Arctic coastline mapping with Sentinel-2 data

Abstract:

Arctic regions are one of the most rapidly changing environments on earth. Especially arctic coastlines are very sensitive to the climate change. Coastal damages can affect communities in those areas. Increasing erosion leads to higher engineering and relocation costs for coastal villages. In addition, erosion releases significant amounts of carbon, which can cause a feedback loop that accelerates climate change and coastal erosion even further. As such, detailed examination of coastal ecosystems, including shoreline types and backshore land cover, is necessary. Specifically a baseline dataset is needed for future continuous monitoring of Arctic coasts.

High spatial resolution datasets are required in order to create and validate such land cover classifications. Multispectral remote sensing has been shown to be a powerful tool for efficient retrieval of landcover. Sentinel-2 offers good spatial (10m) and temporal resolution and is expected to resolve coastal features in detail and thus enable the monitoring of large areas of the Arctic.

In this study a traditional land cover classification method (Minimum Distance Algorithm (MDA)) was compared with a more sophisticated machine learning approach using a Gradient Boosting Machine algorithm. Several study sites have been chosen along the Russian and Canadian coastline. Records from the years 2016- 2019 have been selected (depending on cloud cover). The tests over a range of arctic coast types show that machine learning is superior to traditional approaches for coastal mapping in the Arctic using Sentinel 2 data. Considered coastal landcover included bedrock, sand, sparse vegetation and shrub tundra. However, spectral confusion between classes remained a problem. For example, shadows along steep coasts were classified as water, and heterogenous areas with mixed occurrence of tundra vegetation and water bodies (polygonal tundra with ponds) were classified as bedrock.

The challenge is to attribute further information, such as coast type, automatically in order to assess the sensitivity and erosion potential of arctic coasts while capturing fundamental information for the assessment of climate change impacts and coastal processes in relation to the specificity of arctic coasts. The latter requires a known time stamp for each pixel, as several years of Sentinel-2 need to be combined to obtain cloud free mosaics for the Arctic.

The resulting coastline database is compared to existing records from regional studies available from the literature as well as to the Arctic Coastal Dynamics Database and utility for potential future applications is discussed.