PhD Thesis
Improved intermittent clutter filtering for wind profiler radar
Volker Lehmann (05/2008-05/2010)
Support: Thomas Foken
Ground-based remote measurements of the vertical profile of the horizontal wind vector in the atmosphere by radar wind profiler (RWP) is a technique that has been significantly developed since the first demonstration with the Jicamarca radar by Woodman and Guillen in the early 1970s. Currently, there exist several operational networks of those instruments in the USA, Europe and Japan which provide continuous wind measurements in real-time and most of the data are successfully assimilated in numerical weather prediction models. Although this is an obvious indication of maturity, practical experience has shown that further improvements are both possible and necessary. While the high sensitivity of these clear-air radars is required for receiving the weak atmospheric echoes, it makes them also particularly vulnerable to unwanted radar returns and in-band radio frequency interference. Signal processing must therefore especially deal with the problem of filtering of these unwanted contributions, to avoid associated measurement errors.
A specific difficulty are clutter echoes from various airborne objects, such as aircraft or birds, which generate strong, intermittent contributions to the received signal. The standard RWP signal processing is not able to deal with these signals in an efficient way, because the model assumption on which the processing is based is violated. With the development of sophisticated mathematical tools for the analysis of non-stationary signals in the last two decades and a better understanding of the practically relevant RWP clutter issues, a number of efforts have been made to tackle especially the challenging problem of intermittent clutter returns from migrating birds.
In this dissertation it is shown that the signal structure of RWP raw data contaminated by intermittent clutter is much clearer revealed by a joint time-frequency analysis based on the windowed Fourier transform than by other possible signal descriptions, in particular pure time or frequency representations. An effective intermittent clutter reduction algorithm, called the Gabor filter, is obtained by a combination of a numerically feasible discrete Gabor frame expansion with the statistical test for a stationary Gaussian random signal. This approach is optimized by using near-tight frames and selecting a time-frequency resolution that provides a jointly sparse representation of both atmospheric and clutter signal components. A first evaluation of this approach has shown a superior performance in comparison with hitherto existing methods, but it was also found that additional quality-control of the derived Doppler spectra is still required during extreme bird migration events. The latter is in all likelihood indicative of a principal limit of radar wind profiling during such conditions. However, an effective quality control of the measurement is possible through a combination of a stationarity estimate provided by the Gabor algorithm with a-priori information about typical atmospheric echoes.