Radar signal quality improvement by spectral processing of dual-polarization radar measurements

Transcription

1 Radar signal quality improvement by spectral processing of dual-polarization radar measurements Dmitri Moisseev, Matti Leskinen and Tuomas Aittomäki University of Helsinki, Finland, 1. Introduction Next to calibration, clutter mitigation is the second oldest problem in radar remote sensing. Introduction of Doppler spectral processing has dramatically increased quality of radar measurements, by filtering out ground clutter. Nonetheless, not all clutter types are easy to suppress. Doppler properties of sea clutter, birds and insects, chaff and RF interferences are significantly different from ground clutter and challenge standard Doppler clutter filters. To mitigate those types of clutter a new approach is needed. In this study, spectral analysis of dual-polarization radar measurements is applied to improve quality of radar observations (Moisseev et al., 2000, 2002; Unal and Moisseev, 2004; Moisseev and Chandrasekar, 2009; Bachmann and Zrnić, 2007). The proposed approach gives a greater flexibility for discrimination between different types of scattering sources present in a radar observation volume. The proposed spectral filter design uses a fuzzy logic classification algorithm applied to textures of spectral differential phase and reflectivity and spectral decomposition of co-polar correlation coefficient (Moisseev and Chandrasekar, 2009). Then by rejecting spectral lines that are affected by non- weather signals the clutter filter is defined. The performance of this filter is demonstrated on University of Helsinki C-band radar observations. It is shown that the resulting filter parameters are adaptively defined for each range gate and do not require an assumption on spectral properties of clutter. By comparing to a traditional notch filter it is demonstrated that the new spectral filter improves ground clutter suppression rate and rejects other types of unwanted signals. The gain in the clutter suppression rate can be attributed to the rejection of all parts of spectra that are affected by clutter, and not only filtering around zero frequency. This property of the proposed methodology mitigates effect of system instabilities, i. e. phase noise, which spread clutter to other Doppler frequencies. Moreover, it is demonstrated, on an example of co-polar correlation coefficient observations, that estimates of polarimetric parameters are also improved. 2. Methodology The proposed filtering methodology is based on spectral analysis of dual-polarization radar signals. For this purpose, range-doppler spectrographs are calculated for each radial, where spectra for each range gate are estimated by applying discrete Fourier transform to the windowed correlation functions computed from time-sequence observations. For these studies we use a Chebyshev lag window of length 64 and the side lobe suppression level of 50 db. From calculated crossand power spectra the co-polar coherency spectrum, and textures of differential phase and differential reflectivity are computed. Prior to the co-polar coherency spectrum calculation, the power spectral densities are noise corrected by subtracting system noise spectral density. This step is carried out to minimise effect of low SNR on spectral decomposition of the co-polar correlation coefficient of a precipitation signal. The results of those computations are shown in Figure Fuzzification The proposed classification procedure uses three inputs, i.e. SDZ dr v, SD Ψ dp v and ρ hv v, and produces three classes, namely precipitation, noise and clutter. The fuzzification of the input radar measurements is carried out by using membership functions of Beta functional form. The parameters of the membership functions are defined using observed distributions. The precipitation membership functions are extended over a greater range of values to include variability of the radar observations associated with changes in SNR and precipitation type. 2.2 Inference The inference process architecture is shown in Figure. 2. The scheme is based on additive weighting method, where the rule strength for j-th class, R j, can be written as: R j =w 1j v m j SD Z dr v +w 2j v m j SDY dp v +w 3j v m j ρ hv v (1) where m j is the proposition strength for a class j and w ij v is the weight factor. The weight factors are Doppler velocity dependent and are introduced to slightly force the inference process towards the clutter class in cases where Doppler velocity is close to zero. The weight function for noise and precipitation classes are independent of Doppler velocity and fixed at the weight value of 0.5, as shown in Figure 3. The clutter weight function is defined as: w x2 v =0.4 1 s2p exp v 2 2s, 2 (2) where s =0.4 ms -1

