Abstract In the rapidly advancing realm of artificial intelligence, machine learning, workflow automation, and cutting-edge hardware technologies, the adoption and integration of these innovations within the oil and gas industry has witnessed substantial growth. This is particularly evident in the domain of subsurface reservoir modelling, where these technologies are being leveraged to enhance the accuracy and efficiency of reservoir simulation and optimization processes. Subsurface reservoir models are a critical component in field development planning, as they provide essential insights into both the overall field recovery and the individual reservoir performance under a variety of operational scenarios. These models, which are typically constructed as three-dimensional, physics-based representations of the reservoir, serve as powerful tools for conducting "what-if" analyses, helping to maximize the net present value (NPV) of assets and ensuring the highest possible return on investment. The field of subsurface reservoir science encompasses a broad range of specialized domains, including seismic interpretation, seismic inversion, sedimentological analysis, petrophysical interpretation, geological modelling, dynamic modelling, production facilities modelling, feasibility studies, and economic evaluations. Reservoir models are typically designed as tailored, fit-for-purpose models, integrating data and insights from multiple domains depending on the specific needs of the project. However, as is common with any complex, multi-disciplinary process, the development of reservoir models is often hindered by several challenges related to workflow automation, data consistency across domains, and the time-consuming nature of model building. Some of the key challenges include: Fragmentation: The absence of seamless integration between data, results, and interpretations across different subsurface modelling domains, leading to fragmented and inefficient workflows. Inconsistency: In many cases, results and interpretations derived from one domain may be modified or adjusted in subsequent domains for modelling or historical calibration purposes, often without appropriate feedback to the preceding domains. This lack of consistent updates results in models that are inconsistent across domains, diminishing the overall predictability and reliability of the final model. Deterministic Approach: Traditionally, reservoir modelling processes conclude with the generation of only a few model realizations, often limited to high, medium, and low-case scenarios, which may not fully capture the range of uncertainties inherent in the subsurface. Time-Consuming: The lack of automation in the modelling process significantly prolongs the timeline required to build accurate and reliable models, often delaying decision-making processes. Model Updates: The challenges related to integration and consistency across subsurface domains also complicate the process of updating reservoir models, making it time-consuming task. This paper introduces an innovative solution in the form of an integrated forward modelling workflow—referred to as the Big Loop—which facilitates the seamless propagation of reservoir uncertainties across multiple subsurface domains. The aim of this workflow is to create more reliable, adaptive, and ensemble-based reservoir models, ensuring a more robust and efficient approach to subsurface reservoir management.
Mohammed et al. (Tue,) studied this question.