Enhancing Scenario Assessment through Decomposition: An Epistemological Perspective
Francesco Nappo
9 October 2025
New statistical and machine learning tools promise to enhance the informativeness of multi-model ensemble exercises in climate economics. After introducing ensemble assessment for climate economic models, this talk provides a critical evaluation of one family of ML tools, which adopts decomposition methods to provide fine-grained information about the representation of mitigation strategies across emission pathways in multi-model ensembles. I argue that the appropriate use for these tools is providing insights into the stability of mitigation strategies under alternative scenarios, rather than explaining model behavior (as some authors have suggested). Based on this framing, I articulate three requirements on the quality of stability information in informing climate policy evaluation, namely model diversity, well-assortedness, and regional integration, and exemplify their stringency through the example of selecting an adequate energy mix for achieving ambitious climate change mitigation targets.
