Seismic prediction of soil distribution for the Chang-Bin offshore wind farm in the Taiwan Strait

Abstract

Direct soil measurements are limited to borehole locations and are therefore sparse in the oceans. To effectively characterize the soil distributions for the Chang-Bin offshore wind farm, which is an area with the greatest wind energy potential in the Taiwan Strait, we have developed a workflow to predict the soil distribution in the subsurface based on integrated analysis of seismic data and borehole data. First, we characterize the key seismic units and their seismic response in order to understand the regional stratigraphy. Then, we correlate the soil types to each stratigraphic unit as the constraint for the input and quality control to train a neural network based on seismic multiattribute analysis. Finally, we develop a neural network that is suitable for soil prediction in the Chang-Bin offshore wind farm. Five seismic units identified from the seismic profiles reveal that the regional stratigraphy has been greatly affected by sea-level change and the sediment transportation process. Confirmed by independent in situ borehole data, the neural network is considered reliable up to 60 m below the seafloor, whereas decreased signal-to-noise ratios at greater depths lead to poorer prediction accuracy. Compared to previous studies that mainly are based on high-quality 3D seismic and well logging data, our method can predict the soil distribution by analyzing 2D seismic profiles and simplified soil layers alone. The prediction results reveal detailed lithologic variations that are tested by in situ borehole measurements. Therefore, we are confident that this approach could effectively obtain the soil distribution prediction and thus reduce the costs in offshore engineering applications.