Mapping Retrogressive Thaw Slumps Using Single-Pass TanDEM-X Observations

Vast areas of the Arctic host ice-rich permafrost, which is becoming increasingly vulnerable to terrain-altering thermokarst in a warming climate. Among the most rapid and dramatic changes are retrogressive thaw slumps. These slumps evolve by a retreat of the slump headwall during the summer months,...

Full description

Bibliographic Details
Main Authors: Bernhard, Philipp, Zwieback, Simon, Leinss, Silvan, Hajnsek, Irena
Format: Article in Journal/Newspaper
Language:English
Published: IEEE 2020
Subjects:
Abrupt thaw
Banks island
Digital elevation model (DEM) differencing
DEM generation
Interferometry
Mackenzie river delta
Permafrost
Remote sensing
Retrogressive thaw slumps (RTSs)
Single-pass radar interferometry
Synthetic aperture radar (SAR)
Thermokarst
Arctic
Canada
Mackenzie River
Banks Island
Ice
Mackenzie river
permafrost
Online Access:https://hdl.handle.net/20.500.11850/426695
https://doi.org/10.3929/ethz-b-000426695
Description
Summary:Vast areas of the Arctic host ice-rich permafrost, which is becoming increasingly vulnerable to terrain-altering thermokarst in a warming climate. Among the most rapid and dramatic changes are retrogressive thaw slumps. These slumps evolve by a retreat of the slump headwall during the summer months, making them detectable by comparing digital elevation models over time using the volumetric change as an indicator. Here, we present and assess a method to detect and monitor thaw slumps using time series of elevation models applied on two contrasting study areas in Northern Canada. Our two-step method is tailored to single-pass InSAR observations from the TanDEM-X satellite pair, which have been acquired since 2011. For each acquisition, we derive a digital elevation model and uncertainty estimates. In the first step, we difference digital elevation models and detect the significant elevation changes using a blob-detection algorithm. In the second step, we classify the detections into those due to thaw slumps and other causes using a simple thresholding method (accuracy: 78%), a random forest classifier (87%), and a support vector machine (86%). When our method is applied to other areas, the classifiers should be trained with data from part of the study area or with data obtained from similar areas in terms of topography, vegetation, and thaw slump characteristics to achieve the best performance. The obtained locations of thaw slumps can be used as a starting point to extract important slump properties, such as the headwall height and the volumetric change, which are currently not available on regional scales. ISSN:1939-1404 ISSN:2151-1535

Similar Items