long-term global temperature, the possibility of continued warming and extreme events could cause those efforts to be perceived as a failure if there is an expectation that stabilizing global temperature will also stabilize local and regional climate. Leveraging decarbonization scenarios from multiple global climate models, we find that much of the world faces >30% probability of decadal warming after net-zero CO2 emissions are achieved, with most areas exhibiting sustained probability of extreme hot and wet events. Further, substantial fractions of the global population and gross domestic product could experience post-net-zero warming, including hundreds of millions of people and trillions of dollars in the United States, China and India during the decade following net-zero. This likelihood suggests that some of the most populous, wealthy, and powerful regions may experience climatic conditions that could be perceived—at least in the near-term—to indicate that climate stabilization policies have failed, highlighting the importance of adaptation for ensuring that communities are prepared for the climate variations that will inevitably occur during and after decarbonization.
Export citation and abstract BibTeX RIS
Original content from this work may be used under the terms of the Creative Commons Attribution 4.0 license. Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
1. Introduction
The UN Paris Agreement articulates the goals of holding global warming to 'well below' 2 °C above the pre-industrial baseline, and 'pursuing efforts' towards a 1.5 °C limit (UNFCCC 2015). Given the relationship between cumulative emissions and global warming (Matthews et al 2009), achieving those goals will very likely require reaching net-zero CO2 emissions in the mid-21st century (Tanaka and O'Neill 2018, Matthews and Wynes 2022), and may also require deployment of negative emissions technologies at scale (Tanaka and O'Neill 2018, IPCC 2022a). At present, there remains a large gap between the total emissions reductions that have been committed and the aspirational pace of decarbonization needed to achieve the UN Paris goals (UNEP 2021, Matthews and Wynes 2022, Meinshausen et al 2022). However, many national, subnational and non-state actors have articulated commitments to reach net-zero emissions in the mid-21st-century time-frame (Rogelj et al 2021, Matthews and Wynes 2022).
While it is well established that the long-term global temperature can be stabilized by reducing greenhouse gas emissions to essentially zero (Matthews and Caldeira 2008), the complex, chaotic nature of the coupled climate system creates substantial global, regional and local climate variability (e.g. (Hawkins and Sutton 2009, Deser et al 2012, Rogelj et al 2017)). Because this climate variability exists in the absence of external forcing factors (such as natural variations in solar output or anthropogenic increases in greenhouse gas concentrations), it is often referred to as 'internal variability'. This internal variability can cause local-scale warming, cooling, wetting or drying (e.g. (Deser et al 2012)), and is also the root cause of climate extremes (e.g. (Diffenbaugh et al 2017, Gessner et al 2021)). As a result, it is virtually certain that there will continue to be short-term local climate trends even if ambitious mitigation efforts succeed in stabilizing long-term global temperature. Further, given that internal variability can cause record-setting events in a stationary climate (Gessner et al 2021), it is highly likely that climate extremes will continue to occur after global temperature is stabilized (e.g. (Fischer et al 2021)).
Given the common presumption that reductions in emissions should lead to reversal of warming (e.g. (Sterman and Sweeney 2007)), continued warming and extreme events could cause climate stabilization policies—and the sacrifices needed to achieve them—to be perceived as a failure (Rogelj et al 2017, Keys et al 2022). Such a perception could have significant consequences for continued support of climate policies by decision-makers and the public, particularly in global powers where national-level decarbonization goals are not permanently binding (Allan 2019) and efforts to achieve national commitments have been politically and legally contentious (Pauw et al 2020). Moreover, effectiveness of climate mitigation policy is among the most important criteria for continued support of climate mitigation policies, second only to perceived fairness (Bergquist et al 2022). Such effectiveness is predicated on the general public having a reasonable understanding of how climate policies ought to affect the climate in the first place, yet evidence suggests that there is substantial possibility of misperception (Weber and Stern 2011). Even highly educated observers make fundamental errors in judging how the physical climate system would respond to decarbonization, including misperceptions of the timescales over which a response might occur (Sterman and Sweeney 2007). These potential misperceptions could cause climate policy efforts to be weakened or terminated (Krause et al 2016).
