Debiasing the uncertainties of climate stabilization ensembles

ongoing

Mathematical models have become central tools in global environmental assessments. To serve society well, climate change stabilization assessments need to capture the uncertainties of the deep future, be statistically sound and track near-term disruptions. Up to now, conceptual, computational and data constraints have limited the quantification of uncertainties of climate stabilization pathways to a narrow set, focused on the current century. The statistical interpretation of scenarios generated by multi-model ensembles is problematic due to availability biases and model dependencies. Scenario plausibility assessments are scant. Simplified, single-objective decision criteria frameworks are used to translate decarbonization uncertainties into decision rules whose understanding is not validated. EUNICE aims to transform the methodological and experimental foundations of model-based climate assessments through quantification and debiasing of uncertainties in climate stabilization pathways. Our approach is threefold: construct, consolidate and convert. We first apply simulation and statistical methods for extending scenarios into the deep future (beyond the current century and status quo), quantifying and attributing deep uncertainties. We consolidate model ensembles through machine learning and human ingenuity to eliminate statistical biases, pin down near-term correlates of long-term targets, and identify early signals of scenario plausibility through prediction polls. Finally, we use decision-theoretic methods to convert model-generated maps of the future into resilient recommendations and experimentally test how to communicate them effectively. By advancing the state of the art in mathematical modelling, statistics, and behavioural decision-making, we strengthen the scientific basis of climate assessments, such as those of the IPCC. The approach and insights of EUNICE can be applied to other high-stakes environmental, social and technological evaluations.