Investigating key uncertainties across global and national decarbonisation pathways to 2050; a novel clustering and classification approach

Integrated assessment models are crucial in informing decarbonisation strategies of national and international governing bodies to mitigate the impacts of climate change. Whilst these models have become highly sophisticated over the 21st century, uncertainties are still inherent, both in model structure and input parameters. Uncertainty analysis is therefore vital for understanding plausible scenarios of the energy transition, which have propagating impacts throughout societies concerning economic and energy stability, public health, and equitable access. However the optimal approach to uncertainty analysis is inconclusive; methods remain computationally and time intensive, and each can produce different conclusions based on inferences. This report takes a novel combination-approach, applying machine learning clustering and classification algorithms to a new multi-model and multi- parameter ensemble of 1,632 pathways, and shows a distinct lack of variability between different levels of decarbonisation - 1.5oC, 2oC or warmer world. Despite 2050 net zero targets being considered extremely challenging, this report finds, via mixed parameter K-means clusters and minimal predictive ability of Random Forest classification, that these pathways apply similar strategies for attainment as for higher warming targets. Primary drivers of pathway variability are instead found to be development region and model used, whilst also acknowledging that development year mirrors external macroeconomic factors. These results question the comprehensiveness of pathways and suggests that less probable yet still possible scenarios are not being considered.