Produção Científica
Artigo em Revista
Viscoelastic and viscoacoustic modelling using the Lie product formula We present efficient and accurate modelling of seismic wavefields in anelastic media.We use a firstorder viscoelastic wave equation based on the generalized Robertssonâ€™sformulation. In our work, viscoacoustic and viscoelastic wave equations use the standard linear solid mechanism. To numerically solve the firstorder wave equation, weemployed a scheme derived from the Lie product formula, where the time evolutionoperator of the analytic solution is written as a product of exponential matrices, andeach exponential matrix term is approximated by the Taylor series expansion. Theaccuracy of the proposed scheme is evaluated by comparison with the analytical solution for a homogeneous medium. We also present simulations of some geologicalmodels with different structural complexities, whose results confirm the accuracy ofthe proposed scheme and illustrate the attenuation effect on the seismic energy duringits propagation in the medium. Our results demonstrate that the numerical schemeemployed can be used to extrapolate wavefields stably for even larger time steps thanthose usually used by traditional finitedifference schemes. 

Artigo em Revista
Multiparameter vectoracoustic leastsquares reverse time migration The use of dualsensor acquisitions enabled different studies using the socalledvectoracoustic equations, which admit particle velocity (displacement or acceleration) information instead of solely the pressure wavefield. With cost far fromelastic formulations but comparable with the usually used secondorder acousticequation, previous works involving the use of vectoracoustic equations along withmulticomponent data have been applied to conventional reverse time migration andfullwaveform inversion, always emphasizing the benefits of using wavefields containing directivity information, which make the receiver ghosts interact constructivelywith the backpropagated reflected wavefield. Thus, generated results are superior tothose of conventional singlecomponent data imaging techniques, particularly withspatial subsampling of marine seismic data. To assess whether the effects of applyingthe vectoracoustic equations persist in a linearized inversion, we developed a multiparameter vectoracoustic leastsquares reverse time migration, inverting reflectivitiesassociated with velocity and density. To demonstrate the methodâ€™s performance, weapply it to twodimensional numerical examples and compare the results with thoseobtained by the conventional acoustic leastsquares reverse time migration. Theresults obtained by the vectoracoustic leastsquares reverse time migration methodare accurate for all inverted parameters and also deliver better convergence whencompared with the conventional leastsquares reverse time migration. 

Artigo em Revista
Characterizing seismic facies in a carbonate reservoir, using machine learning offshore Brazil. The authors propose an approach that uses machine learning to characterize carbonate facies in a wildcat (Gato do Mato) prospect in the Santos basin, offshore Brazil. We analyzed different seismic attributes and selected those that best responded to the seismic patterns identified in the study area as input for an unsupervised classification. The classification method used is the selfgrowing neural network (SGNN) technique that consists of the following steps: 1. Seismic pattern identification in seismic amplitude. The main patterns identified are buildup, debris, carbonate platform and bottom lake facies. 2. Generation and analysis of seismic attributes to characterize seismic patterns. We chose Eigen coherence, dipsteered enhancement, relief and relative acoustic impedance to help in seismic characterization. 3. We performed principal component analysis (PCA) of the attributes: amplitude filtered from dipsteered enhancement (seismicdriven structural filtering), Eigen coherence and relief. 4. Unsupervised seismic classification from the PCA of seismic attributes (item 3 above). Using this approach, we associated the classified seismic facies with the patterns identified in the amplitude data. The seismic facies allowed us to differentiate the carbonate platforms from the buildup facies. However, the classification encountered difficulties in identifying the patterns associated with lake bottom facies and the chaotic seismic pattern of debris facies. 

Artigo em Revista
Probabilistic Estimation of Seismically ThinLayer Thicknesses with Application to Evaporite Formations The identification of potassium (K) and magnesium (Mg) salts prior to the well drilling is a key factor to avoid washouts, closing pipes, fluid loss damage, and borehole collapse. The Bayesian classification combines the outcomes from statistical rock physics and seismic inversion, providing the spatial occurrence of the mostprobable salt types. It serves as a facies identifier of Mgâ€“Krich salts (bittern salts) before drilling. Nevertheless, the mostprobable classification is limited to the seismic resolution which may underestimate seismically thinlayer thicknesses. Along with the mostprobable facies, the Bayesian classification renders the facies probability volume. We demonstrate that the facies probability and faciesspecific total thickness highly correlate to each other even under the threshold of seismic resolution. Thus, we employ the bitternsalts probability volume to predict thinbed bitternsalts thickness in undrilled locations. To capture the variability of the seismic estimation, we resort to Monte Carloassisted simulations of wells that emulate the layering patterns of a sitespecific deposition environment. These simulations are crucial to assist the estimation of the joint probability density function between the facies volume and the total thickness. Therefore, given the facies probability, the joint probability density function enables us to derive the conditional expectation and percentiles of thinbed thicknesses. Furthermore, this paper proposes a method to quantify the negative influence of seismic noise in the estimation of thinbed thicknesses. The blind well confirms the consistency of this technique to unfold the uncertainty in the seismic predictability of thin layers. We argue that this procedure is extendable to other facies. 

