Advances in fracture mapping and full 3D modeling have yielded new insights into hydraulic fracture geometry, but it is still impossible to predict height growth. Fracture mapping data collected from a large number of treatments in different basins yield a rule-of-thumb for expected fracture height over fracture length (aspect ratio), but in specific cases fracture design optimization requires a more accurate forecast for height growth. Calibrated models with full 3D fracture geometry will give the best results, but in many projects the available data to calibrate such a model is severely limited. Knowing this, the question this paper attempts to answer is: "Will using a full 3D model give more reliable predictions of fracture geometry (maybe height growth) compared with pseudo-3D models?".

Using data from an instrumented field test and routine fracture treatments, the results of the different fracture models are tested. Even when detailed knowledge of stress and geomechanical properties are available, it is impossible to match observed fracture geometry using only conventional hydraulic fracture physics. So, even a full 3D model does not provide a true prediction of fracture geometry. Both pseudo3D and full 3D fracture models can match observed fracture geometry, but only by introducing additional parameters beyond conventional fracture propagation physics, such as formation lamination or fracture tip pore pressure.

A full 3D model with default input parameters and conventional fracture physics yields a prediction of strong containment, even for modest stress difference between pay and overburden. This agrees in general with average observed geometry, but in specific cases, fracture height growth still occurs, showing that in these cases the model was inadequate and needs to be calibrated. Pseudo-3D models tend to overestimate height growth for default inputs, but that can also be modified to match the stronger containment often seen in practice. Therefore, no benefit is obtained from fully gridded simulation models in routine cases where critical inputs and calibration data are unavailable.

You can access this article if you purchase or spend a download.