Subsurface marine heatwaves (MHWs) represent hidden thermal extremes that can strongly impact marine ecosystems, particularly when they occur as large-scale, long-lasting events. Although they evolve across time, horizontal space, and depth, their global vertical structure and evolution remain poorly understood because of the sparsity of long-record subsurface observations and because subsurface variability has yet to be characterized within a globally consistent framework. Here we apply a four-dimensional event-tracking framework to global ocean temperature reanalysis data to identify and characterize more than 500 large-scale subsurface MHWs in the upper 1,000 m of the near-global ocean (60°N–60°S) from 1993–2020. Using a depth-rescaling approach and clustering analysis, we classify distinct vertical structure and evolution types. We show that large-scale subsurface MHW occurrences are highest in equatorial regions and lowest in the subtropics. Specifically, we capture the extensive subsurface expressions of major historical events, including the northeast Pacific Blob and Caribbean MHWs, which penetrated hundreds of metres below the surface, as well as the Ningaloo Niño, whose subsurface structure extended across the Indian and Pacific Oceans well beyond its traditionally defined surface footprints.  

We identify seven recurring vertical structure types, including mixed-layer, deep, thermocline (warm and cold), and submerged (warm, neutral and cold) MHWs. Five types are significantly associated with El Niño–Southern Oscillation phases, highlighting the role of large-scale climate variability in shaping subsurface MHWs. Persistence-centre analysis further shows that MHWs generally maintain their vertical structure for several months to years, with limited propagation across the mixed layer and thermocline, indicating strong constraints imposed by ocean stratification. Together, our findings show that subsurface MHWs and their evolution are jointly controlled by large-scale climate variability and local ocean stratification, providing a basis for their improved detection and multi-month to multi-year predictability.