A Machine Learning Approach to Predicted Bathymetry
Abstract
Recent and on-going efforts have shown how machine learning (ML) techniques, incorporating more, and more disparate data than can be interpreted manually, can predict seafloor properties, with uncertainty, where they have not been measured directly. We examine here a ML approach to predicted bathymetry. Our approach employs a paradigm of global bathymetry as an integral component of global geology. From a marine geology and geophysics perspective the bathymetry is the thickness of one layer in an ensemble of layers that inter-relate to varying extents vertically and geospatially. The nature of the multidimensional relationships in these layers between bathymetry, gravity, magnetic field, age, and many other global measures is typically geospatially dependent and non-linear. The advantage of using ML is that these relationships need not be stated explicitly, nor do they need to be approximated with a transfer function - the machine learns them via the data. Fundamentally, ML operates by brute-force searching for multidimensional correlations between desired, but sparsely known data values (in this case water depth), and a multitude of (geologic) predictors. Predictors include quantities known extensively such as remotely sensed measurements (i.e. gravity and magnetics), distance from spreading ridge, trench etc., (and spatial statistics based on these quantities). Estimating bathymetry from an approximate transfer function is inherently model, as well as data limited - complex relationships are explicitly ruled out. The ML is a purely data-driven approach, so only the extent and quality of the available observations limit prediction accuracy. This allows for a system in which new data, of a wide variety of types, can be quickly and easily assimilated into updated bathymetry predictions with quantitative posterior uncertainties.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2017
- Bibcode:
- 2017AGUFMOS31C1421W
- Keywords:
-
- 1219 Gravity anomalies and Earth structure;
- GEODESY AND GRAVITY;
- 3002 Continental shelf and slope processes;
- MARINE GEOLOGY AND GEOPHYSICS;
- 3022 Marine sediments: processes and transport;
- MARINE GEOLOGY AND GEOPHYSICS;
- 3045 Seafloor morphology;
- geology;
- and geophysics;
- MARINE GEOLOGY AND GEOPHYSICS