Mapping permafrost landscape features using object-based image classification of multi-temporal SAR images

Microwave imagery has a distinct advantage over optical imagery in high-latitude areas because it allows data to be acquired independently of cloud cover and solar illumination. Synthetic aperture radar (SAR)-based monitoring has become increasingly important for understanding the state and dynamics of permafrost landscapes at the regional scale. This study presents a permafrost landscape mapping method that uses multi-temporal TerraSAR-X backscatter intensity and interferometric coherence information. The proposed method can classify permafrost landscape features and map the two most important features in sub-arctic permafrost environments: permafrost-affected areas and thermokarst ponds. First, a land cover map is generated through the combined use of object-based image analysis (OBIA) and classification and regression tree (CART) analysis. An overall accuracy of 98% is achieved when classifying rock and water bodies, and an accuracy of 79% is achieved when discriminating between different vegetation types with one year of single-polarized acquisitions. Second, the distributions of the permafrost affected areas and thermokarst ponds are derived from the classified landscapes. Permafrost-affected areas are inferred from the relationship between vegetation cover and the existence of permafrost, and thermokarst pond distributions are directly inherited from the land cover map. The two mapped features exhibit good agreement with manually delineated references. The proposed method can produce permafrost landscape maps in complex sub-arctic environments and improve our understanding of the effects of climate change on permafrost landscapes. This classification strategy can be transferred to other time-series SAR datasets, e.g., Sentinel-1, and other heterogeneous environments. (C) 2018 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.