Estimation of effective porosity using geostatistics and multiattribute transforms: A case study

Abstract

The middle Eocene Kalol Formation in the north Cambay Basin of India is producing hydrocarbons in commercial quantity from a series of thin clastic reservoirs. These reservoirs are sandwiched between coal and shale layers, and are discrete in nature. The Kalol Formation has been divided into eleven units (K‐I to K‐XI) from top to bottom. Multipay sands of the K‐IX unit 2–8 m thick are the main hydrocarbon producers in the study area. Apart from their discrete nature, these sands exhibit lithological variation, which affects the porosity distribution. Low‐porosity zones are found devoid of hydrocarbons. In the available 3D seismic data, these sands are not resolved and generate a composite detectable seismic response, making reservoir characterization through seismic attributes impossible. After proper well‐to‐seismic tie, the major stratigraphic markers were tracked in the 3D seismic data volume for structural mapping and carrying out attribute analysis. The 3D seismic volume was inverted to obtain an acoustic impedance volume using a model‐based inversion algorithm, improving the vertical resolution and resolving the K‐IX pay sands. For better reservoir characterization, effective porosity distribution was estimated through different available techniques taking the K‐IX upper sand as an example. Various sample‐based seismic attributes, the impedance volume, and effective porosity logs were used as inputs for this purpose. These techniques are map‐based geostatistical methods using the acoustic impedance volume, stepwise multilinear regression, probabilistic neural networks (PNN) using multiattribute transforms, and a new technique that incorporates both geostatistics and multiattribute transforms (either linear or nonlinear). This paper is an attempt to compare different available techniques for porosity estimation. On comparison, it is found that the PNN‐based approach using ten sample‐based attributes showed highest crosscorrelation (0.9508) between actual and predicted effective porosity logs at eight wells in the study area. After validation, the predicted effective porosity maps for the K‐IX upper sand are generated using different techniques, and a comparison among them is made. The predicted effective porosity map obtained from PNN‐based model provides more meaningful information about the K‐IX upper sand reservoir. In order to give priority to the actual effective porosity values at wells, the predicted effective porosity map obtained from PNN‐based model for the K‐IX upper sand was combined with actual effective porosity values using co‐kriging geostatistical technique. This final map provides geologically more realistic predicted effective porosity distribution and helps in understanding the subsurface image. The implication of this work in exploration and development of hydrocarbons in the study area is discussed.

Tecnologías constituidas, innovaciones en proceso y tecnologías introducidas en el mercado internacional: caso de estudio

“…Para abordar el problema de caracterización no invasiva, se han propuesto algunas técnicas de modelación, tales como el mapeo geostadístico y transformaciones multiatributo para la estimación de la porosidad (Pramanik et al, 2004). Otros enfoques incluyen el uso de algoritmos genéticos (Dorrington & Link, 2004) o redes neuronales (Helle et al, 2001;Bhatt & Helle, 2002).…”

Kriging interpolation in seismic attribute space applied to the South Arne Field, North Sea

“…Existing methods for high-dimensional interpolation include linear regression ͑Hampson et al, 2001;Russell et al, 2002;Hansen et al, 2008͒, spline interpolation, nearest neighbor, cokriging ͑Doyen, 1988͒, and neural networks ͑Hampson et al, 2001Russell et al, 2002;Pramanik et al, 2004;Herrara et al, 2006͒. Some comparative studies of these methods are available.…”

“…Russell et al ͑2002͒ compare generalized regression neural networks and radial basis function networks for prediction of log properties from seismic attributes. Pramanik et al ͑2004͒ compare the application of linear regression, neural networks, cokriging with impedance, and cokriging of actual porosity estimates with a porosity map obtained using a neural network. Herrara et al ͑2006͒ uses a combination of linear regression and neural networks.…”

Prestack inversion and multiattribute analysis for porosity, shale volume, and sand probability in the Havert Formation of the Goliat field, southwest Barents Sea

“…This is done in two steps: (1) a training process in which the target logs and the seismic attribute volumes are analyzed (by applying a least-squares approach) at both wells to derive a statistical function relating the target logs to the attribute volumes and (2) applying the derived function to create corresponding target log values for each trace in the seismic volume. Several authors have shown the benefits of geostatistical multiattribute transforms to predict porosity and lithology in the seismic volume (Hampson et al, 2001;Pramanik et al, 2004;Calderon and Castagna, 2005).…”