Applying Information Theory and Bayesian Inference to Paleoenvironmental Interpretation

Abstract

Gathering and analyzing data in the most unbiased and reproducible fashion is crucial for successful scientific research. However, scientists are influenced by prior experiences and tend to introduce biases in data interpretation when solutions are nonunique. In Earth science, a key task is the reconstruction of paleoenvironmental conditions, which relies on measurable changes in certain attributes of the resultant landscapes or the sedimentary record. The attributes in use commonly have multiple explanations, which may result in biased interpretation. To mitigate the problem, we propose a general conceptual framework for linking the observed attribute changes to their possible causes based on Shannon's information theory. This framework further serves as a hypothesis generator. We then present a rigorous procedure for paleoenvironmental interpretation based on the multiple-hypotheses method and Bayesian inference. Application examples illustrate the benefits of this procedure in paleoenvironmental reconstructions and in understanding environmental signal propagation through sediment routing systems. Plain Language Summary Sedimentary records and landscapes are an important archive of ourplanet's past environmental conditions. Understanding such conditions is important as it can provide us with perspectives of Earth's evolution far beyond historical time scales and facilitate our prediction of future conditions. However, we are faced with the challenge of nonunique solutions since most observed landscape or sedimentary record changes have more than one possible explanation. To tackle the problem, we propose a model that links observed changes to their possible causes based on information theory. We further suggest a procedure for paleoenvironmental condition interpretation based on the multiplehypotheses method and Bayesian inference, which are proven effective ways to reduce human bias in data interpretation. Our study is an important step toward more quantitative, more reproducible, and less biased interpretation practice.

Statistical characterization of a confined submarine fan system: The Pennsylvanian Lower Atoka Formation, Ouachita Mountains, USA

“…However, the lack of high-resolution chronological constraints inhibits further distilling of these signals. Collectively, despite the stochastic nature of submarine fan settings and non-linear filtering during the propagation of environmental signals (Jerolmack & Paola, 2010;Li & Plink-Björklund, 2019;Romans et al, 2016;Straub et al, 2020), it is possible to distill the dominant extrinsic and intrinsic signals with sufficient knowledge of potential control factors (Burgess et al, 2019), although chronological constraints serve as an important validation.…”