Data-Driven Predictions of Peak Warming Under Rapid Decarbonization

Abstract

The severe impacts associated with recent record-setting annual global temperatures elevate the need to accurately predict the hottest conditions that could occur even if the most ambitious decarbonization goals are achieved. We use convolutional neural networks (CNNs) to predict peak global warming from recent observed temperature maps and future cumulative CO₂ emissions. For the SSP1-1.9 decarbonization scenario there is >99% probability that mean global warming exceeds 1.5°C, approximately even odds that it reaches 2°C, and ∼90% probability that the hottest year globally exceeds 2023 by at least 0.5°C. Further, for the SSP2-4.5 decarbonization scenario, there is >90% probability that the hottest annual global temperature anomaly is twice the 2023 anomaly. That our framework makes highly accurate out-of-sample predictions of the hottest historical year provides confidence in the predicted future probabilities, suggesting substantial risks from the extreme local conditions that are likely to result from globally hot years during rapid decarbonization.