Short Note. An application for removing cultural noise from aeromagnetic data

Transcription

1 GEOPHYSICS, VOL. 66, NO. 1 (JANUARY-FEBRUARY 2001); P , 8 FIGS., 1 TABLE. Short Note An application for removing cultural noise from aeromagnetic data Stefan Muszala,Paul L. Stoffa, and L. A. Lawver INTRODUCTION A high-resolution aeromagnetic survey collected over the North Slope of Alaska by World Geoscience was acquired in response to a need for highly detailed data in an area where traditional geophysical techniques are expensive and prohibitive (McConnell, 1995) (Figure 1). These data were recently released to the University of Texas Institute for Geophysics and provide a unique opportunity to investigate the problem of cultural noise suppression in aeromagnetic data. The data contain isolated magnetic anomalies that are presumably from the many drill platforms and their accompanying cultural objects such as buildings and pipe repositories (Figure 2). We present a new, automated method to reduce the amplitude of these cultural anomalies without affecting the magnetic signal from the surrounding geology. The importance of removing cultural noise is important to both processing and interpreting magnetic data. The location and subsequent removal of cultural noise can improve the calculated differences between survey and tie lines, for example, resulting in a more accurately leveled data set. Furthermore, many data sets can be more rigorously interpreted with the absence of cultural noise. A difficulty arises, however, because the cultural objects are randomly located and oriented as well as oddly shaped. Ravat (1996) cites the example of a chemical waste site where steel drums are buried at various depths and have various shapes. In situations such as these, linear inverse modeling only works approximately (Ravat, 1996) and band-pass filtering will not succeed because it will remove naturally occurring magnetic anomalies that happen to have wavelengths comparable to the magnetic anomalies from the cultural objects. In our data, filtering out wavelengths <3500 m (500 fiducials; 1 fiducial is equal to the nominal sample rate, or 7m)reduces the amplitudes of the cultural noise but also affects the mag- netic data away from the noise (Figures 3a and 4a). By applying this filter, we defeat the purpose of the high-resolution aeromagnetic survey by removing the short-wavelength signal that is of interest. Applying a spike rejection filter (such as available with the Geosoft processing system), in which spikes having a width of 350 m (50 fiducials) and amplitudes of >20 nt were removed, proved more satisfactory (Figures 3b and 4b). The results show, however, that anomalies are only trimmed back to the threshold of 20 nt. Lowering this threshold will have the undesirable effect of removing any other anomalies (from real geology or otherwise) that happen to have a width <350 m. The problem then arises whether to use full 3-D modeling methods, which are usually computationally cumbersome. This is where an equivalent source technique could be useful (Dampney, 1969). The idea of the equivalent source is that a magnetic dipole can be used to represent a real geologic or cultural object given that the observer is more than a certain distance away from the object. This distance is that at which the inhomogeneties of the observed object are averaged out and varies according to the scale of the object being viewed (Blakely, 1996). A buried water tank may look like a dipole at the height of an aeromagnetic survey, but it will probably not look like a dipole at the height of a ground survey. The main advantage to using an equivalent source approach instead of a 3-D model is that it is computationally much faster and easier to implement and does not require detailed knowledge of the causative sources. It also seems to be more accurate than a 3-D model, at least when used to model 50-gal drums for an environmental application (Ravat, 1996). We hope to locate the isolated cultural objects; find the attributes of their equivalent source by applying the nonlinear inversion technique, very fast simulated annealing (VFSA) (Ingber, 1989, 1993); and then remove this signal from the Manuscript received by the Editor October 13, 1998; revised manuscript received June 19, Formerly University of Texas at Austin, Institute for Geophysics, 4412 Spicewood Springs Rd., Bldg. 600, Austin, Texas ; presently, Fugro Airborne Surveys, Greoschke Rd., Taxiway A, Bldg. 1A, Suite 100, Houston, Texas stefan@houston.wgc.com.au. Univ. of Texas at Austin, Institute for Geophysics, 4412 Spicewood Springs Rd., Bldg. 600, Austin, TX pauls@utig.ig.utexas.edu; lawver@utig.ig.utexas.edu. c 2001 Society of Exploration Geophysicists. All rights reserved. 213

