Differential Impacts of Proxy Archives on Paleoclimate Reconstruction Performance

Abstract

Reconstructing past climates is challenging because pre-instrumental observations are geographically sparse and proxy records are noisy; data assimilation offers one approach to address these limitations. Although such reconstructions are now widespread, the extent to which different proxy archives determine their accuracy remains unclear. We address this gap using proxy withholding experiments to quantify the marginal impact of eight proxy archives—corals, tree rings, speleothems, documentary records, lake sediments, marine sediments, glaciers, and ground ice—on four reconstructed climate targets: the El Niño–Southern Oscillation, the Indian Ocean Dipole, global mean surface temperature, and global mean precipitation. We also assess their influence on spatial skill patterns and zonal mean skill for temperature and precipitation. Comparing performance across multiple priors shows that assimilation consistently improves skill, with the multi-model ensemble achieving the highest and most stable results by averaging out structural and mean-state biases in single models. Corals emerge as indispensable across all indices, while documentary records and speleothems add key hydroclimate information in monsoon-sensitive, data-sparse regions. Archive-specific fingerprints reveal that some proxy archives exert outsized leverage relative to their geographic coverage, underscoring the importance of proxy location. Together, these results highlight that targeted proxy selection, archive-specific forward modeling, and seasonally resolved assimilation represent promising directions for improving the accuracy and robustness of future hydroclimate reconstructions.