While the magnitude of climate change during decarbonization has been frequently explored at global and other aggregated scales (e.g. Rogelj et al 2017, Marotzke 2019, Samset et al 2020, 2022, Fischer et al 2021, Lee et al 2021, McKenna et al 2021), it is critical to assess how the potential for near-term warming and extreme events overlaps with governance indicators such as national boundaries, population density, and economic activity. We probe these questions using ensemble projections from multiple global climate models (GCMs) under two decarbonization scenarios. Both scenarios are based on Shared Socioeconomic Pathway 1 (SSP1; (O'Neill et al 2014, Riahi et al 2017)). The Intergovernmental Panel on Climate Change (IPCC) identifies the first, 'SSP1-1.9', as the 'Very Low' emissions scenario and the second, 'SSP1-2.6', as the 'Low' emissions scenario (Lee et al 2021). Anthropogenic CO2 emissions reach net-zero in 2056 and 2076 in SSP1-1.9 and SSP1-2.6, respectively, and continue to decline into the late 21st century in both scenarios (figure 1(A)) (Riahi et al 2017). The IPCC has concluded that the probability is 'more likely than not' that global warming would remain less than 1.6 °C above the pre-industrial baseline during the 21st century under SSP1-1.9 (Lee et al 2021). In contrast, the IPCC projects the mean warming to be 2.0 °C (5%–95% range: 1.3 °C–2.8 °C) in the late 21st century under SSP1-2.6 (Lee et al 2021).
We adopt the perceived failure metric of (Keys et al 2022) to assess the probability of local-scale temperature trends during and after decarbonization. For each decarbonization scenario, we measure the probability of near-term warming in the decade after global CO2 emissions reach net-zero, and compare this with the probability early in the decarbonization pathway. Further, given the importance of extreme events for climate impacts (IPCC 2012) and for the perception of climate change (Borick and Rabe 2017, Brügger et al 2021), we also quantify the probability of extreme hot and wet events across the decarbonization period.
2. Materials and methods
2.1. Climate model data
We analyze GCM simulations from Phase 6 of the Coupled Model Intercomparison Project (CMIP6) (Eyring et al 2016). The CMIP6 future scenarios are based on the SSPs, whose future emissions begin in the year 2015 (Riahi et al 2017). To test the sensitivity of our analysis to the pace of decarbonization, we compare SSP1-1.9 and SSP1-2.6, which reach net-zero anthropogenic CO2 emissions in 2056 and 2076, respectively (figure 1). We note that although we use the years 2056 and 2076 as 'net-zero' benchmarks, other greenhouse gases such as methane do not reach net-zero in the same year, and hence total anthropogenic radiative forcing does not reach zero by the end of the 21st century (O'Neill et al 2016).
In order to capture both structural differences between individual climate models and the influence of internal climate variability, we create a 'grand ensemble' that includes five realizations from each of the CMIP6 GCMs that have archived at least five realizations in a given SSP scenario. For SSP1-1.9, there are eight GCMs that have archived monthly temperature from at least 5 realizations, yielding a grand ensemble of 40 realizations. For SSP1-2.6, there are ten GCMs that have archived monthly temperature from at least 5 realizations, yielding a grand ensemble of 50 realizations. We follow the IPCC (IPCC 2013) in using a common 2.5° × 2.5° grid to aggregate the results of the multiple GCMs.
We analyze climate variables on both annual and daily timescales. For the annual timescale, we average the monthly temperature values to generate an annual time series for each realization at each grid point. For the daily timescale, we analyze the subset of GCMs from the SSP1-1.9 grand ensemble for which daily output is also archived for at least five realizations. This yields 30 SSP1-1.9 realizations (from six GCMs) for daily temperature and precipitation.
2.2. Metrics for assessing probability of continued local-scale warming and extreme events during and after decarbonization
We use the 'perceived failure' metric of (Keys et al 2022) to calculate the probability of short-term temperature trends in different stages of decarbonization. This metric is based on four archetypes of regional temperature trends preceding and following climate policy action, and is agnostic about whether the trends are driven by internal variability or forced change. The four archetypes are: 'Rebound Warming' (no warming before, followed by warming after); 'Continued Warming' (warming before, followed by warming after); 'Stabilization' (no warming before, followed by no warming after); and 'Recovery' (warming before, followed by no warming after). Because both Rebound Warming and Continued Warming could potentially be perceived as climate policy failures, Keys et al use climate model simulations to quantify the probability of local-scale warming after the time of policy action (where warming is defined as a 10 year temperature trend of at least +0.1 °C decade−1, or +0.01 °C yr−1). While Keys et al use their perceived failure metric to quantify the probability of local warming following stratospheric aerosol injection, the metric is generalizable to a variety of climate policy actions, including net-zero decarbonization goals.
