This study addresses the assessment of bridge damage risks associated with heavy rainfall, focusing on landslide susceptibility and driftwood generation potential. By integrating convolutional neural networks (CNNs) with traditional machine learning methods, the research develops an advanced predictive framework for estimating driftwood accumulation at river bridges—a recognized challenge in disaster management. Concentrating on the Tokachi River basin in Hokkaido, Japan, the research utilizes diverse environmental and geographical data from authoritative sources. The findings demonstrate that the innovative approach not only enhances the accuracy of driftwood volume predictions but also distinguishes the effectiveness of CNNs compared to conventional methods. Crucially, areas prone to landslides are identified as significant contributors to driftwood generation, impacting bridge safety. The study underscores the potential of machine learning models in improving disaster risk assessment, while suggesting further exploration into real-time data integration and model refinement to adapt to changing climate conditions and ensure long-term infrastructure safety.
Read full abstract