Machine learning and AVO class II workflow for hydrocarbon prospectivity in the Messinian offshore Nile Delta Egypt

Abstract This study presents a comprehensive workflow to detect low seismic amplitude gas fields in hydrocarbon exploration projects, focusing on the West Delta Deep Marine (WDDM) concession, offshore Egypt. The workflow integrates seismic spectral decomposition and machine learning algorithms to identify subtle anomalies, including low seismic amplitude gas sand and background amplitude water sand. Spectral decomposition helps delineate the fairway boundaries and structural features, while Amplitude Versus Offset (AVO) analysis is used to validate gas sand anomalies. The entire seismic volume is classified into facies domains using machine learning, which isolates target features from seismic background data. The study area, covering 1850 km 2 , includes major structures such as the Rosetta fault and Nile Delta offshore anticline, with reservoirs consisting of layered sandstones and mudstones. Over 90 wells, including exploration and development wells, have been drilled in the area. Seismic amplitude data, including full and partial offset stacked, were analyzed to classify gas, water, and shale zones. The workflow’s performance is demonstrated through the successful identification of the low-amplitude Swan-E Messinian anomaly, characterized as a high-risk gas prospect. Machine learning techniques, specifically neural network models, were trained to differentiate seismic features such as low-amplitude gas sand from background-amplitude water sand and shale. By iterating over multiple attributes and validating the models on blind test sets and on a blind section, which excluded a known shallow gas field, the workflow significantly improved the ability to detect potential hydrocarbon reservoirs characterized by low seismic amplitude. The results show that this integrated approach reduces exploration risk, quantifies the chance of success, and enhances decision-making in well placement and hydrocarbon exploration. This method is particularly useful for identifying low seismic amplitude anomalies, which are often challenging to detect with conventional seismic analysis. (1) This study developed a workflow to detect low seismic amplitude gas fields in near-field exploration. (2) It uses a machine learning algorithm to classify and explore low-seismic-amplitude gas sand reservoirs. (3) This approach helps estimate the likelihood of success and reduces the risk associated with hydrocarbon exploration wells.