How historic simulation–observation discrepancy affects future warming projections in a very large model ensemble

Abstract

Projections of future climate made by model-ensembles have credibility because the historic simulations by these models are consistent with, or near-consistent with, historic observations. However, it is not known how small inconsistencies between the ranges of observed and simulated historic climate change affects the future projections made by a model ensemble. Here, the impact of historical simulation–observation inconsistencies on future warming projections is quantified in a 4-million member Monte Carlo ensemble from a new efficient Earth System Model (ESM). Of the 4-million ensemble members, a subset of 182,500 are consistent with historic ranges of warming, heat uptake and carbon uptake simulated by the Climate Model Intercomparison Project 5 (CMIP5) ensemble. This simulation–consistent subset projects similar future warming ranges to the CMIP5 ensemble for all four RCP scenarios, indicating the new ESM represents an efficient tool to explore parameter space for future warming projections based on historic performance. A second subset of 14,500 ensemble members are consistent with historic observations for warming, heat uptake and carbon uptake. This observation–consistent subset projects a narrower range for future warming, with the lower bounds of projected warming still similar to CMIP5, but the upper warming bounds reduced by 20–35 %. These findings suggest that part of the upper range of twenty-first century CMIP5 warming projections may reflect historical simulation–observation inconsistencies. However, the agreement of lower bounds for projected warming implies that the likelihood of warming exceeding dangerous levels over the twenty-first century is unaffected by small discrepancies between CMIP5 models and observations.