The tidal inundation frequency method is commonly employed for rapid remote sensing acquisition of large-scale and high-precision tidal flat topography. However, in certain areas of tidal flats, the natural tidal inundation is partially obstructed by artificial structures, resulting in incomplete correspondence between the inundation frequency and the topography. Therefore, this study proposes the frequency-distance joint retrieval (FDJR) approach to improve the tidal inundation frequency method. The FDJR approach derives the local multi-year lowest tidal level line, characterized by an inundation frequency of 1, from the inundation frequency image. It further incorporates the landward distance between this line and the tidal flat point into the regression model, thereby effectively capturing the overall monotonic variation in the elevation of the tidal flat profile. Taking the Yellow River Delta (YRD) as an example, the retrieval of tidal flat topography was conducted by integrating Sentinel-2 remote sensing images, ICESat-2, ATL08 elevation products and RTK measured terrain. The results show that there exists a high correlation among the landward distance, inundation frequency and elevation. Moreover, the machine learning regression model based on Random Forest and Bagging algorithm effectively characterizes this relationship. Compared with the tidal inundation frequency method, the proposed approach achieves an increase in training accuracy of approximately 30 %-40 % while providing more concentrated retrieval elevation distribution with smoother kernel density curves. The FDJR approach proves applicable to the tidal flats where natural tide inundation is partially obstructed and provides topographic reference for coastal zone studies.
Read full abstract