Navigating Source Attribution Methods for Linking Individual Actors to Climate Change

Attribution science is increasingly extending beyond establishing the role of aggregate anthropogenic forcing in climate change to quantifying contributions from individual sources, such as sectors, nations, income groups, or corporations. This extension of attribution to sources raises fundamental scientific questions about how specific emissions contribute to changes in the global climate system, extreme events, and their impacts. A growing scientific literature now applies multiple, partially overlapping methodological frameworks, yet their assumptions, capabilities, and domains of applicability are not always articulated in a unified manner, potentially confusing the community on these different approaches.

Here, we synthesize and compare source attribution methods across the four connected stages of the climate system: emissions, global climate, local climate, and impacts. Because of the large number of actors, many counterfactual runs have to be computed in source attribution, hindering the direct use of climate models due to their high computational cost. To translate actor-attributed emissions into changes in global climate indicators, reduced-complexity climate emulators are therefore commonly employed. We show that while the choice of emulator itself matters primarily in specific settings, the broader methodological approach has stronger implications, especially for uncertainty treatment and the incorporation of observational constraints. We contrast emulator-based approaches with proportional methods based on fractions of cumulative emissions, highlighting their conceptual simplicity but also their limitations in representing Earth system inertia, non-CO₂ emissions, and non-linear climate responses.

From global climate indicators to local climate, we compare three existing approaches: pattern scaling, spatial climate emulators, and extreme event attribution frameworks. We demonstrate that pattern scaling offers a computationally efficient pathway and facilitates rapid downstream extensions to impact attribution, but is limited to representing central estimates of the local climate. Spatial climate emulators are more sophisticated in this regard, allowing the representation of local climate variability, but this framework still does not represent precisely observed extreme weather events. Extreme event attribution frameworks are capable of representing observed events, but are limited in their ability to inform about future events. We discuss the capabilities of these frameworks to investigate not only the source-attributed changes in intensities, but also in probabilities to inform about causality.

We then illustrate how these methodological differences propagate into impact attribution using heat-related mortality as an example. Linking source-attributed climate changes to epidemiological models reveals that choices made upstream can substantially affect quantitative estimates of attributable impacts. In particular, strong non-linearities in temperature-mortality relationships challenge standard “but-for” counterfactual approaches and require careful methodological adaptations.

The presentation concludes by reflecting on the broader societal relevance of source attribution science. As source attribution is increasingly used to inform assessments of responsibility, including in health impact studies, clarity about methodological foundations, uncertainties, and appropriate interpretation becomes essential. By quantitatively comparing methods across the full attribution chain and illustrating their implications for heat-related mortality, this work aims to strengthen the coherence, transparency, and robustness of source attribution science, and to support its careful and context-appropriate application in policy and legal contexts.