Emissions and emergence: a new index comparing relative contributions to climate change with relative climatic consequences

Projected population and greenhouse gas emissions have been taken from all available Integrated Assessment Models (IAMs) for the five SSPs. Projected estimates of CO₂, CH₄ and N₂O emissions and population are available for each of the five regions of interest for each decade from 2020 to 2090, alongside observed totals for the years between 1990 and 2010.

To better estimate the climate effects of a portfolio of different greenhouse gases, in preference to the more customary global warming potentials, GWP₁₀₀, we use GWP* (Allen et al 2016), since the latter provide a much better mapping between an emissions portfolio and surface temperature impacts (Allen et al 2018). (The 'star' in GWP* is a reflection that GWP* is not a 'new' metric; it is in fact GWP₁₀₀ used in a way that gives a better mapping between emissions and temperature change.) GWP*-weighted annual emissions rates are calculated using a GWP₁₀₀ weighted sum (IPCC-AR5 values) of the annual CO₂ emissions rate, the annual N₂O emissions rate, and the rate of change in the annual CH₄ emissions rate multiplied by a time horizon factor of 100 years. We calculate the rate of change in the annual CH₄ emissions rate using the difference in annual emissions rate relative to those twenty years previously to reflect the timescales of CH₄'s impact on global temperature.

CO₂e* emissions were then calculated for each of the nine decades between 2010 and 2090. The first 20 years of data (1990–2010) is used to calculate GWP*-based estimates of CO₂*-equivalence. These projections of population and CO₂e* emissions are then summed over the nine decades for each region (respectively denoted as _Pi and C_i) and for all five regions together (respectively denoted as P_G and C_G).

There are numerous ways of comparing relative contributions to climate change (Skeie et al 2017), depending on which sectors and emission components are considered, which indicator of climate change is used, which time periods are chosen for emission and evaluation or responses and so on. Though many reasonable combinations are possible, some of these choices make more physical or policy sense than others. For instance, accurately evaluating the role of long-lived and short-lived pollutants is important for a scientifically-accurate estimate of contribution to long-term warming (Allen et al 2018). Also important are choices around baselines and reference periods, where different choices seem reasonable (Millar et al 2017, Millar et al 2018, Schurer et al 2018); and while long baselines are conceptually attractive, uncertainty increases as we move backward in time, and it is not obvious how to treat the pre-independence emissions of previously colonised societies. People may disagree over some of these choices, but it is clear that some sets of choices more coherently map to the temperature target-based climate negotiation framework than others (Skeie et al 2017). Additional innovations regarding the way contributions are assessed are left for future work, but may be important for some potential uses of the index (see the value of climate indices below).

Following previously published methods (Hawkins and Sutton 2012, Frame et al 2017), we calculate S/N for near-SATs, using the CMIP5 simulations for the 25 models that ran each of RCP2.6, 4.5 and 8.5 scenarios, and presented relative to a baseline climate of 1986–2005 (see also supplementary information available online at stacks.iop.org/ERL/14/084009/mmedia). The 'signal' is diagnosed by calculating the global mean SAT and fitting a fourth-order polynomial (GMST) across the period 1950–2100. SATs at each gridpoint are regressed against this smoothed GMST to derive a smoothed gridpoint signal that is proportional to the global mean. The 1986–2005 mean is then removed from the smoothed gridpoint data to produce the change in temperature (S). The N term is the standard deviation of annual mean temperatures in the pre-industrial control simulations at each grid point. The S/N is calculated for each model independently.

To calculate normalised S/N ratios for each of the five regions explicitly represented in the SSPs, and presented in figure 2, we first aggregate, for each model, S/N values averaged over the period 2086-95 for those grid cells which lie within the national boundaries of each of the five regions. We then calculate the mean S/N value for each aggregated region, and divide it by the mean S/N for all five regions aggregated together.

