Euler deconvolution of potential field data
|This tutorial originally appeared as a featured article in The Leading Edge in April 2014 — see issue.|
In this tutorial (second article in the TLE series) we'll talk about a widely used method of interpretation for potential-field data called Euler deconvolution. Our goal is to demonstrate its usefulness and, most importantly, call attention to some pitfalls encountered in the interpretation of the results. The code and synthetic data required to reproduce our results and figures can be found in the accompanying IPython notebooks at figshare or GitHub. The notebooks also expand the analysis presented in this article. We encourage you to download the data and try it on your software of choice. For this tutorial we'll use the implementation in the open-source Python package Fatiando a Terra.
Euler deconvolution was first developed by Thompson (1982) and later extended by Reid et al. (1990). Since then, it has been adapted and improved upon by Keating (1998), Mushayandebvu et al. (2004), and many others. This popularity is largely due to its great simplicity of implementation and use, making it the tool of choice for a quick initial interpretation.
In many cases, maps of gravity and magnetic data (and transformations thereof) provide good constraints on the horizontal location of an anomaly source. Euler deconvolution adds an extra dimension to the interpretation. It estimates a set of (x, y, z) points that, ideally, fall inside the source of the anomaly. Euler deconvolution requires the x-, y-, and z-derivatives of the data and a parameter called the structural index (SI). The SI is an integer number that is related to the homogeneity of the potential field and varies for different fields and source types (Stavrev and Reid, 2007). For example, in the case of total field magnetic anomaly data, a sphere is represented by an SI of 3, whereas a dike is represented by an SI of 1. There are methods that can estimate the SI and we refer the reader to Barbosa et al. (1999) and Melo et al. (2013) for more information.
An example with synthetic data
The best way to fully grasp the usefulness and limitations of any geophysical method is to apply it in a controlled setting using synthetic data. To make things more interesting, we'll apply the Euler deconvolution on data generated by a model that differs slightly from ideal sources (Figure 1).
First, we need to calculate the 3 directional derivatives of the data. For gridded data, this can be done with the Fourier transform using module
fourier from Fatiando a Terra. Assuming that the data has been loaded into Numpy arrays
z with the respective x-, y-, z-coordinates of each data point and the array
data with our magnetic data (see the accompanying notebook):
from fatiando.gravmag import fourier from fatiando import utils # Need data in SI units data_si = utils.nt2si(data) # shape = (ny, nx) number of points in y and x dx = fourier.derivx(x, y, data_si, shape) dy = fourier.derivy(x, y, data_si, shape) dz = fourier.derivz(x, y, data_si, shape)
Now that we have calculated the derivatives, we have to create a solver to run the deconvolution.
Fatiando provides the
Classic solver which uses the whole dataset to estimate a single point (source location).
The usual strategy for multiple sources, introduced by Reid et al. (1990), is to use a moving window over the data. A point is estimated for each window using the data that fall inside it, producing a large number of solutions. The challenge then becomes separating the good solutions from the spurious ones.
FitzGerald et al. (2004) provide an overview of different selection criteria.
Fatiando provides the
MovingWindow solver that can run a given Euler solver on a moving window (see the accompanying notebook for more details on the selection criterion used):
from fatiando.gravmag.euler import Classic, MovingWindow classic = Classic(x, y, z, data_si, dx, dy, dz, struct_index) solver = MovingWindow(classic, windows=(50, 50), size=(1000, 1000)) solver.fit()
solver.fit() runs the deconvolution using the specified 50 x 50 windows of size 1 km. The estimated points are stored in the
For comparison, we ran the deconvolution using various different combinations of structural index and window size. Figure 2 shows scatter plots of the solutions (the typical way Euler solutions are presented). Notice how changing the window size influences the spread of the solutions and can affect our interpretation. For example, for a 1 km window and structural index 3, we cannot tell apart the two model bodies in the left. Furthermore, increasing the structural index will increase the depths of the solutions.
