Using "old" dogs for new tricks: Exploratory machine learning to predict submarine slope instability
Abstract
Machine learning algorithms (MLAs) are now highly accessible and easily deployable, and as such have been recently applied to a broad range of earth science problems. This study highlights one application: geospatially predicting the presence and magnitude of submarine slope instability through various MLAs, under the broader purvey of NRL's Global Predictive Seabed Model (GPSM).
The approach builds on other geospatial seabed prediction efforts through GPSM, and therefore follows a similar protocol. Various proxies of seabed instability, including difference of depth (DoD) between bathymetric surveys, morphological evidence of past seabed failure (such as evacuation scarps and downslope debris fans), and in situ recording of seabed movement are used as training and validation data. Various well-established MLAs, such as K-Nearest Neighbor, Random Decision Forest, and Support Vector Machine, are then used to geospatially predict the location and magnitude of submarine slope instability based on qualitatively correlated geological, physical, and morphological parameters, including depth, slope, grain size, clay fraction, shear strength, and significant wave height. Predictions of submarine slope instability is the focus on two disparate environments in geological parameter space: the subaqueous Mississippi River Delta Front (MRDF) and the US Atlantic Margin (USAM). Slope failures on the MRDF tend to present as mudflows in unconsolidated, organic-rich fine sediments, while USAM slope failures are typically "classic" submarine landslides wherein a clearly defined evacuation scarp and debris lobe is produced. GPSM successfully predict the location of slope instability in these dissimilar environments, but misfit between observed and predicted events indicates more data and/or predictors may yield better results. We will expand the approach presented here to other submarine slope failure-prone environments in pursuit of an eventual global-scale prediction of seafloor movement probability.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2018
- Bibcode:
- 2018AGUFMEP51E1866O
- Keywords:
-
- 1824 Geomorphology: general;
- HYDROLOGYDE: 1942 Machine learning;
- INFORMATICSDE: 4217 Coastal processes;
- OCEANOGRAPHY: GENERALDE: 4558 Sediment transport;
- OCEANOGRAPHY: PHYSICAL