TMCnet News
Ground Clutter Mitigation for Weather Radars Using the Autocorrelation Spectral Density [Journal of Atmospheric and Oceanic Technology](Journal of Atmospheric and Oceanic Technology Via Acquire Media NewsEdge) ABSTRACT Radar returns from the ground, known as ground clutter, can contaminate weather signals, often resulting in severely biased meteorological estimates. If not removed, these contaminants may artificially inflate quantitative precipitation estimates and obscure polarimetric and Doppler signatures of weather. A groundclutter filter is typically employed to mitigate this contamination and provide less biased meteorologicalvariable estimates. This paper introduces a novel adaptive filter based on the autocorrelation spectral density, which is capable of mitigating the adverse effects of ground clutter without unnecessarily degrading the quality of the meteorological data. The so-called Clutter Environment Analysis using Adaptive Processing (CLEAN-AP) filter adjusts its suppression characteristics in real time to match dynamic atmospheric environments and meets Next Generation Weather Radar (NEXRAD) clutter-suppression requirements. (ProQuest: ... denotes formulae omitted.) 1. Introduction Radar returns from the ground (known as ground clutter) can greatly affect the performance of weather radars by contaminating weather signals, introducing biases in the meteorological variables, and potentially impeding the recognition of weather phenomena (Doviak and Zrnic 1993). Leftuntreated, ground-clutter contamination leads to biases in estimates of the spectralmoments and the polarimetric variables (e.g., Friedrich et al. 2009). For example, under strong contamination conditions, reflectivity estimates are artificially inflated and Doppler velocities and spectrum widths are shifted toward zero. If not removed, these biases could propagate to algorithms and models that rely on radar data (e.g., Chrisman et al. 1994; Fornasiero et al. 2005). Ground-clutter filters (GCF) are typically employed to mitigate the negative effects of returns from the ground, which are characterized by near-zero Doppler velocities and narrow spectrum widths (Sirmans 1987, 1992). An effective GCF suppresses most of the ground clutter signal while retaining most of the weather signal, a task that is more suitably accomplished at the signal-processing level, that is, by operating on the received complex voltages [also referred to as the in-phase and quadrature (I and Q) time series data]. Processing limitations of early pulsed Doppler weather radars led to the use of simple, nonrecursive time-domain filters (e.g., Sirmans and Dooley 1980). But as radar signal processors became more powerful, complex techniques ranging from recursive (e.g., Hamidi and Zrnic 1981) and regression filters (e.g., Torres and Zrnic 1999) to adaptive frequency-domain filters (e.g., Passarelli et al. 1981) have become feasible for implementation in operational systems. The reader is referred to Hubbert et al. (2009) and Moisseev and Chandrasekar (2009) for excellent literature reviews on ground-clutter filters for weather radars. Designing a filter that achieves the required ground clutter suppression under operational conditions is only one side of the problem. The other side consists of devising an effective mechanism to selectively apply the filter only when it is needed. This is important because weather signals may exhibit similar spectral characteristics as ground clutter signals, and also because the ground clutter environment can quickly change over time because of varying anomalous propagation (AP) conditions (Pamment and Conway 1998). Thus, the mitigation of ground-clutter contamination can be more effective if the degree of ground clutter contamination (or lack thereof) is dynamically recognized and a suitable filter (or no filter) is adaptively applied. On the Weather Surveillance Radar-1988 Doppler (WSR-88D), ground clutter mitigation was initially accomplished with a bank of time-domain, five-pole elliptic filters and a static clutter map that controlled the degree of suppression (none, low, medium, or high) at every radar resolution volume (Sirmans 1992; Chrisman et al. 1994). While effective at suppressing contamination from ground clutter returns, this mitigation scheme only allowed for adaptation to the varying clutter environment when the maps were augmented with real-time input from radar operators or entirely recreated on a seasonal basis. Improvements in this area were made feasible only after the upgrade of the legacy signal processor as part of the Open Radar Data Acquisition (ORDA) project (Crum et al. 1998). The Gaussian model adaptive processing (GMAP) filter (Siggia and Passarelli 2004) replaced the legacy bank of elliptic filters and included the ability to adapt its suppression to the clutter environment in real time, which resulted in overall improved performance (Ice et al. 2004). Thus, the legacy static map was simplified to a (still static) binary map that controlled the application (or not) of the GMAP filter. In a recent upgrade of the ORDA, the static clutter map was complemented by the clutter mitigation decision (CMD) algorithm (Hubbert et al. 2009). CMD uses the so-called clutter phase alignment and various spatial features in a fuzzy logic scheme to perform automatic clutter detection in real time and thus control the application of the GMAP filter without operator intervention (Ice et al. 2009). Although significant improvements were realized with the implementation of CMD and GMAP on the WSR-88D, more can be achieved by integrating filtering and its control into a single algorithm. Bachmann et al. (2008) had one of the earliest successes in this regard, and their work has served as inspiration to extend the use of spectral phase information to improve ground clutter mitigation for weather radars. In general, techniques that operate on I and Q samples from a single radar-resolution volume are better suited for integration with other signal-processing functions and operational implementation. However, one of the drawbacks of this approach is that, in the absence of additional information, weather signals with near-zero Doppler velocities and narrow spectrum widths may be incorrectly attenuated because they exhibit characteristics similar to those of ground clutter returns. The use of information from neighboring resolution volumes (e.g., spatial textures) has been shown to add skill in the discrimination between weather signals in the so-called zero-velocity isodop (i.e., echoes with zeroDoppler velocity) and those from ground clutter (e.g., Fabry and Gadoury 2009). Nevertheless, detection and filtering functions are most effective when integrated into a single algorithm (e.g., Moisseev and Chandrasekar 2009). The technique described in this paper is referred to as the Clutter Environment Analysis using Adaptive Processing (CLEAN-AP1) filter. It operates on I and Q samples from a single radar-resolution volume and integrates both aspects of the ground clutter mitigation process: selective filter application and effective clutter suppression. At the core of CLEAN-AP is the autocorrelation