Innovative methods for cavity detection in levees using ground-penetrating radar and supervised machine learning

Abstract

Animals can compromise the structural integrity of a levee by digging burrows within it. Since these cavities are located in the subsurface, a geophysical survey method is required for detection. One such method is ground-penetrating radar (GPR). This research determined with supervised machine learning (ML) the conditions under which GPR can detect cavities in levees and developed a methodology based on a transverse electric (TE) amplitude-vs-incidence angle (AVA) analysis and wave polarity that can distinguish cavity reflections from non-cavity reflections such as those resulting from buried debris or changes in lithology. The results indicate that GPR is only effective for detecting small-scale burrows with a radius of 0.15 m located 1.5 m or less below the surface in levees in which the electrical conductivity of the outer layer is less than 0.037 S/m. This implies that GPR will be a successful survey method for detecting such small-scale cavities in sanddominated levees but that the method should not be employed for detecting such voids in claydominated levees. A fieldwork conducted to validate the research was also unable to detect visually identified animal burrows in a clay-dominated levee in Everdingen, The Netherlands. However, this is most likely due to the tunnels having collapsed and not a result of the clay being too conductive. The AVA analysis further produced distinct theoretical curves for different anomaly compositions in poorly conducting levees. By combining these AVA curves with wave polarity, the composition of an anomaly can be uniquely determined. This enables differentiating cavities, filled with either air or water, from other subsurface features such as sand lenses. This helps to assess the structural integrity of a levee and determines if further safety measures are required. However, practical implementation is complicated by subsurface heterogeneity and interference from the environment. Future research should focus on creating a database of AVA curves for common subsurface materials and on enhancing AVA analysis by calculating transverse magnetic (TM) AVA curves to enhance practical application.