1. Introduction
Improving radar-based quantitative precipitation estimation (QPE) has been one of the major goals of the radar-meteorology community for more than 50 years. Retrieval of rain rate R requires accurate radar measurements and a relation that relates radar observations to rain rate. For example, Marshall and Palmer (1948) utilized the measured raindrop size distribution (RSD), the simulated reflectivity Z, and a power-law relation Z = aRb (Z–R) to estimate rainfall rate. However, Z is approximated by the sixth moment of the RSD, and R is approximated by the 3.67th moment. The Z–R relation varies vastly in convective and stratiform precipitation and in different climatological regions because of the natural variability in RSD (Battan 1973).
Seliga and Bringi (1976) proposed utilizing the dual-polarization radar measurement known as differential reflectivity ZDR (dB) to reduce variability in the radar-based QPE. The ZDR estimates the mean raindrop size by measuring the ratio between the horizontal reflectivity ZHH and the vertical reflectivity ZVV. The axis ratios of raindrops decrease as the raindrop sizes become larger. The radar-based QPE is thus improved by including the RSD information from the ZDR (Seliga et al. 1981; Gorgucci et al. 1995; Ryzhkov and Zrnić 1995). Furthermore, the dual-polarization radar measurement known as specific differential phase KDP (° km−1) that is proportional to the fourth moment of the RSD is found to improve the accuracy of rainfall estimation (Ryzhkov and Zrnić 1996; Bringi and Chandrasekar 2001). The KDP is unbiased by the transmitted power of the radar, and it is ideal for estimating rain rate when rain is mixed with hail or randomly tumbling ice particles (Bringi and Chandrasekar 2001). With the inclusion of ρhv (copolar correlation coefficient), the radar-data quality control is significantly improved through distinguishing rain echoes from ground clutter and other nonmeteorological signals. Thus, overestimations of the radar-based QPE due to these artifacts are significantly reduced when the ρhv are used.
A number of dual-polarization QPE algorithms are proposed in the literature. These algorithms in various “power law” forms attempt to utilize the advantage of dual-polarization moments (DPMs) by using one parameter [R(KDP): Gorgucci et al. 1999; Bringi and Chandrasekar 2001; Ruzanski and Chandrasekar 2012], two parameters [R(ZHH, ZDR): Ryzhkov et al. 2005a; Illingworth and Thompson 2005; R(KDP, ZDR): Ryzhkov and Zrnić 1995; Ryzhkov et al. 2005a; R(ZHH, KDP): Lee 2006], or three parameters [R(ZHH, KDP, ZDR): Gorgucci et al. 2001; Bringi et al. 2002]. These various forms of dual-polarization QPE algorithm have shown pronounced improvements relative to Z–R relations.
To construct the aforementioned algorithms, specific RSD models (e.g., gamma distribution) or observed RSDs are combined with a rigorous electromagnetic-scattering model (Vivekanandan et al. 1991; Bringi and Chandrasekar 2001). The coefficients (e.g., a and b of R = aZb) of these power-law QPE algorithms are thus obtained by linear regression. Because of various RSDs and assumptions in the microphysical model (e.g., axis ratio and canting angle), a variety of power-law QPE algorithms with different combinations of DPMs are reported in the literature (Ryzhkov et al. 2005a; Lee 2006). Lee (2006) has shown that RSD variability is one of the major sources of error in radar-based QPE. Therefore, for reducing error in QPE, these power-law algorithms require additional “empirical tuning” to take into account spatiotemporal RSD variations as a function of climatological regions.
Errors in DPM measurements lead to errors in QPE. The ZDR can be noisy because of the low sampling rate of a fast-scanning radar (Bringi and Chandrasekar 2001; Melnikov 2004). The KDP is estimated from the measurements of differential propagation phase ΦDP as a range-derivative variable. The artifacts due to the fluctuations in ΦDP from backscattering Mie phase shift δ introduced into KDP are reduced by the filtering technique (Hubbert et al. 1993; Hubbert and Bringi 1995; Bringi and Chandrasekar 2001; Ryzhkov et al. 2005a; Wang and Chandrasekar 2009; Giangrande et al. 2013), but the filtering techniques introduce additional uncertainty in KDP-related QPEs. Also, in the case of light rain, KDP is too small to be estimated with sufficient accuracy. It is not uncommon that a particular dual-polarization QPE algorithm performs well in one rain event but fails in another event because of unsatisfactory DPM data quality.
