Advanced Permeability Estimation in Heterogeneous Carbonate Reservoirs: Integrating Machine Learning and Petrophysics in the Kangan Formation, Iran

Permeability estimation in carbonate reservoirs remains challenging due to heterogeneity. This study aims to present the most comprehensive flowchart possible for permeability estimation using well-logging data. This study evaluates five permeability estimation methods Artificial Neural Network (ANN), FZI-Stoneley (FZI-ST), improved FZI-ST, Timur-Coates, and SDR for the Kangan carbonate reservoir in Iran. The improved FZI-ST method achieved superior accuracy (R² = 0.88) by addressing reservoir heterogeneity through multi-resolution graph-based clustering (MRGC), outperforming traditional methods (Timur-Coates: R² = 0.36; SDR: R² = 0.27). ANN also showed strong performance (R² = 0.86). NMR-based methods underperformed due to T₂ cutoff instability in carbonates. Our workflow offers a cost-effective, log-based solution for complex reservoirs, with global applicability. It can be applied to other heterogeneous carbonate reservoirs, such as those in the Middle East or North America, where similar diagenetic processes (e.g., dolomitization, fracturing) are prevalent. This broad applicability makes the method a valuable tool for global reservoir characterization.