Automatic coastline extraction from Landsat 9 satellite images and raster maps

Abstract

Due to its dynamic nature, coastline requires a continuous monitoring action, especially in presence of natural phenomena such as erosion and sea level rise. Earth observation from space is definitely useful for monitoring coastline variations, particularly in areas that are difficult to access. In addition to satellite images, other sources of information are maps that are very useful to support studies aimed at reconstructing the dynamics underway starting from periods even prior to remote sensing from space. This article aims to define a new methodological approach to automate the process of extracting the coastline, both from maps and satellite images, so as to eliminate the human error related to photointerpretation and manual vectorization. Experiments are carried out on Landsat 9 satellite images with cell resolution 30 m concerning the North-West of Campania Region and two raster maps in scale 1:25,000 included in the same area. The layer resulting from the application of Normalized Difference Water Index (NDWI) to the satellite dataset is submitted to K-means algorithm for distinguishing land and sea. The coastline is automatically extracted as separation between those two classes. The same unsupervised process is applied to the raster maps with the same objective. Tests on the accuracy of the resulting coastlines confirm the quality of the proposed approach.