2 Figure 1. Top three panels show power (reflectivity), differential reflectivity and differential phase spectra. The lower three panels show co-polar coherency spectrum and textures of differential reflectivity and phase spectrographs. The clutter weight function mimics an expected clutter spectrum shape in the region around zero Doppler velocity and is constant outside this region. The constant part of the weight function is introduced to include such effects as a phase noise that can spread clutter signal to other frequencies. Figure 2. The inference process architecture. At the final stage of the inference process a class with maximum rule strength for each cell in the range-doppler domain is found. That results in assigning one class out of three to each range-doppler cell. An example of the

3 Figure 3. The weight factor, defined as a function of Doppler velocity. classification is shown in Figure. 4. This figure is obtained from one ray of data collected on Oct 16, 2006 by CSU-CHILL radar. One can see that the classification scheme correctly detects three classes. Also it is important to note that not all cells around zero Doppler frequency identified as clutter. For example for ranges larger than 50 km, where the precipitation echo has zero frequency components, clutter is detected only for cells that are affected by clutter. This is an important result that demonstrates that the proposed methodology allows for adaptive clutter filtering. 2.3 Filtering At the final step of the procedure weather echo spectral moments are calculated. For the moments calculation only spectral lines that were classified as precipitation are used. We have used spectral processing (Bringi and Chandrasekar (2001), p. 287) for estimation of the mean velocity and spectrum width. Then for a given rage gate the mean velocity and spectrum width can be found as: v k S hh k s 2 N v= S hh k s 2 N (3) v k v 2 S hh k s 2 N s 2 v = S hh k s 2 N 2 where s N is the noise power spectral density, and is the precipitation mask that is equal to 1 if a spectral line is classified as a precipitation signal and zero otherwise. The can be regarded as a frequency response of the clutter and noise rejection filter. To reduce the noise influence on the calculated moments even further the noise power spectral density subtraction is done. The reflectivity is calculated as a sum of precipitation power spectral densities Z h = S hh k s 2 N, (4) and the polarimetric variables can be calculated as follows

4 Figure 4. An example of the classification. Z dr = ρ hv 0= N 1 S hh k s N 2 S vv k s N 2 N 1 k= 0 S hv k S hh k s 2 N S vv k s 2 N (5) It should be noted that clutter and noise filtering do not bias the polarimetric variables. Therefore, calculation of polarimetric variable does not require interpolation over notched parts of the spectra. To reduce possible reflectivity calculation errors one can use interpolation procedure (Siggia and Passarelli, 2004; Bharadwaj et al., 2007). However, the main advantage of this methodology, over the procedures presented in literature, is the adaptive selection of the filter properties that minimizes weather signal loss.

5 3. Application to other types of clutter To investigate dual-polarization spectral properties time-series data in a variety of conditions were collected. Below, three cases are presented, i.e. sea clutter, insects and birds, and chaff Sea clutter Figure 5. Sea clutter observations by University of Helsinki Kumpula radar. Figure 6. Distributions of sea clutter co-polar coherency spectrum, correlation coefficient, spectral differential phase and reflectivity textures. 3.2 Insects Figure 7. PPI of Kumpula radar observations that shows mainly insects.

6 Figure 8. Same as Figure 6, only for insects. 3.3 Chaff Figure 9. PPI of reflectivity and linear depolarization ratio observed on 06 May 2008 by Kumpula radar. Figure 10. Same as Figure 8, only for chaff. It is interesting to note that in all those cases, there is a distinct difference between distribution of values of co-polar correlation coefficient and co-polar coherency spectrum. This difference indicates that in all those cases, we are dealing with scattering from discrete scatterers. 4. Results The proposed dual-polarization spectral filter was tested on University of Helsinki Kumpula radar observations of a precipitation event that took place on May 27, The observations contain echoes from precipitation as well as from ground clutter, insects and possibly sea clutter. The observations were collected in the PPI mode with an elevation angle of 0.4 degrees. For this measurement 64 pulses were transmitted per radial. In Figure 11, unfiltered and filtered radar data is shown. On the example of the co-polar correlation coefficient observations, it can be seen that the proposed methodology results in a more accurate estimates of polarimetric observables. This improvement is related to both clutter and noise suppression. It should be also observed that the proposed filter also removes high Zdr areas that corresponds to insect echoes.