The combining of various dual-polarization QPE algorithms using decision-tree logic is proposed (Cifelli et al. 2002, 2011; Ryzhkov et al. 2005a,b; Giangrande and Ryzhkov 2008). This composite method has been shown to be more accurate than an individual dual-polarization algorithm (Ryzhkov et al. 2005a). This method helps one to avoid using the KDP for light-rain estimation. For example, R(Z) < 6 mm h−1 is considered to be light rain in Ryzhkov et al. (2005b). Thus, R(KDP) is only applied when R(Z) > 6 mm h−1. Cifelli et al. (2011) suggest a more sophisticated method for detecting light rain using multiple thresholds: KDP < 0.3° km−1, ZHH < 38 dBZ, and ZDR < 0.5 dB. Nevertheless, the threshold for differentiating light rainfall from moderate rainfall is dependent on the data quality of a particular radar and the RSD characteristics of a specific climatological value or rain event. Therefore, these thresholds may not be applicable across the board for all radars or for various rain climatological datasets for which RSDs change as a function of time and space. Another major drawback of the decision-tree-logic QPE algorithm is that the transitions between individual algorithms introduce discontinuity in the rainfall estimate.
Pepler et al. (2011) and Pepler and May (2012) introduced a blended technique that combined various QPE algorithms objectively according to their respective theoretical error characteristics. In this technique, observational error characteristics of each DPM are estimated theoretically as a function of radar samples and Doppler spectrum width (Bringi and Chandrasekar 2001). Thus, the theoretical errors of various QPE algorithms are derived accordingly. The optimal rain estimation is subsequently obtained by combining various rainfall estimations weighted by their respective errors. They found that the weighted combination of dual-polarization algorithms offers modest improvements over the decision-tree-logic method.
The quality of various DPMs is not only determined by the number of radar samples and Doppler spectrum width, however; it is also influenced by a range of factors under various measurement conditions. Possible uncertainties in ZDR of up to ±0.7 dB and in ΦDP of up to ±10° in the presence of nonuniform beamfilling (Ryzhkov 2007) introduce significant error in QPE (Bringi and Chandrasekar 2001; Lee 2006). In addition, the measurement volume may contain mixed-phase hydrometers (e.g., coexistence of rain and hail) or various artifacts (e.g., birds, insects, and elevated ground clutter via sidelobes). Hubbert et al. (2010a,b) demonstrated in the case of a simultaneously transmitted mode that the orientation of scatters causes cross coupling between horizontally and vertically polarized signals. Furthermore, ZDR can be contaminated by a wet radome (Gorgucci et al. 2013). Consequently, even the advanced power-law-based QPE algorithms that make use of decision-tree and blended techniques become vulnerable to the aforementioned observation errors.
In summary, these power-law-based dual-polarization QPE algorithms are affected by two major factors: 1) natural RSD variability and 2) observational errors in DPMs. The variational algorithm that minimizes the above-mentioned two factors has the potential to improve the accuracy of dual-polarization QPE. A 1D variational scheme for retrieving rainfall rate using S-band polarimetric radar is introduced by Furness (2005) and Hogan (2007; hereinafter H07). This variational scheme dynamically adapts to natural variability of RSD, observational error, and rain climatological information. H07 shows that the variational algorithm can successfully utilize the DPMs and the background information concurrently for rainfall estimation. Also, it is relatively less sensitive to observation error than are power-law-based dual-polarization QPE algorithms.
Figueras i Ventura et al. (2010) applied the variational algorithm to French C-band radar network measurements, but the QPE validation against the rain gauges was unsatisfactory. They concluded that the higher attenuation in C band (especially in heavy-precipitation cases) and improper treatment of measurement error variance might have caused unsatisfactory QPE. Performance of the variational algorithm critically depends on the accuracy of the error variances of DPMs and background information.
To objectively estimate spatiotemporal error variances of DPMs, a statistical diagnostic method (Desroziers et al. 2005) was proposed by Chang et al. (2014; hereinafter C14). The usage of proper error variances of DPMs greatly improved the accuracy of the attenuation correction of X-band polarimetric radar data. Yet, the improvement to the QPE due to the usage of proper error variances was not investigated in C14. The main goal of this study is to apply the variational scheme from C14 (hereinafter referred to as variational QPE) and to investigate its applicability to QPE. The performance of the variational QPE is evaluated by comparing it with one of the most widely used “conventional” power-law dual-polarization QPE algorithms [hereinafter referred to as “dual-pol QPE,” or R(ZHH, ZDR)]. An epic Colorado flood event that produced nearly 300 mm of rainfall in 48 h between 11 and 12 September 2013 (Hamill 2014) and that has data from operational S-band dual-polarization radar and rain gauges will be carefully examined in this study.