2 214 Muszala et al. magnetic data. Nonlinear inversion such as VFSA involves finding a correlation between real and synthetic data by iteratively varying model parameters until a suitable fit is found. The initial model parameters are chosen based on a temperaturedependent Cauchy distribution within a predefined range. The declination, for example, was allowed to range between 0 and 360.The synthetic data are produced as part of the forward problem where the forward model (one dipole, in this case) is used to make measurements that can be compared directly to the real data. APPLICATION TO SYNTHETIC DATA FIG. 1. Base map showing the extent of the World Geoscience aeromagnetic data from the North Slope of Alaska (large dashed polygon). Inset map shows regional location. Before applying this technique to the real aeromagnetic data, a synthetic dipole model was formulated to test how well the proposed inversion scheme works. The program used was a modification of code from Sen and Stoffa (1995) and magnetic dipole modeling subroutines from Blakely (1996). The tests involved modeling one dipole and then multiple dipoles with and without random noise. We used the normalized correlation FIG. 2. Magnetic profiles of the North Slope study area. The magnetic data has been detrended. Inset map (not to scale) shows the location of wells, drill platforms, and other cultural objects. This shows the correspondence between noise in the aeromagnetic data and the cultural features. Map projection is WGS-84, UTM zone 6N.

3 Cultural Noise in Aeromagnetic Data 215 and a total error defined as (Dobs D syn ) 2 C = (Dobs + D syn ) 2 e = 1.0 C, (1a) (1b) where D obs is the observed data and D syn is the model data. The total error, e, derived from the correlation of the synthetic and observed data, is a measure of the degree of similarity of the two data sets. Using equation (1b), an error of 0.0 implies no correlation between the data sets, while an error of 1.0 implies perfect correlation. The first test of this method was to construct a model space of one dipole, run the VFSA to find the dipole parameters, and then evaluate the correlations obtained. Table 1 shows the results of this test run at six different iterations. The correlation by the th iteration, 0.99, was excellent, and all of the dipole attributes were effectively obtained. Figure 5 shows the evolution of the correlation coefficient and the X, Z position, magnetization as well as dipole and field inclination and declination. Looking at this plot, there is a trade-off between trying to obtain the best error and computation time. If computation time were a concern, this VFSA run could have been stopped well before iteration 7500 without significant loss of accuracy. From here a linear inversion technique could have been used to obtain the final correlation. Figure 6 graphically shows the progression of the VFSA program and the initial success of the technique. This VFSA trial used 15 moves per temperature and 1000 temperatures for a total of iterations. The second test case involved modeling five dipoles, first with no random noise (Figure 7a). Two dipoles were placed very FIG. 3. Single profiles of survey line (A) The original magnetic data (black) overlain by a 3500-m low-pass filter result (gray). Although the amplitude of the cultural noise was reduced, magnetic data in areas where no cultural objects existed were also smoothed. (B) The original magnetic data (black) overlain by the results from a 350-m-wide, 20-nT threshhold spike-removal filter (gray). The filter has introduced high-frequency noise where amplitudes have been reduced, and the filter will never be able to work on negative amplitudes. (C) VFSA results (gray) with the original magnetic data plotted in black. The VFSA has reduced the amplitudes and wavelengths of the two cultural noise spikes near fiducials 6800 and 9501, respectively. It has introduced high-frequency noise near fiducial FIG. 4. Single profiles of survey line (A) Original magnetic data (black) overlain by the 3500-m low-pass filter result (gray). Similarly to line , the amplitude of the cultural noise was reduced, but magnetic data in areas where no cultural objects existed were oversmoothed. (B) Original magnetic data (black) overlain by the results from a 350-m-wide, 20-nT threshhold spike-removal filter (gray). The filter has introduced high-frequency noise where amplitudes have been reduced. (C) VFSA results (gray) with the original magnetic data (black). The VFSA has reduced the amplitude and wavelength of the cultural noise between near fiducial 5250, although it also has introduced some high-frequency noise.

