A Comparative Analysis of Classical Approaches vs. Machine Learning Methodologies for Property Modeling Using EMBER

Abstract This study evaluates the effectiveness of the Embedded Model Estimator (EMBER), a machine learning-based tool, designed to improve porosity modeling within petrophysical workflows. EMBER’s performance is tested in two field development plan (FDP) projects: a clastic reservoir in the Sirte Basin and a carbonate reservoir in the Intisar Field, both located in Libya. Due to their significant geological heterogeneity, these reservoirs provide a a good example for comparing EMBER withtraditional geostatistical modeling techniques. In addition to the porosity in this study we incorporates facies models and stratigraphic seismic attributes, including acoustic impedance, into the EMBER workflowto have better representation for the porosity distribution. The clastic reservoir lies within a structurally complex area between the Zaltan and Al Jahamag platforms, featuring faulted anticlines and a stratigraphic succession comprising rocks spanning from the Cambro-Ordovician Gargaf to the Upper Cretaceous Bahi Formations. In contrast, the carbonate reservoir in the Intisar Field comprises pinnacle reef structures belonging to theZaltan carbonate shelf, characterized byhydrocarbon accumulation in the highly heterogeneous Upper Sabil limestones.Results demonstrate that the EMBER algorithmconsistently produces porosity models that closely align with well log measurements while preserving geological continuity. In the carbonate reservoir, the EMBER algorithm slightly outperforms classical methods in terms of accuracy and distribution fidelity, while in the clastic reservoir, EMBER outperforms the classical methods. Across both formations, the method significantly reduces modeling time - by over 70% - without compromising data integrity. This time efficiency, coupled with its high predictive performance, highlights the value of the EMBER’s algorithm especially in complex geological settings. Unlike conventional approaches that rely on fixed statistical assumptions, EMBER uses machine learning algorithms to learn adaptive relationships between geological and geophysical inputs. This allows more accurate and realistic modeling of subsurface heterogeneity. The study provesEMBER’s advantages in enhancing modeling accuracy, operational efficiency, and reducing geological uncertainty, reinforcing its value in hydrocarbon reservoir characterization and development planning.