Artigo em Revista
Mineralogy based classification of carbonate rocks using elastic parameters: A case study from Buzios Field Brazilian presalt carbonate reservoirs are highly heterogeneous. This feature is mostly justified by the nature of the original depositional system and subsequently diagenetic processes. Consequently, reserve estimates and production forecasting are under large uncertainties. In this geologic context, it is of great relevance to develop techniques that helps to obtain a detailed description on the spatial distribution of these different rocks. In doing so, it contributes to the understanding of presalt carbonate sedimentary deposits, providing inputs for more predictive reservoir models. Traditionally, these carbonates are grouped into three classes, from which only one exhibits reservoir properties. Using a dataset from Buzios Field, this work proposes a characterization of presalt carbonate reservoir rocks by grouping them in terms of their mineral composition. Taking advantage of rock physics concepts, we aim to potentialize the use of elastic parameters for multiple rock type discrimination. We explored several attempts for rock classification by using a Bayesian approach. Among all the tested propositions, a twostep workflow for five lithotypes classification, emerges as the most appropriate for the Buzios Field. In this scheme, three lithotypes represent goodquality reservoirs and the other two are lowporosity and Mgclayrich carbonates. The average rootmeansquare error of the most likely a posteriori rock proportions is around 8.4%, only approximately 1% higher than the conventional three lithotypes configuration. To support that, we compared different methodologies for Bayesian classification at welllog scale through acoustic impedance and compressional to shear velocity ratio. Potential applicability of the proposed methodology at field scale is reinforced by similar results achieved using welllogs filtered to the seismic bandwidth. 

Artigo em Revista
The influence of subseismicscale fracture interconnectivity on fluid flow in fracture corridors of the BrejĂµes carbonate karst system, Brazil The present study used a multitool approach to characterize fractures of several orders of magnitude in large fracture corridors, caves, and canyons to investigate their impact on fluid flow in carbonate units. The study area is the BrejĂµes carbonate karst system that is located in the Neoproterozoic Salitre Formation in the IrecĂŞ Basin, SĂŁo Francisco Craton, Brazil. The approach included satellite imagery, used for interpreting the regional structural context, Unmanned Aerial Vehicle (UAV) and groundbased Light Detection And Ranging (LiDAR) imagery, used for detailed structural interpretation. Regional interpretation revealed that fracture corridors, caves and canyons occur along a Nâ€“Soriented anticline hinge. An advanced stage of karstification caused fracture enlargement and intrabed dissolution, and the formation of caves and canyons. A river captured by the highly fractured zone along the anticline hinge played an important role as an erosive agent. Detailed characterization of fracture corridors comprised structural analysis, topological studies, persistence estimations, powerlaw fitting of fracture trace length distributions, and identification of network backbones. Our results indicate that fracture corridors comprise four subvertical fracture sets: Nâ€“S and EW and a conjugate pair, NNESSW and NWSE. Fractures observed in the caves show the same dominant directions. Fracture directions are consistent with a common origin associated with the anticline folding. Fracture traces range from 1.0 m to 300 m, comprising both subseismic (<50 m) and seismic scale fractures (>50 m). Networks have dominance of node terminations Y and X (notably Y), CB values higher than 1.8, high P20 and P21 persistence values, and highly interconnected backbones. Fracture network connectivity is associated with powerlaw exponents greater than 2.5 for the fracture trace distributions, indicating large influence of subseismicscale fractures on fluid flow. As the final result of folding and karstification, large volumes of secondary macroporosity were created, particularly in the zone of maximum fracture intensity around the hinge zone of the anticline. This scenario can be used to understand better oil reservoirs formed in similar structural controls in nearsurface conditions. 