spectral density (ASD) introduced byWarde and Torres (2014): an extension of the classical power spectral density (PSD) that includes spectral phase information. CLEAN-AP combines the use of adaptive data windows and the argument of the lag-1 ASD to provide a suitable compromise between clutter suppression and the quality of themeteorological variables obtained after filtering. In this paper, simulations and real data recorded from two operational WSR-88D radars are used to evaluate and illustrate the performance of the CLEAN-AP filter, which is shown to adapt its suppression to the detected clutter environment. As a result, it meets the groundclutter- suppression requirements for the WSR-88D (NEXRAD Radar Operations Center 1996). That is, it provides effective suppression of ground clutter returns up to (at least) 50 dB stronger than the weather returns with small biases in the resultingmeteorological variables (except, as allowed in the requirements, for weather signals with near-zero Doppler velocities and very narrow spectrum widths). The rest of the paper is structured as follows. The CLEAN-AP filter is described in section 2 and its performance is analyzed using simulations in section 3. In section 4, the technique is applied to radar observations from the KTLX radar in Twin Lakes, Oklahoma (prior to its upgrade to dual polarization), and the KVNX radar at Vance Air Force Base in Enid, Oklahoma (after its dual-polarization upgrade). Section 5 ends with a short summary and plans for future work. 2. The CLEAN-AP filter The CLEAN-AP filter operates on a single-resolutionvolume, single-dwell set received complex voltages, V(m) (0#m,M).Here,mindexes sample time at pulse repetition time (PRT) increments Ts andMis the number of samples per dwell (the range-time index is irrelevant and omitted herein). The CLEAN-AP filter is equally suited for single- and dual-polarization weather radars, and variations in processing between these implementations are identified in the algorithm description that follows. In a nutshell, the CLEAN-AP filter consists of four steps: data window selection, lag-1ASDcomputation, clutter-extent determination, and filtering. Whereas the processing path on a single-polarization implementation is straightforward, in a dual-polarization implementation the choice is between the simpler master-slave channel configuration and the combined-channel configuration. In the master-slave channel configuration, all adaptation is based on information fromonly one of the polarimetric channels (typically the horizontal one, as done on the WSR-88D). Conversely, in the combined-channel configuration, adaptation is based on information from both channels, which could be more effective in cases where ground clutter and weather returns exhibit markedly different polarimetric characteristics. Herein, the dualpolarimetric master-slave channel configuration is chosen for the simulations and real-data processing-still, both options are described for the adaptive steps of the algorithm in this section. For dual-polarization radars, only the simultaneous (or hybrid) transmission mode (Bringi and Chandrasekar 2001) is considered; the reader is referred toWarde and Torres (2013) for an extension of the CLEAN-AP filter to dual-polarization radars employing the alternating transmission mode. a. Data window selection A data window should achieve a suitable compromise between clutter suppression and the variance of meteorological-variable estimates after filtering. That is, more tapered windows lead to larger variance of estimates but reduce the spectral leakage from strong ground clutter signals and result in improved ground clutter mitigation. The opposite holds for less tapered windows; thus, the degree of tapering may be tailored to the dynamic range of the signals under analysis. To estimate the required amount of tapering, the direct current (DC) power (a proxy for clutter power) is compared to the noise power per frequency bin as ... where the N is the system noise power on singlepolarimetric radars or the channel-specific noise power on dual-polarimetric radars. It should be noted that this quantity is only a rough estimate of the actual clutter-tonoise ratio (CNR). However, its use is confined to the window selection process, for which it is adequate. The ''best'' data window is the least tapered window that has sidelobes below d CNR (on a dual-polarization implementation, this is computed for the master channel in a master-slave configuration or as the maximum between the two channels for a combined-channel configuration). Whereas such a data window could be found using optimization techniques (e.g., Adams 1991), it is simpler (and not significantly detrimental) to choose between data windows that are readily available in realtime signal-processing environments. For this implementation, the data window is adaptively selected between the rectangular, von Hann, Blackman, and Blackman-Nuttall windows, which have mainlobe-tomaximum- sidelobe levels of 213, 232, 258, and 298 dB, respectively (Harris 1978; Nuttall 1981). Thus, the rectangular window is selected for d CNR values less than 13 dB, the von Hann window between 13 and 32 dB, the Blackman window between 32 and 58 dB, and the Blackman-Nuttall window for values above 58 dB. Finally, all data windows are normalized to be power preserving; thus, no correction is needed for the quantities computed from the windowed I and Q samples. Figure 1 shows histograms (and their means) of data window selection for a clutter-contaminated weather signal as a function of the clutter-to-signal ratio (CSR). The y axis corresponds to the different windows available for selection, and the gray intensities represent the frequency of selection as a percentage of the total number of simulated realizations. As expected, the selected window becomes progressively more tapered as the clutter contamination gets stronger relative to the weather signal. Thus, the more aggressive Blackman-Nuttall window is selected at the highest CSRs, which typically correspond to radar volumes close to the radar from which ground-clutter contamination is the strongest. b. Lag-1 ASD computation The ASD was introduced by Warde and Torres (2014) as a generalization of the periodogram and an alternate tool for spectral analysis of weather signals. The argument of the ASD provides information that is not readily available from the complex spectrum (recall that the argument of the Fourier transform of the complex voltages is a sequence of random variables, each uniformly distributed between 2p and p) and that is completely absent from the PSD (a real function); it can be used to detect strong narrowband signals, such as those from ground clutter returns. The lag-1 ASD is at the core of the CLEAN-AP filter; it is computed as ... where the superscript asterisk (*) denotes complex conjugation and ... where l 5 0, 1 for each of the factors on the right-hand side of (2) and d is the (M 2 1)-sample data window obtained in the previous step. In a dual-polarization implementation using the master-slave channel configuration, only the master-channel ASD is computed; otherwise, using the