Shoreline delineation and shoreline change detection are expensive processes in data source acquisition and manual shoreline delineation. These costs confine the frequency and interval of shoreline mapping periods. In this dissertation, a new shoreline delineation approach was developed targeting on lowering the data source cost and reducing human labor. To lower the cost of data sources, we used the public domain LiDAR data sets and satellite images to delineate shorelines without the requirement of data sets being acquired simultaneously, which is a new concept in this field. To reduce the labor cost, we made improvements in classifying LiDAR points and satellite images. Analyzing shadow relations with topography to improve the satellite image classification performance is also a brand-new concept. The extracted shoreline of the proposed approach could achieve an accuracy of 1.495 m RMSE, or 4.452m at the 95% confidence level. Consequently, the proposed approach could successfully lower the cost and shorten the processing time, in other words, to increase the shoreline mapping frequency with a reasonable accuracy. However, the extracted shoreline may not compete with the shoreline extracted by aerial photogrammetric procedures in the aspect of accuracy. Hence, this is a trade-off between cost and accuracy.
This approach consists of three phases, first, a shoreline extraction procedure based mainly on LiDAR point cloud data with multispectral information from satellite images. Second, an object oriented shoreline extraction procedure to delineate shoreline solely from satellite images; in this case WorldView-2 images were used. Third, a shoreline integration procedure combining these two shorelines based on actual shoreline changes and physical terrain properties. The actual data source cost would only be from the acquisition of satellite images. On the other hand, only two processes needed human attention. First, the shoreline within harbor areas needed to be manually connected, for its length was less than 3% of the total shoreline length in our dataset. Secondly, the parameters for satellite image classification needed to be manually determined. The need for manpower was significantly less compared to the ground surveying or aerial photogrammetry.
The first phase of shoreline extraction was to utilize Normalized Difference Vegetation Index (NDVI), Mean-Shift segmentation on the coordinate (X, Y, Z), and attributes (multispectral bands from satellite images) of the LiDAR points to classify each LiDAR point into land or water surface. Boundary of the land points were then traced to create the shoreline. The second phase of shoreline extraction solely from satellite images utilized spectrum, NDVI, and shadow analysis to classify the satellite images into classes. These classes were then refined by mean-shift segmentation on the panchromatic band. By tracing the boundary of the water surface, the shoreline can be created. Since these two shorelines may represent different shoreline instances in time, evaluating the changes of shoreline was the first to be done. Then an independent scenario analysis and a procedure are performed for the shoreline of each of the three conditions: in the process of erosion, in the process of accession, and remaining the same. With these three conditions, we could analysis the actual terrain type and correct the classification errors to obtain a more accurate shoreline.