Electromagnetic backscatter modelling of icebergs at c-band in an ocean environment

• Main Author: Ferdous, Md. Saimoom
• Format: Text
• Language: English
• Published: Memorial University of Newfoundland 2020
• Subjects: Newfoundland; Canada; The Sentinel
• Online Access: https://dx.doi.org/10.48336/eqhw-sf10; https://research.library.mun.ca/14454/

This thesis outlines the development of an electromagnetic (EM) backscatter model of icebergs. It is a necessary first step for the generation of in-house synthetic aperture radar (SAR) data of icebergs to support optimum iceberg/ship classifier design. The EM modelling was developed in three stages. At first, an EM backscatter model was developed to generate simulated SAR data chips of iceberg targets at small incidence angles. The model parameters were set to mimic a dual polarized dataset collected at C-Band with the Sentinel-1A satellite. The simulated SAR data chips were compared with signatures and radiometric properties of the satellite data, including total radar cross section (TRCS). A second EM model was developed to mimic the parameters of a second SAR data collection with RADARSAT-2; this second data collection was at larger incidence angles and was fully polarimetric (four channels and interchannel phase). The full polarimetric SAR data allowed for a comparison of modelled TRCS and polarimetric decompositions. Finally, the EM backscatter models were tested in the context of iceberg/ship classification by comparing the performance of various computer vision classifiers using both simulated and real SAR image data of iceberg and vessel targets. This step is critical to check the compatibility of simulated data with the real data, and the ability to mix real and simulated SAR imagery for the generation of skilled classifiers. An EM backscatter modelling tool called GRECOSAR was used for the modelling work. GRECOSAR includes the ability to generate small scenes of the ocean using Pierson-Moskowitz spectral parameters. It also allows the placement of a 3D target shape into that ocean scene. Therefore, GRECOSAR is very useful for simulating SAR targets, however it can only model single layer scattering from the targets. This was found to be limiting in that EM scattering throughout volume of the iceberg could not be generated. This resulted in EM models that included only surface scattering of the iceberg. In order to generate realistic SAR scenes of icebergs on the ocean, 3D models of icebergs were captured in a series of field programs off the coast of Newfoundland and Labrador, Canada. The 3D models of the icebergs were obtained using a light detection and ranging (LiDAR) and multi-beam sonar data from a specially equipped vessel by a team of C-CORE. While profiling the iceberg targets, SAR images from satellites were captured for comparison with the simulated SAR images. The analysis of the real and simulated SAR imagery included comparisons of TRCS, SAR signature morphology and polarimetric decompositions of the targets. In general, these comparisons showed a good consistency between the simulated and real SAR scene. Simulations were also performed with varying target orientation and sea conditions (i.e., wind speed and direction). A wide variability of the TRCS and SAR signature morphology was observed with varying scene parameters. Icebergs were modelled using a high dielectric constant to mimic melting iceberg surfaces as seen during field work. Given that GRECOSAR could only generate surface backscatter, a mathematical model was developed to quantify the effect of melt water on the amount of surface and volume backscatter that might be expected from the icebergs. It was found that the icebergs in a high state of melt should produce predominantly surface scatter, thus validating the use of GRECOSAR for icebergs in this condition. Once the simulated SAR targets were validated against the real SAR data collections, a large dataset of simulated SAR chips of ships and icebergs were created specifically for the purpose of target classification. SAR chips were generated at varying imaging parameters and target sizes and passed on to an iceberg/ship classifier. Real and simulated SAR chips were combined in varying quantities (or targets) resulting in a series of different classifiers of varying skill. A good agreement between the classifier’s performance was found. This indicates the compatibility of the simulated SAR imagery with this application and provides an indication that the simulated data set captures all the necessary physical properties of icebergs for ship and iceberg classification.