To test the sensitivity to the pace of decarbonization, we calculate the metric in the SSP1-1.9 and SSP1-2.6 scenarios. In our primary analysis, we calculate the metric for each grid point for two time horizons: the near-future period (2025–2034) and the decade after CO2 emissions reach net-zero (2056–2065 for SSP1-1.9 and 2076–2085 for SSP1-2.6). To test the sensitivity of the results to the length of time over which the trend is calculated, we repeat this analysis using progressively longer temperature trends. Finally, to constrain the 'irreducible uncertainty' in the perceived failure metric, we calculate the metric during the 19th century of the CMIP6 Historical forcing simulations.
We also use this metric to quantify exposure of population and economic activity to warming trends during and after decarbonization. Following the approach described in (Keys et al 2022), we download gridded projections of population (Gao 2017, 2020) and gross domestic product (GDP, in units of purchasing power parity or PPP; (Crespo Cuaresma 2017, Riahi et al 2017)) for future decades in SSP1, and interpolate those data to the common 2.5° × 2.5° grid. We then calculate the fractions of country-level population, global population, and global gross domestic product (GDP) that are exposed to warming in each decade of each climate model realization.
Other climate variables in addition to annual temperature trends are also important indicators of the potential for perceived success or failure of climate policy. For example, given the wide range of impacts that extreme events have on people and ecosystems (IPCC 2012, 2022b), the occurrence of extremes is likely to be a strong influence on the perception of climate policy actions (since that perception is linked to the perceived harmfulness of climate change; e.g. (Sterman and Sweeney 2007)). Hence, we calculate—for each grid point on the common 2.5° × 2.5° grid—the fraction of years in which the hottest (wettest) day of the year is at least as hot (wet) as the hottest (wettest) day of the early industrial baseline period. We define this baseline threshold for each GCM using the hottest (wettest) day of the 1851–1870 period of the Historical experiment. Because there are five realizations of each GCM, the baseline threshold value is the hottest (wettest) day of 100 years of simulation in the 1851–1870 forcing, calculated respectively at each grid point.
We calculate the fraction of years in which the hottest (wettest) day of the year meets or exceeds this early industrial threshold for five periods of the CMIP6 simulations: the period of the CMIP6 Historical experiment directly following the period over which the baseline threshold is calculated (1871–1920); the last decade of the CMIP6 Historical experiment (2005–2014); a near-future decade of the CMIP6 SSP1-1.9 experiment (2025–2034); the decade after reaching net-zero CO2 emissions in the CMIP6 SSP1-1.9 experiment (2056–2065); and the last decade of the 21st century in the CMIP6 SSP1-1.9 experiment (2091–2100, at which time substantial negative emissions have been deployed for multiple decades; figure 1).
3. Results
The GCM ensemble-mean 'forced response' suggests that, in addition to halting long-term global warming (figures 1(B) and (C)), decarbonization is also effective at stabilizing mean temperature at regional and local scales (figures 1(D) and (E)). For example, in SSP1-1.9, 83% of the global surface area exhibits ensemble-mean warming over the period prior to net-zero emissions (2015–2055; figure 1(D)), while 0% exhibits ensemble-mean warming over the subsequent period (2056–2100; figure 1(E)). However, because there is only one realization of earth's actual climate system, it is the individual GCM realizations—rather than the ensemble-mean—that reflect the magnitude and spatial heterogeneity of temperature trends that can occur within the context of decarbonization.
