Forecasting production and estimating reserves with confidence in unconventional resources, particularly shale reservoirs, is a major unsolved problem in the petroleum industry. Many would prefer a rigorous analytical or numerical model to be used as the basis for production forecasting and reserves estimation, but we are limited by two factors: (1) we may not understand the physics of hydraulic fracture stimulation and fluid transport in shales well enough to ensure that we are using a model with appropriate physics and appropriate fracture and formation characteristics, and (2) no long-term production histories are available in shales reservoirs with which we can calibrate our models. The oldest of the major shales plays, the Barnett Shale in Texas, began to use horizontal wells with multi-stage fracturing only in 2004, so eight years or less of production data are available for wells drilled and completed with the most modern technology, and the wells may have 30-year or longer lives. At this point, we do not know the location of the cracks in the rock and we don’t know the conductivity of proppant in these cracks. Even less is known about oil reservoirs because of more limited production histories.

As a result of these problems, most operators in ultra-low permeability reservoirs use empirical methods, especially decline curve analysis or type curves based on decline trends, for production forecasting and reserves estimates. The most commonly used method is the traditional Arps hyperbolic decline model coupled with a minimum terminal (exponential) decline rate. This approach, while widely used, has problems: (1) the Arps model assumes stabilized flow with unchanging decline constant ‘b’, whereas the flow in shales wells is transient for long periods, perhaps even the life of the well in some instances; and (2) we have no long-term production data which allow to determine the appropriate terminal minimum decline rate.

Much field evidence supports a decline model in which flow is linear in a hydraulically fractured shale reservoir for at least a few years, followed by boundary-dominated flow at the time interference between adjacent hydraulic fractures occurs. This may be followed by linear flow from the formation beyond the stimulated reservoir volume. Unfortunately, we do not have reliable methods to predict the end of linear flow and the appropriate flow model at this time. Two recent decline models, the Stretched Exponential Model and the Duong Model, can accommodate linear flow and may be able to accommodate flow after the time of fracture interference without resorting to uncertain estimates of formation and fracture properties.