7 References Bachmann, S., and D. Zrnić, 2007: Spectral Density of Polarimetric Variables Separating Biological Scatterers in the VAD Display. J. Atmos. Oceanic Technol., 24, Bharadwaj, N., V. Chandrasekar, and F. Junyent, 2007: Evaluation of First Generation CASA Radar Waveform in the IP1 Testbed. Preprints, International Geoscience and Remote Sensing Symposium, Barcelona, Spain, IEEE, CDROM Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar: Principles and Applications. Cambridge University Press, 636 pp. Chandrasekar V., V. Bringi, and P. Brockwell, 1986: Statistical properties of dual-polarized radar signals. Preprints, 23d Conf. on Radar Meteorology, Snowmass, CO, Amer. Meteor. Soc., Doviak R. J., and D. S. Zrnić, 1993: Doppler Radar and Weather Observations. Academic Press, 562 pp. Giuli D., M. Gherardelli, A. Freni, T. A. Seliga, and K. Aydin, 1991: Rainfall and clutter discrimination by means of dual-linear polarization radar measurements. J. Atmos. Oceanic Technol., 8, Gourley, J.J., P. Tabary, and J. Parent du Chatelet, 2007: A Fuzzy Logic Algorithm for the Separation of Precipitating from Nonprecipitating Echoes Using Polarimetric Radar Observations. J. Atmos. Oceanic Technol., 24, Groginsky, H. L., and K. M. Glover, 1980: Weather radar canceler design, Proc. Of 19th Radar Meteorol. Conf., Passarelli R. E. Jr., P. Romanik, S. G. Geotis, and A. D. Siggia, 1981: Ground clutter rejection in the frequency domain. Preprints, 20th Conf. on Radar Meteorology, Boston, MA, Amer. Meteor. Soc., Ryzhkov, A. V., and D. S. Zrnić, 1998: Polarimetric rainfall estimation in the presence of anomalous propagation. J. Atmos. Oceanic Technol., 15, Sachidananda, M., and D. Zrnić, 1986: Zdr measurement considerations for a fast scan capability radar. Radio Sci., 20, Sachidananda, M., and D. Zrnić, 1989: Efficient Processing of Alternately Polarized Radar Signals. J. Atmos. Oceanic Technol., 6, Siggia, A. D., and R. E. Passarelli, 2004: Gaussian model adaptive processing (GMAP) for improved ground clutter cancellation and moment calculation. Proc. Of 3rd European Conf. On Radar in Meteorol. And Hydrology (CDROM), Visby, Sweden. Da Silveira R. B., and A. R.Holt, 1997: A neural network application to discriminate between clutter and precipitation using polarization information as feature space. Proc. 28th Conf. Radar Meteorology Austin, TX, Moisseev, D., C. Unal, H. Russchenberg, and L.Ligthart, 2000: Doppler polarimetric ground clutter identification and suppression for atmospheric radars based on co-polar correlation. Proc. MIKON, Wroclaw, Poland, Moisseev, D., C. Unal, H. Russchenberg, and L. Ligthart, 2002:A new method to separate ground clutter and atmospheric reflections in the case of similar Doppler velocities. IEEE Trans. Geosci. Remote Sens, 40, Moisseev, D. N. and V. Chandrasekar, 2009: Polarimetric spectral filter for adaptive ground clutter suppression. J. Atmos. Oceanic Technol., 26 (2) Nguyen, C. M., D. N. Moisseev, and V. Chandrasekar, 2007: A parametric time domain method for spectral moment estimation and clutter mitigation for weather radars. Accepted to J. Atmos. Oceanic Technol. Torres S. M., and D. S. Zrnic, 1999: Ground clutter canceling with a regression filter. J. Atmos. Oceanic Technol, 16, Unal C. M. H., and D. N. Moisseev, 2004: Combined Doppler and polarimetric radar measurements; correction for spectrum aliasing and non simultaneous polarimetric measurements. J. Atmos. Oceanic Technol., 21, p

8 Figure 11. Before and after application of the spectral polarimetric clutter filter. The left column shows Kumpula radar observations before the filtering. The right column shows results of the filtering.