The Paleoclimate Uncertainty Cascade: Tracking Proxy Errors Via Proxy System Models.
Abstract:Paleoclimatic observations are, by nature, imperfect recorders of climate variables. Empirical approaches to their calibration are challenged by the presence of multiple sources of uncertainty, which may confound the interpretation of signals and the identifiability of the noise. In this talk, I will demonstrate the utility of proxy system models (PSMs, Evans et al, 2013, 10.1016/j.quascirev.2013.05.024) to quantify the impact of all known sources of uncertainty. PSMs explicitly encode the mechanistic knowledge of the physical, chemical, biological and geological processes from which paleoclimatic observations arise. PSMs may be divided into sensor, archive and observation components, all of which may conspire to obscure climate signals in actual paleo-observations.
As an example, we couple a PSM for the δ18O of speleothem calcite to an isotope-enabled climate model (Dee et al, submitted) to analyze the potential of this measurement as a proxy for precipitation amount. A simple soil/karst model (Partin et al, 2013, 10.1130/G34718.1) is used as sensor model, while a hiatus-permitting chronological model (Haslett & Parnell, 2008, 10.1111/j.1467-9876.2008.00623.x) is used as part of the observation model. This subdivision allows us to explicitly model the transformation from precipitation amount to speleothem calcite δ18O as a multi-stage process via a physical and chemical sensor model, and a stochastic archive model. By illustrating the PSM's behavior within the context of the climate simulations, we show how estimates of climate variability may be affected by each submodel's transformation of the signal. By specifying idealized climate signals(periodic vs. episodic, slow vs. fast) to the PSM, we investigate how frequency and amplitude patterns are modulated by sensor and archive submodels. To the extent that the PSM and the climate models are representative of real world processes, then the results may help us more accurately interpret existing paleodata, characterize their uncertainties, and design sampling strategies that exploit their strengths while mitigating their weaknesses.