The Study On The Automatic Interpretation Methods Of Potential Field (Gravity&Magnetic) And Its Gradients

As the increasing of the exploration efficiency, exploration precision, data parameters andvolume of the geophysical equipment, many traditional automatic interpretation methods ofpotential fields cannot satisfy the requirement of current geophysical exploration because ofthe disadvantages of low precision, complicated computation and human interference, so thefast and high-precision automatic interpretation methods have been get hotpot issues. In thispaper, I present many new automatic methods that interpret potential field and gradient data,mainly include small sub-domain filtering, edge detection filter, Euler deconvolution method,analytic signal method, local wavenumber method, property inversion and correlationimaging method et al, and get more satisfactory results.Measured gravity and magnetic anomalies are the comprehensive reflection of theuneven distribution of all the sources from the surface to the deep, and we should use the lowpass filter to process the anomaly to get the anomaly of the target source. Spectrum analysistechnique is a commonly used method in the separation of gravity and magnetic anomalies,but this kind of method usually blurs the edges of anomaly. Small sub-domain filter is a filedseparation method that can keep the edges of the anomalies, but the standard sub-domainfilter doesn’t fully consider the cases that the calculation point is in the regional field andremove weak anomalies. I present optimized small sub-domain filter, which improve thedistribution of sub-domain and selection criterion, and can accomplish separation taskeffectively. The new method is demonstrated on theoretical model, and the outputs of theoptimized small sub-domain filter are more reasonable and precise compared to the previousmethods, which improved the shortcomings of the traditional method effectively. At last, Iapply it to real data, which can finish the field separation, and keep the edges of theanomalies more effectively compared to the traditional method.Edge detection is an indispensable task in the interpretation of potential field data, whichcan show the edges of the stratums and the horizontal locations of the sources clearly. Manyprevious edge detection filters only can identify the edges of shallow bodies, and therecognized edges of deep sources are very blurred. In order to improve this problem, I presentusing the correlation coefficient of horizontal derivative and vertical derivative to recognizethe edges of the sources. I demonstrated this method on synthetic potential field data, whichcan display the edges of the shallow and deep bodies simultaneously, and the recognizededges are clearer than the edges recognized by the other filters. I apply the correlationcoefficient method to real data, and obtain the distribution of the faults and the edges of thestructures, which are consistent with the real geology. The computation of vertical derivativeswill increase the effect of noise and disturb the stability of results, so I propose thenormalized total horizontal derivative (NTHD) method to recognize the edges of the sources.The theoretical tests show that the NTHD can recognize the edges of the shallow and deepsources simultaneously, and are more stable compared to the other methods, is insensitive to noise. I also applied this method to real gravity data, and obtain the distribution feature offaults. However, the edges recognized by the correlation coefficient and NTHD methods arediffuse, which is hard to divide the locations of the sources easily based on the results. Inorder to get clearer edges, I present enhanced balancing filters, which are the combinations ofdifferent order derivatives, and introduce a stable algorithm about the vertical derivativesbased on the Laplace equation to low the effect of noise. The model tests show that theenhanced balancing filters can display the edges more precisely and clearly, and canrecognize weak anomalies more effectively. I apply this method to real data, and the locationsand trend of the faults recognized by the enhanced balancing filters are consistent with realgeology, and discover more subtle details. The enhanced balancing filters need thecomputation of third order derivatives, and the high-order derivatives are unstable when thespace of the data is big. So I present improved local phase (ILP) filter, which use thenon-linear combinations of first order and second-order horizontal derivatives to recognizethe edges of the geologic bodies. The ILP filter is demonstrated on synthetic potential fielddata, which can display the edges of shallow and deep sources clearly, and is more stable thanthe enhanced balancing filter. I apply it to real data, and the results show the stratigraphicmarks and locations of the sources clearly. It found that the existed edge detection filters willoccur the unwanted edges when process the anomalies generated by the positive and negativedensity bodies. In this paper, I present optimized edge detection filter, which add a constantvalue in the equation to remove the unwanted edges, and the model and real data test showthat the optimized edge detection filter is effective. I plug the high-resolution edge detectionfilters into the software platform-Geosoft, and use to implement and enhance the function ofprocessing of interpretation, and realize the transformation from the study to the applicationmodule, and accomplish the joint with the foreign superior software platform.As the increasing of the data volume, the geophysical interpreter would like to applyautomatic interpretation methods to interpret the data. However, the existing methods havethe shortcomings of lack of clarity, low precision and complicated computation. I improve thedisadvantages existed in Euler deconvolution, analytic signal method and local wavenumbermethod of automatic interpretation method, and make the inversion results more stable andprecise. Extended (gradient) Euler convolution has the advantage that is insensitive toregional field, and is applied more extensively compared to standard Euler deconvolution.This method have two solution strategies, and the first way is to estimate the depth