Comparing Surface and Subsurface Marine Heatwaves in Ocean Reanalyses

Abstract. Marine heatwaves (MHWs) are increasingly recognised as major drivers of ecological and socio‑economic disruption, yet their subsurface expression remains poorly constrained due to sparse in situ observations. Here, we characterise properties such as the intensity, duration, and heterogeneity of subsurface MHWs using the Copernicus Global Ocean Reanalysis Ensemble Product (GREP). MHWs are detected across multiple depths (0.5–200 m) using a consistent percentile-based thresholding, applied to three reanalysis members (CGLORSv7, GLORYS2v4, ORAS5) and evaluated against well‑documented historical events. All reanalyses successfully capture the timing and general evolution of major, well-known surface expressions of MHW events when compared against the ESA Climate Change Initiative SST dataset in given areas of interest, such as, for example, events of 2011 in Western Australia, of 2015 in the Northeast Pacific Blob, and of 2006 in the Mediterranean Sea. Here, it is also shown that anomalies frequently penetrate below the mixed layer, down to 100–150 m. On a global scale, there is a positive bias of about 15 days in the mean duration for all the events with respect to satellite SST data, consistent with the limited eddy-resolving capability of the eddy-permitting GREP reanalyses, the additional smoothing introduced by ensemble averaging and the difference between the different representative depths of SST in satellite observations and models. Global diagnostics reveal that MHWs extending into the subsurface are more persistent, spatially heterogeneous, and characterised by sharper onset and decline rates than shallow events, particularly in the tropics, western boundary currents, and the Southern Ocean. Mixed‑layer variability strongly modulates the vertical propagation of anomalies, underscoring the role of vertical mixing in shaping subsurface extremes. Our results highlight both the value and limitations of current reanalyses for subsurface MHW characterisation, emphasising the need for ensemble‑based approaches to quantify uncertainty.