Climate scenarios with probabilities via maximum entropy and indirect elicitation

The authors address a major gap in climate scenario analysis by proposing a framework to assign probabilities to future climate scenarios, rather than treating them as equally plausible narratives. They develop two complementary approaches: one based on the maximum entropy principle and another based on the elicitation of economists' estimates of the Social Cost of Carbon (SCC), which they translate into probabilities of future emissions-abatement policies. After adjusting for the persistent gap between economists' recommendations and actual climate policies, they derive probability distributions for end-of-century temperature outcomes. Their results are robust across a wide range of modelling assumptions and suggest that limiting warming to the Paris Agreement targets is unlikely under current policy trajectories, while warming above 3°C remains a significant risk.

Because widely used climate scenarios (such as the IPCC SSP-RCP pathways) are generally not accompanied by probabilities, the authors argue that their framework provides a more useful basis for investors, regulators, and policymakers who need probabilistic assessments for risk management, asset valuation, and long-term decision-making.

KEY FINDINGS

• The authors develop a methodology to assign probabilities to climate scenarios using both maximum entropy and expert elicitation of the Social Cost of Carbon, producing robust probability distributions for future warming outcomes.
• They find that the probability of limiting end-of-century warming to 1.5°C is very low, while the median projected warming is around 2.8–2.9°C, with a substantial probability (roughly 35–40%) of temperatures exceeding 3°C by 2100.
• The authors show that there is a remarkably stable relationship between the Social Cost of Carbon and the aggressiveness of optimal emissions-abatement policies, allowing expert estimates of carbon pricing to be translated into probabilities of future climate policy.
• They conclude that the commonly used SSP5-RCP8.5 scenario represents a low-probability but non-negligible tail risk that should continue to be considered in climate stress testing and long-term risk analysis.
• The findings imply that investors, financial institutions, and regulators should complement deterministic climate scenarios with probabilistic assessments, enabling more realistic evaluations of physical climate risks, transition risks, and long-term economic impacts.

ABSTRACT: 

What are the probabilities of a C, C or C world? In the last decades, researchers and policymakers have worked with a wide range of future climate scenarios, but the likelihood of their realisations has never been quantified. We propose two ways to obtain a probability distribution for the aggressiveness of emission abatement policies: one method is based on the elicitation from economists of estimates of the optimal social cost of carbon; the second on the principle of maximum entropy. In both cases we make use of information on the technological and fiscal feasibility of various policies to bound the distribution, and of the observed difference between average recommended and implemented policies to adjust its mean. We find that the results are robust to methodological choices. They suggest that the likelihood of achieving the Paris Agreement target is small; that there is a significant probability of an end-of-century temperature anomaly above C; that the much-studied 8.5 W/ forcing has low probability, but should not be neglected, at least in tail-event studies. If one considers, as we do, the distribution of end-of-century temperatures we obtain as ‘too dangerous’, the disconnect between economists’ recommendations and policy action should be substantially reduced.