Unsupervised Clustering as a Tool for Geomorphological Interpretation: A Case Study from a Sarmatian Delta in Romania’s Dacian Basin

Seismic attributes are a valuable tool in identifying sedimentary facies. Machine learning (ML) techniques can be helpful in combining information coming from different attributes. The unsupervised clustering algorithm called self-organising map (SOM) has already been successfully applied in paleo deep-water settings in many basins defining seismic facies. To evaluate the applicability of this ML method to a paleo shallow-water setting the SOM workflow is applied to 3D seismic data acquired onshore Romania over the south Carpathian foredeep. Many attributes are extracted along an interpreted horizon. Principal Component Analysis (PCA) is performed on the list of attributes in two ways. Firstly, the main principal components are defined to use them as input for the first run of SOM clustering (SOM 1). Secondly, the most important attributes are identified and used as input for the second run of SOM clustering (SOM 2). The geomorphological interpretation of the SOM results is compared with the interpretation of conventional attributes. The results show that the SOM is a powerful tool in defining seismic facies as it allows the definition of details not discernible using other tools. This exercise shows also that ML techniques can be easily implemented by exploration geophysicists using standard interpretation software’s and open-source Python libraries.