Impact of Unsupervised Machine Learning and Seismic Attributes on Mapping Stratigraphic Traps
Maitham Alabbad, Hussain Alkhalifah, Sami Awfi · International Petroleum Technology Conference · 2024
Mapping lateral variations of reservoir and seal is a key step for stratigraphic trap exploration. Hard data such as well testing results, fluid samples, pressure data, core analysis, and wireline log signatures are often used as evidence of lateral fluid barriers and facies boundaries which provide the lateral seal element of stratigraphic traps. Mapping these boundaries is challenging and requires the use of seismic data, especially in clastic reservoirs where sand bodies with different pore pressures or hydrocarbon columns can be mapped and correlated laterally, although they may not be interconnected. One of the common workflows is to rely on seismic amplitude variations as a proxy for reservoir quality. Another method uses absolute acoustic impedance (AI) volumes with specific cutoffs derived from the rock physics template for the lateral seal definition. Also, red-green-blue (RGB) color blends of decomposed frequencies can be used to show the geomorphology of the reservoir unit and indicate lateral variations in depositional environments. These techniques require a prior knowledge of facies character on seismic to define suitable cutoff values. A good well dataset covering all expected facies is needed for calibration. Unsupervised machine learning techniques, on the other hand, are useful tools that guide seismic interpretation of large 3D datasets and reveal subtle trends automatically. The different unsupervised machine learning algorithms and their application on seismic data have been discussed in the literature. Mingjun and Cheng (2018), Lubo-Robles et al. (2019), and Owusu and Raef (2022) describe its use in facies classifications. In this paper, two different seismic waveform classification algorithms are used to map potential stratigraphic lateral seal based on conventional land 3D seismic data. The first algorithm is Hierarchical Classification. The second one is based on K-mean Clustering. Methodologies, strengths, and weaknesses of these waveform classifications are discussed in this study. The derived seismic facies are compared with other products from conventional seismic methods such as seismic amplitude extraction maps, AI geobody extractions, and frequency spectral decomposition RGB blending.