Sneaky Submarine Landslides, and how to Quantify them: A Case Study from the Mississippi River Delta Front Contrasting Geophysical and Machine Learning Techniques
Abstract
The highly publicized subsidence and decline of the Mississippi River Delta Front's (MRDF) subaerial section has recently precipitated studies of the subaqueous MRDF to assess whether it too is subsiding and regressing landward. These studies have largely focused on the area offshore the most active current distributary of the Mississippi River, Southwest Pass, during a decade (post-Hurricane Rita 2005-2014) of relatively quiescent Gulf of Mexico hurricane activity. Utilizing repeat swath bathymetric surveys, it was determined that submarine landslides not associated with major (category ≥ 3) passage are important drivers of downslope sediment transport on the MRDF. Volumetrically, sediment flux downslope without major hurricane influence is approximately half that during a given hurricane-influenced year (5.5 x 105 and 1.1 x 106 m3, respectively). This finding is notable and warrants comparison with other settings to assess the global impact on the source-to-sink budget of small but frequent landslides, but the resource-intensive repeat geophysical surveys required make it a prohibitive option at the margin and global scale. One option to quantify small-scale submarine slope failures while reducing required data acquisition is to utilize machine learning algorithms (MLAs) to intelligently estimate the occurrence and magnitude of submarine landslides based on correlated physical and geological parameters. Here, the MRDF volumetric changes described above are parsed into training and validation data, and physical and geological parameters associated with slope failure (such as porosity, steep slopes, high rates of sedimentation, and presence of gas in pore water) known from prior coring and seafloor mapping expeditions serve as potential predictive variables. The resulting submarine landslide spatial distribution and magnitude maps output by the MLAs are compared to those obtained through geophysical surveys, providing a proof of concept that machine learning can complement and expand the reach of previously acquired geophysical and geological data.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFMNH44B..03O
- Keywords:
-
- 3022 Marine sediments: processes and transport;
- MARINE GEOLOGY AND GEOPHYSICS;
- 3070 Submarine landslides;
- MARINE GEOLOGY AND GEOPHYSICS;
- 4304 Oceanic;
- NATURAL HAZARDS;
- 4219 Continental shelf and slope processes;
- OCEANOGRAPHY: GENERAL