Seismic facies analysis involves the interpretation of reflection patterns from seismic data to provide insights into subsurface sedimentary environments,depositional processes,and lithological variations,aiding geore...Seismic facies analysis involves the interpretation of reflection patterns from seismic data to provide insights into subsurface sedimentary environments,depositional processes,and lithological variations,aiding georesources exploration.This study evaluates unsupervised machine learning(ML)models for seismic facies mapping within the Barail group from the Amguri region of the Upper Assam basin in northeast(NE)India.Utilizing high-quality three-dimensional seismic data,a comprehensive set of seismic attributes,including amplitude-based,instantaneous,spectral,geometric,and textural measures,is extracted,optimally selected,and integrated using two unsupervised models:the self-organizing map(SOM)and the generative topographic mapping(GTM).The two models are compared to identify the most effective approach for discerning seismic facies patterns within the Barail-Coal-Shale(BCS)and Barail-Main-Sand(BMS)units.Results indicate that GTM outperforms SOM by providing improved cluster separation,enhanced facies continuity,and greater geological consistency across the target interval in the study area.GTM-derived facies insights are validated with borehole and field data,facilitating an in-depth interpretation of the sedimentary environment.The BCS interval predominantly consists of coaly shale,coal-shale alternations,and minor sandstone facies,indicative of a swampy deltaic setting conducive to periodic vegetation accumulation.In contrast,the BMS interval primarily comprises sandstone with occasional shale and coal-shale intercalations,reflecting a fluvial-dominated environment.Post-depositional tectonic processes contributed to the deformation and structural complexity within these intervals.The proposed methodology(specifically data enhancement,attribute optimization,ML model comparison and integration of interpretations with different geoscientific data)demonstrates potential for application in other geological settings to enhance subsurface interpretation.展开更多
文摘Seismic facies analysis involves the interpretation of reflection patterns from seismic data to provide insights into subsurface sedimentary environments,depositional processes,and lithological variations,aiding georesources exploration.This study evaluates unsupervised machine learning(ML)models for seismic facies mapping within the Barail group from the Amguri region of the Upper Assam basin in northeast(NE)India.Utilizing high-quality three-dimensional seismic data,a comprehensive set of seismic attributes,including amplitude-based,instantaneous,spectral,geometric,and textural measures,is extracted,optimally selected,and integrated using two unsupervised models:the self-organizing map(SOM)and the generative topographic mapping(GTM).The two models are compared to identify the most effective approach for discerning seismic facies patterns within the Barail-Coal-Shale(BCS)and Barail-Main-Sand(BMS)units.Results indicate that GTM outperforms SOM by providing improved cluster separation,enhanced facies continuity,and greater geological consistency across the target interval in the study area.GTM-derived facies insights are validated with borehole and field data,facilitating an in-depth interpretation of the sedimentary environment.The BCS interval predominantly consists of coaly shale,coal-shale alternations,and minor sandstone facies,indicative of a swampy deltaic setting conducive to periodic vegetation accumulation.In contrast,the BMS interval primarily comprises sandstone with occasional shale and coal-shale intercalations,reflecting a fluvial-dominated environment.Post-depositional tectonic processes contributed to the deformation and structural complexity within these intervals.The proposed methodology(specifically data enhancement,attribute optimization,ML model comparison and integration of interpretations with different geoscientific data)demonstrates potential for application in other geological settings to enhance subsurface interpretation.
