Recent scientific advances make it possible to assign extreme events to human-induced climate change and historical emissions. These developments allow losses and damage associated with such events to be assigned country-level responsibility.
The Paris Agreement of December 2015 ruled out the possibility that “addressing loss and damage associated with the adverse effects of climate change”1 “should provide a basis for liability or compensation”2. While pre-empting the issue of state-level liability between Parties to the Agreement, the careful language of Article 8, recalling that “climate change” is explicitly defined as “a change of climate which is attributed directly or indirectly to human activity”3, indicates that attribution remains at the heart of this topic. Assigning historical responsibility is of relevance not only for financial interests, but also for climate justice through contributing to reconciliation and moral repair4,5,6.
It is now possible to quantitatively determine the contribution of individual countries to global mean temperature change7,8 (GSMT). However, GMST does not directly incur economic or societal losses, which instead result from extreme weather and climate-related events, rendering these of particular interest in the context of liability and anthropogenic climate change9. Methodological developments have increased robustness and confidence in event attribution10, and it is now possible to say with high confidence that the likelihood of occurrence of individual classes of extreme weather events has increased due to anthropogenic climate change (for example, ref. 11). Here, however, we go a step further and highlight the potential of assigning individual countries' greenhouse gas and aerosol emissions to specific extreme weather events.
Assigning historical contributions
As an example, we use the summer 2013–2014 heatwave in Argentina, during which time large parts of the country witnessed the highest temperatures on record. This particular heatwave was chosen on the basis of its comparatively large anthropogenic signal, which previous studies reveal made the event five times (400%) more likely12.
Using the climate model data described in ref. 12 (including discussion on associated uncertainties), we simulate two ensembles of possible summer temperatures in Argentina: 2013–2014 conditions as observed, and counterfactual conditions without anthropogenic climate change. Based on these ensembles of raw climate model output, we apply two different statistical methods to estimate the change in the frequency of this event attributable to individual regions' greenhouse gas emissions.
The underlying assumption in both cases is that the GMST contribution of individual countries can be linearly transferred to quantify their contribution to the Argentinian heatwave. While a strong assumption, the only way to explicitly test this would be to employ large ensembles of high-resolution coupled climate models where countries' individual emissions could be removed. Lacking the capability for such a test, we assume that the assumption holds for extreme events where the anthropogenic contribution is mainly through thermodynamics, of which large-scale heatwaves are good examples. If, however, dynamic changes are a main driver of the change in risk, the assumption might be less justifiable. It can also be argued that by focussing on global mean temperature as the main 'responsibility indicator', we follow the pragmatic choice taken in global climate policies where the target is given in global mean temperature. There are two straightforward statistical methods to implement this choice.
The first methodology (hereafter 'distribution method') fits a distribution to the raw model data for both ensembles and estimates the percentage change in the distribution characteristics for individual regions by applying the contributions to GMST from Skeie et al. (Supplementary Table 1 in ref. 8). The second approach (hereafter 'gradient method') differs in that the return-time curve is used directly to calculate the gradient of the curve at the threshold of the event in both ensembles, and scaled between the two gradients according to the percentages given in Skeie et al. Using this methodology, no assumptions about the shape of the distribution (as in the first approach) are made. However, analysing a normally distributed quantity such as temperature, the gradient of the return time curve is highly sensitive to small changes in the temperature (Supplementary Fig. 1).
The Argentina case study
Using both methods, we estimate the contribution of an individual country, asking the question: whether, and to what extent, does the likelihood of the Argentinian heatwave occurring change when we remove the individual countries' greenhouse gas contributions from present day simulations? Figure 1 shows results for 10 of the 20 regions from Skeie et al.8; data for all 20 regions is shown in Supplementary Fig. 4.
As revealed by the distribution method, excluding the greenhouse gas contribution from the EU28 makes the Argentinian heatwave a 1-in-15-year event, compared to a 1-in-12-year event under the current climate that includes all emissions. Such a 37% increase in risk using the distribution method compares well with, and is within the uncertainty estimates of, the 31% reported by the gradient method (Fig. 1). For comparison, the contribution from the US is assessed at 34% (20–54% uncertainty) for the distribution method, and 28% (19–45% uncertainty) for the gradient method. In contrast, smaller traditional emitters such as China and India are assessed to have increased Argentinian heatwave risk by 21% and 18% (China), and 11% and 10% (India), for the two methods, respectively (Fig. 1). The difference between the methods increases with more extreme events, leading to large differences in the individual contribution shares in the overall change in risk (see Supplementary Fig. 2).
