Subduction zones seen by GOCE gravity gradients

In this study, the GOCE (Gravity field and steady state Ocean Circulation Explorer) gradiometry data were used to study geologic structures and mass variations within the lithosphere in areas of known subduction zones. The advantage of gravity gradiometry over other gravity methods is that gradients are extremely sensitive to localized density contrasts within regional geological settings, which makes it ideally suited for detecting subduction zones.

Second order gravity gradients of disturbing potential were extracted from global geopotential model, the fifth release GOCE model ‘EGM_TIM_RL05’. In order to remove the signal which mainly corresponds to the gravity signal of the lower mantle, long wavelength part of the gravity signal was removed up to degree and order 60.

Because the areas with notable topography differences coincide with subduction zones, topography correction was also performed. Few pattern recognition methods were tested on all 6 gravity gradient tensor components represented as global scale maps with resolution of 100km (corresponds to the resolution of the GOCE satellite data).

By adjusting pattern recognition methods’ features and optimizing various input patterns, the best method was applied. That is a combination of methods based on SURF (Speeded Up Robust Features) and MSER (Maximally Stable Extremal Regions) algorithms provided in MATLAB’s Computer Vision System Toolbox.

Based on 6 gravity gradient components, the global gradient anomaly maps were produced and used as starting point for analysis based on image processing. On obtained maps, locations of known subduction zones were represented with characteristic elongated patterns and cross-sections. Cross sections of well-known subduction zones were used as input patterns for pattern recognition method on global maps.

The search for discrete point correspondences between these images was divided into three main steps: Interest point detection, interest point description and matching between images. Resulting routine compares vertical gravity gradient anomaly signal in the areas with known subduction zones with all locations on the Earth (covered by GOCE gravity gradients).

Searching, comparing and detecting the compatible signal lead to correct detection of all known subduction zones but also gave indications for locations of unknown subduction. Apart from subduction zones, certain geological features were detected and studied. The method proved its advantages and should be easily adjusted and conducted on other datasets with similar representations.