A brief description of the variational QPE algorithm is presented in section 2. The variational QPE is first evaluated with simulated DPMs by using observing system simulation experiments (OSSE) in section 3. Practical applicability of this method is demonstrated in the case of an epic Colorado flood event. The data used in this study are described in section 4. A comparison of the variational QPE and the polarimetric Denver (KFTG) Next Generation Weather Radar (NEXRAD) level-3 QPE products with gauge measurements is presented in section 5. Section 6 summarizes the results and describes future work.
2. Description of variational QPE
a. Variational method
In the power-law relation Z = aRb, parameters a and b represent variations in RSD (Steiner and Smith 2000; Steiner et al. 2004). The multiplicative parameter a is much more sensitive to RSD variations than is the exponent parameter b (Bringi and Chandrasekar 2001; H07). Previous studies (Furness 2005; H07; C14) have shown that it is adequate to fix b and that a value of 1.5 is satisfactory. Hence, the parameter a of Z = aR1.5 represents the RSD variations (H07; C14). The variational method retrieves the “target” parameter a. In practice, the natural logarithm of coefficient a of Z–R [i.e., ln(a)] is retrieved to avoid an unrealistic negative value of a. Hereinafter, the symbol 
For a specified RSD, the rigorous electromagnetic-scattering and wave-propagation model computes ZHH, ZDR, KDP, AH, and ADP at S-band frequency and the “referenced” rainfall rate is derived. To construct the forward models, the values of gamma parameters are varied for emulating natural variability of the RSD. The median volume diameter D0 (mm) varies from 0.1 to 6 mm, and the value of the intercept parameter log10(N0) (mm−1−μ m−3) varies from 1.0 to 16.0. The shape parameter μ (dimensionless) is fixed at 5 (H07; C14). The forward models are implemented as lookup tables for numerical computations (Fig. 1). Cao et al. (2013) used a constrained-gamma RSD for implementing the forward model as a table lookup. The variational scheme considers rain below freezing level in this study. Mixed-phase precipitation, such as rain–hail or rain–snow, is not considered.
The S-band forward models for ZDR, KDP, AH, and ADP as a function of Z/R.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
For given first guess of x, the measured ZHH is first corrected for attenuation along the radar beam by a predicted AH from the forward model. Using the corrected ZHH, the forward model predicts an intrinsic ZDR, ADP, and KDP. Subsequently, the intrinsic ZDR is reduced by the amount of ADP. The predicted ΦDP is integrated from the predicted KDP along the range. These predicted ZDR and ΦDP values are compared with corresponding measured values (i.e., y). The optimal estimations of x (i.e., 
The background term xb is included as an a priori so that an appropriate value of 
b. Estimation of the observation and background error variance
The performance of the variational QPE critically depends on the accuracy of the error covariance matrices (i.e., 
To estimate the 
The 
3. Observing system simulation experiments
As discussed in H07, the performance of the variational scheme depends on the relative weights given to the observations and the a priori (i.e., background term). The improvements in the variational QPE due to the usage of objectively derived 
The 96 NCAR S-Pol plan position indicator (PPI) data from the Dynamics of the Madden–Julian Oscillation (DYNAMO) field campaign (Yoneyama et al. 2013) were used to generate the “pseudo” DPMs—mainly, 
The pseudo radar observables of a single PPI scan: (a) reflectivity 
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
a. Sensitivity of variational-based QPE to the background term
In the light–medium-rain region (ZHH < 40 dBZ), S-band radar measurements exhibit weak ΦDP increment. Despite the fact that the attenuation effect is often considered to be undesirable in the application of DPMs, the estimation of 
The sensitivity of the variational QPE to a set of 
As shown in Fig. 3a, the standard deviation of the retrieved 
The standard deviations of (a) retrieved 
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
b. The performance of variational QPE with regard to radar measurement errors
The histogram of coefficients a1, b1, and c1 of 
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
