Bayesian Errors‐in‐Variables Estimation of Specific Climate Sensitivity.

Estimation of climate sensitivity is fundamental to assessing how global climate will warm as atmospheric CO2 ${\mathrm{C}\mathrm{O}}_{2}$ concentration increases. Geological archives of environmental change provide insights into Earth's past climate, but the incomplete nature of paleoclimate reconstructions and their inherent uncertainties make estimation of climate sensitivity challenging. Thus, quantifying climate sensitivity and assessing how it changed through geological time requires statistical frameworks that can handle data uncertainties in a principled fashion. Here we demonstrate some of the hurdles to estimating climate sensitivity, with a focus on current statistical techniques that may underestimate both climate sensitivity and its associated uncertainty. To solve these issues, we present a Bayesian error‐in‐variables regression model, which can yield estimates of climate sensitivity without bias. The regression model is flexible and can account for data point uncertainties with a known parametric form. The utility of this approach is demonstrated by estimating specific climate sensitivity with uncertainty for the Eocene. Plain Language Summary: As atmospheric CO2 ${\mathrm{C}\mathrm{O}}_{2}$ increases due to human activities, the Earth will warm. But how much warming can be expected? Climate sensitivity describes how much global average surface temperature will warm with a given increase in atmospheric CO2 ${\mathrm{C}\mathrm{O}}_{2}$. While this is a simple definition, estimating climate sensitivity is difficult because Earth's climate system is complex with a number of poorly understood interacting parts. One approach to estimating climate sensitivity is to quantify how Earth's climate changed as a result of variations in atmospheric CO2 ${\mathrm{C}\mathrm{O}}_{2}$ through geological time. This information is invaluable, but it is patchy and has large uncertainties that make estimating climate sensitivity challenging. In particular, existing statistical techniques may underestimate climate sensitivity and, thus, underestimate future warming. In this paper we present an alternative approach to determining climate sensitivity that overcomes the underestimation problem and demonstrate its performance using geological data from the Eocene epoch. Key Points: We demonstrate that regression‐based estimates of specific climate sensitivity may be biased toward zero because of data uncertaintiesA Bayesian error‐in‐variables approach is developed that accounts for data uncertainties in regression‐based climate sensitivity estimatesEocene specific climate sensitivity is estimated to demonstrate the utility of Bayesian error‐in‐variables regression [ABSTRACT FROM AUTHOR]