Unraveling remagnetization sources using statistical learning

Abstract

The paleomagnetic archive provides invaluable insights into Earth’s history, but its records are often obscuredby various geological processes. A prime example is remagnetization, which can replace the original naturalremanent magnetization. Although magnetic overprints can be detected by traditional paleomagnetic tests,the mechanisms responsible for them often remain elusive because linking bulk magnetic properties to theirmicroscopic sources is inherently challenging. Here, we bridge this gap by pairing an extensive rock magneticand geochemical dataset with statistical learning techniques for the first time. Using a Random Forest regressortrained on geochemical data, we accurately predict the growth of fine-grained magnetite in an undeformed lateEdiacaran section of remagnetized carbonate rocks from Paraguay. Our modeling results identify the K/Al ratio alongside K and Sr contents—as key predictors of this remagnetization mechanism. Notably, clay mineralogyanalyses further link the K/Al ratio to enhanced clay authigenesis (illitization) driven by K-feldspar dissolutionand albitization—processes that also release iron. Together, these findings indicate that remagnetization occurredvia authigenic magnetite formation under isochemical diagenesis—without the involvement of external fluids.This novel application of statistical learning to uncover the geochemical drivers of chemical remagnetizationsprovides a robust framework to investigate and understand these events. It could also open new avenues for theirdirect dating, thereby significantly enriching the global paleomagnetic record.