Reservoir Quality Mapping with Geostatistical and Neural Network Techniques

This case study shows techniques to map quantitatively a Triassic Halfway reservoir in NE British Columbia, Canada. The sand is at a depth of roughly 1000 meters and has a reservoir thickness that is within the seismic tuning thickness. It demonstrates how significant seismic attributes can be extracted from 20 seismic data . These attributes include amplitudes, instantaneous seismic parameters, inversion results, and AVO parameters including delta p, delta s, fluid stack, and Lame's parameters. In total more than 30 attributes are analyzed and correlated with net sand thickness and fluid content of the reservoir. However, not all attributes are independent. Some significant attributes are selected to predict net sand thickness and fluid content.

Even a single attribute such as amplitude yields a good approach to mapping the sand thickness in a qualitative way. A quantitative estimate is derived with cokriging and simulation techniques. More reliable prediction can be achieved with multiple linear regression or a backpropagation neural network approach. The neural network yields the best results. The derived net sand thickness distribution fits well with the geological understanding of this area.

For the prediction of the fluid content we discriminate between four cases, i.e., d&a , gassy (water with some gas), gas, and oil production (which is basically a mixture of oil, gas, and water). Single and multiattribute regression analysis are ambiguous. This was to be expected because we cannot necessarily assume a linear relation between seismic attributes and fluid content. Only the neural network approach yields useful results. However, only the cases d&a, oil production or gassy can be determined reliably. The identification of the gas only case is not successful, because there is not enough gas only cases in the training set for the neural network approach. The derived map for net sand thickness and fluid content greatly support development and exploration in the investigated area.