combined-channel configuration requires two ASDs, one for each channel. For consistency with the ASD, the PSD can be computed as the average of PSDs from F0 and F1 (i.e., Welch's method). This ensures that both spectral representations are on the same frequency axis and are derived fromall available samples. c. Ground-clutter extent determination In this step, spectral leakage in the argument of the lag-1 ASD is used to identify the spectral coefficients around zero Doppler velocity with significant groundclutter contamination. As described by Warde and Torres (2014), the argument of the ASD of a narrowband signal (such as ground clutter returns) locally ''flattens'' (i.e., there is a bias compared to the ideal or expected argument) around the magnitude spectral peak due to the large power gradient. On the other hand, the argument of the ASD of a wideband signal appears nearly unbiased thanks to the relatively small power gradient. Biases in the argument of the lag-1 ASD due to the spectral leakage of ground clutter signals can be predicted by modeling the argument of the lag-1 ASD for a ground clutter return. The so-called clutter model is based on an ideal Gaussian spectrum with zero mean Doppler velocity and an adjustable spectrum width, which controls the filter's maximum suppression. For this work, the spectrum width of the clutter model is set to 0.4ms21, which matches the value used by the GMAP filter on the WSR-88D. The clutter-model lag-1 ASD is generated for each data window by convolving a finely sampled, ''ideal'' ground clutter lag-1 ASD (the magnitude is Gaussian with zero mean Doppler velocity and 0.4ms21 spectrum width, and the argument is linear with unit slope) with the power spectral density of the (M 2 1)-sample data window (zero padded to the same number of samples as the ideal clutter ASD, which is 512 in our case). Figure 2 shows the clutter-model lag-1 ASD for the four possible data windows and the ideal ground clutter lag-1 ASD. In practice, the clutter-model lag-1 ASD can be precomputed for given values of M and Nyquist velocity, and themaximumarguments (one value for each datawindow) can be stored to be used in the realtime determination of clutter extent as described next. A spectral coefficient is identified as clutter contaminated if the corresponding lag-1 ASD coefficient has a magnitude above the spectral noise floor and an absolute argument less than the maximum value predicted by the clutter model. For a more robust determination, the argument of the ASD is previously smoothed with a running median filter to minimize the impact of statistical fluctuations (a three-point filter was empirically selected based on simulation results). In addition, up to two ASD coefficients that do not meet the argument requirement are allowed inside (but not at the end) of each half of the clutter extent when partitioned into positive and negative Doppler velocities. In a dual-polarization implementation using the combined-channel configuration, both ASDs are subjected to this process, and the clutter extent is defined as themaximumbetween the two channels. Figure 3 shows an example of clutter-extent determination using the spectral leakage in the argument of the lag-1 ASD and the corresponding thresholds from the clutter model. Figure 4 shows the mean extent of clutter contamination determined by this procedure as a function of the CSR (same parameters as in Fig. 1). It is not surprising that, as theCSR increases, the clutter extent gradually increases. The ''elbow'' in the curve at a CSR of ;75dB corresponds to a situation where, even with the most tapered data window available, small increases in clutter power cause large increases in the extent of clutter. This roughly corresponds to a CNR of ;95 dB [the signal-to-noise ratio (SNR) is 20 dB], where spectral leakage can no longer be effectively contained by the Blackman-Nuttall window. d. Filtering Having identified the spectral coefficients with significant ground-clutter contamination in the lag-1 ASD, the next step is to remove the contamination from the corresponding coefficients in the other spectral densities. Using the ASD as a basis for this makes sense in a singlepolarization implementation because, in the mean, the magnitude of the ASD equals the PSD (Warde and Torres 2014), so the clutter extent is the same on both spectral representations. However, in a dual-polarization implementation, the horizontal- and vertical-channel PSDs and the ASDs and the cross-spectral density (CSD) may not have the same distribution of powers as a function of Doppler velocity. Thus, in a master-slave configuration, using the clutter extent as determined from the masterchannel ASD may lead to inadequate suppression for the slave channel. However, this is an issue only in radar volumes where the polarimetric characteristics of ground clutter returns aremarkedly different from those ofweather returns (e.g., ZDR of clutter returns is negative and that of weather returns is positive). In principle, a combinedchannel configuration in which the clutter extent is determined using information from both polarimetric\ channels would mitigate this problem at the price of additional computations. Figure 5 illustrates the results of the complete CLEAN-AP process and shows the magnitude and argument of the unfiltered (gray) and filtered (black) lag-1 ASD for data collected with the KOUN radar in Norman, Oklahoma, on 19 March 2006. Data correspond to three different resolution volumes (at different ranges) for the same antenna position: one with ground clutter (left), one with hydrometeors (middle), and one with a mix of ground clutter and hydrometeors (right). For all cases,M566 and the Nyquist velocity ya528ms21. The Blackman-Nuttall window was selected for the clutteronly case, the rectangular window for the weather-only case, and the Blackman window for the weather-andclutter case. The number of spectral coefficients identified as having clutter contamination and the amount of clutter suppression are listed for each case. In theweatheronly case, only the zero-velocity coefficient is identified as having clutter contamination and no filtering is applied. In this implementation, filtering is applied if at least three spectral coefficients are identified as having clutter contamination, and the clutter extent is widened to include the same number of spectral coefficients about zero velocity. Filtering is done in two different ways depending on whichmeteorological variable is to be estimated fromthe filtered spectra. The spectral moments are typically derived from estimates of the autocorrelation function (R) at a few small lags (Doviak and Zrnic 1993); without loss of generality, we assume that only the first two lags, R(0) and R(1), are required. In this case, the coefficients identified as clutter contaminated are removed from S (the PSD) and S1 (the lag-1 ASD), which is straightforward because the two spectral representations are on the same frequency axis. Then, a linear interpolation (independently applied to the magnitudes of the PSD and ASD in logarithmic units and to the argument of the ASD) is used to compensate for the potential