GIS based analysis and accuracy assessment of low-resolution satellite imagery for coastline monitoring

It is well known that in order to study the shoreline evolution in an accurate way, successive time series of high-resolution images are needed. However, the availability of historical high-resolution data is limited to analogue air photos while the more recent high-resolution satellite images are quite expensive. Thus, many researchers have based their coastal evolution studies on low-resolution images like Sentinel and Landsat data. As a result, a main question is raised about the accuracy of the results of such studies. As the Landsat program has started in 1972, there is archival data spanning almost 50 years. This is a great tool for long-term shoreline change monitoring in researchers' hands. Another advantage of the specific data set is the existence of diverse multispectral bands and the opportunity to extract band ratios sensitive to existence or not of water. The biggest challenge for using this archive for shoreline monitoring is its limited spatial resolution (30m). In the current study, the accuracy of low-resolution satellite data such as Sentinel-2 MSI and Landsat TM, ETM+ and 8 OLI for coastal monitoring is under control. Many low-resolution images were digitally processed, and Normalized Difference Vegetation Index (NDVI) as well as on-screen digitizing are used in order to automatically and manually separate the sea from the land respectively and extract the coastline in different periods. Then, the shoreline vectors derived from the Landsat and Sentinel-2 data were compared to the respective shorelines derived from high resolution satellite data such as Worldview-2 (0.5 m resolution) and orthomosaics created from digital airphotos with 1m spatial resolution. The study area is located in the Gulf of Patras in the North Peloponnese, Greece. The accuracy control covered three different periods from 1996 to 2018 and in every case, shorelines extracted from low resolution data sets were compared to shorelines created from high resolution data sets. Statistical analysis was performed, and the results are presented and discussed in the current paper.