Closed access
Author
Date
2019-12-11Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
Forests cover about 30 % of the Earth's land surface. They are an important element of the carbon and water cycle as they can act as both sinks and sources. As a result, they can influence the local and global climate. At the same time, forests are an important economic factor as they provide wood and food, and of high ecological importance as they provide habitat for numerous plants and animals. Due to its importance, the mapping, inventory and assessment of forests and their changes worldwide become a priority for scientists and governments, and is the motivation for this thesis. In particular, the thesis focuses on the study of the three dimensional (3-D) forest structure, as it is a key parameter of the forest ecosystem useful for many applications like biomass estimation, plant biodiversity or forest management and productivity.
Since decades, the most common way to characterize forest structure has been the use of single-tree information collected in situ. However, this type of measurement is typically limited to small and accessible forest areas. Remote sensing techniques overcome these limitations allowing large coverage and the collection of information without physical contact. Across the different remote sensing techniques, synthetic aperture radars (SAR) play an essential role. They provide unique ability to systematically image large areas at high spatial and temporal resolution, while at the same time they allow penetrating into forest volumes (especially at low frequencies) providing information about its 3-D structure.
SAR tomography (TomoSAR) uses multiple SAR acquisitions at slightly different positions to reconstruct the 3-D radar reflectivity, by means of imaging techniques without the need of using scattering models. However, the potential of TomoSAR configurations to extract 3-D structural information with ecological meaning has not been fully assessed yet. In this context, the first goal of this thesis is to establish a link between TomoSAR measurements and 3-D forest structure. In this sense, a framework that allows the qualitative and quantitative characterization of the 3-D forest structure from tomographic SAR profiles has been proposed. From the statistical distribution of the local maxima of the TomoSAR profiles, two indices are proposed to describe the horizontal density and the vertical complexity of a forest. This approach has been evaluated on a TomoSAR L-band data set over a temperate forest in Germany, together with the structure obtained from single-tree ground measurements and Lidar data. The results show a high correlation between the structures derived from the three sources of data, which confirms the ecological significance of the proposed method. Moreover, the same methodology has been used to demonstrate the ability of TomoSAR profiles to detect and characterize structure changes of forest stands.
The second goal of the thesis is to evaluate different TomoSAR algorithms and configurations with a focus on forest structure applications. Fourier beamforming, Capon beamforming and compressive sensing have been analyzed with simulated and real TomoSAR data at L-band. The results indicate that compressive sensing is the most appropriate one for the characterization of forest structure as proposed in this thesis, but it has the drawback of introducing sometimes false local maxima. Furthermore, it has been found that in tandem-like (bistatic) TomoSAR implementations the reconstruction of the profiles is robust to temporal changes of the scattering on a longer time span, in contrast to repeat-pass (monostatic) implementations. Finally, the requirements of TomoSAR acquisitions for distinguishing different structure types have been evaluated in terms of vertical resolution, height of ambiguity and peak side-lobe level of the point spread function. The results show that, for a limited number of images (5 to 7), a non-uniform distribution of the tracks with low vertical resolution (10 to 15 m) is the best compromise to avoid higher values of the peak to sidelobe level that would degrade the quality of the TomoSAR profiles and their interpretation.
The results of the thesis underline the potential of TomoSAR as a 3-D imaging tool for forest structure applications and contribute to better understand the TomoSAR profiles. Moreover, they open the door to further evaluate, investigate and exploit the enormous amount of TomoSAR data that the two upcoming SAR missions (ESA’s BIOMASS and DLR’s Tandem-L) will provide over all forests of the Earth. Show more
Publication status
publishedExternal links
Search print copy at ETH Library
Contributors
Examiner: Hajnsek, Irena
Examiner: Huth, Andreas
Examiner: Tebaldini, Stefano
Examiner: Ulander, Lars M. H.
Publisher
ETH ZurichSubject
Synthetic aperture radar (SAR); Tomographic imaging; Forest structureOrganisational unit
03849 - Hajnsek, Irena / Hajnsek, Irena
More
Show all metadata
ETH Bibliography
yes
Altmetrics