weathersignal loss around zero Doppler velocity. Despite its apparent lack of sophistication, this linear interpolation scheme has an overall performance similar to that of the GMAP's iterative Gaussian interpolation scheme, so it was preferred because of its computational simplicity. Because correlations and spectral densities (the PSD and the ASD) are Fourier transform pairs, R(0) and R(1) can be computed fromthe filtered PSDandASD, S and S1, as ... respectively. Note that R(0) could also be computed as the sum of the magnitudes of the filtered lag-1 ASD. Other spectral-moment estimators based on higher autocorrelation lags (e.g., the hybrid spectrum width estimator described by Meymaris et al. 2009) can be similarly computed by using the ASD at higher lags, which increases the computational complexity, or from the lag-1 ASD as ... which does not significantly increase the computational complexity but requires bias corrections as described by Torres et al. (2007). As implied earlier, in this work we use the classical spectrum width estimator based on the ratio R(0)/R(1) (Doviak and Zrnic 1993). The polarimetric variables are typically derived from estimates of lag-zero autocorrelations, Rh(0) and Ry(0), and cross correlation, Rhv(0) (Bringi and Chandrasekar 2001). In this case, the coefficients identified as clutter contaminated are simply removed (zeroed out) from Sh, Sy (the PSDs computed from Vh and Vy, respectively), and Shv (the CSD computed from Vh and Vy). A spectral reconstruction to mitigate potential signal loss around zero Doppler velocity is not needed nor desired in this case, as it would bias the polarimetric variable estimates (Zrnic et al. 2008). Finally, Rh(0), Ry(0), and Rhy(0) are computed as the sums of filtered densities Sh, Sy, and Shv, respectively [cf. (4)], and other estimators based on higher correlation lags (e.g., Lei et al. 2012) can be computed using expressions analogous to (6). 3. Simulations In this section, the performance of the CLEAN-AP filter under a wide variety of conditions is characterized using simulations. Instead of comparing to one of the many existing ground-clutter filtering techniques, we will show that the CLEAN-AP filtermeets (and inmost cases exceeds) the set of WSR-88D operational requirements for ground-clutter suppression (NEXRAD Radar Operations Center 1996). WSR-88D clutter-suppression requirements are defined for a clutter-contaminated weather signal model in which the weather signal has an SNR of 20dB and a spectrum width of 4ms21 and the ground clutter signal has a Doppler velocity of 0ms21. Not specified in this model is the spectrum width of the ground clutter signal, which we set to 0.28ms21 as in previous similar analyses (e.g., Ice et al. 2004; Sirmans 1992). Also not specified are the polarimetric characteristics of weather and clutter signals in themodel. In fact,WSR-88D clutter-suppression requirements are only given in the context of single polarization, where the expectation is that the performance obtained from naturally extending the GCF to the polarimetric variables is acceptable. As described in the previous section, the CLEAN-AP filter is extended to dual-polarization radars in the same manner that the GMAP filter was extended for the WSR-88D (Zrnic et al. 2008). Thus, the performance of theCLEANAP filter will be quantitatively assessed using only the available single-polarization requirements. Nevertheless, qualitative performance of CLEAN-AP will be shown using data from both single- and dual-polarization radars in section 4. For each set of model parameters, 1000 realizations of time series data are simulated using the procedure described by Zrnic (1975) with 64 samples, a PRT of 1 ms, and a radar frequency of 2850MHz. Variable model parameters (i.e., those not specified in the requirements) include the weather-signal Doppler velocity and spectrum width, and the CSR. In this analysis, the weather-signal Doppler velocity (v) is varied from 0 to 26.3ms21 in 55 steps of 0.48ms21, and the spectrum width (sy) is varied a total of 20 steps: from 0.1 to 1ms21 in steps of 0.1ms21, from 1 to 4ms21 in steps of 0.5ms21, and from 4 to 8ms21 in steps of 1ms21. The CSR is varied from 230 to 70 dB in 21 steps of 5 dB. As dictated by the WSR-88D clutter-suppression requirements, the performance of the CLEAN-AP filter is assessed in terms of suppression and also in terms of biases and standard deviations of reflectivity, Doppler velocity, and spectrum width estimates after filtering. The required clutter suppression, defined as the ratio of unfiltered to filtered powers, is 30 dB for reflectivity estimates and either 20 dB(lowsuppression), 30 dB(medium suppression), or 50 dB (high suppression) for Doppler velocity and spectrum width estimates. However, clutter suppression for modern filters adapts to the amount of clutter present in the received echoes, so it is not tied to fixed levels. Ideally, clutter suppression should be given by the ratio of the sum of weather- and clutter-signal powers to the weather-signal power; this assumes a cluttercontaminated signal at the input of the filter and perfect clutter removal at its output. In practice, it is seldom possible to completely remove the clutter signal without affecting the weather signal, so a clutter suppression of 0 dB (input and output powers are the same) is acceptable for clutter contamination that, if unfiltered, would not induce significant estimate biases. The level of clutter contamination for which this holds is different for eachmeteorological variable, and the reader is referred to Friedrich et al. (2009) for an analysis of meteorologicalestimate biases due to ground-clutter contamination. Because a GCF induces biases that depend on the velocity and spectrum width of the weather signal, maximum allowable biases are normally specified for weather signalswithDoppler velocities in the filter's passband and different spectrum widths (Sirmans 1992). For the WSR- 88D, the filter's passband is defined as all Doppler velocities outside the interval (2y0, y0), where y0 (the passband edge velocity) should be no larger than 2, 3, and 4ms21 for clutter-suppression levels of 20, 30, and 50 dB, respectively. For the CLEAN-AP filter, the passband is defined by the spectral extent of ground clutter (or the filter's notch width). For a clutter-free condition (i.e., a CSR of 230 dB in the model), WSR-88D cluttersuppression requirements specify that application of the filter should result in maximum reflectivity biases no larger than 10, 2, and 1 dB (in absolute value) for all velocities in the passband and for spectrum widths of 1, 2, and larger than 3ms21, respectively. Additionally, velocity and spectrum width biases and standard deviations should not exceed 2ms21 each for weather signals in the passband. Throughout this analysis and to simplify the validation of requirements, filtered estimate biases and standard deviations are computed with respect to the true simulation parameters. Thus, in addition to the effects of the filtering process, they include any errors inherent to the meteorological-variable estimators. To facilitate the assessment of the filter's performance, we