Rewrite High-income groups disproportionately contribute to climate extremes worldwide as a headline for a science magazine post, using no more than 8 words

We quantified intensity and frequency changes in extremely hot (dry) months attributable to specific emitter groups. The methodological framework relies on three steps (Fig. 1): first, we constructed counterfactual emissions pathways (that is, emissions pathways with and without the emissions of selected population groups); second, we translated emissions into gridded temperature, precipitation and potential drought data via a chain of computationally efficient emulators; and third, we built on the framework of extreme event attribution to quantify changes in the grid-cell-level distributions of the climatic variables.

We relied on the SPI-3 to identify meteorological droughts²⁰. The SPI-3 is computed from precipitation data only, meaning that it does not account for changes in soil- and plant-based water demands. As climate-driven precipitation signals are dominated by natural variability and intermodel disagreement²⁴, climate change-induced trends in our drought indicator are probably a conservative estimate of actual changes²⁰. Therefore, we also computed the SPEI-3⁴⁸. The SPEI-3 takes changes in water demands via potential evapotranspiration (PET) into account. Ideally, PET is estimated from temperatures, radiation, wind speed and humidity via the Penman–Monteith equation^20,49. Given that our emulation framework only depicts temperature and precipitation, we relied on the Thornthwaite method to compute PET from temperature data only⁵⁰. However, PET estimates via the Thornthwaite method are prone to overestimations in terms of magnitude and temporal trends⁵¹. This left us with an indicator for meteorological droughts (SPI-3) that probably underestimates drought risks and an additional indicator for potential droughts (SPEI-3 via the Thornthwaite method) that probably provides an overestimation. We used the conservative estimates in the main part of our analysis and show potential drought risks in Supplementary Section 5.

Counterfactual emissions pathways

We assessed what our climate today would look like if the wealthiest 10%, 1% and 0.1% globally, as well as in the United States, EU27, India and China, had not contributed to global emissions between 1990 and 2019. We followed ref. ⁵² to construct a time series of historic baseline emissions from 1850–2019 resolved by gas. Next, we removed emitter-specific contributions from these baseline emissions (Fig. 1). To do so, we relied on a dataset of consumption-based CO₂e emissions categorized by country and income decile between 1990 and 2019³. The estimates relate to all emissions except those from agriculture, forestry and other land use. Our analysis required us to make assumptions about how to disaggregate the reported basket emissions into individual gases. We focused on decomposing emissions into CO₂, nitrogen oxide (N₂O) and CH₄. These three gases make up 98.7% of the total global GHG emissions (excluding agriculture, forestry and other land use)⁵³. The composition of production-side GHG emissions varies strongly by country, ranging from primarily CO₂-based emissions (for example, Singapore) to almost equal shares of CO₂ and CH₄/N₂O (for example, Qatar) and, in low-income countries in particular, primarily CH₄/N₂O (for example, Chad)⁵⁴. The carbon inequality dataset from ref. ³ employs input–output tables that redistribute production-side emissions to consumers across countries. About one-half of global CH₄ emissions are embodied in global trade, with household consumption dominating the final demand category⁵⁵. Given these considerations, and a lack of alternative data, we chose to apply the same decomposition assumptions across countries and emitter groups. For our central estimate, we assumed that emissions for each GHG scale proportionally with the globally aggregated emissions. We tested the sensitivity to this assumption by providing two extreme cases in which the wealthy emitters (1) solely emit carbon (CO₂case) or (2) solely emit CH₄ and N₂O (non-CO₂case). Note that in the non-CO₂ case, the emissions associated with the global top 10% are larger than the total global CH₄ and N₂O emissions combined, and we removed the excessive emissions from the CO₂ time series. We converted between individual GHGs and CO₂e using the Global Warming Potential 100.

Emulator-based modelling approach

We transformed counterfactual emissions into grid-cell-level distributions of temperature and precipitation using emulators and subsequently computed drought measures from the emulated data. The emulation consisted of two steps: first, converting emissions into GMT; and second, translating GMT into grid-cell-level monthly mean temperature and precipitation distributions (Fig. 1). The first translation step was carried out with MAGICC^10,56. MAGICC is a simple, computationally efficient climate model for global climate indicators. Our temperature outcomes were calculated with MAGICC v7.5 in a probabilistic setting that reflects the assessed uncertainty ranges from the IPCC’s Sixth Assessment Report²⁴. We generated 600 GMT trajectories for each scenario. The second translation step was carried out using MESMER-M-TP¹¹. MESMER-M-TP combines parametric approaches and stochastic sampling to approximate the behaviour of individual climate models. For any climate model, the emulator can be calibrated with a small set of actual climate model data and then used to generate gridded temperature and precipitation data that statistically resemble the climate model data. Here we calibrated MESMER-M-TP with 24 different models from the Phase Six of the Coupled Model Intercomparison Project (Supplementary Table 4). Subsequently, we converted each GMT trajectory into a single gridded time series of temperature and precipitation. We computed the SPI-3/SPEI-3 indicator following ref. ⁴⁸ and used the gamma distribution for normalization. This provided us with a dataset containing 4 variables × 600 realizations × 2,652 grid points × 170 years × 12 months for each scenario.

Attribution framework

Traditional attribution studies typically aim to understand how climate change altered the statistics of a specified observed extreme. Our study deviates from this approach. We were interested in understanding the extent to which changes in a broad class of historic extremes can be related to emissions from specific emitter groups. We therefore used the framework for event attribution as a guideline¹⁴ but modified it according to our research questions. Most importantly, our analysis fully relied on modelled data, meaning that we were not taking observational data into account. Hence, the event attribution framework was reduced to three essential steps: first, we defined extreme events; second, we performed an analysis using emulated (climate model) data; and third, we synthesized the hazards into an attribution statement.

Extreme event definition

We defined extreme events relative to the reference period 1850–1900 and focused on 1-in-100-year (main text) and 1-in-50/1-in-10,000-year (Supplementary Information) events.

Climate model analysis and hazard synthesis

In a first step, we tested whether changes in the grid-cell-level distribution of a climatic variable under a given counterfactual scenario were significantly different from its present-day distribution. To this end, we computed the differences between the present-day and counterfactual present-day distributions and employed a Student’s t-test⁵⁷ to verify that the distribution was significantly different from zero. If this was the case, we proceeded with the actual attribution. We used the modelled distribution of climatic variables over the reference period to derive grid-cell-specific intensity thresholds for our defined events. To assess frequency changes, we counted how many times the reference intensity threshold was exceeded in a present-day (2020) climate and in a counterfactual 2020 climate, and attributed the difference to a specific emitter group. Similarly, we quantify intensity changes by assessing how hot (dry) a specific extreme would be in a present-day climate as compared to a counterfactual climate, and attribute the difference in values.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.