Inferring seafloor topography from gravity anomaly currently is the dominant method to obtain a global view of the oceans. This study deals with the methods of inferring this topography from gravity gradients, which is more sensitive to topography at short wavelengths than gravity anomaly. Two methods, one in the spectral domain and the other in the spatial domain, were developed and tested using vertical gravity gradients derived from satellite altimetry.

The spectral domain method is based on the linear approximation to the Parker's infinite series, which is for the convenience of inversion the admittance that relates topography and gravity gradient. Only 15--160 km wavelength topography was estimated from the vertical gravity gradient, for which the inversion was stable. The long wavelengths were obtained by low-pass filtering existing bathymetric depths, and the wavelengths shorter than 15 km were omitted. This method was tested in a 2° x 2° area in the West Pacific Ocean centered at (21° N, 157° E). There, the seafloor topography estimation has a root mean square error of +/-268 m.

Through a numerical test, it was found that the nonlinear terrain effect was not negligible in rugged areas. Algorithmic analysis through the coherency showed that estimation accuracy at high frequencies cannot be improved by refining the resolution of gravity gradients, due to the linear approximation.

To remove the linear approximation in the modeled relationship between gravity gradients and topography, the simulated annealing, a global optimization technique that can process nonlinear inverse problems, was used to estimate the seafloor topography parameters in a forward model by minimizing the difference between the observed and forward-computed vertical gravity gradients. Careful treatments like conducting truncation error analysis, and padding the vicinity of the study area with a known topography model, were required for successful estimation. A numerical test for the same study area generated an estimation with root mean square error of +/-236 m. This improves the results from the spectral domain method by 12%. Compared with the global topography model version 18 as released by the Scripps Institution of Oceanography, the estimation accuracy is improved by 22% over the study area.

The simulated annealing approach developed in this study may be used to update the global seafloor model, especially in rugged areas. Besides, this approach has no restrictions on data distribution, as required in Parker's infinite series model, thus enabling the use of data from an airborne gravity gradiometer, whose resolution is high but flight trajectory may be irregular. This method is developed under uniform density assumption. Therefore, its performance at places of complex sub-surface geology would be poor, in general.