We can view these solutions in 3D using the Fatiando module
myv, a wrapper for the Mayavi visualization software (http://code.enthought.com/projects/mayavi):
from fatiando.vis import myv myv.figure() myv.points(solver.estimate_) myv.axes(myv.outline(model_bounds)) myv.show()
This will create something similar to Figure 3. Notice that the Euler solutions do fall approximately inside the sources. However, it is curious how the solutions make arch-like structures that have no relation to the actual forms of the sources. In fact, Silva et al. (2001) have shown that Euler solutions are biased and that the bias depends on the moving-window scheme and selection criterion used.
Last few words of caution
We repeat here for emphasis: Euler deconvolution solutions do not represent the three-dimensional outline of the sources. They indicate, at most, an approximate source location. It is crucial to remember that Euler solutions are subject to non-uniqueness and bias just like any other geophysical inverse problem. Tests using synthetic data are indispensable to assert the plausibility of our interpretations.
- Thompson, D. T. (1982), EULDPH: A new technique for making computer-assisted depth estimates from magnetic data, Geophysics, 47, 31–37, http://dx.doi.org/10.1190/1.1441278
- Reid, A. B., J. M. Allsop, H. Granser, A. J. Millett, and I. W. Somerton (1990), Magnetic interpretation in three dimensions using Euler deconvolution, Geophysics, 55, 80–91, http://dx.doi.org/10.1190/1.1442774
- Keating, P. B. (1998), Weighted Euler deconvolution of gravity data, Geophysics, 63, 1595–1603 http://dx.doi.org/10.1190/1.1444456
- Mushayandebvu, M. F., V. Lesur, A. B. Reid, and J. D. Fairhead (2004), Grid Euler deconvolution with constraints for 2D structures, Geophysics, 69, 489–496, http://dx.doi.org/10.1190/1.1707069
- Stavrev, P., and A. Reid (2007), Degrees of homogeneity of potential fields and structural indices of Euler deconvolution, Geophysics, 72, L1-L12, http://dx.doi.org/10.1190/1.2400010
- Barbosa, V. C. F., J. B. C. Silva, and W. E. Medeiros (1999), Stability analysis and improvement of structural index estimation in Euler deconvolution, Geophysics, 64, 48–60, http://dx.doi.org/10.1190/1.1444529
- Melo, F. F., V. C. F. Barbosa, L. Uieda, V. C. Oliveira Jr, and J. B. C. Silva (2013), Estimating the nature and the horizontal and vertical positions of 3D magnetic sources using Euler deconvolution, Geophysics, 78, J87-J98, http://dx.doi.org/10.1190/geo2012-0515.1
- FitzGerald, D., A. Reid, and P. McInerney (2004), New discrimination techniques for Euler deconvolution, Computers & Geosciences, 30, 461–469, http://dx.doi.org/10.1016/j.cageo.2004.03.006
- Silva, J. B. C., V. C. F. Barbosa, and W. E. Medeiros (2001), Scattering, symmetry, and bias analysis of source-position estimates in Euler deconvolution and its practical implications, Geophysics, 66, 1149–1156, http://dx.doi.org/10.1190/1.1487062
- IPython notebook that generates the synthetic data - Uses the forward modeling in Fatiando a Terra to create the synthetic data.
- IPython notebook for Euler deconvolution - Notebook that runs the Euler deconvolution on the synthetic data and produces the figures. Has a more detailed explanation of the method and generates a few more results.
- Tutorials repository - The GitHub repository for the Geophysical tutorials section of TLE has the notebooks from all tutorials published so far.
- GitHub repository - Has the text of the tutorial, IPython notebook (source code), synthetic data, and figures.
- figshare repository - A persistent archive for the synthetic data. Use this to download the data and for citation purposes.
- Fatiando a Terra - The open-source Python package used to produce the data and results in this tutorial.
- Corresponding author: Leonardo Uieda,