will draw attention to specific situations where estimator errors dominate the results. In the presence of ground clutter (interpreted to mean for CSRs above 230 dB), the WSR-88D requirements allow for additional errors due to clutter residue. For reflectivity, the allowed bias is 1 dB for spectrum widths greater than or equal to 4ms21 when the output CSR is 210 dB. For velocity and spectrum width, the allowed biases and standard deviations are 1ms21 each for a spectrumwidth of 4ms21 and velocity of ya/2 with output CSRs of 211 and 215 dB, respectively. Although not a requirement, a goal of ,2-dB standard deviation of reflectivity estimates after averaging four 250-m gates in range is also desirable (e.g., Ice et al. 2004). For typical range correlations, this goal means that the standard deviation of nonrange-averaged reflectivity estimates should be below ;2.74 dB. The performance of the CLEANAP filter is evaluated against these requirements and goals next. a. Ground clutter suppression Figure 6 shows the CLEAN-AP filter clutter suppression (lefty axis) and reflectivity bias (right y axis) for different CSRs (x axis), evaluated using the weather-andclutter model described above. For each CSR value, the average performance of 55 000 realizations is shown (i.e., the 55 weather-signal velocities, each with 1000 realizations, are combined). Note that, in average, the suppression of the CLEAN-AP filter follows the ideal behavior given by 10 log10[(Pc 1Pw)/Pw], where the unfiltered power is the sum of clutter- and weather-signal powers, Pc 1 Pw, and the filtered power is the expected weather-signal power Pw. The CLEAN-AP suppression curve shows that there is no mitigation for CSRs below 215dB and a one-to-one correspondence (dotted line) for CSRs above 10 dB. Reflectivity estimates after the application of the CLEAN-AP filter (including cases both in the filter's stopband and passband) exhibit very small negative biases for all CSRs. A more detailed analysis of the filter effects on the bias and standard deviation of spectral-moment estimates as a function of Doppler velocity and spectrum width is provided next. b. Biases and standard deviations of spectral-moment estimates Biases in the meteorological variables after filtering quantify the negative effects that the filter has on the weather signal. WSR-88D clutter-suppression requirements are written to assess the effects of the filter when the weather signal is not contaminated by ground clutter (the CSR is set to 230 dB in the weather-andclutter model). In Fig. 7a, the bias of reflectivity estimates is shown for each of the weather-signal spectrum widths (y axis) and velocities (x axis) in the simulated range. The color scale of reflectivity bias (dB) ranges from ,220 (maroon) to 0 (white) to 1 (pink). Note that a positive reflectivity bias indicates a situation in which the power at the output of the filter is larger than at its input, which can happen because of the spectral interpolation process. The asterisks in Fig. 7a mark the WSR-88D requirement benchmarks at weather-signal spectrum widths of 1, 2, and 3ms21 and a velocity of 0ms21, where observed reflectivity biases are 24.0, 20.9, and 20.5 dB, respectively. That is, the magnitudes of the reflectivity biases meet the established requirements. Moreover, it can be seen in this figure that the filter performance exceeds operational requirements where the magnitude of the reflectivity bias after filtering stays under 1 dB (light blue to pink) for weather-signal spectrum widths above 1.5ms21 or velocities larger than 1.5ms21, under 2 dB (dark blue to pink) for spectrum widths above 1ms21 or velocities larger than 1.0ms21, and under 10 dB (dark red to pink) for spectrum widths . 0.6ms21 or velocities. 0.5ms21. As expected, the magnitude of the reflectivity bias for weather-signal spectrum widths below 1ms21 increases rapidly for small velocity values.This behavior is typical ofGCFs and is sometimes referred to as the ''zero-velocity isodop loss.'' The standard deviation of reflectivity estimates is shown in Fig. 7b where the color scale (dB) ranges from 0 (white) to 4 (maroon). Recall that the interpreted goal is a standard deviation of less than ;2.74dB for the nonrange-averaged reflectivity estimates. For narrow spectrum widths, larger standard deviations are inherent to the estimator (e.g., Fig. 6.2 of Doviak and Zrnic 1993). The standard deviations at the benchmark spectrum widths of 1, 2, and 3ms21 and zero velocity (asterisks) are ;1.4, ;1.6, and ;1.5 dB, respectively. The reader should note that the decrease in the standard deviation for nearzero velocities and very narrow spectrum widths is accompanied by a large increase in the magnitude of the biases (Fig. 7a); thus, it should not be interpreted as a performance improvement. Figure 7c is similar to Fig. 7a, except that the bias of velocity estimates is shown. The color scale of velocity bias (m s21) ranges from ,22 (maroon) to 0 (white) to .2 (dark blue). An asterisk at a velocity of 2ms21 (i.e., the filter's passband edge velocity) and a spectrum width of 4ms21 marks the benchmark for which filtered velocity bias magnitudes are required to be below 2ms21. Velocity biases are well within the allowable limit of 2ms21 for the entire range of simulated parameters. In fact, the absolute bias of velocity estimates is less than 0.6ms21 for all velocities larger than 1ms21 or sy larger than 0.2ms21. The maximum bias is 1.4ms21 at y 5 0.5ms21 and sy 5 0.2ms21, and the minimum bias is 20.6ms21 at y 5 0.5ms21 and sy 5 0.6ms21. At the benchmark, the velocity bias is 20.17ms21. Similar to Fig. 7b, the standard deviation of velocity estimates after filtering is shown in Fig. 7d. An asterisk marks the benchmark at a velocity of 2ms21 and spectrum width of 4ms21 where the standard deviation of velocity estimates is 1.1ms21. It is seen that the estimate statistics are well within operational requirements with values of ,2ms21 for y . 1.0ms21 and sy . 0.8ms21, with a maximum of 1.84ms21 occurring at y 5 0.5ms21 and sy 5 8.0ms21. One can see in Fig. 7d significant contributions to the standard deviation inherent to the estimator. The standard deviation of velocity estimates using this estimator increases as a function of spectrum width (e.g., Fig. 6.5 of Doviak and Zrnic 1993); however, the increase as the velocity approaches zero is not related to the estimator but to the filtering process. For example, for a spectrum width of 8ms21, increases in the standard deviation of velocity estimates attributed to the filtering process systematically reduce from a maximum of 0.5 to 0ms21 for velocities from 0 to ya/2 as compared to corresponding values above ya/2. Using similar comparisons, the increase due to filtering for the standard deviation of velocity estimates can be seen to be less than 1ms21 except for sy , 1ms21 and y ,1.5ms21. The standard deviation of velocity estimates at the benchmark is 1.1ms21; this represents an increase of 0.2ms21 as compared to 0.9ms21 at ya. Figure 7e shows the bias of spectrum width estimates. As before, an asterisk at a velocity of 2ms21 and a spectrum width of 4ms21 marks the benchmark for which filtered spectrum width biases are required to be below 2ms21. This figure shows that the magnitude of the bias for filtered spectrum width estimates is more than 2ms21 only for weather-signal spectrum widths below 0.7ms21 and velocities smaller than ;1ms21 with a maximum of 8.6ms21 at y and sy 5 0ms21. For other conditions, the spectrum width biases are well below 1ms21 (light red to light blue) with a bias of 20.02ms21 at the benchmark. Figure 7f shows the standard deviation of spectrum width estimates. Increases in the spectrum width standard deviation for narrow spectrum widths and large spectrum widths are inherent to the estimator (e.g., Fig. 6.6 of Doviak and Zrnic 1993). At a CSR of230 dB, the standard deviation of spectrum width estimates is below 2ms21 for sy . 