Machine Learning–Based Digital Integration of Geotechnical and Ultrahigh–Frequency Geophysical Data for Offshore Site Characterizations

Geophysical data play a vital role in offshore site characterizations, especially during the planning phase of the geotechnical investigation, in order to define the scope to be performed at the site most effectively, but also to aid interpretation once geotechnical data are acquired. Although geophysical data are advantageous in imaging the subsurface conditions over large offshore areas and can reveal important information about seabed features and the sediment depositional history, the information obtained is usually qualitative from a geotechnical perspective. In today's practice, the engineering properties of soils for design purposes are determined by established ground-truth methods (e.g., offshore sampling, seabed drilling, and in situ testing, among others) applied directly at locations of proposed infrastructure. However, recent advances in seismic acquisition and interpretation methods, machine learning algorithms as well as computational power, provide opportunities to fully integrate geophysical and geotechnical (G&G) data in a digital format to quantitatively assess the site variability and design parameters with limited geotechnical data. Therefore, the motivations of this paper are to advance the quantitative G&G integration to benefit the broader engineering community and to provide thoughts on improving this technique for industrial applications. The scope of the paper is to demonstrate a successful G&G integration workflow that mainly consists of acoustic impedance inversion on ultrahigh–frequency (>2,000 Hz) geophysical data, and the integration with cone penetration test (CPT) data using artificial neural networks to create synthetic CPT profiles. The proposed workflow has been blindly tested and satisfactory results are achieved for predicting the CPT profiles when compared with the actual data.