From manganese mineral evolution history to atmospheric oxygen reconstruction

Abstract The evolutionary record of redox-sensitive Mn minerals encodes critical information about Earth’s oxygenation history. By building a global Mn mineral dataset (144,200 entries across 25 feature dimensions), we developed a URD (Unequal-size feature matrix, Re-coupling relationship, Disaccord labels) deep-learning model to reconstruct continuous atmospheric oxygen level (pO2) changes over 4.0 billion years. Our results provide robust mineralogical evidence linking the timing and tempo of oxygenation to planetary-scale tectonics and biosphere evolution. Specially, the reconstruction reveals two distinct oxygenation modes: a protracted and gradual increase during the Paleoproterozoic-Mesoproterozoic, reflected in the moderately progressive evolution of Mn mineral assemblages; and a more rapid rise preceding and following the Neoproterozoic, coincided with supercontinent breakup and convergence, respectively−a pattern potentially driven by tectonic modulation of Mn supply and demand. This study introduces a mineral-informatic framework for decoding complex, high-dimensional mineral records, offering a transformative approach for systematically interrogating Earth’s long-term evolution.