A wind tunnel experiment was carried out to investigate the movement characteristics of bed sediment grains in turbulent river flows through the partial assistance of machine learning. For gravel-bed rivers, the turbulent processes of bed-load transport was obtained by grain tracking analyses for various flow velocities. Near the critical flow velocity, about 50 % of total sediment grains moved on a bed surface in low time-probabilities. Over the critical flow velocity, in contrast, almost all sediment grains moved on a bed surface, and then, their time-probability rapidly increased. In particular, the turbulence intensities of bed sediment grain movement velocity were non-homogeneous, because the Stokes number was large. The parameter R (= streamwise turbulence intensity / spanwise turbulence intensity) varied complicatedly with an increase in the normalized Shields number τ*/τ*c, such as, (1) taking a maximum value (the critical flow : τ*/τ*c = 1.0) and decreasing rapidly, (2) increasing mildly, and (3) converging to a constant value(= about 1.85).
抄録全体を表示