andstructural index simultaneously, and the second way is that first estimate the depth and thencompute the structural index. Now, the interpreters usually used the first way to estimate theparameters of the source. I derive the inversion equation of second solution way, andcompared the contrast of the inversion results computed by the first and second solution way,and the results show that the precision of the second way is higher. I applied the second wayto real gradient magnetic data, and obtain the distribution of ore bodies, and this research hasstrong directive significance for the application of extended Euler deconvolution method.Analytic signal is insensitive to magnetization direction, which is a widely used in theinterpretation of magnetic data. The existed analytic signal methods need the computation of third order derivatives to compute the location and structural index of source, and this willincrease the effect of noise and low the reliability of the results. I present three optimizedanalytic signal methods, and the first two methods require the computation of second-orderderivatives, and the third method only compute the first order derivatives, so the optimizedmethods can increase the accuracy and stability of the inversion results. I demonstrated theimproved methods on synthetic and real potential field data, and they can finish theinterpretation of magnetic data successfully, and the precision of them is higher than thesimilar analytic signal methods, especially the inversion results of the third method are verystable and accurate. Local wavenumber method is a usually used automatic method in recentyears, which mostly use the derivatives of local wavenumber to estimate the depth andstructural index of the source. In this paper, I present using the combinations of the differentlocal wavenumbers to estimate the source parameters, which can low the interference of noiseand increase the accuracy of the inversion results. The tests on synthetic and real magneticdata show that the precision of the proposed local wavenumber methods is higher. Otherwise,I derive another local wavenumber method that can estimate the depth and structural indexsimultaneously, which has advantage over the big volume data. I demonstrate the optimizedlocal wavenumber methods on synthetic and real magnetic data, and the precision of the newmethod is not lower than the other local wavenumber methods, and has good real applicationeffect. The composite application of above three automatic interpretation methods can finishthe interpretation of complicate data, and low the instability and risk of the results.In this paper, I present using the fast simulated annealing (FSA) method to accomplish theproperty inversion of potential field data, which can avoid the computation of big matrix, andincrease the efficiency. I use the mean square error (MSE) and the parameters change as theiteration stopping condition, and the new method can increase the accuracy of inversionresults. The theoretical tests show that this method can finish the inversion of density andsusceptibility correctly, and are consistent with real geologic models. I apply the proposedmethod to gravity anomaly of cavities in Liaoyuan, and obtain the locations of cavities, whichis consistent with the electricity results. Potential field correlation imaging can obtain thedistribution of the sources quickly based on the correlation coefficient between the syntheticanomaly generated by assumed source and measured anomaly. The existing correlationimaging method use the sphere to simulate the shape of the subsurface geologic bodies, whenthe shape of the real source is different from sphere, the inversion results will have big errors.In order to improve this problem, I present enhanced correlation imaging method, whichcomputes the correlation coefficients between the data generated by different kinds of sourcesand the real data, respectively, and the model that can make the correlation coefficients getmaximum value is consistent with the real geological model, and the location of themaximum value is corresponding to the location of source. The enhanced correlation imagingmethod can both estimate the location parameters and the structural index of the source. I usethe analytic signal to compute the correlation coefficients for the magnetic data, which isinsensitive to magnetization direction, and the computation equation is simpler. Idemonstrated this method on synthetic data, and this method can successfully finish the inversion of potential field data, and is more stable. At last, I apply this method to realmagnetic data of Shanghai, and obtain the distribution of unexploited body.Tensor measurement is a new geophysical exploration tool, which can provide the gradientreflection in different directions of the geologic body, and can delineate the feature of thegeologic bodies more correctly. But there are no systematic researches about the tensorinterpretation method. In this paper, I present tensor local wavenumber method and directionanalytic signal method to interpret full tensor gradient data. Tensor local wavenumber methoduses the matrix consist of tensor local wavenumber to estimate the location parameters of thesource. The feasibility of this method is illustrated by comparing with the conventional localwavenumber method, and can get more accurate results. I also apply this method to real datafrom American, and get the depth of the source, and the inversion results are coincident withthe results computed by the other methods. Direction analytic signal method uses the analyticsignals in the x, y and z directions to interpret magnetic tensor data, and use the horizontalanalytic signals to recognize the source of the edges. I demonstrated it on theoreticalexamples, and the results show that the direction analytic signal method can finish theinversion of magnetic tensor data effectively, and the results and recognized edges areinsensitive to magnetization direction. I also apply it to real magnetic tensor data, and obtainthe distribution of ore bodies. In order to complete high-precision gravity and magnetic dataprocessing and interpretation workflow of Geosoft, and plug the tensor interpretation methodto the software platform.