Coastline boundaries are constantly changing due to natural or human-induced events that take place in the world. Therefore it is necessary to correctly observe coastline boundaries. Remote sensing is one of the most frequently used methods to monitor the changes in coastal areas. In this study, it is aimed to solve the problem of choosing the right method for coastal change observation. This paper introduces a spatial pixel-based and object-based image classification approach to recognize changing areas in the coastline. The coastline boundary changes occurred in a part of Yamula Dam Lake in Kayseri province were examined using three multispectral Landsat 8 satellite images of March, August and November 2016. Firstly, image-to-image registration processing was performed to register the three satellite images. Then, each satellite image was classified into two information classes either ‘Lake’ and ‘Other Field’ by using pixel-based Artificial Neural Networks (ANNs) and object-based K-Nearest Neighbor (KNN) method. Classification accuracies for ANNs method were obtained 99.97%, 99.90% and 99.80% respectively in March, August and November. As for the accuracies of the classification for the KNN method, in March, August and November were obtained 99.99%, 99.93% and 99.92% respectively. The change images were formed for March-August and August-November pairs by using the obtained classification images. The post-classification comparison method was used to determine the changes in coastline boundaries. At the end of the study, seasonal changes from water to land and from land to water were detected. According to the result of the changes, there is a 5,67 km2 increase from March to August and a 3,14 km2 decrease from August to November in Yamula Dam Lake.
Keywords: Artificial Neural Networks; k-nearest Neighbor; Change Detection; Landsat 8