SYNTHETIC SEISMIC DATA GENERATION FOR PETROPHYSICAL WORKFLOW AUGMENTATION USING MACHINE LEARNING CASE STUDY IN THE NIGER DELTA

ABSTRACT

The Niger Delta's complex geology and limited seismic data hinder exploration and production activities due to the logistic and financial challenges associated with data acquisition hindered by the extensive swampy terrain, dense vegetation present in the region as well as the presence of shallow gas pockets and irregular near-surface conditions that degrades seismic data quality, introducing noise and complicating imaging. This study presents an integrated approach to generate synthetic seismic data for the region by combining geological modelling with GemPy, seismic wavefield simulation using Devito, and machine learning techniques. A detailed geological model capturing the stratigraphic architecture and structural features is constructed with GemPy. Devito simulates realistic seismic wavefields based on this model, augmenting the available real data. The synthetic seismic data is preprocessed and used to train machine learning models for tasks like seismic interpretation and reservoir characterization. Results demonstrate the potential of this approach to improve model performance by leveraging domain knowledge from geological modelling and seismic simulation. Seismic data generated is of good enough quality for training and interpretation tasks. The resultant model interprets a seismic dataset 13 times faster than traditional manual interpretation methods and model processes seismic data at a faster rate in terms of traces per second, compared to manual interpretation which can take several minutes to hours to process a single trace. The model interpretes multiple seismic datasets simultaneously, increasing the throughput and efficiency of the interpretation process. Implications for exploration in the Niger Delta and future research directions are discussed.