Detection and characterization of icebergs in Kongsfjorden (Svalbard) based on ground-based radar images and additional remote sensing data

• Main Author: Linge, Stefan
• Other Authors: Lauknes, Tom Rune, Rouyet, Line, Insinööritieteiden korkeakoulu, Rautiainen, Miina, Aalto-yliopisto, Aalto University
• Format: Master Thesis
• Language: English
• Published: 2019
• Subjects: GPRI; detection; iceberg; remote sensing; satellite SAR; UAV; Blomstrandhalvøya; Ny-Ålesund; Svalbard; Kongsfjord*; Kongsfjorden; Ny Ålesund
• Online Access: https://aaltodoc.aalto.fi/handle/123456789/39821

This thesis focuses on the exploitation of ground-based radar images to detect icebergs. Additional remote sensing data from space-borne Synthetic Aperture Radar (SAR), Unmanned Aerial Vehicle (UAV) and in-situ boat tracks has been used to compare and validate the results. The investigation site is located in Kongsfjorden (Svalbard) and the combined data acquisition took place during a two-week campaign in April 2018. Five tidewater glaciers terminate in Kongsfjorden and produce a large number of icebergs of different sizes and shapes. The ground-based radar had an elevated position in Ny-Ålesund to overview a several kilometer-wide section of the fjord. The ground-based radar used during the campaign is the GAMMA Portable Radar Interferometer (GPRI). The 5 min temporal resolution of the dataset allows one to make comparisons with the above mentioned auxiliary remote sensing data. The software Python was used to process the GPRI data. Firstly, the GPRI images were pre-processed to account for the decreasing performance in range resolution. Secondly, an area of interest located between Ny-Ålesund and Blomstrandhalvøya was chosen. Hereby, it is important to focus only on the sea region and to leave out lagoons and other coastal lines. The area of interest covers approximately 2 km long region with only water and icebergs passing by while leaving Kongsfjorden. Thirdly, a threshold was applied to the GPRI images in order to separate potential icebergs from the sea background. Analysing histograms of both iceberg and sea background is important to find the appropriate threshold. This makes sure to include as many true positive as possible. In general, we can choose between two threshold modes, namely the automated and the manual threshold methods. The automated threshold method relies on the 99.93th percentile and shows the best compromise between all GPRI images. The automated threshold method is efficient and preferably used for big amounts of data and small time slots, because one loses small icebergs or detect ...