For the sake of illustration, we select a single member from our SSP1-1.9 grand ensemble (member #13; figures 1(F) and (G)). Within this arbitrarily chosen realization, temperature trends exceed +0.1 °C yr−1 over some regions. In addition, the magnitude of local trends can be very similar before and after net-zero emissions are reached. Further, regions can experience warming both before and after net-zero emissions (e.g. over central North America or the eastern tropical Pacific in this particular realization), or a reversal from cooling to warming (e.g. over Northern Africa or eastern Eurasia in this particular realization). This one climate model realization thus illustrates that the combination of internal variability, non-CO2 anthropogenic forcings, and climate system inertia can cause substantial decadal-scale warming at individual locations during and after decarbonization.
To explore this possibility more systematically across the larger CMIP6 ensemble, we use the 'perceived failure' framework of (Keys et al 2022), which measures the probability of 10 year warming trends that exceed +0.01 °C yr−1 (see materials and methods). We find that during the initial period of decarbonization (2025–2034), both the SSP1-1.9 and SSP1-2.6 scenarios exhibit >50% probability of warming trends over most of the global surface area (figures 2(A) and (B)). However, the two scenarios exhibit differences across the decarbonization trajectory. For example, the more slowly decarbonizing SSP1-2.6 scenario exhibits larger areas with >70% probability of warming trends during the 2025–2034 period. Similarly, in the decade after net-zero, the more rapidly decarbonizing SSP1-1.9 scenario exhibits considerably larger areas with <30% probability, particularly in the tropics (figures 2(C) and (D)). Still, both scenarios exhibit large areas of the mid- and high-latitudes with >30% probability of warming in the decade after net-zero, including >50% probability over areas of the northeast Pacific and western North America in both scenarios. We find broadly similar pattern and magnitude in a single-model large ensemble (figures S3(A) and (B)) as in the multi-model CMIP6 ensemble (figures 2(C) and (D)).
Download figure:
Standard image High-resolution imageTo constrain the lower bound—or 'irreducible uncertainty'—in the metric, we quantify the probability of 10 year warming trends that exceed +0.01 °C yr−1 during the 19th century of the CMIP6 Historical simulations. We find that, absent large forcing from anthropogenic emissions or large volcanic eruptions, the probability mostly remains between 30%–50% in the low-latitudes and between 40%–60% in the mid- and high-latitudes (figure 3(A)). The probabilities during the first decade of SSP1-1.9 (figure 3(C)) and the first two decades of SSP1-2.6 (figure 3(E)) are clearly larger than this 'unforced' lower bound. In contrast, the probabilities at the time that net-zero emissions are reached are similar to this unforced lower bound across most latitudes in both SSP1-1.9 and SSP1-2.6 (figures 3(C) and (E)).
Download figure:
Standard image High-resolution imageGiven the substantial probability of 10 year warming trends after net-zero emissions are reached and in the unforced climate, we also investigate the required length of time (over which the trends are calculated) for the trends to reach a low probability of warming. We find that, by the end of the third decade in the pre-Krakatoa period of the CMIP6 Historical simulations, <10% of realizations exhibit a 30 year warming trend at the equator (figure 3(B)). However, at that point there is still >20% probability of a 30 year warming trend in the mid-latitudes, and >40% probability in the northern-hemisphere high-latitudes. In contrast, at the time that CO2 emissions reach net-zero in SSP1-1.9, there is >30% probability of a 30 year warming trend across all latitudes, including >60% in the northern-hemisphere mid-latitudes and >70% in the northern-hemisphere high-latitudes. The probabilities of multi-decadal warming trends are even larger at the time that CO2 emissions reach net-zero in SSP1-2.6, including >50% probability of a 50 year warming trend in the tropics and southern-hemisphere high-latitudes, and >90% probability in the northern-hemisphere high-latitudes.
This latitudinal pattern generally matches the spatial pattern of the forced response, which exhibits faster warming in the northern hemisphere than the southern hemisphere, and fastest warming over the northern-hemisphere high-latitudes (Lee et al 2021). It also generally matches the spatial pattern of internal variability, which is highest in the northern-hemisphere high-latitudes (e.g. (Diffenbaugh and Charland 2016)), leading to more substantial long-term temperature trends by random chance (figure 3(B)). As a result, the probability that warming trends—and hence the possibility of perceived failure—persist over progressively longer time windows is greatest in the high latitudes (figures S1 and S2).
