Neural network modeling of tidal flat terrain based on lidar survey data

Abstract

The southern yellow sea radial submarine sand ridges are in the central Jiangsu coast, where sediment dynamics is complex and the tidal ridges and channels are changing. The purpose of this paper is to model tidal flat terrain. Based on the regularity and variability characteristics of the tidal flats combined with remote sensing and LiDAR survey data, this research focuses on tidal flat terrain modeling with a neural network method. Firstly, the network structure and the parameters involved, such as weights and offset values of neurons, are determined by the BP Neural Network calculation using the 2006 LiDAR DEM in this area. Secondly, the characteristic lines, which are boundary lines of tidal basins, skeleton lines of tidal creeks and a series of waterlines, and so on are extracted from TM images of the no-data region similar to the area of study. Combining with survey data, the elevation data of characteristic lines are obtained. At last, the terrain of the region without elevation data is generated by the model. The test shows the terrain calculated by the model is very close to the surveyed terrain. The residual distribution is normal. The study is significant in getting a dynamic tidal flat terrain fast and efficiently.