Arctic Thin Sea Ice Thickness Regression Models for Sentinel-2

• Main Author: Skogvold, Øystein Fredriksen
• Format: Master Thesis
• Language: English
• Published: UiT Norges arktiske universitet 2019
• Subjects: VDP::Mathematics and natural science: 400::Physics: 430::Electromagnetism; acoustics; optics: 434; VDP::Matematikk og Naturvitenskap: 400::Fysikk: 430::Elektromagnetisme; akustikk; optikk: 434; VDP::Mathematics and natural science: 400::Mathematics: 410::Analysis: 411; VDP::Matematikk og Naturvitenskap: 400::Matematikk: 410::Analyse: 411; VDP::Mathematics and natural science: 400::Mathematics: 410::Statistics: 412; VDP::Matematikk og Naturvitenskap: 400::Matematikk: 410::Statistikk: 412; Remote Sensing; Multispectral data; Thin sea ice thickness; Regression; Machine Learning; FYS-3931; Arctic; Sea ice
• Online Access: https://hdl.handle.net/10037/15758

Sea ice thickness is an important parameter for modelling the sea ice mass balance, momentum and gas exchanges, and global energy budget. The interest of studies into thin sea ice has increased as trends in recent years show a increasing abundance in thin first year ice. Existing thin sea ice thickness products operate at resolutions down to 750 meters. Very high resolution (less than 100 meters) retrieval of sea ice parameters is of particular interest due to maritime navigation and model parametrization of physical processes at meter-scaled resolutions that usually requires in-situ measurements. The Norwegian Meteorological Institute provided a 500 meter resolution thin sea ice thickness product developed by the Norwegian Computing Centre for the Norwegian Space Agency’s "Sentinel4ThinIce" project. The product is derived from Sentinel-3’s SLSTR sensor. Using overlapping multispectral optical data from Sentinel-2’s MultiSpectral Instrument at metre-scaled resolutions, we retrieved multiple regression models for thin sea ice thickness for Sentinel-2 data. The models included three univariate models for three different spectral band combinations using non-linear least squares method, and one multivariate model for three different band reflectance data-sets using a gradient boosting regression tree. The optical band reflectance data increased monotonically with sea ice thickness and saturated for thicker ice, proving a clear correlation between thin sea ice thickness and Sentinel-2’s band reflectance. The multivariate model produces overall best results compared to the univariate models. The reliability of the models couldn’t be trusted due to inaccurate atmospheric correction procedures and not enough temporal and geographical variance in the data-set. Proper calibration of Sentinel-2 data is of high priority in order to extend Sentinel-2’s platform further into Arctic research.