Complex Empiricism and the Quantification of Uncertainty in Paleoclimate Reconstructions

Abstract:

Because the global climate cannot be observed directly, and because of vast and noisy data sets, climate science is a rich field to study how computational statistics informs what it means to do empirical science. Traditionally held virtues of empirical science and empirical methods like reproducibility, independence, and straightforward observation are complicated by representational choices involved in statistical modeling and data handling. Examining how climate reconstructions instantiate complicated empirical relationships between model, data, and predictions reveals that the path from data to prediction does not match traditional conceptions of empirical inference either. Rather, the empirical inferences involved are “complex” in that they require articulation of a good deal of statistical processing wherein assumptions are adopted and representational decisions made, often in the face of substantial uncertainties. Proxy reconstructions are both statistical and paleoclimate science activities aimed at using a variety of proxies to reconstruct past climate behavior. Paleoclimate proxy reconstructions also involve complex data handling and statistical refinement, leading to the current emphasis in the field on the quantification of uncertainty in reconstructions. In this presentation I explore how the processing needed for the correlation of diverse, large, and messy data sets necessitate the explicit quantification of the uncertainties stemming from wrangling proxies into manageable suites. I also address how semi-empirical pseudo-proxy methods allow for the exploration of signal detection in data sets, and as intermediary steps for statistical experimentation.