Abstract: To solve the problem of dynamic extraction for tidal-flat structure caused by periodic tide level and long sequences of human activities,the paper combines the tide level with the spectral index quartile features to construct an algorithm for monitoring the dynamics of tidal-flat structure.With utilization of MNDWI and NDWI mean ordering of the training samples for Landsat images,the quality mosaic and medium mosaic are used to fuse the image stacks into tide level images,and utilized the Random Forest classification algorithm to identify the spatio-temporal dynamics of tidal-flat in Bay from 1988 to 2021.The results shown that the average overall accuracy and Kappa coefficient of the algorithm are 0.85 and 0.78,respectively,realizing the simultaneous extraction of the dynamic structure and transitions of tidal-flat.On the whole,the tidal-flat in Shenzhen Bay show a dynamic pattern of "narrow in the north and wide in the south,shrinking in the west and expanding in the east"; the northern construction land in the reclaimed land area and the artificial greenland are massively expanded to the supratidal and intertidal zones; the eastern development of tidal-flat in the ecological protection zone has prompted the seaward extension of the beaches and the mangroves; and the southern beaches in the potential development zone and the water bodies show a fluctuating change pattern.The development of beaches and water bodies in the southern potential development zone showed trade-off dynamics.