Open Collections

Refining low-quality digital elevation models using synthetic aperture radar interferometry

University of British Columbia

Two-pass synthetic aperture radar (SAR) interferometry (InSAR) is a technique for processing the phase difference between coincident SAR images to obtain the range difference from the two radars to a common point on the earth's surface. The accuracy of the range difference measurement is in the order of one millimeter, and this range information can be processed to obtain digital elevation models (DEMs) of the surface topography. The digital processing required to make the DEM is quite complicated, mainly due to two factors. Firstly, the phase information is obtained from complexvalued data and therefore lies between -π and +π whereas the complete phase information is needed. To obtain this, the phase must be "unwrapped" where the missing integer number of 2π are estimated for each data sample. Secondly, the geometry of the satellite passes relative to each other must be known to an accuracy of a few millimeters in order to obtain the surface height values to the required accuracy (about 10 m). Both of these steps require supplemental information and manual guidance to be performed correctly. Phase unwrapping is difficult because of noise and undersampling inherent in the measurements. The geometry estimates are difficult to make because the orbit is only known to an accuracy of a few meters, and the received phase data is a non-linear function of the satellite geometry. In the past, the geometry estimates have been made using known ground control points (GCPs), which requires a considerable manual effort, and has its own set of errors. The objective of this thesis is to use supplemental information in the form of a coarse DEM to make the InSAR processing more accurate and more automatic. We achieve this objective by developing a new algorithm which incorporates the coarse DEM directly into the processing stream, with the result that phase unwrapping and geometry estimation are performed accurately and reliably. In effect, the input DEM points serve as a large, dense set of GCPs. While the accuracy of each input DEM point is not very high, the large number of them provide adequate geometric accuracy, particularly as an automatic algorithm can register them directly to the radar data. There are two key steps in the new algorithm, which are interwoven in an iterative framework. First of all, the satellite geometry is estimated from the DEM and interferometric phase. This is done with a non-linear, iterative optimization algorithm without having to unwrap the phase. Avoiding phase unwrapping is important, as phase unwrapping errors can significantly bias the geometry estimates. Second, the input DEM along with the refined satellite geometry are used to create a model of the unwrapped interferogram phase that should be received from the two satellite passes. When this phase is wrapped, and compared with the measured phase, a differential interferogram is obtained which represents the difference between the coarse input DEM and the topography as measured by the satellite. This differential interferogram has a relatively low bandwidth, which means that it can be filtered and unwrapped reliably and accurately. Finally, the information in the unwrapped interferogram is used to refine the grid spacing and vertical accuracy of the coarse DEM. We have used mathematical analysis and simulation to develop the algorithm, to obtain statistical quality measures and to understand what system parameters affect the accuracy of the DEM results. We find that the main factors affecting accuracy are the interferometer's sensitivity of phase to height and the number of available DEM points, including the size and variability of the input DEMs' errors. We have successfully applied the DEM refinement algorithm to ERS Tandem Mission and RADARS AT-1 data. The generated InSAR DEMs had standard deviations of 12 to 20 meters compared to a control DEM with approximately 3 meters standard deviation. The output InSAR-enhanced DEMs had two to four times improvement in height accuracy compared with the input DEMs. In this way we have demonstrated that one can generate reliable estimates of topography for standard SAR scenes without having access to precision orbit data. Thus we have shown that the processing bottlenecks in dealing with repeatpass satellite InSAR data can be overcome, and useful topographic information can be obtained from the vast supply of existing InSAR data sets.

Thesis/Dissertation

application/pdf

2009-07-03

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/10094

Electrical and Computer Engineering

University of British Columbia

UBCV

DSpace