How much of the uncertainty in paleoclimate reconstructions arises from uncertain climate physics? Quantifying structural uncertainties with an application to the Last Millennium

Abstract:

Reconstructing paleoclimate conditions relies on information from both climate proxy data and statistical or dynamical relationships that relate those data to climate variables across space and time. Here we explore structural uncertainties in paleoclimate reconstructions arising from limitations in our knowledge of how climate quantities covary with one another. We develop a data assimilation procedure to assess these uncertainties using covariance relationships from several climate models, and then apply the approach to paleoclimate networks and observational uncertainties representative of data for the last millennium. We find that structural uncertainties arising from uncertain climate (model) physics contribute roughly 10-20% of the total uncertainty in paleoclimate reconstructions of temperature variability, both at small (~200km) and continental and hemispheric length scales, with larger errors in regions where the impact of long-range climate covariances is in disagreement among models. A parallel analysis using instrumental data supports the hypothesis that model-model physics differences provide a useful measure of errors arising when model physics are used to reconstruct the real climate system. The results point to the importance of understanding and mitigating structural uncertainties arising from different representations of climate physics in paleoclimate reconstructions.