UC San Diego

The ocean floor, more than 70% of the solid earth surface, remains mostly unmapped by conventional echo-sounding methods. Measurements from satellite altimetry precipitated maps of marine gravity with uniform global coverage which have improved markedly as repeated measurements have reduced noise. Such maps allow us to see the ocean floor that was previously hidden.

The structures revealed on the seafloor reflect processes of plate tectonics. We describe a new type of tectonic feature revealed in maps of marine vertical gravity gradient (VGG). So- called seesaw propagators reflect ridge propagation that reverses direction. SSPs are ubiquitous on seafloor that formed at half spreading rates between 10–40 mm/yr, their propagation directions do not appear to be correlated over large length scales, and they occur where the ridge offset is less than 30 km. This suggests the yield strength of the lithosphere at large MOR offsets prevents propagation. We develop a model framework based on a force balance wherein forces driving propagation must exceed the integrated shear strength of the offset zone. We apply this model framework to 4 major propagating ridges, 55 seesaw propagating ridges, and 69 transform faults. The model correctly predicts the migration of major propagating ridges and the stability of transform faults, but the results for SSPs are less accurate. This model framework simplifies deformation in the shear zone, but can possibly explain why non-transform deformation is preferred at short offsets.

Finally, we give a new method for predicting bathymetry from gravity using a machine learning approach. We design and train a neural network on a collection of 50 million depth soundings to predict bathymetry globally using gravity anomalies. Our final predicted depth model improves on the old predicted model rms by 16%, from 165 m to 138 m. Additionally, we recommend a strategy for partitioning depth sounding data such that the problem conforms with the assumption of data independence required by many machine learning algorithms.