The high probability of decadal warming during decarbonization (figures 2, 3 and S1–S3) means that large fractions of the global population and global economy could be exposed to short-term warming even if climate stabilization policies are successfully enacted (figure 4). For example, in the more rapidly decarbonizing SSP1-1.9 scenario, there is >50% probability that more than half of the global population is exposed to near-term warming conditions over the 2025–2034 period (figure 4(C)). That probability is lower in the decade after net-zero emissions are reached (2056–2065; figure 4(C)), although large populations in China, India and the United States all exhibit at least 75% probability of being exposed to near-term warming over the 2056–2065 period (figure 4(A)). Regions encompassing similarly large fractions of global economic activity are also subject to warming during and after decarbonization (figure 4(B)). Indeed, even in 2085–2094 of SSP1-1.9, well after negative CO2 emissions have been achieved (figure 1), a minimum of 20% of the global economic activity is exposed to warming of >0.01 °C yr−1 over that decade (figure 4(B)).
Download figure:
Standard image High-resolution imageGiven the importance of extreme events for the impacts of climate change (e.g. (IPCC 2012)), we also analyze the probability of extreme hot and wet events during the CMIP6 Historical and SSP1-1.9 simulations (figures 5, S4 and S5). We find that the probability of years in which the hottest day meets or exceeds the hottest day in the early-industrial baseline remains relatively stable over most of the globe throughout the SSP1-1.9 scenario, including the near future (2025–2034), the decade following net-zero (2056–2065), and the last decade of the 21st century (2091–2100) (figures 5 and S4). We find that these probabilities during and after decarbonization are higher than the probabilities during the last decade of the CMIP6 Historical experiment (2005–2014) over most areas of the globe (figure S4). For example, no areas exhibit >90% probability during the late Historical period, but much of the tropics exhibits >90% probability during all three periods of SSP1-1.9. Further, the probabilities in both the late Historical period and all three periods of SSP1-1.9 are an order of magnitude greater than the probabilities in late 19th/early 20th century.
Download figure:
Standard image High-resolution imageWe find similar persistence of probabilities for extreme wet events during the SSP1-1.9 decarbonization scenario, though the magnitude of the probabilities is generally smaller (figures 5 and S5). Like the hottest day, the probability of years in which the wettest day meets or exceeds the wettest day in the early-industrial baseline remains relatively stable over most of the globe throughout the near future period, the decade following net-zero, and the last decade of the 21st century (figure S5). Likewise, the area exhibiting >10% probability is substantially larger in all three periods of the SSP1-1.9 decarbonization scenario than the last decade of the Historical simulations. Finally, the probabilities during the decarbonization scenario exceed the probabilities in late 19th/early 20th century over most of the globe, although some areas do exhibit probabilities of extreme wet events that are similar to those of the late 19th/early 20th century.
4. Discussion and conclusions
Our analysis suggests that much of the globe faces >50% probability of near-term warming exceeding +0.01 °C yr−1 early in the decarbonization period, and >30% probability after net-zero CO2 emissions are reached under the lowest SSP emissions scenario. Given the strong imprint of internal variability on decadal-scale regional temperature trends (e.g. (Deser et al 2012)), it should not be surprising that large fractions of the globe exhibit substantial probability of 10 year warming trends during decarbonization (figures 2 and 3). However, because our threshold for defining a warming trend is relatively large (+0.01 °C yr−1), the null hypothesis for an unforced climate is not necessarily the 50% probability that would be expected for a threshold of >0 °C yr−1. In order to constrain the null hypothesis, we quantify the probability of trends >0.01 °C yr−1 in the 1850–1900 period of the CMIP6 Historical simulations (figure 3). We find that there is substantial heterogeneity across latitudes, with higher probability of warming trends at higher latitudes. In addition, although the unforced probability is <50% for most latitudes for both decadal and multi-decadal warming trends, the unforced probabilities are still substantial, including >30% for a 10 year warming trend. Further, the probabilities of warming trends after CO2 emissions reach net-zero in SSP1-1.9 and SSP1-2.6 are similar to the unforced probabilities.
