Risk assessment for regional submarine landslides integrating AutoML with AHP-EWM approach: a case study from the South-West Iberian sea

Assessing the spatial risk of submarine landslides is critical for marine geo-disaster prevention and control. Traditional methods, including numerical simulations and conventional machine learning approaches, are constrained by costly exploration and inefficient modelling, struggling in complex marine geoenvironments. In this study, a multi-dimensional ‘susceptibility-hazard-vulnerability’ methodology integrated an automated machine learning (AutoML) algorithm with the Analytic Hierarchy Process (AHP)-Entropy weighting model (EWM) approach is proposed to assess submarine landslide risk in the southwestern Iberian Sea, utilising morphological, geological and marine environmental features of 1552 submarine landslides. The results demonstrate that the innovative coupled methodology is more suitable for submarine landslide risk assessment. Compared to classic machine learning, ensemble learning and deep learning models, AutoML performs outstandingly in susceptibility modelling, achieving an AUC > 96%. The resulting risk map indicates that very high risk and high risk zones under the influence of landslides are mainly distributed in three canyons and two strips, respectively. In these zones, the most sensitive triggering factor is slope, while earthquakes and faults are of secondary importance. The primary threat posed by submarine landslides is to subsea cables. This assessment provides decision support for marine engineering site selection and geo-disaster early-warning.