From catastrophic collapse to multi-phase deposition: flow transformation, seafloor interaction and triggered eruption following a volcanic-island landslide

The current understanding of tsunamis generated by volcanic-island landslides is reliant on numerical models benchmarked against reconstructions of past events. As the largest historical event with timed tsunami observations, the 1888 sector collapse of Ritter Island, Papua New Guinea provides an outstanding opportunity to better understand the linked process of landslide emplacement and tsunami generation. Here, we use a combination of geophysical imaging, bathymetric mapping, seafloor observations and sampling to demonstrate that the Ritter landslide deposits are spatially and stratigraphically heterogeneous, reflecting a complex evolution of mass-flow processes. The primary landslide mass was dominated by well-bedded scoriaceous deposits, which rapidly disintegrated to form an erosive volcaniclastic flow that incised the substrate over much of its pathway. The major proportion of this initial flow is inferred to have been deposited up to 80 km from Ritter. The initial flow was followed by secondary failure of seafloor sediment, over 40 km from Ritter. The most distal part of the 1888 deposit has parallel internal boundaries, suggesting that multiple discrete units were deposited by a series of mass-flow processes initiated by the primary collapse. The last of these flows was derived from a submarine eruption triggered by the collapse. This syn-collapse eruption deposit is compositionally distinct from pre- and post-collapse eruptive products, suggesting that the collapse immediately destabilised the underlying magma reservoir. Subsequent eruptions have been fed by a modified plumbing system, constructing a submarine volcanic cone within the collapse scar through at least six post-collapse eruptions. Our results show that the initial tsunami-generating landslide at Ritter generated a stratigraphically complex set of deposits with a total volume that is several times larger than the initial failure. Given the potential for such complexity, there is no simple relationship between the volume of the tsunamigenic phase of a volcanic-island landslide and the final deposit volume, and deposit area or run-out cannot be used to infer primary landslide magnitude. The tsunamigenic potential of prehistoric sector-collapse deposits cannot, therefore, be assessed simply from surface mapping, but requires internal geophysical imaging and direct sampling to reconstruct the event.