Monitoring coastal erosion in Arctic regions with SAR data : Erfassung von Küstenerosion in der Arktis mit Satellitendaten
The annual coastal erosion rates of Arctic coasts are among the highest in the world, and the rates are increasing because of climate change. Monitoring of these mass movements with optical images is challenging due to frequent cloud cover of the Arctic. Synthetic Aperture Radars (SAR) are barely af...
| Main Author: | |
|---|---|
| Format: | Text |
| Language: | English |
| Published: |
TU Wien
2019
|
| Subjects: | |
| Online Access: | https://dx.doi.org/10.34726/hss.2019.53920 https://repositum.tuwien.at/handle/20.500.12708/2664 |
| Summary: | The annual coastal erosion rates of Arctic coasts are among the highest in the world, and the rates are increasing because of climate change. Monitoring of these mass movements with optical images is challenging due to frequent cloud cover of the Arctic. Synthetic Aperture Radars (SAR) are barely affected by the atmosphere, but the commonly used interferometry methods are not effective for the rapidly changing Arctic coastline. Therefore researchers introduced a backscatter-threshold based method with high-resolution TerraSAR-X images. The aim of this study was to advance SAR data analysis for coastal erosion measurements. Therefore, this study applies threshold classification to a variety of Arctic SAR images. TerraSAR-X X-band, PALSAR and PALSAR-2 L-band, and Sentinel-1 C-band ellipsoid corrected images were analyzed. The images had spatial resolutions of 0.62 to 20 m and various polarizations. The thresholds were tested with and without filtering in study areas along the Yukon Coast, the Bykovsky Peninsula, and the Barents Sea Coast. The analysis showed only weak effects of the incidence angle on the backscatter. All sample distributions were modeled well with linear threshold functions. During the error assessment all steep coast test samples were classified correctly by the threshold functions (100% producer's accuracy). Misclassification of land and water occurred for all threshold functions. Overall, the threshold functions for filtered, co-polarized images had a slightly higher classification accuracy, with Kappa Coefficients between 83.52% and 99.84%. Misclassifications were mainly caused by wet snow, wide sand beaches, and infrastructure. The classification results were further used to calculate seasonal, annual, and multi-year coastline change rates. The coastline was identified based on steep cliff classifications or the border between water and land classifications. For regions near Kay Point on the Yukon Coast, the near-zero calculated seasonal and annual shoreline change rates matched optical indications that erosion processes are not active in that area. For the west coast of Herschel Island, erosion rates calculated based on steep cliff classifications matched results of previous studies and optical images. However, the erosion estimates from land-water boundary did not match well, probably because snow interfered with the land classification. The annual and multi-year land-water rates for a region at the Barents Seas Coast showed good accordance with previous estimates. A comparison between the annual and multi-annual results based on steep coast classifications showed overall the same coastline movement tendencies. Greater differences become apparent when the results are split into smaller areas, which could be because of the high uncertainty of the annual rates or changes of the erosion processes over the years. Annual results for same regions are similar but not identical because of high rate uncertainties and possibly small differences between the chosen transects and orbit parameters of the compared images. |
|---|