Derivation of Environmental Parameters of Arctic Tundra Landscapes with SAR


Contact
Jennifer.Sobiech [ at ] awi.de

Abstract

Climate warming twice as large as the global average has been observed in the Arctic. In regions underlain by permafrost, the warming leads to changes in the timing of snowmelt and active layer thaw onset towards earlier dates in the year and later dates for the refreeze in fall. This results in larger active layer depth, which in turn results in changes of the hydrology and vegetation cover. Lakes and rivers show earlier ice-off and later refreeze dates. Remote sensing is a valuable tool to observe the state of the surface in near real time. Sensors operating with microwaves are especially eligible, as the are - in contrast to optical sensors - able to penetrate clouds and to work independent of sunlight. Synthetic Aperture Radars (SAR) are active microwave systems, which send a pulse to the earth and then record the backscattered energy. SAR systems can operate at different frequencies and send and receive the pulses either in horizontal (H) or vertical (V) polarizations. The purpose of this work was to validate the potential of imaging radar to derive environmental parameters in Arctic tundra landscapes at the example of the Lena River Delta in northern Siberia. The area is composed of three different tundra ecosystems and characterized by a high amount of lakes and river channels. A time-series of 12 HH-polarized TerraSAR-X (9.65 GHz) and 20 quad-polarized RADARSAT-2 (5.405 GHz) images, acquired from April to November 2011, were available for this study. During an expedition to the Lena Delta in summer 2010, extensive mapping of the vegetation and the moisture of the top soil layer was performed. Four automatic weather stations (AWS) were set up to record the climate conditions in the Lena Delta all year round. This field data were used to support the analyses of the SAR images. With regard to the observed warming of the Arctic, the study concentrated on the detection of the shifts of the thermal state (thaw and refreeze) of the soils and the water bodies as well as the timing of the snowmelt and the extension of the snow cover during spring. For correct interpretation of the SAR data, the scattering mechanisms of the radar waves at the surface were investigated. First, the timing and the duration of the ice-decay on lakes and river channels were determined, which is important as the albedo and heat fluxes change with melt onset, and evaporation and the exchange of further gases rise. Several methods for the classification of ice and water were applied to the SAR data. It was found that the unsupervised k-means classification followed by a morphological closing filter is preferable for the classification of ice and water fractions on lakes, while the application of a low-pass filter followed by the k-means classification is the best method for river channels. Ice-melt in 2011 was first visible in the TerraSAR-X image acquired on June, 6th. Total ice-off occurred after June, 26th. Since the presence or absence of snow highly influences the albedo and since a snow cover has an insulating effect between the soils and the atmosphere, the spatial extend of the snow cover and the timing of the snowmelt in spring have a high impact on the ecosystem. These parameters were investigated using the SAR images acquired during spring. Here, two snow-melt events were detected. The first snowmelt took place at the end of April, the second event in mid May. These events were identified via change detection of the backscatter intensities in the radar images and are in agreement with the climate data recorded at the AWSs. A polarimetric decomposition of the RADARSAT-2 images in the parameters entropy, anisotropy and alpha angle was used to map the spatial extent of the melting snow. This was successful, as the scattering mechanisms are different between wet snow and wet soils. With active layer melt onset and refreeze, the energy cycle in the tundra is almost inverted. Thus it is important to know the dates. With melt onset, the backscatter intensity rises by 3 to 5 dB, with refreeze the intensities drop. Melt onset was identified via change detection from the SAR images at May, 25th and refreeze at October, 10th, which agrees with the in-situ observations. Land cover classification is necessary as thresholds of the backscattering intensity need to be set for each landscape unit separately (for example to flag freeze-thaw events). At the test site, land cover classification according to the different geomorphologic units was impossible with single-polarized data, whereas good classification results were achieved with quad-polarized data. In this case, the phase difference, the entropy, anisotropy and alpha values were taken into account as well. To further support information useful for the upscaling of energy and water vapor fluxes, the potential of X- and C-band SAR to monitor moisture changes during summer was investigated. No significant backscatter changes during summer could be observed, although SAR data acquired under different weather conditions were present. Thus, X- and C-band SAR data are not suitable to deliver information on moisture changes in these special environments. Altogether, the results give a comprehensive overview on the potential of X- and C-band SAR data to derive environmental parameters in Arctic tundra landscapes.



Item Type
Thesis (PhD)
Authors
Divisions
Primary Division
Programs
Primary Topic
Peer revision
Not peer-reviewed
Publication Status
Published
Eprint ID
33332
Cite as
Sobiech, J. (2012): Derivation of Environmental Parameters of Arctic Tundra Landscapes with SAR , PhD thesis, Westfälische Wilhelms-Universität Münster, Institut für Landschaftsökologie.


Share

Research Platforms

Campaigns


Actions
Edit Item Edit Item