Interwell Property Mapping Using Crosswell Seismic Attributes

Abstract

Abstract We have used crosswell seismic data and an artificial neural network to predict interwell reservoir properties in the Sulimar Queen field. In this study, we first predict the gamma ray response between two wells where the crosswell seismic was shot. The gamma ray response is predicted between the wells by training an artificial neural network with actual gamma ray logs available at the wells and the interwell seismic data. Using the a priori information that the porosity is well correlated with the gamma ray response, we estimate the interwell porosity distribution. Introduction Reservoir engineers struggle daily to estimate reservoir properties on a spatial scale from wellbore data. This is true even though geostatistical methods have helped reduce the uncertainty associated with spatial predictions from pseudo-point supports (wells). The current motivation for using seismic data, 3D or otherwise, to predict reservoir properties is due to the inherent large areal coverage of the seismic measurement. The trade-off between the seismic surveys, which allow excellent areal coverage, and the wellbore measurements is in the vertical resolution. Wellbore measurements, such as logs, tend to have higher vertical resolution, to the order of a few inches. On the other hand, seismic surveys which require acoustic energy to be transmitted over large depths are plagued with poor vertical resolution. This is due to the large wavelengths, to the order of a few feet (30 ft.), necessary for the seismic acoustic signal. The pertinent question is: Can we can exploit the obvious advantages - vertical resolution and areal coverage - of the two measurements to our benefit? The current modus operandi of predicting reservoir properties on a field scale is to solve the inverse problem. In this approach, the unknown reservoir parameters are modified under the dynamic data constraints, which are usually pressure and production in the field. Although this method has been successful, especially under the auspices of automatic history matching, there are a few drawbacks. The inverse problem is often ill-posed and results in non- unique solutions. These solutions may yield unrealistic and/or multiple realizations of the property distributions that honor the imposed constraints. Recently, researchers have constrained the models further with seismic or well test data. These additional constraints improve the spatial property predictions but the non-uniqueness of the realizations continues to be a problem. To resolve the issue of non-uniqueness, other researchers have focused their efforts on deciphering the underlying physics that explicitly correlates seismic attributes to reservoir properties. Although far from their objective, they suggest some good alternatives to utilize the available seismic data. A typical approach relies on correlating porosity with seismic impedance, and then applying the developed correlation across the field impedance map to generate a porosity distribution. The permeability distribution is then obtained from a core porosity-permeability relationship. Advanced techniques, such as collocated cokriging and annealing cosimulation, use geostatistics and geostatistics combined with optimization, respectively, to generate petrophysical distributions. These techniques are adequate for areal predictions of petrophysical properties but are limited in their vertical precision since the seismic scale is not always honored. Deutsch et al. recently proposed a modified simulated annealing technique that explicitly allows for vertical averaging of the seismic data. An interesting approach proposed by Behrens et al. is Sequential Gaussian Simulation with Block Kriging (SGSBK). In this approach, the authors combine information from well logs and a vertical variogram model (for vertical variability) with a seismic map and an areal variogram (for spatial variability) to generate 3D realizations. Alternatively, the relationship between a sonic log shot in the wellbore and the crosswell seismic velocities is exploited to correlate seismic velocities with petrophysical properties. Although this approach is relatively successful, it requires a sonic log at the wellbore The interested reader is referred to Schultz et al., and Bunch and Dromgoole. P. 467