Geostatistics offers a robust way to estimate the spatial distribution of reservoir properties. Geostatistical methods, such as kriging, cokriging, and sequential simulation have been applied to integrate well-log data and seismic attributes. However, conventional deterministic methods of geostatistics, kriging and cokriging often have difficulty identifying the characteristics of lithologic reservoirs because only one secondary attribute is incorporated. To decrease the uncertainty and improve the definition of the final estimate, two modified techniques, cokriging with multiple secondary attributes and block cokriging with multiple secondary attributes, are implemented. However, these deterministic methods can only provide one predicted result, which has trouble capturing the natural heterogeneity of reservoirs and assessing the uncertainty of the predicted map. To solve this issue, an improved stochastic technique, sequential simulation using multi-variable cokriging, is presented.
All these presented techniques are applied to real datasets. Case studies are presented to predict the thickness of the reservoir, total organic carbon, and porosity. The final predicted maps demonstrate that these methods can enhance the lateral resolution. Leave-one-out cross-validation is used to evaluate the construction models, and shows that the uncertainty of the estimate can be reduced due to the use of more seismic attributes than traditionally implemented, while still optimizing cross-validation.