eservoir characterization is the process of assessing reservoir properties and its condition, using the available data from different sources such as core samples, log data, seismic surveys (3D and 4D) and production data. This is done in different stages of the E&P process from high grading reservoirs in exploration to their delineation for their development, as well as their description for optimum production to assessing their evolution in their stimulation for enhanced oil/gas recovery to extend their economic life.

An integrated approach for reservoir characterization bridges the traditional disciplinary divides, leading to better handling of uncertainties and improvement of the reservoir model for field development.

Among the main difficulties in reservoir characterization is the “SURE challenge.” Figure 1 demonstrates the complications involved in integrating different data types with different scale, uncertainty, resolution and environment (SURE). The top left illustrates three key data types: core, well log and seismic data (referred to as a data pyramid). The base of the pyramid is the seismic with very large coverage but with limited resolution and lesser level of certainty. The top of the pyramid is the core data with very little coverage (only at a particular well location involving a fraction of the well) but with high level of certainty and resolution.

Effective integration of all the data types, in spite of the SURE challenge, is the objective of reservoir characterization. Artificial intelligence (AI) and data analytics (DA) can play key roles in offering solutions to the SURE challenge.

AI-DA has been gaining popularity in many aspects of E&P recently, and the expectation is that it will become an integral part of the tool box for many of our applications. DA, which is the systematic use of computational analysis of the data for making decisions, is an appropriate tool to address the need to deal with large amounts of data (Figure 1). The DA engine is energized by the power of AI and its machine learning (ML) and deep learning (DL) subsets. AI-DA may prove to be the exact medicine to address the SURE challenge.

FIGURE 1. The SURE challenge highlights the difficulties in integrating data with wide ranging differences in terms of scale, uncertainty, resolution and environment. (Source: AIM-DEEP/University of Houston)

The bottom right of Figure 1 shows a pyramid comprising different aspects of integration. Vast amounts of Big Data with their 4V characteristics (volume, velocity, variety and veracity) need to be combined with technical knowledge and experience from domain experts to perform effective data mining and ultimately reservoir characterization. This requires designing a human machine interface, perhaps based on fuzzy logic and natural language processing to facilitate flow of data and information between the two.

An integrated reservoir characterization starts with collecting data from geological, petrophysical, seismic and engineering data. A multidisciplinary data analysis process creates a model of different reservoir properties including reservoir architecture, lithologies and facies. The geometry of the flow units is established (physical rock properties such as porosities and permeabilities of flow units). Three properties are related to the pore space:

• Porosity: the fraction of the entire volume part occupied by pores, cracks and fractures;
• Internal surface: the magnitude of the surface of pores as related to the rock mass pore volume and controls interface—effects at the boundary grain—pore fluid; and
• Permeability: the ability to flow fluid through rock pores.

Given different levels of uncertainty and other aspects of the SURE challenge, the estimates of reservoir properties should also be accompanied with their respective levels of uncertainty. This is derived from the calibration process and the extent of the match between estimated models with the ground truth (well/production data). This necessitates integration of physics-based and data-based approaches, also referred to as hybrid methods. Reservoir description is an iterative process from the input data to the process (e.g., well data, seismic data and production data). High-performance computers, both for their computing power and memory capacity, are crucial for performing data mining and iterations in a timely fashion, especially for real-time reservoir monitoring.

AI-DA offers a natural toolbox for reservoir property estimation and their uncertainties. ML and DL methods perform much like a human brain. They can receive a variety of data from many different sources with drastically different characteristics and undertake necessary evaluations, and they can eventually make the right decisions and/or solve complicated problems. For example, DL finds particular features in the data that could be useful for classification of facies or prediction of different reservoir properties (Figure 2). They are well equipped to handle the issues highlighted under the SURE challenge.

Nevertheless, human intelligence (engineers and geoscientists) will always have a superior performance with qualitative data than computers that are better dealing with quantitative data. Thus, we should design effective human-machine interfaces to create hybrid solutions based on combining machine intelligence with human intelligence.