Zoom In Zoom Out Reset image size

Previous studies investigating the increasing frequency and severity of extreme heat have shown similar spatial patterns of results to those represented by the S/N calculations used here. Examples have been demonstrated across annual (Diffenbaugh and Scherer 2011, Mahlstein et al 2011, Hawkins and Sutton 2012, Lehner and Stocker 2015), seasonal (Anderson 2011, 2012, Mahlstein et al 2011, Davis and Diffenbaugh 2016, Mueller et al 2016), monthly (Coumou and Robinson 2013, Sippel et al 2015, Mueller et al 2016) and daily (Fischer et al 2013, 2014, Andrew et al 2015, Fischer and Knutti 2015, Angélil et al 2016, 2017, Luke et al 2016, Pfahl et al 2017) timescales, as well as for a variety of heatwave metrics (Simone et al 2016, Nicholas et al 2017), with all studies sharing a common framing of climate change emergence in the context of pre-existing local variability.

This ranking of emergence correlates with several of the inputs to climate change vulnerability, as well as composite indicators captured by the ND-GAIN index (Notre Dame Global Adaptation Initiative) (Chen et al 2015) (figure S29), so it seems reasonable to conclude that the emergence pattern reflects important climate change vulnerabilities. National averages can mask domestic heterogeneity, which may be significant (Green 2016)—however, a positive correlation is found between the magnitude of sub-national income inequality (as measured with a Gini coefficient) and the severity of temperature emergence (figure S27(b)). In addition, a robust anti-correlation also exists between the magnitude of temperature emergence and metrics of both progress towards achieving the United Nations Sustainable Development Goals, and per capita national incomes (figures S28(b) and S26(b) respectively). The focus on patterns of temperature-driven emergence is supported by previous results which highlight the links between increasing heat extremes and reduced crop yields (Lobell and Burke 2008, Battisti and Naylor 2009, Lobell et al 2011, Asseng et al 2014, Liu et al 2016), as well as impacts on ectotherms (Deutsch et al 2008), even if slow-emerging impacts, like changes to ecosystem zones (Mahlstein et al 2011) and more-frequent precipitation extremes (Andrew et al 2015), will not necessarily be well captured with a focus on temperature S/N ratios. Thus the emergence pattern does not capture all important dimensions of impacts, but it does capture many important ones, and as characterisation of the emergence of other variables develops (e.g. Zhang et al 2018, Rojas et al 2019) we can look to include these in future revisions. Significantly, spatial patterns similar to the emergence patterns we identify are also evident when comparing the temperature emergence literature with other climate vulnerability indices (Althor et al 2016).

Figure 2(a) shows the S/N ratios, normalised relative to the global average, for five regions and three RCP scenarios, with regional aggregations following those used in the SSPs. The different forcing scenarios lead to very overall different levels of climate change, both in terms of temperature change above pre-industrial, and in terms of the S/N ratios expected by the end of the century (Frame et al 2017). However, when the S/N ratios are normalised relative to the global average S/N for each of those scenarios, a very consistent order of relative emergence becomes apparent across all three scenarios: the Middle East and Africa experiences the largest relative climate change, followed by Latin America and Asia, with the Organisation of Economic Cooperation and Development (OECD) and reforming economies experiencing slower relative climate change under all scenarios. Despite substantial model uncertainty in the S/N ratios, this general sequence in which regions experience emergence of the climate signal above pre-existing variability faster than others remains strongly robust, and is largely insensitive to the choice of model (table S1). This lack of scenario uncertainty (Hawkins and Sutton 2009) therefore suggests that normalised S/N ratios represent a socioeconomically robust variable with which to construct an overall measure of the distribution of important climate impacts.

In terms of assessing the relative roles of different forcing agents on temperatures, for illustration we use the SSP dataset, which implies using production emissions and using 2010 as the start date for counting emissions (choosing of a different start date would make a difference of a few percent to contributions to warming) (Skeie et al 2017). The long-standing convention of using production emissions rather than consumption emissions (Davis and Caldeira 2010) is noted, and this clearly matters for discussions about responsibility. With appropriate data inputs, the EEI could easily be tweaked to incorporate a consumption-based approach instead or, indeed, some hybrid partitioning between consumption and production. In line with recent research (Allen et al 2016), we weight emissions by GWP* because this is a better predictor of temperature development than is GWP (the basis of CO₂-e emissions).