0.8ms21 and y . 1ms21, and it is 0.46ms21 at the benchmark. As critical as it is to study the negative effects of the filter to the weather signal when there is no clutter contamination (CSR5230 dB), the WSR-88D cluttersuppression requirements address the performance of the filter under more demanding clutter-contamination conditions (CSR 5 50 dB). For spectrum widths at or greater than 4ms21, the reflectivity bias is required to be less than 1 dB. Although not a requirement, one should expect reduced biases in the passband of the filter. From Fig. 4, the mean extent of ground clutter at a CSR of 50 dB is ;0.15 of the Nyquist co-interval, which can be equated to a passband edge velocity of about 64ms21 for the simulation parameters used here. The content and layout of Figs. 8a-f is the same as Figs. 7a -f, except that the CSR is 50 dB. Figure 8a shows that, at the passband edge velocity of 4ms21 and for spectrum widths of 1, 2, and 3ms21, observed reflectivity biases are 21.0, 21.7, and21.1 dB, respectively.Note that themagnitude of the bias of reflectivity estimates after filtering stays under 1 dB (very light blue to pink) throughout the Nyquist interval for spectrum widths above 3.0ms21 or velocities above 4.4ms21, under 2 dB (purple to pink) for spectrum widths above 2.5ms21 or velocities . 3.4ms21, and under 10 dB (blue-green to pink) for spectrumwidths above 1ms21 or velocities . 2.5ms21. In this case, small positive biases (pink) are less than 1 dB with a maximum of 0.94 dB at y 5 4.4ms21 and sy 5 0.2ms21; asmentioned before, these are attributed to the spectral interpolation used in the filtering process. In Fig. 8b, the standard deviation of reflectivity estimates is larger than the levels seen in Fig. 7b but the increase is still within the goal of ;2.74 dB for spectrum widths above 0.8ms21. As in Fig. 7b, Fig. 8b shows artificially reduced standard deviations for spectrum widths in the filter stopband; these are accompanied by large biases (Fig. 8a) and should not be construed as a performance improvement. The velocity biases observed in Fig. 8c are under 2ms21 except for the estimates at sy 50.3ms21 and y 52.3ms21 where the observed bias is 2.2ms21. At the benchmark of sy 5 4ms21 and y 5 4ms21, the velocity bias is 20.8ms21. In fact, the magnitude of the bias of velocity estimates is less than 1ms21 for all velocities in the filter passband and for all spectrum widths above 3ms21. In Fig. 8d, the standard deviation of velocity estimates is below 2ms21 for all velocities with spectrum widths between 1.5 and 7ms21, and is 1.5ms21 at the benchmark (sy 5 4ms21 and y 5 4ms21). As noted before, inherent to the estimator is an expected increase in the standard deviation of velocity estimates as the spectrum width increases. Using the same analysis as with Fig. 7d, the filter-induced standard deviation increases are,2ms21 forsy.1.0ms21 and y . 2.5ms21 and ,1ms21 for sy . 1.5ms21 and y . 2.9ms21 with a 0.45ms21 increase at the benchmark. Seen in Fig. 8e, the magnitude of the biases for filtered spectrum widths are ,2ms21 for sy . 1ms21 or for y . 2.4ms21, and ,1ms21 for sy . 1.5ms21 or for y . 3.4ms21. At the benchmark (sy 5 4ms21, y 5 4ms21), the bias is20.03ms21. InFig. 8f, the standard deviation of spectrum width estimates is below 2ms21 for sy .1ms21 or y . 2.4ms21, and it is 0.82ms21 at the benchmark. As seen in the statistical analyses using simulations, the performance of the CLEAN-AP filter comfortably meets the WSR-88D operational requirements for ground-clutter suppression, bias, and standard deviation of spectral-moment estimates. Whereas two extreme contamination cases are mainly discussed in this section (no contamination at a CSR of 230dB and strong contamination at a CSR of 50 dB), the performance of the CLEAN-AP filter was comprehensively evaluated at CSRs ranging from 230 to 70 dB (e.g., see Fig. 6; Torres et al. 2010, 2012). Through this evaluation it was determined that the integrated filter design provides automatic adjustment of clutter suppression while recovering meteorological estimates in all expected operational clutter environments: from no contamination to a CSR of (at least) 50 dB. 4. Application to WSR-88D data In this section, the performance of the CLEAN-AP filter is assessed using data collected with operational WSR-88D radars. Whereas only limited to the examples shown, qualitative assessments of meteorologicalvariable fields are used to illustrate that the performance of the CLEAN-AP filter is as expected and in support of the quantitative results presented in the previous sections. The performance of the CLEAN-AP filter with strong (dominant) clutter contamination is assessed by focusing on regions of the filtered fields where the corresponding unfiltered fields clearly exhibit textures and values typical of ground clutter returns (e.g., elevated and spatially inconsistent reflectivities, nearzero Doppler velocities, and narrow spectrum widths for single-polarization data). Supporting the results in Fig. 8, it is expected that the filtered fields in these regions will show smoother textures that are typical of weather returns and values that are consistent with those in neighboring regions with no apparent ground-clutter contamination. To assess the performance of the CLEAN-AP filter for the case of little-to-no groundclutter contamination, the focus will be on regions of the filtered fields where the corresponding unfiltered fields clearly exhibit textures and values typical of weather returns. Supporting the results in Fig. 7, it is expected that the unfiltered and filtered fields in these regions will be almost identical except perhaps along the zero isodop. The CLEAN-AP filter was applied to I andQsamples collected with the KTLX radar in Twin Lakes on 27 October 2006 at an elevation angle of 0.58; 18 samples per dwell at a PRT of 3.107 ms were processed for reflectivity estimates, and 53 samples per dwell at a PRT of 0.987 ms (ya 5 26ms21) were processed (independently) for Doppler velocity and spectrum width estimates. Range overlaid echoes in the short-PRT data (e.g., purple in Doppler velocity and spectrum width displays) are identified using the powers from the long- PRT data. Figure 9 shows plan position indicator (PPI) displays of unfiltered (leftpanels) and filtered (right panels) spectral-moment estimates. As mentioned before, dominant ground-clutter contamination can be