You are on page 1of 4

On the use of different methods for estimating magnetic depth

Xiong Li* Fugro-LCT Inc., Houston, Texas


Summary This presentation explains an experience of programming and/or testing the following different magnetic depth estimate techniques: the manual Peters half-slope and Bean ratio methods, the modern (semi-) automatic Naudy, Werner, Euler and SPITM (local wavenumber) methods. In all but the original SPITM method, the moving window concept is critical for reducing the working time possibly from months to hours. The modern methods are superior to the manual methods even if we leave the moving window concept behind. The depth estimate is often inaccurate mainly because the observed data is imperfect, and the real sources are different from the idealized. Different methods produce results with different accuracies because they depend differently on those two factors. This work will explore assumptions behind different methods, and demonstrate using examples that a better solution can be obtained only when a proper method is selected according to the data quality and the nature of ones particular geological problem. Role of Depth Estimation A complete quantitative interpretation of potential filed data is to estimate three types of information about sources of geological interest: the depth, dimension and contrast in physical property. Such an interpretation suffers from the inherent ambiguity. It is impossible to obtain all three types of information simultaneously without other a prior information. In many applications, we are often interested in depth more than dimension and physical property contrast. Many different techniques have been thus developed, over half a century, to estimate the magnetic depth. These techniques are independent of the susceptibility contrast and work for different simplified source geometries (dimensions). Depths estimated by a technique can be used as the final solution in some extreme situations: the anomaly is well isolated and the noise is insignificant or well removed. Estimated depths may also provide a good starting point for a genuine structural interpretation using, e.g., an interactive modeling or a constrained inversion. Different Techniques and Assumptions The very common assumption behind any techniques is that the magnetic data is perfect. Observed data contains noise. It is always required that noise be well removed from signal. Since such a removal can never be perfect, the accuracy of a technique is strongly dependent on its treatment for noise. The techniques investigated in this work interpret profile data. They all require that in theory, the profile be perpendicular to the strike of 2D (two-dimensional) bodies, and the length along strike be infinite. Otherwise, the estimated depth should be multiplied by the sine of the actual angle between the profile and the strike, which is called the strike direction correction. Relatively speaking, the strike length correction is much more difficult. A simple formula doesnt exist. However, the depth estimate can be directly accepted, when the length of the body is five times the width or more. The slope method is one of the earliest magnetic depth estimate techniques. Peters half-slope method (Peters, 1949) estimates depth by finding the horizontal distance between two parallel lines that pass through the maximum and minimum of an anomaly and have a slope equal to one half of the maximum horizontal gradient of the anomaly. It can work on the cylinder and sphere models. In practice, it is applied widely to a 2D tabular body (dike) that has vertical sides, great depth extent, and is with uniform and vertical magnetization. The depth to the top of the body is proportional to the horizontal distance. The proportionality factor varies from 1/1.2 for very thin bodies to 1/2 for very thick bodies. The Bean ratio method (Bean, 1966) has improved the Peters method by introducing a regional slope and thus removing the effect of a local regional field. The slope method is called a manual method because it works on a single anomaly each time and the extrema need to be selected manully. The Naudy method (Naudy, 1971) is one of the earliest (semi-) automatic techniques and still in use today. It scans a profile, end-to-end, forming the so-called moving window concept. Through a folding technique (i.e., to split the field curve into symmetric and anti-symmetric components) it first locates anomaly centers and then estimates depths by matching the symmetric component to responses due to a given, vertically magnetized, twodimensional model. The estimate is done via a look-up table technique. The accuracy is closely associated with these tables. The method works on either the total magnetic anomaly or its vertical gradient. It assumes the source to be one of the three: dike, horizontal plate and contact. The Werner method (e.g., Ku and Sharp, 1983) was originally designed to solve the dipping thin dike problem. This simplified geometry leads to linearization of a

Different magnetic depth estimators


Naudy and Werner methods, an advantage of the SPITM method is that it avoids the use of the troublesome moving window concept. The depth is estimated from the reciprocal of the local wavenumber. The very disadvantage is that it needs the second-order derivatives of the magnetic anomaly field. This method has been generalized by introducing the second-order local wavenumber and the concept of a multimodel wavenumber. The term multi-model applies because this quantity gives a depth estimate independent of the 2D source geometry (a horizontal cylinder, thin dike, contact, thick dike and finite step). However, the second-order local wavenumber requires calculation of third-order derivatives of the magnetic field. Manual versus Automatic Methods The primary goal of automation is to speed up the depth estimation, for example, to reduce the processing time, of a project with tens of thousands of line kilometers, from months to hours. Some one may still prefer the Peters or Bean ratio method to the Naudy, Werner, Euler, or SPITM method. This is probably only because machine can never replace human. The concept of moving window is the key to realize the automation. Automatic methods are superior to manual methods even if we leave the moving window behind. In other words, like a manual method, we apply the Naudy, Werner, Euler or SPITM method to a selected individual anomaly. This can be seen from the following four aspects. First, a manual method uses only two (in the Peters method) or four (in the Bean ratio method) characteristic points of an anomaly curve to determine depth but an automatic method can use as many points as one wants (at least in theory). The more information the method uses, the more accurate result the method will generate potentially. Second, a manual method is hardly capable of dealing with noise in data but an automatic method may be. In particular, an automatic method involving a least-squares inversion can greatly reduce the effect of random noise. Third, a manual method is dependent of the direction of magnetization, and often requires a reduction-to-pole process. On the contrary, the automatic methods arent and dont. Fourth, due to the magnetization dependence, a manual method may place incorrectly the horizontal position of a source. However, most automatic methods estimate these positions accurately. Determination of Dip and Susceptibility The automatic methods have been extended, by different groups, to determine the dip and susceptibility. In fact, one can estimate the dip and susceptibility only when the

