Analysis of Riverbed Sediment Distribution and Key Contributing Factors: A Deep Learning-based Approach

초록

This study quantitatively analyzed the interactions between riverbed sediment distribution and key geomorphological and riverine variables in Korea’s four major river basins (Geum, Nakdong, Yeongsan, and Han Rivers) using deep learning models integrated with attention mechanisms. Traditional statistical and theoretical approaches often failed to account for the complex nonlinear interactions between variables and regional differences. To address these limitations, this study modeled the relationships between five key geomorphological variables—river length, topographic elevation, slope, flatness, and channel width—and the thicknesses of clay, sand, and sand-gravel layers, systematically evaluating the factors influencing sediment formation and distribution. The results revealed that river length is the most critical factor determining sediment distribution across all basins, while its effects were modulated by unique basin-specific characteristics, such as elevation, slope, flatness, and channel width. In basins with longer rivers, reduced flow velocity and fine sediment accumulation led to thicker clay and sand layers downstream. In contrast, upstream regions with higher elevations and steep slopes exhibited faster flow velocities, promoting the formation of sand-gravel layers. The attention mechanism employed in this study enhanced model interpretability by visually identifying the relative importance of variables during predictions, addressing the “black-box” nature of deep learning models, and enabling deeper insights into variable interactions. These findings underscore the role of geomorphological factors, such as river length, slope, and flatness, in shaping distinct sedimentary characteristics across basins and provide a scientific basis for developing region-specific river management and restoration strategies that reflect local conditions.

키워드