readily distinguished by its biasing effects on weather signals as compared with the surrounding regions of uncontaminated weather signals. For this case, dominant clutter contamination can be observed in the unfiltered data as high reflectivity values near the radar (top-leftPPI) collocated with near-zero-velocity values (middle-leftPPI) and very narrow spectrum width values (bottom-leftPPI). After the CLEAN-AP filter is applied, clutter contamination is mitigated, which is evident by the spatially smooth reflectivity, velocity, and spectrum width fields (right panels) that are typical of weather returns. Whereas these images cannot be used to accurately measure filtering biases, the qualitative assessment of spatial textures and magnitudes supports performance expectations for the CLEAN-AP filter. That is, in most cases, it appears that the spectral moments of the underlying weather signal are reasonably recovered. There does seem to be some loss along the zero-velocity isodop close to the radar where the velocity appears to have a small gradient (i.e., negative velocities transitioning to positive velocities). However, the spatial texture and magnitudes of the filtered reflectivity field in regions where the velocity field is near zero are consistent with those in neighboring regions where the velocity field is away from zero. Thus, as expected from the results in Fig. 6, reflectivity biases introduced by the CLEAN-AP filter in this case are small and not visually apparent. As another example, the CLEAN-AP filter was applied to I and Q samples collected with the dual-polarized KVNX at Vance Air Force Base, Oklahoma, on 8March 2011 at an elevation angle of 0.58; 17 samples per dwell at a PRT of 3.107ms were processed for reflectivity and polarimetric-variable estimates, and 61 samples per dwell at a PRT of 0.847 ms (ya 5 30.6ms21) were processed (independently) for Doppler velocity and spectrum width estimates. Figure 10a shows PPI displays (enlarged to show coverage out to ;65km from the radar) of unfiltered (left) and filtered (right) reflectivity (top), velocity (middle), and spectrum width (bottom) estimates. Extensive strong ground-clutter contamination can be observed close to the radar in the unfiltered data where moderateto- high reflectivity values are collocated with near-zerovelocity and very narrow spectrumwidth values. However, spatial textures and magnitudes of the corresponding filtered fields are smooth and consistent with those of regions with no apparent ground-clutter contamination. Thus, similar to the previous example, the fields of filtered data qualitatively confirm that the CLEAN-AP filter can mitigate ground clutter signals in regions with varying degrees of contamination. Further comparisons show that, as expected, the impact of the filtering process in uncontaminated regions (e.g., the region north of the radar) is not visually apparent. Figure 10b contains PPI displays of differential reflectivity, differential phase, and correlation coefficient for the same region as in Fig. 10a. More evidence of ground-clutter contamination for this case can be seen in the increased variability (textures) and biases of unfiltered polarimetric-variable fields over clutter-contaminated regions [the reader is referred to Park et al. (2009) for a good synopsis of the signatures of ground clutter in the fields of the different polarimetric variables]. After filtering, the same regions exhibit textures that are more consistent with those seen in regions devoid of clutter contamination (e.g., north of the radar). Thus, the qualitative assessment of these examples supports the quantitative simulation results in the previous sections and confirms that the CLEAN-AP filter can effectively mitigate ground-clutter contamination for all meteorological variables in both single- and dualpolarization weather cases. 5. Conclusions This paper introduced theCLEAN-APfilter as a novel approach for ground clutter mitigation on single- and dual-polarimetric weather radars. The proposed technique operates on samples froma single radar-resolution volume and addresses both aspects of the cluttermitigation process: selective filter application and effective clutter suppression. The CLEAN-AP filter works by identifying the spectral leakage caused by ground clutter signals in the argument of the lag-1 autocorrelation spectral density. Being coupled with an adaptive data window selection mechanism, the CLEAN-AP filter is capable ofmitigating the adverse effects of ground clutter without unnecessarily degrading the quality of the meteorological data. Simulations and real data were used to evaluate and illustrate the performance of the CLEAN-AP filter under a wide variety of conditions. It was shown that the CLEAN-AP filter comfortably meets the ground-clutter suppression requirements of the WSR-88D. That is, in the majority of cases, it provides effective suppression of ground clutter returns for clutter-to-signal ratios of up to (at least) 50 dB with minimal biases in the resulting meteorological variables. However, as with most filters that operate on a single-radar-resolution volume, the CLEAN-AP filter exhibits biases in the meteorologicalvariable estimates corresponding to weather signals with near-zero Doppler velocities and very narrow spectrum widths (the so-called zero-velocity isodop). Further research is underway to exploit polarimetric information to improve the robustness of the spectral clutter-extent determination (Kilambi et al. 2013). In addition, work is ongoing to extend the functionality of the CLEAN-AP filter to other radar waveforms, such as systematic phase coding and staggered PRT. Although the CLEAN-AP filter has been shown to meet the ground-clutter-suppression requirements of the WSR-88D, more comparisons with the current operational techniques are needed. Nevertheless, the analysis documented in this work confirms that the CLEAN-AP filter can provide effective mitigation of ground-clutter contaminationwhen implemented on operationalweather radars. Acknowledgments. The authors thank Christopher Curtis, Igor Ivic, and three anonymous reviewers for providing comments to improve the manuscript. Funding was provided by NOAA's Office of Oceanic and AtmosphericResearch underNOAA-University ofOklahoma Cooperative Agreement NA11OAR4320072, U.S. Department of Commerce. 1CLEAN-AP 2009 Board of Regents of the University of Oklahoma. REFERENCES Adams, J. W., 1991:Anew optimal window[signal processing]. IEEE Trans. Signal Process., 39, 1753-1769, doi:10.1109/78.91146. Bachmann, S., D. Zrnic, and C. Curtis, 2008: Spectral identification and suppression of ground clutter contributions for phased array radar. 24th Conf. on Interactive Information Processing Systems for Meteorology, Oceanography, and Hydrology, New Orleans, LA, Amer. Meteor. Soc., P9A.4. [Available online at https:// ams.confex.com/ams/88Annual/techprogram/paper_131493.htm.] Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar: Principles and Applications. Cambridge University Press, 636 pp. Chrisman, J., D. Rinderknecht, and R. Hamilton, 1994: WSR-88D clutter suppression and its impact on meteorological data interpretation. Preprints, First WSR-88D User's Conf., Norman, OK, WSR-88D Operational Support Facility, 9-20. Crum, T. D., R. E. Saffle, and J. W. Wilson, 1998: An update on the NEXRAD program and future WSR-88D support to operations. Wea. Forecasting, 13, 253-262, doi:10.1175/ 1520-0434(1998)013,0253:AUOTNP.2.0.CO;2. Doviak, R., and D. Zrnic, 1993: Doppler Radar and Weather Observations. 2nd ed. Academic Press, 562 pp. Fabry, F., and J. Gadoury, 2009: Another method for ground target identification and filtering using spectral processing of dualpolarization returns. 34th Conf. on Radar Meteorology, Williamsburg, VA, Amer. Meteor. Soc., P5.19. [Available online at https://ams.confex.com/ams/34Radar/techprogram/ paper_156127.htm.] Fornasiero, A., P. P. Alberoni, R. Amorati, L. Ferraris, and A. C. Taramasso, 2005: Effects of propagation conditions on radar beam-ground interaction: Impact on data quality. Adv. Geosci., 2, 201-208, doi:10.5194/adgeo-2-201-2005. Friedrich, K., U. Germann, and P. Tabary, 2009: Influence of ground clutter contamination on polarimetric radar parameters. J. Atmos. Oceanic Technol., 26, 251-269, doi:10.1175/ 2008JTECHA1092.1. Hamidi, S., and D. S. Zrnic, 1981: Considerations for the design of ground clutter cancelers for weather radars. Preprints, 20th Conf. onRadarMeteorology, Boston,MA, Amer.Meteor. Soc., 319-326. Harris, F. J., 1978: On the use of windows for harmonic analysis with the discrete Fourier transform. Proc. IEEE, 66, 51-83, doi:10.1109/PROC.1978.10837. Hubbert, J. C., M. Dixon, and S. M. Ellis, 2009: Weather radar ground clutter. Part II: Real-time identification and filtering. J. Atmos. Oceanic Technol., 26, 1181-1197, doi:10.1175/ 2009JTECHA1160.1. Ice, R., D. Warde, D. Sirmans, and D. Rachel, 2004: Report on Open RDA: RVP-8 signal processing. Part I: Simulation study. WSR-88D Radar Operations Center Rep., 87 pp. _____, and Coauthors, 2009: Automatic clutter mitigation in the WSR-88D, design, evaluation, and implementation. 34th Conf. on Radar Meteorology, Williamsburg, VA, Amer. Meteor. Soc., P5.3. [Available online at https://ams.confex.com/ ams/34Radar/techprogram/paper_155409.htm.] Kilambi, A., F. Fabry, and S. M. Torres, 2013: Improving polarimetric radar parameter estimates and target identification: A comparison of different approaches. 36th Conf. on Radar Meteorology, Breckenridge, CO, Amer. Meteor. Soc., 328. [Available online at https://ams.confex.com/ams/36Radar/ webprogram/Paper229001.html.] Lei, L., G. Zhang, R. J. Doviak, R. Palmer, B. L. Cheong, M. Xue, Q. Cao, and Y. Li, 2012: Multilag correlation estimators for polarimetric radar measurements in the presence of noise. J. Atmos. Oceanic Technol., 29, 772-795, doi:10.1175/ JTECH-D-11-00010.1. Meymaris, G., J. K. Williams, and J. C. Hubbert, 2009: Performance of a proposed hybrid spectrum width estimator for the NEXRAD ORDA. 25th Int. Conf. on Interactive Information and Processing Systems for Meteorology, Oceanography and Hydrology, Phoenix, AZ, Amer. Meteor. Soc., 11B.1. [Available online at https://ams.confex.com/ams/89annual/ techprogram/paper_145958.htm.] Moisseev, D. N., and V. Chandrasekar, 2009: Polarimetric spectral filter for adaptive clutter and noise suppression. J. Atmos. Oceanic Technol., 26, 215-228, doi:10.1175/2008JTECHA1119.1. NEXRAD Radar Operations Center, 1996: WSR-88D system specification. Specification 2810000A, NEXRAD Operational Support Facility, 274 pp. [Available online at http:// www.roc.noaa.gov/wsr88d/PublicDocs/NTR96.pdf.] Nuttall, A. H., 1981: Some windows with very good sidelobe behavior. IEEE Trans. Acoust. Speech Signal Process., 29, 84-91, doi:10.1109/TASSP.1981.1163506. Pamment, J. A., and B. J. Conway, 1998: Objective identification of echoes due to anomalous propagation in weather radar data. J. Atmos. Oceanic Technol., 15, 98-113, doi:10.1175/ 1520-0426(1998)015,0098:OIOEDT.2.0.CO;2. Park, H., A. V. Ryzhkov, D. S. Zrnic, and K. E. Kim, 2009: The hydrometeor classification algorithm for the polarimetric WSR-88D: Description and application to an MCS. Wea. Forecasting, 24, 730-748, doi:10.1175/2008WAF2222205.1. 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., 295-300. Siggia, A., and J. R. Passarelli, 2004: Gaussian model adaptive processing (GMAP) for improved ground clutter cancellation and moment calculation. Proc. Third European Conf. on Radar in Meteorology and Hydrology, Visby, Gotland, Sweden, ERAD, 67-73. Sirmans, D., 1987: NEXRAD suppression of land clutter echo due to anomalous microwave propagation-Part 1. NSSL Rep. to NEXRAD JSPO, 73 pp. -, 1992: Clutter filtering in the WSR-88D. OSF Tech. Rep., 125 pp. -, and J. Dooley, 1980: Ground clutter statistics of a 10 cm ground based radar. Preprints, 19th Conf. on Radar Meteorology, Miami Beach, FL, Amer. Meteor. Soc., 184-191. Torres, S. M., and D. S. Zrnic, 1999: Ground clutter canceling with a regression filter. J. Atmos. Oceanic Technol., 16, 1364-1372, doi:10.1175/1520-0426(1999)016,1364:GCCWAR.2.0.CO;2. -, C. D. Curtis, D. S. Zrnic, and M. Jain, 2007: Analysis of the new NEXRAD spectrum width estimator. 33rd Int. Conf. on Radar Meteorology, Cairns, QLD, Australia, Amer. Meteor. Soc., P7.8. [Available online at https://ams.confex.com/ams/ 33Radar/techprogram/paper_123048.htm.] -, D. Warde, B. Gallardo, K. Le, and D. Zrnic, 2010: Signal design and processing techniques for WSR-88D ambiguity resolution: Staggered PRT algorithm updates, the CLEANAP filter, and the hybrid spectrum width estimator. Part 14, NSSL Rep., 145 pp. [Available online at http://cimms.ou.edu/; torres/Documents/NSSL Ambiguity Report - Part 14.pdf.] -,-, andD.Zrnic, 2012: Signal design and processing techniques for WSR-88D ambiguity resolution: The CLEAN-AP filter. Part 15, NSSL Rep., 65 pp. [Available online at http://cimms.ou.edu/; torres/Documents/NSSL Ambiguity Report - Part 15.pdf.] Warde, D., and S. Torres, 2013: Automated real-time mitigation of ground clutter contamination for dual-polarization Doppler weather radars using the alternating transmission mode. 29th Conf. on Environmental Information Processing Technologies, Austin, TX, Amer. Meteor. Soc., 17. [Available online at https://ams.confex.com/ams/93Annual/webprogram/ Paper217703.html.] -, and -, 2014: The autocorrelation spectral density for Doppler-weather-radar signal analysis. IEEE Trans. Geosci. Remote Sens., 52, 508-518, doi:10.1109/TGRS.2013.2241775. Zrnic, D. S., 1975: Simulation of weatherlike Doppler spectra and signals. J. Appl. Meteor., 14, 619-620, doi:10.1175/ 1520-0450(1975)014,0619:SOWDSA.2.0.CO;2. -, V. M. Melnikov, and I. Ivic, 2008: Processing to obtain polarimetric variables on the ORDA. NOAA/NSSL Rep., 55 pp. [Available online at http://www.nssl.noaa.gov/ publications/wsr88d_reports/RPT_DoNotDestroy.doc.] SEBASTIÁN M. TORRES AND DAVID A. WARDE Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, and NOAA/OAR /National Severe Storms Laboratory, Norman, Oklahoma (Manuscript received 13 May 2013, in final form 28 March 2014) Corresponding author address: Sebastián Torres, National Weather Center, 120 David L. Boren Blvd., Norman, OK 73072. E-mail: [email protected] DOI: 10.1175/JTECH-D-13-00117.1 (c) 2014 American Meteorological Society |