2024 Volume 80 Issue 17 Article ID: 24-17163
In big data (BD) methodologies for monitoring seafloor topography over long periods, irregularly spaced spatial data from fishing vessels are allocated onto grid points to generate bathymetric data. Interpolation of large amounts of depth data is necessitated for this purpose. Kriging, a widely used method for terrain data interpolation, requires computational costs to eliminate anomalies, posing a challenge in the BD methodologies dealing with massive data. On the other hand, Support Vector Regression (SVR), a machine learning technique, is suitable for estimating surfaces with minimal influence from anomalies in spatial domains. In this study, we applied SVR to interpolation in the BD methodologies. The demonstration showed that it achieves accuracy equivalent to Kriging while reducing computational time for generating terrain with 50 m grid spacing to approximately 2.5 % of one-point-removed Kriging.
