A New Paradigm for Automated Ground Clutter Removal: Global Regression Filtering

Ground clutter filtering is an important and necessary step for quality control of ground-based weather radars, and this is widely done in the spectral domain via use of the discrete Fourier transform (DFT). To accomplish spectral clutter filtering, the I (in phase) and Q (quadrature) time series are multiplied by a window function, such as the Blackman window, which suppresses clutter leakage. Subsequently, the clutter is filtered from the Doppler spectrum by setting spectral coefficients to zero where there is a clutter signal. Recently, it has been shown with simulations that a regression clutter filter is a viable alternative. To demonstrate the regression filter, it is applied to radar data from both National Science Foundation-National Center for Atmospheric Research (NSF-NCAR) S-band polarimetric radar (S-Pol) and Next Generation Weather Radar (NEXRAD). A regression filter can be applied directly to the nonwindowed time series data, thus avoiding the signal attenuation from a window function, which is required for spectral domain clutter filterssuchasGaussian Model Adaptive Processing (GMAP). It is shown that the regression filter rejects clutter as effectively as the spectral technique, but has the distinct advantage that standard error of estimates of the radar variables are in general improved by about 25%–50% over a comparable spectral domain filter. The regression filter is straightforward and can be executed in real time. SIGNIFICANCE STATEMENT: Filtering of short discrete data sequences has been almost exclusively done by either finite impulse response (FIR), infinite impulse response (IIR), or spectral-based filters. Each has its drawbacks. For weather radars, spectral clutter filtering is used by the vast majority even though heavily attenuating time-domain window functions are required, which reduce data quality. Global regression filtering has been by and large ignored mainly because it has traditionally been treated and thought of as a statistical tool for estimating data trends and residuals and used for forecasting. Furthermore, digital signal processing textbooks do not address global regression as a filter with a frequency response. Regression filtering does not require data windows and also has distinct advantages over IIR and FIR filters. We believe global regression filtering is greatly underused and should have many applications for short-length sequences.