4 216 Muszala et al. close together so as to appear as one dipole in an attempt to trick the VFSA (located at 161 distance units in Figure 7a). The program ran successfully, and a final correlation of 0.99 was obtained. Although all of the dipoles were successfully removed from the grid, the two dipoles placed close to each other were modeled with one dipole. This shows that one strength of the equivalent source technique lies in allowing for a wide range of model parameter values and that the exact solution is not needed to achieve excellent results. The third and final test case involved adding random noise to the synthetic line. Since each of the four dipoles produced magnetic fields of different amplitudes, the percent of noise ranged from 12% to 31% (Figure 7b). After applying the VFSA, a 0.74 correlation was obtained. These last two test trials were achieved with 1000 temperatures, using 15 moves per temperature for a total of iterations. APPLICATION TO REAL DATA Next, the VFSA and equivalent source technique were applied to the real data. We applied the technique to each individual survey line, allowing five passes per line with one dipole being modeled per pass. Break points based on correlation and the line data range were added in the code to increase efficiency. Based on our own tests, applying this technique to line data was much more efficient than applying it to gridded data, effectively allowing us to apply the technique to the entire survey as opposed to only a small portion given the same amount of processing time. Finally, line-based data would not contain any biases that gridding algorithms would introduce to the data. The technique was successful in removing much of the cultural noise while maintaining signal from the underlying geology. Lines and (Figures 3c and 4c) achieved Table 1. Attribute evolution for the single-dipole VFSA test case. Selected iterations correspond to the profiles in Figure 5. X Z (depth) Field Field (arbitrary (arbitrary Magnetization Inclination Declination inclination declination Iteration # units) units) (A/m) ( ) ( ) ( ) ( ) Correlation Original dipole FIG. 5. Synthetic data plot of iteration number versus correlation and model parameters for one dipole. There is no significant change in the model parameter values after the 6700th and 1400th iterations in the correlation value. Stopping the VFSA run near the 1400th iteration would save 93% of the computing time; stopping the program at the 6700th iteration would save approximately 45% of the time. Parameters are scaled to better show variation. Final values are in parentheses. Magnetization/30, fdec/10, finc/10, inc/10, dec/10, correlation 10. See Table 1 for final unscaled values.

5 Cultural Noise in Aeromagnetic Data 217 correlations of 0.67 and 0.93, respectively. Line was modeled by four dipoles (four passes), while line was modeled with one dipole (one pass). The entire aeromagnetic survey is shown after the application of VFSA in Figure 8. Amplitudes associated with cultural objects were successfully reduced. The VFSA and equivalent source technique suppresses cultural noise without unduly affecting the magnetic signals that arise from real geologic features. The technique provides superior culture suppression to low-pass filtering and produced results comparable to the spike-rejection filter. Although the VFSA introduces short-wavelength noise similar to the spikerejection filter, the strength of the technique lies in its flexibility to use different forward models. Spike-rejection filters do not have this flexibility, nor can they operate on cultural noise that has a response more complicated than that of a dipole. The removal of high-amplitude anomalies from drill platforms is important for interpretation since a more realistic picture of the earth s magnetic field can be obtained. Once the cultural signal is removed, automatic depth determina- tion techniques such as Euler deconvolution (Thompson, 1982; Reid et al., 1990) can be applied to these decultured data with more accurate results. CONCLUSIONS The combination of an equivalent source technique and VFSA yields a promising new and automated method to remove cultural noise in magnetic data. The magnetic signatures from drill platforms and other cultural artifacts have been drastically reduced. This technique has advantages over band-pass filtering because it does not affect the magnetic signal resulting from real geology. It has advantages over spike finding filters in that it is controlled by a model that can be changed based on information about the cultural artifact and it is not controlled by arbitrary wavelength and amplitude thresholds. Furthermore, the spike-finding filter will never remove negative amplitudes. A weakness of the technique is that it introduces high-frequency noise in areas where the signal from the equivalent source does not match the real data very well. FIG. 6. Synthetic data plots showing the evolution of the VFSA run for one dipole (see Table 1). The original dipole response is shown in black, while the best-fit model is plotted with a dashed light gray line. The difference is plotted in dark gray. (A) Iteration 1, the initial model; correlation = (B) Iteration 150, correlation = (C) Iteration 50, correlation = (D) Iteration 750, correlation = (E) Iteration 1050, correlation = (F) Iteration , correlation = Comparing (F) to (A) shows that the inversion was successful.

