Reproducible Workflow for "Target Limitations in Machine-Learning Reconstruction of Coral-Derived Seawater δ18O from Paired Sr/Ca–δ18O Archives"

This repository provides the reproducible code and workflow for the study “Target Limitations in Machine-Learning Reconstruction of Coral-Derived Seawater δ18O from Paired Sr/Ca–δ18O Archives.” The pipeline processes paired CoralHydro2k Sr/Ca–δ18O records, evaluates PAGES seawater δ18O target availability, constructs EN4 sea-surface-salinity-derived monthly δ18Osw pseudo-targets, and benchmarks machine-learning residual correction against traditional paired-proxy estimates and simple baselines using leave-one-record-out validation. The workflow also generates the manuscript figures, supplementary tables, regression diagnostics, and extended-window sensitivity test for 1950–2016. The archive is designed to support transparent reproduction of the manuscript results and to document the target-data limitations, amplitude mismatch, and residual-cancellation behavior that constrain current supervised ML reconstruction of coral-derived monthly seawater δ18O.