Considering that the overall risk of the heatwave was made 400% more likely due to total anthropogenic emissions12, an increase in risk of 37% due to the largest historical emitter, the EU28, appears rather small (Fig. 1). However, these values are comparable to results of overall changes in other attribution analyses — for example, extreme precipitation events13 — and are significantly different from zero, thus representing a robust result. It is important to note, however, that all temperature-related attribution statements crucially depend on the definition of the event, and hence the threshold of evaluating the return times (for rainfall events that follow a Gumbel distribution this dependency is much weaker (see Supplementary Fig. 2 and, for example, ref. 13).
The importance of choosing the baseline
As regional contributions are calculated by individually subtracting their effect, the changes in individual likelihoods illustrated in Fig. 1 do not add up to the total (see also Supplementary Fig. 4). To do so requires contributions to be subtracted cumulatively. Given that individual contributions add up to the total change in likelihood, it could be argued that this would be the most realistic method. However, the change in likelihood assigned to an individual country strongly depends on the change in risk comparing the present-day climate to the pre-industrial one. Such a cumulative approach requires countries to be ordered, the choice of which is somewhat arbitrary; for example, a country with higher emissions in the pre-industrial climate would be assigned a higher responsibility compared to a country with the same share of emissions but where these emissions were released in more recent times. It is, however, important to note that the timing of emissions is also taken into account when accounting for the regions' contributions to GMST8; if a cumulative approach would be applied it needs to be assured that the accounting period is the same. In the context of extreme events (excluding tipping points) we can conclude that early emissions matter more.
If we assume the US were the first emitters, the US would get assigned a responsibility for an over 100% change in likelihood of the Argentinian heatwave occurring (light blue bars) while the same region is only assigned responsibility for a 28% change if the order is reversed (medium blue) and the US were the last emitter. This shows that calculating the contributions in a cumulative way would be very sensitive to the order of countries, making it hard to justify without a clear temporal order of emissions.
This also highlights that the decision of whether to use the current climate as the baseline and calculate the change in likelihood if the individual country would not emit but all others would, as done in the examples above, is crucial for the overall magnitude of the assigned contribution. If the pre-industrial climate were taken as baseline and it is assumed only the region in question would emit, but all others do not emit, values for each countries' contributions would be considerably higher.
Figure 1 thus indicates that timing of emissions matters more than the choice of methodology. This means that the choice of methodology amounts to differences much smaller in magnitude than the size of the overall contribution of that region to the event. Thus, a key choice that matters for an event with a robust anthropogenic signal such as the Argentinian heatwave is how the question is formulated: 'How would the likelihood of the event change if only the region in question has emitted?' versus 'How would the likelihood of the event change if the region of interest had not emitted?'
Summary and policy implications
While the increase in GMST provides the most robust evidence of anthropogenic climate change, it does not directly cause damage and loss of life and livelihood, which are instead primarily the result of extreme weather. With rapid developments in the science of extreme event attribution10, it is now possible to robustly assess the extent to which anthropogenic climate change has altered the likelihood of such events. The role of scientific evidence in addressing climate-related loss and damage is presently unclear14. The considerations above, however, highlight that historical responsibility of individual countries and regions can now be quantified for specific extreme events, as shown here for an Argentinian heatwave during 2013–2014. By combining the upper and lower bounds of the two methodologies utilised, it is found that EU28 emissions made the Argentinian heatwave 19–60% more likely, while total anthropogenic emissions increased the likelihood by 400%.
For comparison, the Arctic heatwave in December 2016 was made more than 1,000% more likely (following an analysis from ref. 15) leading to individual contributions from the EU28 and US more than doubling the risk (not shown). By contrast, the extreme rainfall event in the UK during January 2014 (ref. 13) was made ∼40% more likely due to total anthropogenic emissions, but the EU28 contribution was only ∼3% (see Supplementary Fig. 3). Given the sensitivity of historical responsibility to the baseline definition and the specific method used, it is imperative that these decisions are transparently presented.
It is unlikely that international or multi-national agreements on loss and damage will include or ask for quantitative information in the near future. However, the fact that it is possible to provide such quantifications will greatly advance the possibility of an informed discussion on whether such information is useful, necessary, and should be included in multi-national agreements. Furthermore, the possibility of assigning contributions of individual regions to damage could have the potential to reshape environmental litigation9, raising questions regarding damage and responsibility in national jurisdictions, and thus climate justice.
Supplementary Figures and Tables
About this article
Nature Communications (2018)