Figure 2(b) shows normalised cumulative CO₂e* emissions per capita between 2010 and 2090, for each of five regions resolved in the SSPs under a range of different IAMs. The width of the bars represents inter-IAM spread. Because different regions could follow different development pathways in the future (i.e. development more similar to different SSPs in different regions), we cannot make the same simple pairwise comparison regarding the constancy of the relative contribution to warming in the future that we make for normalised emergence.

In essence, the S/N or emergence elements of climate change are determined by global concentrations of GHG, and are largely insensitive to the national origin of emissions. On the other hand, contributions to climate change are determined by the national origin of emissions (at least insofar as nations provide the usual way of determining contributions). We can use estimates of past GHG emissions to determine contributions, but to estimate future scenarios we must consider the possible patterns of future GHG emissions. This is why it is sufficient to consider only global concentrations for emergence, but why we must resolve emissions at regional or national scale.

We can, however, examine the extent to which differing scenarios of future emissions indicate a reduction or exacerbation of existing differences in terms of emissions per capita. Some SSPs pull regions towards unity (i.e. relative emissions parity); others push them away from it. Most IAMs find that global SSP1, SSP2, or SSP5 trajectories imply a diminution of existing inequalities between the OECD and the rest of the world. The reasons are different in each case: in SSP1 the OECD countries take the lead in emissions reductions and decarbonise their economies much faster than economies elsewhere; by contrast, in the high carbon SSP3 and SSP5 worlds, OECD emissions revert towards the global per capita average because other regions catch up to the OECD's (high) levels. In the intermediate SSPs, emissions per capita inequalities remain high. Interestingly, under the mitigation-oriented SSP1 the Middle East and Africa actually exacerbates existing inequalities in terms of per capita emissions; if everyone mitigates then there is contraction, but no convergence, of relative responsibilities for climate change.

There are of course important caveats, such as the limited number of IAMs with diverse abilities to represent energy-economy in different groups of countries. At more refined levels of aggregation—those at which national policies are set—the picture becomes more variegated. SSPs are indicative, rather than prescriptive, normative, or predictive. As the developeds of SSPs have noted (O'Neill et al 2014), 'SSPs are only examples of the kinds of socioeconomic futures that can produce particular challenges to adaptation and mitigation'. In the normalisation we employ, we interpret the SSPs as place-holders for future emissions trajectories to illustrate the point that future fossil fuel emission use will have implications for the pattern of relative contributions to climate change.

Combining the information from figures 2(a) and (b) to define the EEI enables a novel method of expressing, relative to the global median, relative contributions to change, alongside the relative emergence of impacts (compared to a baseline local climate).

Figure 3 displays an estimate of historical EEIs for all countries with populations above one million people, comparing normalised cumulative GWP* weighted emissions per capita for 1970–2012 against normalised signal-to-noise ratios (using the average of all models across all RCP scenarios from figure 1(a)). Because the emergence pattern is relatively insensitive to the amplitude of the forcing, the horizontal ordering of countries is relatively insensitive to whether the world follows a high or low emissions trajectory—because they are robust spatial patterns, and because we are normalising the emergence pattern to pick out national variations, it matters little whether we use emergence patterns to date or diagnose them from future forcing trajectories. EEI values range from as high as 8—for slow-emerging and prosperous Northern European countries—to well below 1/100 for populous low income countries, such as Burundi. There is also more diversity in the position of individual nations (table S4), with Singapore and Malaysia being both disproportionate contributors to emissions and disproportionately impacted in terms of how fast their climates are changing. Collectively however, nearly all of the highest and lowest income nations exhibit EEI estimates above 2 or below 1/2 respectively, with few exceptions.

Zoom In Zoom Out Reset image size