Automated Multivariate Well Log Normalization With Gaussian Mixture Modeling: Honoring Joint Log Properties And Geological Context

High-quality well-log data are essential for reliable petrophysical interpretation.Inregional loganalysis,this means macro-scale geological trends are preserved, formations with similar properties observe comparable measurements, and divergent measurements reflect true changes in petrophysical properties. Achieving this is challenging when routine log measurements are systematically affected by tool calibration, vintage, design, and vendor variability, which are not easily addressed with physics-based approaches-as for example, with nuclear logs. These pathologies become more pronounced as petrophysical characterization expands to larger scales when modern interpretations are applied to legacy datasets. Weakly informed normalization risks over homogenizing geology or over imposing variation, making objective normalization workflows increasingly critical. To this end, we introduce a novel multi-log normalization approach using Gaussian Mixture Modeling (GMM) to honor the joint statistical properties of well logs, minimize subjectivity, and preserve geological trends. Additionally, we show its application with three case studies from different basins. Our workflow models log datasets from reference and application wells as multivariate probability distributions, from several measured logs, using GMM. Normalization is performed by applying an affine transformation, which includes scaling, rotation, and shearing, to targeted logs while preserving a subset of logs that provide geological context. Transformation parameters are optimized to minimize the Jensen-Shannon Divergence between the normalized and reference distributions, to achieve statistical similarity. Unlike traditional univariate methods, this approach accounts for covariance between logs and enables robust differentiation between genuine geological variation and non-petrophysical effects on measurements. The workflow accommodates normalization between wells with only partially shared depth coverage and supports reweighting of populations observed in the reference well, allowing normalization even when the proportions of petrophysical properties such as lithology, fluid content, and texture differ between wells. This work shows that the proposedapproachscales fromsinglewell to regional studies. We applied the workflow to datasets from several basins with diverse lithologies and reservoir properties. The method consistently normalizes logs in neighboring offset wells to closely match reference well multivariate distributions. In distant wells, normalized logs demonstrated no undue bias toward reference populations and aligned with nuclear log measurements reconstructed from measured core-plug chemistry, indicating no over-normalization was occurring. Furthermore, advanced log measurements corroborate regional interpretations in these wells. In one study, the regionally interpreted carbonate fraction relative error was reduced from 20% to 4% after applying our approach. This has direct impacts on subsequent estimates of rock strength and stress estimates for geomechanics. Additionally, geological trends and log relationships were preserved even with changing petrophysical properties or with only partially common stratigraphic overlap. These studies highlight the impact of the proposed workflow since standard univariate normalization often fails to correctly normalize logs, resulting in misleading petrophysical answer products. The results presented demonstrate that the introduced automated multi-log normalization workflow offers an objective and scalable solution for conditioning well-log 1 SPWLA 67th Annual Logging Symposium, May 16-20, 2026 data across diverse geological settings and petrophysical to variations in tool design, vendor, or vintage frequently properties. By modeling logs as multivariate remain. These then propagate to downstream distributions and optimizing transformations to align reference and application datasets, the approach preserves geological trends and relationships between logs, even when wells have only partially shared depth coverage or differing petrophysical characteristics. Case studies show that this method reduces errors in mineralogy, porosity, and saturation, and improves the reliability of regional interpretations. Unlike traditional univariate normalization, our introduced workflow adapts to variations in petrophysical properties, supporting more accurate and robust analysis, especially when extending regional interpretations from data rich wells to those with routine logging programs.