<p>Submarine canyons transport large volumes of sediment from continental margins to deep-ocean basins, primarily through erosive gravity flows originating in canyon heads. The source characteristics of coseismic landslides initiating in shallow upper canyon zones remain poorly constrained due to the prevalence of coarse resolution (&gt; 10&#xa0;m) or single capture bathymetric datasets, which bias mapping toward large, aggregated scars, obscure diagnostic morphology, and can lead to misidentification of failure types and inaccurate volume estimates. The 2016 Mw 7.8 Kaikōura earthquake generated widespread shallow submarine landslides in the upper Kaikōura Canyon, New Zealand, providing a rare pre- and post-event dataset. For the first time in a canyon head setting, we applied tectonic deformation corrections to the post-earthquake digital elevation model (DEM) and classified landslides from high-resolution morphology. We mapped 853 landslides across a 3.5 km² study area; most of which occurred as small, shallow debris slides within post-glacial sediment overlying bedrock. Individual failures have a mean area of 2250 m², mean volume of 9910 m³, and a median failure depth of 4&#xa0;m, with &lt; 4% deep-seated. This dataset provides a detailed characterization of coseismic landslides in canyon heads and shows that high-resolution, multitemporal bathymetry can resolve failures 3–7 orders of magnitude smaller than those reported in inventories derived from coarser or single-capture datasets. These advances directly support hazard applications in shallow coastal environments, including tsunami modeling, seabed‑infrastructure risk evaluation, and planning for cable routes and coastal assets.</p>