Submarine landslide susceptibility assessment integrating frequency ratio with supervised machine learning approach

Marine geological hazard assessment is crucial for the development and utilization of marine resources, among which submarine landslide susceptibility assessment constitutes a key and primary stage. However, current research, especially the application of supervised machine learning in this field remains limited. In this study, nine submarine landslide-related factors in the South-West Iberian margin were gained; including bathymetry, slope, curvature, earthquake magnitude density, distance to fault, distance to volcano, sediment type, pipeline density, and vessel density, and then a submarine landslide inventory was compiled. By combining the frequency ratio with representative supervised machine learning algorithms (logistic regression, random forest, and artificial neural network), the large-scale submarine landslide susceptibility assessment was conducted. The susceptibility result was categorized into five levels utilizing the Jenks breakpoint method, ranging from very low to very high. Meanwhile, all models were evaluated from the perspective of probability characteristics and machine learning. The results showed that the frequency ratio-based supervised machine learning models have more reasonable statistical characteristics and exhibit better accuracy, with the frequency ratio-based artificial neural network model emerging as the most capable of assessing submarine landslide susceptibility in the study area, delivering the most precise results. This study provides a reference for the application of supervised machine learning in submarine landslide susceptibility assessment. The methodology and research findings have the potential to enhance the awareness of submarine landslide risks in this or other regions and facilitate the development of effective risk management strategies.