6 218 Muszala et al. The VFSA method has the potential to accurately locate many different forms of cultural artifacts, such as finding the number of buried drums in a waste-contaminant site. The technique can be extended to use exact forward modeling as opposed to just the equivalent source method. All computer runs were performed on a Sun SPARCstation- 20 with a 50-MHz processor. Typical run times for the singledipole model were well under 1 minute, and those for the multiple dipole tests (with and without noise) were just under 2 minutes. A typical run time for an application to this real data was 11 hours ( 3 minutes/survey line). However, we allowed the VFSA programs to run well beyond the point where significant change was occurring in the correlation coefficient. Stopping the programs before this point would save an appreciable amount of computational time. ACKNOWLEDGMENTS We thank J. D. Phillips of World Geoscience Inc. for use of the North Slope aeromagnetic data and also Terrance McConnell, formerly of the World Geoscience Corp. Thanks to Tim Ryherd of the USGS for the digital well location data and Odile Delabeaujarderie of the NSF. This work was supported by NSF grant OPP REFERENCES FIG. 7. (A) VFSA test to synthetic data with no noise and five dipoles. The location of two dipoles that were placed very close to each other is near distance unit 161. The final correlation for the synthetic data with no noise was 0.99, with four dipoles being used to model the data. (B) The synthetic test with added random noise. Percent noise was 22% with respect to the dipole located at 22, 18% with respect to the dipoles located at 61, 31% with respect to the dipole at 102, and 12% with respect to the dipoles found near unit 161. For this case a final correlation of 0.74 was achieved and four dipoles were used to model the data. The method s strength lies in its ability to solve for a wide range of model parameter perturbations without having to solve for the exact model. Blakely, R. J., 1995, Potential theory in gravity and magnetic applications: Cambridge Univ. Press. Dampney, C. N. G., 1969, The equivalent source technique: Geophysics, 34, Ingber, L., 1989, Very fast simulated re-annealing: Math. Comp. Modeling, 12, No. 8, , Simulated annealing: Practice versus theory: Math. Comp. Modeling, 18, No. 11, McConnell, Terence J., 1995, North Slope, Alaska: A case study of sedimentary aeromagnetics (SAM): 65th Ann. Internat. Mtg., Soc. Expl. Geophys., Expanded Abstracts, Ravat, R., 1996, Magnetic properties of unrusted steel drums from laboratory and field-magnetic measurements: Geophysics, 61, Reid, A. B., Allsop, J. M., Granser, H., Millett, A. J., and Somerton, I. W., 1990, Magnetic interpretation in three dimensions using Euler deconvolution: Geophysics, 45, Sen, M. K., and Stoffa, P. L., 1995, Global optimization methods in geophysical inversion: Elsevier Science Publ. Co., Inc.. Thompson, D. T., 1982, EULDPH: A new technique for making computer-assisted depth estimates from magnetic data: Geophysics, 47,

7 Cultural Noise in Aeromagnetic Data 219 FIG. 8. Decultured magnetic profiles of the North Slope study area. Like the magnetic profiles in Figure 2, these data have been detrended. Inset map shows the location of wells, drill platforms, and other cultural objects. The VFSA and equivalent source technique have successfully removed much of the cultural noise signal from the data without affecting the signal from natural sources. Map projection is WGS-84, UTM zone 6N. Inset map is not to scale.