Paleoclimate Data Assimilation over the Common Era

Abstract

Past climates provide key insights into the drivers and behavior of the Earth’s climate system, and such insight is highly valuable in the context of anthropogenic climate change. The Common Era is a particularly useful period for scientific inquiry because it represents the baseline climate prior to the industrial revolution and provides a comparison point for understanding observed climates. Climate reconstructions, in which climate proxy records are leveraged to estimate past climate fields and variability, are a powerful tool for studying past climates. Recently, data assimilation (DA) has emerged as a promising reconstruction technique. Unlike traditional reconstruction methods, DA integrates climate proxy records with climate model output, thereby leveraging strengths of both information sources. This dissertation presents three studies that develop DA reconstruction methodologies and leverage DA for Common Era climate reconstructions. The first study examines the consequences of applying DA to a small, highly sensitive proxy network. We find that the method underestimates temporal variance as the proxy network becomes sparse. We also observe that DA is sensitive to climate model biases but find that the use of multi-model ensembles helps retain reconstruction skill. Using these insights, we then use DA to reconstruct summer temperatures in the Northern Hemisphere spatially over the last millennium. The second study uses a DA framework to reconstruct the Southern Annular Mode, a major mode of climate variability. We optimize the assimilation using insights from the first study, and also extend DA methodologies to accommodate gridded, spatially-covarying drought atlases as proxy records. We also adapt DA into an optimal sensor framework, which allows us to quantify the influence of individual proxy records on the reconstruction. Our reconstruction reduces major uncertainties inherent in existing SAM reconstructions and extends those reconstructions by a full millennium. In the final study, we present DASH, a MATLAB toolbox designed to facilitate paleoclimate data assimilation. This toolbox is motivated by the practical difficulty of implementing DA methods for generalized paleoclimate analyses and provides command-line and scripting routines that implement common tasks in DA workflows. DASH is highly modular so supports paleoclimate DA for a diverse range of time periods, spatial regions, proxy networks, and algorithms. This toolbox consolidates and clarifies complex DA methods and can serve as a tool for paleoclimatologists with varying areas of expertise. Overall, these papers help develop and establish DA as powerful tool for paleoclimate reconstruction.