Data analytics and geostatistical workflows for modeling uncertainty in unconventional reservoirs

New methods are required to support unconventional reservoir uncertainty modeling. Unconventional plays add additional complexity with greater uncertainty in reservoir measures (e.g. unreliable permeability measures in low permeability rock) and weakened relationships between measurable reservoir properties and production results (production mechanisms may not be well understood). As a result, unconventional plays are often referred to as "statistical plays", suggesting the reliance on statistical characterization of production spatial distributions. Various methods have been proposed for probabilistic modeling with statistical plays. Regardless of workflow, it is critical to account for spatial context, including: production spatial continuity; local conditioning from well-based production; local secondary information; and boundaries of the area of interest. Methods that are insensitive to spatial context are unreliable for decision-making. To these methods, the uncertainty in the aggregate production over the next set of wells is the same regardless of their respective locations. Even in unconventional reservoirs, the spatial context still matters. There are theoretical methods to explore the uncertainty in the production rates of wells within a pad and of the aggregate production of wells over pads within a development block. Yet, the most flexible methods are based on empirical model resampling. These methods extract multiple samples from actual reservoir models to simulate the drilling strategy over multiple realizations and scenarios of the subsurface uncertainty model, an ensemble of possible models. These methods integrate all available information sources while further leveraging the uncertainty model that is routinely built for reservoir forecasting. The aim of this paper is to demonstrate the ability of the resampling method to answer reservoir development questions, such as: how much variability in well production is predicted between wells in a single pad; how much variability in well aggregate production is predicted between pads; how much information does the first well's production provide about the total pad production; and when is it best to abandon a pad? Knowing the answer to these questions improves reservoir development decision-making. This paper advocates for new data analytics, geostatistical methods and workflows to support the best use of geoscience and engineering practice.