3D-seismic channel–floodplain geomorphology of the Rhaetian Distributive Fluvial System in the North German Basin: an integrated workflow with machine learning for geothermal prospects

• New 3D seismic data interpretation workflow for geothermal sandstone reservoirs. • Workflow combines amplitude and attribute analysis with unsupervised ML techniques. • Case-study reveals fluvial morphology of meandering channel-floodplain systems. • Amplitude-based classifications identified fluvial landforms and elements. • Attribute-based clustering reveals internal variability of channel fills. This study addresses the challenge of identifying fluvial reservoir facies within 3D-seismic data of the Rhaetian Distributive Fluvial System (Upper Keuper, Triassic), a major geothermal target in the North German Basin. A workflow integrating seismic amplitude analysis, seismic attributes, and unsupervised machine learning was developed. Seismic amplitude data revealed key fluvial geomorphological features in vertical transects and horizon slices. Principal component analysis (PCA) was used to select the most relevant seismic attributes for clustering. Combining different attribute classes improved fluvial facies classification, enabling more detailed facies maps that clearly delineate key features. Two clustering cases were evaluated: one using seismic amplitude data and another using the selected seismic attributes as input for two clustering methods, PCA and Self-Organizing Maps (SOM), leading to four scenarios. The analysis revealed key fluvial elements of a meandering channel-floodplain system, providing significant insights into fluvial styles, lateral migration patterns, avulsion of channels, and crevasse splay formation. While amplitude-based classification effectively identified fluvial elements, attribute-based clustering provided a finer delineation of internal facies variability and a better discrimination between sand- and mud-prone facies. The workflow remains constrained by seismic resolution and unsupervised learning bias toward dominant features. Nevertheless, the results emphasize the importance of the initial amplitude analysis in guiding the interpretation of seismic attributes and clustering results. This study highlights the value of integrating machine learning into traditional seismic interpretation and enables more accurate morphology-based identification of fluvial reservoir facies, thereby reducing the exploration risk of geothermal targets in the North German Basin.