A comparison between optical and SAR classification results for thin sea ice in Storfjorden

• Main Author: Quigley, Cornelius
• Format: Master Thesis
• Language: English
• Published: UiT Norges arktiske universitet 2017
• Subjects: VDP::Mathematics and natural science: 400::Mathematics: 410; VDP::Matematikk og Naturvitenskap: 400::Matematikk: 410; VDP::Mathematics and natural science: 400::Physics: 430::Electromagnetism; acoustics; optics: 434; VDP::Matematikk og Naturvitenskap: 400::Fysikk: 430::Elektromagnetisme; akustikk; optikk: 434; Synthetic Aperture Radar; Sea ice; Earth observation; classification; FYS-3900; Arctic; Climate change
• Online Access: https://hdl.handle.net/10037/11895

According to the scientific consensus, the Arctic is currently in a state of unprecedented change. In recent years, climate change has been identified as the main cause of Arctic sea ice decline. For this reason, the need to have access to timely and cost effective data is of great importance. Remote sensing via Earth orbiting satellites offers extensive data coverage in areas that are typically inaccessible due to their remote location and extreme weather conditions. As well as this, having knowledge of sea ice conditions aids in exploration and navigation. This thesis is concerned with classifying thin sea ice in Storfjorden using data acquired from both MODIS and Radarsat-2 in order to determine if data from either sources can be considered complimentary to each other. For this purpose, four comparisons were made. These included comparing MODISs 36 band data set with data from Radarsat-2 ScanSAR Narrow and Wide modes. As well as this, a comparison between MODISs 36 band data and data from Radarsat-2 QuadPol mode was made. HEM thickness measurements are also available from a helicopter campaign around the same time the data was taken. From laser altimeter data that accompanied the thickness measurements, a roughness characteristic was derived that was compared against the HEM measured thicknesses. All MODIS data were screened for corrupted bands. The resulting bands were transformed into a new space via Principal Component Analysis (PCA). The first few components that contained most of the variance of the transformed data set were kept for segmentation. The SAR data was multilooked and feature extracted. The features that were chosen are a set of six basic features that have shown reasonably good results in the segmentation of sea ice previously and are known as the Extended Polarimetric Feature Space (EPFS). This set of features is composed of five polarimetric features plus a feature for non-Gaussianity. All features were segmented using a Mixture of Gaussian algorithm with Markov Random Field based ...