Artigo em Revista
A constrained version for the stereology inverse problem: Honoring power law and persistences of the fracture traces exposed on arbitrary surfaces We present a stereological study in a cave setting that is part of a karstic carbonate system located in the SĂŁo Francisco Craton, Brazil. Using a Lagrangian approach, a constrained version of the nonlinear inverse problem of stereology is solved. Besides the classical demand of fitting the histogram of fracture traces measured on arbitrary exposed surface, it is imposed that the solution honors also measures of surface intensity (p21) and power law exponent obtained from fracture traces on the same exposed surfaces. Estimates of volumetric intensities (p32) of conjugate fracture pairs might be also imposed to be close values. The resulting cost functional is minimized using the Particle Swarm Optimization (PSO) method. The implemented version of PSO furnishes the best solution and a set of suboptimal quasisolutions, from which the solution uncertainty is evaluated. A key aspect of the implemented approach is that all terms composing the cost functional are normalized, obtaining as a result, robustness for the weighting parameters to changes in the input data. The resulting stochastically simulated discrete fracture networks honor all statistical observations but, in general, do not reproduce the positions of the observed fracture traces. The methodology is applied to synthetic and field data examples. All obtained solutions are stable and geologically reliable. 

Artigo em Revista
Well log analysis for lithology and fluid contacts in Rovuma Basin â€“ Mozambique: Application of cluster and discriminant analyses This study applies the cluster and discriminant analyses in geophysical well log data from the Rovuma sedimentary Basin  Mozambique. The main objective was to determine the lithological profile and fluid contacts in reservoirs. Well log data from five wells drilled on the same basin were used. For the discrimination, a reference well was chosen for training, and the obtained functions from it were then applied to the remaining wells. The classification process comprehended three main phases, namely, the separation of shale/nonshale layers along the entire logged section, separation of water/hydrocarbon within reservoirs and the separation of oil/gas within hydrocarbon bearing zones. The two methods, cluster analysis and discriminant analysis, were applied in parallel and the results are compared in each classification phase. The quality of reservoirs was also assessed by applying cutoffs in relation to shale content and effective porosity, delineating net reservoirs. In general, both methods converged to the same lithological model and fluidtypes in reservoirs. Gas has been indicated as the most predominant hydrocarbon in the basin. 

Artigo em Revista
Comparison between oil spill images and lookalikes: an evaluation of SARderived observations of the 2019 oil spill incident along Brazilian waters Three SARderived observations of dark surface patches along the Northeastern Brazilian coastline by the end of 2019 were misreported in the Brazilian media as oil spillrelated. Unfortunately, these observations were misled by false positives or lookalikes. Therefore, this paper aims to technically evaluate these lookalike classes by analyzing image attributes found to be helpful to the identification of ocean targets, including oil spills, rain cells, biofilms, and low wind conditions. We use image augmentation to extend our dataset size and create the probability density function curves. The processing includes image segmentation, optimal attribute extraction, and classification with random forest classifiers. Our results contrast with the opensource oil spill detection system and patch classifier methodology called â€śRIOSS.â€ť Analysis of the feature probability density functions based on optimal attributes is promising since we could capture most of the false positive targets in the three SARreported images in 2019. The only exception was the biofilm slick observed on October 28th, where the RIOSS mistakenly classified this organic patch as a low wind region with oil spots. This pitfall is acceptable at this project stage since we had only five biogenic film samples to train the algorithm. 

Artigo em Revista
Lowrank seismic data reconstruction and denoising by CUR matrix decompositions. Lowrank reconstruction methods assume that noiseless and complete seismic data can be represented as lowrank matrices or tensors. Therefore, denoising and recovery of missing traces require a reducedrank approximation of the data matrix/tensor. To calculate such approximation, we explore the CUR matrix decompositions, which use actual columns and rows of the data matrix, instead of the costly singular vectors derived from singular value decomposition. By allowing oversampling columns and rows, CUR decompositions obviate the need for the exact rank. We evaluate three different procedures for randomly selecting columns and rows to obtain the CUR. Once the lowrank approximation is estimated, data reconstruction is achieved by an iterative optimization scheme. To demonstrate the effectiveness of CUR matrix decompositions for multidimensional seismic data recovery, we present examples of 3D and 4D synthetic and field data. Results derived by CUR compare well to conventional eigenimagefamily methods. 