complex, non-linear, magnetic inverse problem. The solution of a linear system can be then used to estimate the horizontal position, depth, product of susceptibility and thickness of the dike, and dipping angle of the dike. When the interference in the magnetic anomaly is expressed in the form of a quadratic polynomial, the coefficients of polynomial as unknown parameters can be estimated simultaneously in solving the system of equations. The Werner method has been extended to solve the sloping contact problem. A contact can be considered to consist of a large number of parallel thin dikes. In theory, the horizontal gradient of the total field caused by a sloping contact is equivalent to the total field caused by a dipping thin dike. In other words, we solve the thin dike problem by working on the total field anomaly, and the contact problem on the horizontal gradient of the total field. The Werner and Euler methods are often termed Werner deconvolution and Euler deconvolution although neither is a deconvolution in the classical signal processing sense. The Euler method is probably the most-popularly studied and used method of determining the horizontal location and depth of magnetic sources. The very reason may be that the method can work on both profile and gridded data. Correspondingly, we have 2D and 3D Euler methods. Even so, the sources assumed and used in 2D and 3D methods are not different. The Euler method uses Eulers homogeneity equation to construct a system of linear equations, and then determines through a least-squares inversion one time for one window the (vertical and horizontal) position of a single source for a given source geometry. It requires calculation of the horizontal and vertical derivatives of the magnetic field if they are not observed. Thompson (1982) called the fall-off rate (i.e., the negative of the degree of homogeneity) the structural index (SI). Eulers homogeneity equation is valid for bodies of arbitrary shape. In practice, the Euler method assumes idealized 2D structures such as contact, thin sheet (dike), vertical or horizontal cylinder, and 3D sphere. For such geometries, the SI is an integer. Giving a source geometry means assigning an SI. It makes no big sense to assume the SI to be not an integer (e.g., 0.5 or 1.3). First, it can only indicate that the assumed body is between this and that. Second, in theory there is not a constant, non-integer SI at all. A body with a non-integer SI has actually a distance between the observation and the body. The SPITM (local wavenumber) method is a technique using an extension of the complex analytical signal to estimate magnetic depths. The original SPITM method (Thurston and Smith, 1997) works for two models: a dipping thin dike and a sloping contact. Compared to the
2 ) ) #10'  '#&  $#! "!     ( %       

Different magnetic depth estimators

source location is known and the source geometry (SI) is known or correctly assumed. In the case of a contact, one can then solve for the dip and susceptibility contrast. In the case of a thin sheet (dike), the dip and the susceptibilitythickness product can be estimated. In the case of a horizontal cylinder, it is possible to determine the product of its cross-section area and the susceptibility. Some Difficulties with Automatic Methods The depth estimate is inaccurate or even fails because of many different factors. Two important ones are the noise effect and the source model assumption. An automatic method is further impaired by a choice of the size of the moving window. The moving window The original SPITM method doesnt involve this concept. All the other automatic methods require that there be only one source within one window. The multi-source Werner or Euler method overcomes this limitation but generates new difficulty. The n-source Werner method involves a solution of nth-order non-linear equations. The n-source Euler method requires calculation of nth-order derivatives of the magnetic field. For a single anomaly, a smaller window results in a more scattered distribution of solutions, and a larger window produces a tighter grouping. If the window is large enough, a cluster may become a single point eventually. For a multi-source problem, an optimum window size is small enough to see only a single anomaly yet large enough to cover sufficient variations in slopes or curvatures in the magnetic field. In practice, sources are located at different depths and their resulting anomalies have different wavelengths. The interference makes a correct choice of the window size difficult. One way to reduce clustering may be to ignore the moving window concept, pick, separate and then work on a single anomaly each time, with the cost of a significant increase in time. Choosing a model As explained above, each method works for a few simplified models. In the Werner method, for the thin dike model the total field must be used; for the contact the horizontal gradient of the total field must be used. Figure 1 shows a combination model and its Werner solutions. Evidently, when the horizontal gradient is used, the depth estimate of the thin dike is too shallow although the depth estimates of the faults are very accurate. On the contrary, when the total field is used, the depth estimates of the faults are too deep and even their horizontal positions are also incorrect although the depth estimate of the thin dike is very accurate. The original SPITM method also works on the thin dike and contact models. Thurston and Smith (1997) clearly

explained that a priori information and typically a judgment of an interpreter are required to determine which one of the two models is appropriate. In practice, the following thumb of rule may help make a correct choice. Assume a thin dike source if there is a single TGR (the magnitude of the total gradient) maximum and two HGR (the magnitude of the horizontal gradient) maximum. Assume a contact if there is a single TGR maximum and a single HGR maximum.

