Electromagnetic methods, particularly magnetotelluric (MT) sounding, are often used to detect fluids in the crust and upper mantle. Recently, MT studies on land have focused on the deep structure of the San Andreas fault (SAF) in central California. One of these studies found a region of high conductivity, interpreted as a local abundance of pore fluid, southwest of the fault at depths corresponding to the lower crust and upper mantle. The study also identified a change in the connection of this deep conductor to the root of the SAF as their measurements moved along the strike of the fault. It was proposed that a change in access of fluids to the deep extension of the fault, as evidenced by the modeled resistivity of the region, is responsible for the observed change in behavior of both earthquakes and non-volcanic tremor along the central Californian segment of the SAF. As an effort to extend knowledge of the resistivity structure into the offshore environment of central California, as well as to better constrain the interpretations made onshore, the Marine Electromagnetic Research Group at the Scripps Institution of Oceanography collected marine MT and controlled-source electromagnetic (CSEM) data there. Our single profile of data extends an existing profile of land data. In this dissertation I interpret both profiles of MT data, land and marine, jointly. Inversions of the amphibious dataset produce a resistivity model that is similar to previous results. However, upon closer inspection it was found that the marine data and the land data solicit antithetical structures. I suspected this to be caused by distortion from bathymetry that varies in three dimensions (3D), which is a problem because our 2D models assume that the earth is uniform in the horizontal dimension perpendicular to our profile. Few numerical methods exist to calculate the effects of 3D bathymetry in the necessary detail, so I improved upon an established thin-sheet technique to allow for higher resolution models. With this method I estimated how the effects of 3D bathymetry along our profile deviate from those predicted by an analogous 2D model. The effects are significant relative to the data errors used in inversion. As a test, I created synthetic data from both 2D and 3D bathymetric models, and invert them with a 2D inversion algorithm. Surprisingly, the inversion of 2D data did not recover the simple model used to create the synthetic data. Anomalous structures were introduced that weakly resembled those found in the models derived from actual data. Inversion of the 3D data made matters worse; the same artifacts found with 2D data became more extreme and their resemblance to the features in the results from actual data was enhanced. I concluded that the results from applying the established 2D inversion scheme on the actual data cannot be trusted, as is likely the case for the work on the same data that was previously published by others. The thin-sheet modeling suggests that future interpretations must at least take into account the known 3D bathymetric variations that occur off our profile