Emilson Pereira Leite

Probabilistic neural networks applied to mineral potential mapping for platinum group elements in the Serra Leste region, Carajás Mineral Province, Brazil

This work presents an application of probabilistic neural networks to map the potential for platinum group elements (PGE) mineralization sites in the northeast portion of the Carajas Mineral Province (CMP), Brazilian Amazon. Geological and geophysical gamma-spectrometric and magnetic data were used to generate evidential maps to derive input feature vectors. Feature vectors representing known mineralized locations were used as training data. The networks were created based on the training dataset and the evidential maps were classified in terms of probabilities using these networks. We have produced mineral potential models that depict classes with high, moderate and low favorability for Au-PGE mineralization sites and a model with high and low favorability classes for Cr-PGE mineralization sites. The cut-off values for each class were selected as the inflexion points of the curves of favorability against cumulative percentage of the study area. These curves were also used to check for the efficiency of the models by plotting the favorability values at the training sites. Leave-one-out tests were applied to validate the models and the overall accuracy is 87.5%. For Au-PGE mineralization sites, the high favorability areas accounts for 0.57% of the study area and are comprised mainly within meta-pelites and meta-siltites. For Cr-PGE mineralization sites, the high favorability areas are much more restrict and accounts for only 0.17% of the study area, being associated chiefly with mafic and ultramafic rocks. These mineral potential maps can be used as reconnaissance guides for future detailed ground surveys of possible new PGE occurrences, which is of critical importance to shorten exploration time and costs in such densely forested Amazonian terrains.

Well log denoising and geological enhancement based on discrete wavelet transform and hybrid thresholding

Well logging is an important tool for the characterization of subsurface rocks, being commonly used in the study of reservoir geology. It is well known that signals obtained as responses from geological media contain noise that can affect their interpretation, and that wavelet transform is more suitable than the Fourier transform to denoise non-stationary signals, as the ones obtained from well logs. On the other hand, there are several parameters that must be considered when working with wavelet transform, such as the choice of the wavelet basis function (mother wavelet), the decomposition level and also the function and rules that “control” which and how the coefficients will be used for signal reconstruction. This study analyzes the process of denoising well log data by discrete wavelet transform. Since the well log data are usually used in lithological classification, we propose a method associated with the k-nearest neighbor classification algorithm to investigate how different combinations of parameters affect the output signals and its performance in the classification, thus making it a data driven process. We propose a new thresholding function that shows better results when compared with traditional ones. The potential of wavelet transform as a tool to aid geological interpretation is evidenced by the identification of important geological features of the Namorado Field, Campos Basin, Brazil.

TEXTNN-A MATLAB program for textural classification using neural networks

A new MATLAB code that provides tools to perform classification of textural images for applications in the geosciences is presented in this paper. The program, here coined as textural neural network (TEXTNN), comprises the computation of variogram maps in the frequency domain for specific lag distances in the neighborhood of a pixel. The result is then converted back to spatial domain, where directional or omni-directional semivariograms are extracted. Feature vectors are built with textural information composed of semivariance values at these lag distances and, moreover, with histogram measures of mean, standard deviation and weighted-rank fill ratio. This procedure is applied to a selected group of pixels or to all pixels in an image using a moving window. A feed-forward back-propagation neural network can then be designed and trained on feature vectors of predefined classes (training set). The training phase minimizes the mean-squared error on the training set. Additionally, at each iteration, the mean-squared error for every validation is assessed and a test set is evaluated. The program also calculates contingency matrices, global accuracy and kappa coefficient for the training, validation and test sets, allowing a quantitative appraisal of the predictive power of the neural network models. The interpreter is able to select the best model obtained from a k-fold cross-validation or to use a unique split-sample dataset for classification of all pixels in a given textural image. The performance of the algorithms and the end-user program were tested using synthetic images, orbital synthetic aperture radar (SAR) (RADARSAT) imagery for oil-seepage detection, and airborne, multi-polarized SAR imagery for geologic mapping, and the overall results are considered quite positive.

Independent component spectral analysis

AbstractSpectral decomposition techniques can break down the broadband seismic records into a series of frequency components that are useful for seismic interpretation and reservoir characterization. However, it is laborious and time-consuming to analyze and to interpret each seismic frequency volume taking all the usable seismic bandwidth. In this context, we propose a multivariate technique based on independent component analysis (ICA) with the goal of choosing the spectral components that best represent the whole seismic spectrum while keeping the main geological information. The ICA-based method goes beyond the Gaussian assumption and takes advantage of higher order statistics to find a new set of variables that are independent of each other. The independence between two components is a more general statistical concept than the noncorrelation and, in principle, allows the extraction of more significant information from the data. We have tested four different contrast functions to estimate the independ...

Maximum Autocorrelation Factors applied to electrofacies classification

A vast amount of data is obtained during the development of a petroleum field. Seismic data, well logs, core and production data, all contribute to a better reservoir characterization and modeling. Several methods of multivariate data analysis can be used to support its interpretation, helping in important tasks as the identification of lithological facies. The most used and widely known of those methods is Principal Component Analysis (PCA) which intends to reduce data dimension while keeping as much as possible of their variance. Data dimension reduction can also be performed with the method of Maximum Autocorrelation Factors (MAF) which seeks to keep the spatial autocorrelation in data. In this work both methods were applied to data from well logs of the Namorado field, testing their performances in the classification of electrofacies. Following data dimension reduction, supervised classification methods known as k-nearest neighbors (kNN) and weighted k-nearest neighbors (wk-NN) were applied, and the results obtained were compared by cross-validation. MAF showed to be more efficient than PCA in reducing data dimension, while keeping relevant information. The wk-NN performed a little better in classifying electrofacies than the usual k-NN. According to these results, the combination of MAF and wk-NN can be a valuable tool for classifying the facies of uncored wells from their logs.