(a)

(b)
Figure 1. A synthetic model consisting of a thin dike and faults, and the final depths estimated by the Werner method. (a) the horizontal gradient solution and (b) the total field solution.

Figure 2. Comparison of magnetic anomalies from a broad dike and a narrow lense (thin horizontal plate).

As another example, the Naudy method produces three depth estimates, at one run, corresponding to three models: the dike, horizontal plate and contact. Figure 2 compares the magnetic anomalies due to a broad dike and a narrow horizontal plate. The central portions of the anomalies are

Different magnetic depth estimators

almost identical although their depths are very different. The key to distinguish them lies in the outer parts, where the thin plate develops a negative effect. However, this is exactly the part of the curve mostly subject to interference from neighboring features. In practice and from one profile alone, it is almost impossible to choose between a deep thin plate and a shallow broad dike model. Again, the experience of an interpreter, who is familiar with magnetic responses from various geological situations, becomes invaluable. Besides, two helpful rules may be as follows. First, the anomalies frequently exhibit typical patterns that dictate the choice of the model. For instance, intrasedimentary volcanic flows may be easily identified, and the interpreter should then choose a thin plate model. Linear features very often correspond to dikes or faults. Second, the criterion of consistency should be widely applied. Each value must be in reasonable agreement with those found on adjacent lines, e.g., those values obtained from strong anomalies almost certainly related to dikes. In the Euler method, the analysis is done for a given SI. When an incorrect SI is chosen a structure will be poorly incorrect. A too low index gives a too small depth, and a too high index results in a too large depth. The source geometry (SI) and depth may be defined simultaneously in some exceptional cases: a perfect anomaly caused by an idealized source body and isolated from any noise and interference. In general, such a simultaneous determination is impossible although different approaches have been proposed, e.g., the Euler method of the magnitude of the analytical signal, or of the nth-order derivative of the observed field instead of the field itself. One may interpret that the use of nth-order derivatives removes a background or interference field of the nth-order polynomial. However, for real, arbitrarilythe depth and SI are really dependent on one another. In particular, the approaches are impractical because of the noise effect. Noise effects A method produces a correct depth estimate only when (1) anomaly is precisely observed; (2) derivatives (if required and not observed) are precisely calculated; (3) residual anomaly is precisely isolated. Observed data contains noise. It makes the derivative calculation and anomaly separation inaccurate. The Naudy method can work on either the total magnetic anomaly or its vertical derivative. Tests show that the Naudy method for the vertical gradient gives more accurate depth estimates where data is accurate, but is generally unsatisfactory where there is strong noise in data. Noise
 '   00#! &) 0  0'#! '%% $!  ( 0#( ' ) 2 ( (  (  ' 2 % !  # 1! 12#)10   "#  ( )0' ( %  2 2 (  ' 2  #) 2

must be well removed in order to compute derivatives accurately. Different algorithms for denoising and derivative calculation have been tested. It is found that none can calculate second-order derivatives and higher of real data accurately enough for depth estimation. Depth or depth estimate is very sensitive to the magnitude of derivatives. However, any denoising or low-pass filtering has difficult to recover the magnitude correctly. Conclusions The magnetic depth estimate is strongly dependent on the data quality and source geometry assumption. The Naudy and Werner methods may not require calculation of any derivatives and will potentially produce better depth estimates when the data is heavily noise-contaminated. However, users need to make a right decision about the source geometry, e.g., according to geology. Unless the data quality is excellent, the estimates given by the SPITM and Euler methods involving second- and third-order derivatives are unreliable even if very accurate algorithms are used to calculate these derivatives. Acknowledgements I thank Fugro-LCT Inc for supporting this work and allowing for publication. SPITM is a trademark of Fugro Airborne Surveys. References Bean, R. J., 1966, A rapid graphical solution for the aeromagnetic anomaly of the two-dimensional tabular body: Geophysics, 31, 963-970. Ku, C. C., and Sharp, J. A., 1983, Werner method for automated magnetic interpretation and its refinement using Marquardt inverse modeling: Geophysics, 48, 754-774. Naudy, H., 1971, Automatic determination of depth on aeromagnetic profile: Geophysics, 36, 717-722. Peters, L. J., 1949, The direct approach to magnetic interpretation and its practical application: Geophysics, 14, 290-320. Thompson, D. T., 1982, EULDPH - A new technique for making computer-assisted depth estimates from magnetic data: Geophysics, 47, 31-37. Thurston, J. B., and Smith, R. S., 1997, Automatic conversion of magnetic data to depth, dip, and susceptibility contrast using the SPITM method: Geophysics, 62, 807-813.

You might also like