Using Last Glacial Maximum proxy data and a large ensemble of climate simulations to constrain an estimate of equilibrium climate sensitivity

While recent "emergent constraint" approaches provide useful insight into the precursors that predict a climate response to a doubling of CO2, a complete assessment of associated radiative forcing and feedback uncertainty must be considered in calculating equilibrium climate sensitivity (ECS). Therefore, past climates under markedly different radiative forcings may provide a useful complementary constraint on ECS and future climate change. The Last Glacial Maximum (LGM, 21,000 years before present) was the most recent time period of large climate change, with global mean cooling associated with perturbations in several radiative forcings, including a large reduction in greenhouse gas concentrations. In addition, a wealth of proxy data from ocean cores, ice cores, and other terrestrial records have been compiled that constrain the magnitude of LGM cooling in a variety of locations around the globe. As such, the LGM provides a useful target to constrain climate simulations and the associated ECS of a particular model. However, the previous LGM-based constraint on ECS by Schmittner et al. (2011) did not include an appropriate assessment of feedback uncertainty. For example, cloud feedbacks, which are the largest source of uncertainty in comprehensive climate models, had not been included. Here, we present a new estimate of ECS using a model-data comparison from a large ensemble of simulations (>1800) with the University of Victoria Earth System Climate Model (UVic-ESCM). We introduce a simple but spatially resolved parameterization of cloud feedbacks that have been diagnosed from comprehensive climate models of the CMIP5/PMIP3 archive. In addition, we explore parametric uncertainties in dust forcing, snow albedo, and atmospheric diffusivities, which all impact important feedbacks in UVic-ECSM. Finally, we compare results from these climate simulations with proxy data using a Bayesian statistical approach that better incorporates the dominant sources of model and data uncertainty.