Application of Decision Tree Algorithm on Landsat 9 OLI-2 Images for Coastline Extraction

The extraction of the coastline from aerial and satellite images constitutes a basic task of remote sensing that finds a powerful operational tool in Machine Learning techniques. The various algorithms present in the literature, such as K-Means, Decision Tree (DT), Support Vector Machine, can be applied directly to one of the available multispectral bands or to a combination of them; alternatively, two or more bands can be previously processed using specific indices aimed at highlighting the different spectral response of water pixels compared to others of a different nature, i.e. vegetation and/or bare soil, present in the analyzed scene. This paper aims to verify the effectiveness of the DT algorithm applied to satellite Landsat 9 OLI multispectral imagery concerning a large part of the Tyrrhenian Calabrian coast (Italy). Specifically the following datasets are considered: Near Infrared (NIR) band, RGB true color composition (RGB), combination of RGB and NIR (RGB+NIR), Normalize Difference Vegetation Index (NDVI), Normalize Difference Water Index (NDWI), Modified Difference Water Index (MNDWI), SWIR Minus Blue Index (SMBI). DT is run on MATLAB, while all remaining operations are performed using Q-GIS software. The extracted coastlines are compared with the reference one resulting from manual vectorization to establish the most performing approach. The best result is derived by DT applications to MNDWI.