A semi-automated framework for global detection of previously undocumented submarine calderas

Submarine calderas are far less documented than their subaerial counterparts, yet recent events highlight their potential for impactful geohazards. Their detection has been limited by the scarcity of high-resolution bathymetry data. Here we apply a hybrid framework to the low-resolution but globally available General-Bathymetric-Chart-of-the-Ocean, combining an automated algorithm for detecting depressions on the seafloor, expert (user-based) classification, and Principal Component Analysis validation to systematically identify previously undocumented submarine calderas. We identify 78 calderas spanning a wide range of water depths (to 5,600 m), diameters (to 20 km), and tectonic settings (divergent, convergent, intraplate), of which only five were previously known. A sensitivity test shows that under-detection of known calderas is largely associated with low circularity indexes ( < 50%). Despite this limitation, our dataset fills a major observational gap and provides a reproducible, upgradeable framework for submarine volcano characterisation, underscoring the need to incorporate submarine calderas into future global volcanic assessments. The detection of 73 previously undocumented submarine calderas worldwide was achieved through a hybrid framework integrating automated bathymetric depression analysis, expert classification, and statistical validation.