Nowcasting submarine slope instability at local, margin, and global scales using machine learning
Jeffrey Obelcz, Warren T. Wood, Benjamin J. Phrampus, Taylor R. Lee
Received Date: 4th November 19
Submarine slope instability (SSI) is a broad term for events ranging from 100 km3 instantaneous open slope failures on continental margins to 0.001 km3 creeping mudflows on heavily sedimented river deltas. SSI events such as the 2018 Sunda Strait and 1929 Grand Banks submarine landslides extract high societal tolls, yet SSI predictive capability is limited. SSI observational studies are resource intensive and cover small areas, necessitating a method for estimating SSI geohazard potential where it is not measured. Here we use machine learning and a compilation of SSI observations and predictors to make the most comprehensive geospatial SSI predictions, or nowcasts, to date. Nowcasts on regional (100 km2) to global spatial scales identify SSI hotspots on glaciated and volcanic margins (global), offshore of lowstand depocenters (margin), and mudflow gullies on the Mississippi River Delta (regional). Global SSI nowcasts, normalized by total sediment deposited since the start of the Pleistocene (2.58 Ma), show that passive margins are on average composed of approximately twice as much SSI-derived sediment compared to active margins, affirming the prominent role seismic strengthening plays in millennial-scale slope stability. We also demonstrate the particular suitability of time-series bathymetric surveys for training SSI predictions in instability-prone areas; using machine learning prediction to interpolate between resurvey transects can decrease the necessary data volume required for hazard assessment by a factor of three.
Read in full at EarthArXiv.
This is an abstract of a preprint hosted on an independent third party site. It has not been peer reviewed but is currently under consideration at Nature Communications.