Our results thus pinpoint an 'irreducible uncertainty' in short-term local-scale warming trends that must be considered even after successful climate mitigation has been achieved, if climate risk management decisions are to be robust (e.g. (Mankin et al 2020)). In the case of our analysis, the fact that post-net-zero probabilities of 10 year warming trends are similar to this irreducible uncertainty raises important questions about what outcomes policy makers, decision makers, and the general public expect from decarbonization policies––and on what timescales. That there should be continued likelihood of local-scale warming trends after reaching net-zero emissions is already well known in the climate science community: The internal variability of the coupled climate system will clearly create periods of warming and cooling within a stationary climate. This is particularly true for shorter-term trends, and indeed we find the probability of warming trends following net-zero to be generally smaller for progressively longer time periods over most areas of the globe (figures S1 and S2). Further, it has long been established that inertia in the coupled climate system will delay the response of temperature to greenhouse gas mitigation (Meehl et al 2005), including via the response of modes of variability such as the Atlantic Meridional Overturning Circulation (Tilmes et al 2020, An et al 2021). Thus, our focus on 10 year trends in our primary analysis does not arise from a physical expectation that decarbonization could remove decadal variability. Rather, it is motivated by the timescale over which people may incorporate local trends into their perception of global warming (Shao et al 2016), and the reality that regardless of what is actually occurring in the climate system, public expectations of net-zero achievement are likely to be that the climate ought to respond in short order (Sterman and Sweeney 2007). (That is, that temperatures ought to begin going down rapidly.)
Our results further reinforce the above literature on global-scale climate system inertia (e.g. (Meehl et al 2005, An et al 2021)) by illustrating that, although reaching net-zero—or even negative—emissions stabilizes the frequency of extreme hot and wet events, it does not appreciably reduce the frequency (figures 5, S4 and S5). The potential for continued extreme events poses risks for natural and human systems, and has become a key motivator for decarbonization (IPCC 2018). While our analysis only includes the hottest and wettest days of the year, these can be indicators of other kinds of extremes (e.g. (IPCC 2012)). Hence, the persistence of high probability of extreme hot and wet events throughout the SSP1-1.9 decarbonization scenario (figures 5, S4 and S5) suggests that other types of extremes are also likely to persist.
As discussed in the Introduction, given that changes in local climate conditions are a key indicator of individual perceptions of global warming (Shao et al 2016), and the expectation that it is possible to quickly reverse warming (Sterman and Sweeney 2007), continued local-scale warming could lead to the perception that climate stabilization efforts have failed (Keys et al 2022). This risk is particularly pronounced in the event that those changes overlap with large populations and/or with loci of political power.
In the case of our analysis, we find that substantial fractions of the global population and global economic activity could experience 10 year periods of local-scale warming well after global CO2 emissions reach net-zero (figure 4). In particular, using the IPCC uncertainty thresholds (Mastrandrea et al 2011), hundreds of millions of people and trillions of dollars of GDP in the United States, China and India are 'likely' to experience warming conditions in the decade after net-zero emissions are achieved. This likelihood highlights the importance of effective adaptation policies, both to ensure that communities are prepared for the variations in local climate that will inevitably occur during decarbonization, and to ensure that those variations are not perceived as an indication that decarbonization efforts have failed.
Acknowledgments
We thank two anonymous reviewers for insightful and constructive feedback. We acknowledge the World Climate Research Programme, which, through its Working Group on Coupled Modelling, coordinated and promoted CMIP6. We thank the climate modeling groups for producing and making available their model output, the Earth System Grid Federation (ESGF) for archiving the data and providing access, and the multiple funding agencies who support CMIP6 and ESGF. Computational resources were provided by CEES and SRCC at Stanford University, as well as Colorado State's Walter Scott, Jr College of Engineering Asha Cluster. NSD acknowledges financial support from Stanford University. EAB and PWK acknowledge financial support from Colorado State University.
Data availability statement
The datasets analyzed during the current study are available in the Coupled Model Intercomparison Project (CMIP6) data repository from the Earth System Grid Federation (ESGF) at https://esgf-node.llnl.gov/projects/cmip6/.
All data that support the findings of this study are included within the article (and any supplementary information files).
Code availability
Code for analyzing the Keys et al perceived failure metric in the CMIP6 climate model ensemble is archived on Zenodo at the following DOI: https://doi.org/10.5281/zenodo.7865351.
Supplementary data (12.3 MB PDF)