Jump to Search Jump to Navigation Jump to Content

Depth estimation of positive mass anomalies from FTG gravity data

One advantage of full tensor gradient (FTG) gravity surveys over conventional Gz surveys is that it allows small shallow masses to be differentiated from deep large masses. We use FTG data to estimate the depths of mass anomalies.

We have developed an algorithm that uses FTG data to estimate the potential 3D locations of mass anomalies that have a higher density than the surrounding rock body. The videos below illustrate the performance on the algorithm on a synthetic dataset where various mass anomalies are modelled over a field FTG survey. The use of synthetic data in these examples allows us to examine the performance of the algorithm under various levels of added noise. We are currently exploring the application of this algorithm to other types of potential field datasets, and adapting it to locate bodies of lower density than the surrounding rock.

We envisage that the output of this method could be used for various applications, such as:

  • targeting specific locations for further exploration, e.g. drilling;
  • providing an initial state for inversion;
  • creating geological maps;
  • producing height-fields onto which gridded data or other visualisations can be texturemapped.

This web page contains movies generated by a prototypes of the algorithm. You will need a current browser that supports HTML5 and a good broadband connection to view them.

The dataset used in generating the videos below contains forward-modelled synthetic bodies overlaid on a field FTG survey. The line data was gridded before being provided as input into the algorithm. There are five anomalies of various densities, shapes and depths, as shown in the following Gz image:

Label Object shape Relative density Depth range
A Block +0.2g/cc 500–900m
B Sheet +0.15g/cc 0–1000m
C Sphere +0.3g/cc 50–350m
D Cone -0.4g/cc 150–350m
E Tunnel -2.7g/cc 3–7m

The following videos have been generated from FTG components from the synthetic dataset, each independently corrupted with Gaussian noise. No noise removal process has been attempted, e.g. the tensor property that the tensor has zero trace is not enforced. These results may therefore be improved upon.  The videos show an overhead view stepping through different depths of the Earth; the depth shown in each frame is shown within the video. Bright points indicate the presence of anomalies as embedded in the FTG data. This video presents results for running the algorithm on the synthetic dataset with Gaussian noise of 1E added independently to each FTG component. The image is aligned as per the above Gz image.


(If the above video doesn't play, try downloading the video as an MP4 or OGG file and playing it on your computer.)

Note that the linear feature B appears dimmer than other features. This is an artefact of the algorithm, which detects the feature's presence along the entire length of the feature. For more localised mass anomalies, such as the mass sources A and C, the intensity is concentrated at a single location, hence the increase in brightness.

Our method is not yet able to directly detect anomalies of lower density than the surrounding rock mass, however, anomaly D is indirectly visible by a "hole" in the output around its corresponding position within the 3D model.

This second video demonstrates adding Gaussian noise of 2E to each FTG component: 

(If the above video doesn't play, try downloading the video as an MP4 or OGG and playing it on your computer.)

Gaussian noise of 5E was added to each component used in generating this video: 

(If the above video doesn't play, try downloading the video as an MP4 or OGG and playing it on your computer.)

The above videos demonstrate that the algorithm detects the depths of the anomalies robustly in the presence of noise. The following video has positioned the above videos side-by-side to highlight the similarities in performance at different levels of noise. 

(If the above video doesn't play, try downloading the video as an MP4 or OGG and playing it on your computer.)

Note that the videos look very similar, demonstrating that this algorithm performs well with various levels of noise. We have only provided results for synthetic datasets here as it is difficult to validate the algorithm's performance on field surveys.






Paul Johnston (UWA Physics)
Chris Wijns


UWA (Research Development Award 2012)