Abstract

In this paper, we explore the use of non-linear multivariable regression to correlate statistically selected seismic attributes to reservoir porosity (Φ), water saturation (Sw), and net pay, for the Cherry Canyon and Brushy Canyon in Nash Draw Field southeast New Mexico. Seismic attributes have recently been the focus of renewed interest for evaluating reservoir properties. Well data gives very precise information on the reservoir properties at specific field locations with a high degree of vertical resolution. 3-D seismic surveys can cover large areas of the field, but reservoir properties are not directly observable, in part due to relatively poor vertical resolution. This paper presents a method for relating interval reservoir properties at the well-bore to sets of seismic attributes, in order to predict Φ, Sw, and net pay across the whole field.

There is a bewildering array (350 and growing) of seismic attributes that can potentially be used in a regression analysis of reservoir properties. Using all attributes is not feasible (i.e., it is beyond computing abilities) and labor intensive, so the first step involves statistical analysis of the attributes with respect to each reservoir property to be evaluated. We use Fuzzy-curve analysis, a statistical technique, to rank attributes according to their value in correlating to each reservoir property, and select the top three ranked attributes for use in developing regression equations for each reservoir property.

Non-linear regression is used because individual attributes had low correlation coefficients when cross-plotted with reservoir properties. A neural network architecture was developed to relate the three selected attributes to each property. In each case the output data used for training was a reservoir property, Φ, Sw, or net pay, from nineteen wells in the field. The validity of the non-linear regressions was tested by removing several wells from the training data, re-computing the weights and predicting the three absent points. We did this three times for each reservoir property, with different points removed. Each network accurately predicted these nine test points and the solutions are therefore considered robust.

Maps of Φ, Sw, and net pay were generated using the regression relationships and seismic attributes at each seismic bin location. Map of computed Φh (porosity thickness) and hΦSo (hydrocarbon pore volume) were generated from the reservoir property maps. The techniques that we have developed maximize information from both the well control and seismic data, and generated useful maps for targeted drilling programs in the Nash Draw field.