3D acoustic impedance and porosity mapping from seismic inversion and neural networks

In this work we have applied a two-fold methodology to obtain porosity 3D maps from post-stack 3D seismic amplitude data and well log data. Firstly, acoustic impedance values are derived from seismic amplitudes by applying a L1-norm sparse-spike inversion algorithm in the time domain, followed by a recursive inversion performed in the frequency domain, where low-frequency impedance trends estimated at well-logs are used as constraints. Secondly, a feed-forward Neural Network (NN) is trained, validated and tested using effective porosity data observed at the well locations as input. The trained NN is applied for the whole reservoir volume to obtain a 3D effective porosity map. The results suggest that this simple workflow can be applied successfully specially in reservoirs for which the observed relationship between porosity and acoustic impedance is non-linear.

Mapas auto-organizáveis aplicados ao mapeamento do potencial mineral na região de serra leste, província mineral de Carajás, Pará

A Self-Organizing Map (SOM) was designed with the aim of integrating and searching for patterns in airborne geological and geophysical gamma-spectrometric and magnetic data of the Serra Leste region, Carajas Mineral Province. SOM is an unsupervised Artificial Neural Network method that performs a non-linear mapping from a high-dimensional data space to a 2-dimensional grid, whereas preserving the topological relations in the original data. The SOM grid can be efficiently used in an integrated visualization and understanding of the internal relationships in the data. The K-means algorithm is applied to the SOM grid to reduce the number of mapped patterns so as to facilitate interpretation. Unfolding of the clustered SOM grid associates each mapped pattern with the spatial position of each data point. The SOM reclassified map was compared with a classified map obtained with the Fuzzy C-means method for the same input data and with the same number of classes. The results show the potentiality of SOM in producing higher quality integrated maps to support mineral exploration.

Avaliação do uso de dados aerogeofísicos para mapeamento geológico e prospecção mineral em terrenos intemperizados: o exemplo de Serra Leste, província mineral de Carajás

A regiao Amazonica e caracterizada por floresta equatorial densa, espessa cobertura de solos e clima chuvoso. Este contexto torna os aerolevantamentos geofisicos ferramentas-chave para mapeamentos litologicos e estruturais e para revisao de mapas geologicos pre-existentes. Este artigo apresenta um estudo de aplicacao de dados gamaespectrometricos e magnetometricos na regiao de Serra Leste, Provincia Mineral de Carajas, proporcionando discussoes sobre as assinaturas geofisicas de litotipos especificos e estrategias para realce de alvos favoraveis a exploracao de depositos de Au-Pd-Pt, Cr-EGP, Cu-Mo-Au e Fe.

3-D Geological modelling: a siliciclastic reservoir case study from Campos Basin, Brazil

Reservoir static modelling plays a fundamental role in the evaluation phase of a petroleum field. Integrated modelling allows a better understanding of how the local geology and depositional systems are related through the distribution of facies and petrophysical properties within the reservoir. In this study, geological static models of the siliciclastic Carapebus Formation of Campos Basin were built using subsurface data. The applied methodology was divided into five phases: (1) establishment of a conceptual model, (2) building of a structural model, (3) generation of 100 realizations of lithofacies using sequential indicator simulation, (4) generation of 100 realizations of porosity and permeability using sequential Gaussian simulation, and (5) validation of models by targeting both statistical and geological consistency. The obtained models are consistent and honor the conditioning data. A lithofacies constraint is crucial to better characterize the petrophysical properties distribution of the reservoir. A Dykstra-Parsons coefficient of V=0.52 characterizes this reservoir as moderately homogeneous.

Facies recognition using a smoothing process through Fast Independent Component Analysis and Discrete Cosine Transform

Abstract We propose a preprocessing methodology for well-log geophysical data based on Fast Independent Component Analysis (FastICA) and Discrete Cosine Transform (DCT), in order to improve the success rate of the K-NN automatic classifier. The K-NN have been commonly applied to facies recognition in well-log geophysical data for hydrocarbon reservoir modeling and characterization. The preprocess was made in two different levels. In the first level, a FastICA based dimenstion reduction was applied, maintaining much of the information, and its results were classified; In second level, FastICA and DCT were applied in smoothing level, where the data points are modified, so individual points have their distance reduced, keeping just the primordial information. The results were compared to identify the best classification cases. We have applied the proposed methodology to well-log data from a petroleum field of Campos Basin, Brazil. Sonic, gamma-ray, density, neutron porosity and deep induction logs were preprocessed with FastICA and DCT, and the product was classified with K-NN. The success rates in recognition were calculated by appling the method to log intervals where core data were available. The results were compared to those of automatic recognition of the original well-log data set with and without the removal of high frequency noise. We conclude that the application of the proposed methodology significantly improves the success rate of facies recognition by K-NN.