Reservoir-scale 3-D models used for planning horizontal wells are usually unsuitable to provide quantitative support for drilling decisions that are required to optimize well placement. This is a consequence of the scales at which well and seismic data are collected and represented in these prior models. Offset well data frequently form the basis for horizontal well prognoses but lateral variation, not adequately sampled by existing wells, may limit the application of these well-derived models at the proposed locations. Surface 3-D seismic data often have insufficient frequency content to delineate small displacement faults in the reservoir and are subject to uncertainties associated with depth conversion. Consequently, unforeseen geological variation is the most common cause of sub-optimal well placement.
We have developed a modelling and visualization tool that combines geometric data from wells and seismic to produce 3- D geological reservoir models. This tool enables rapid updating of prior 3-D geological models with logging-while-drilling (LWD) data to make local adjustments to dip and displacement of geological surfaces. The drilling team may then visualize the results and obtain quantitative information on projected formation dip, as well as distance of the bit from formation boundaries and faults. Both the computational framework and graphical rendering are 3-D. The continuous updating of these models maximizes the utilization of LWD data and prior models to support real-time drilling decisions that are required to optimize horizontal well placement.
Reservoir management decisions are increasingly being made with the aid of 3-D numerical reservoir models. Very often these models contain apparently realistic geological detail in the volumes between the wells. However, much of this geological structure is generated using statistical algorithms rather than local measurements at the proposed well location. These models reportedly give better predictions of flow behaviour than more traditional "homogenous" models but are driven by statistical algorithms that honour field-wide statistics such as net-to-gross ratio. Consequently, each model is described as an "equiprobable realization" of the actual subsurface. However, this property makes the models poor predictors of the actual subsurface geology that will be encountered during drilling of a horizontal well at any specific location.
Well prognoses usually are based on an expected depth of penetration and an expected reservoir quality. For horizontal wells, the absolute depth is frequently less important than the relative distance from anticipated gas-oil, gas-water or oil-water contacts and/or formation boundaries. These predictions are based on seismic and/or offset well data. By using well data alone, uncertainties in the depth prognosis may result from faults or flexures not adequately sampled by the wells (Fig. 1). Seismic data may reduce this uncertainty but errors in depth conversion due to velocity variations or uncertainties in time picks frequently give rise to a remaining uncertainty of metres or tens of metres.
Predictions of reservoir quality are usually based on interpolation of offset well data that may be augmented by information provided by seismic data. Very often these predictions are based on the presumption of "layer cake" geology between the wells, i.e., the expectation that the stratigraphy encountered in the offset well extends unchanged to the target well location (Fig. 2). Unfortunately, in many reservoirs lateral change over length scales of tens of metres is more common than lateral persistence. For these reasons unexpected geology is the principle cause of disappointing horizontal wells. For example, the well may exit the target formation (Fig. 1) or encounter an unexpectedly poor quality reservoir (Fig. 2). Consequently, geosteering of horizontal wells has become increasingly widespread. This methodology uses measurements collected by logging-while-drilling (LWD) sensors, located in the drill string, to transmit data concerning the formation to the surface in real-time.