The dual-pol and variational QPE algorithms were applied to the 96 PPIs of pseudo DPMs. The computed rainfall rates Rsim from the retrieved RSDs at each gate were considered to be the referenced rainfall rate. Plots of the number density function (NDF) between the estimated and the referenced rainfall rate are shown in Fig. 5. The rainfall estimation from the variational algorithm (Fig. 5a) shows less scatter than does that from the dual-pol algorithm (Fig. 5b). Because the variational algorithm utilizes the ZDR, ΦDP, and 
The NDFs of retrieved rainfall rate from the referenced rainfall rate and (a) variational QPE or (b) dual-polarization radar-based QPE. The numbers are shown in logarithmic scale.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
Three different rainfall estimates were retrieved from the pseudo DPMs for comparison: 1) the variational QPE with an improper 
Both of the variational QPEs (VAR_d_O and VAR_i_O) used the same 
The instantaneous rainfall estimates from 96 PPIs were obtained for evaluating the performances of the dual-pol and the variational QPE algorithms. As shown in Fig. 6a, the VAR_d_O has the lowest RRMSE for all rain intensities. The VAR_i_O has an RRMSE that is comparable to that of R(ZHH, ZDR) and is much higher than the RRMSE of VAR_d_O. This suggests that the diagnosed 
The (left) RRMSE and (right) RBIAS for variational QPE with improperly assumed 
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
In Fig. 6b, the RBIAS of the VAR_d_O is the lowest for all of the rain intensities > 0.2 mm h−1. It varies between 0% and −6% as a function of rain intensity. The RBIAS of VAR_i_O are between 17% and 21%. The R(ZHH, ZDR) overestimates by 17% for rain intensities > 0.2 mm h−1. The RBIAS of R(ZHH, ZDR) decreases with increasing rain intensity and underestimates by approximately 1% for rain intensity > 60 mm h−1.
To quantify the effect of observation error, the rainfall estimation was obtained using ideal DPMs (
The consistently smaller values of RRMSE and RBIAS indicate that the variational algorithm is less vulnerable to ZDR observation error than is the R(ZHH, ZDR) power-law algorithm. Previous studies (Blackman and Illingworth 1997; Lee 2006) have suggested that spatiotemporal smoothing reduces uncertainty in ZDR and hence errors in hourly and daily averaged QPE. In Fig. 6c, the hourly VAR_d_O has the lowest RRMSE (21%–50%). On the other hand, the RRMSE of hourly VAR_i_O and R(ZHH, ZDR) are 27%–75% and 24%–57%, respectively. The R(ZHH, ZDR) has smaller RRMSE than does the VAR_i_O, except for rain intensity >55 mm h−1. In Fig. 6d, RBIAS of the VAR_d_O shows no significant change for instantaneous and hourly rain estimations since the bias could not be removed by the smoothing.
The OSSE results for the daily rainfall estimates are shown in Figs. 6e and 6f. The RRMSE for VAR_d_O is the lowest (12%–14%). The RRMSE for VAR_i_O (27%–32%) is the highest for all rain intensities. The RBIAS values for VAR_i_O are similar to the instantaneous and hourly rainfall estimates. The values of RBIAS are 0.0%–2.5% for VAR_d_O, 14%–16% for R(ZHH, ZDR), and 22%–23% for VAR_i_O. The lowest RBIAS of VAR_d_O is noteworthy since all three estimates used the same radar measurements.
The OSSEs show that the observational errors of simulated DPMs have a distinct influence on various QPE algorithms. The variational algorithm consistently performs better than R(ZHH, ZDR) in terms of the RRMSE and RBIAS when dynamically varying spatiotemporal errors were used. With the increase of time averaging in the hourly and the daily rainfall, the R(ZHH, ZDR) outperformed the variational algorithm when an improper 
c. Accuracy of variational QPE as a function of ZDR observation error
Sensitivity of the variational QPE to the ZDR observation error was examined by varying the standard deviation of random observation error of ZDR from 0.2 to 0.5 dB. The observation error of ZHH and ΦDP remained the same at 1 dB and 2°, respectively. As shown in Fig. 7a, the RRMSE of instantaneous QPE increases with an increase in the observation error of ZDR for both the variational and R(ZHH, ZDR) algorithms. The RRMSE of R(ZHH, ZDR) is 150%–750% for Rsim > 0.2 mm h−1 and 100%−400% for Rsim > 10 mm h−1. The RRMSE of the variational algorithm with the diagnosed 
The (a) RRMSE and (b) RBIAS of the variational and R(ZHH, ZDR) QPEs as a function of ZDR observation errors. The red solid and dashed lines represent the variational algorithm for rain-intensity thresholds 0.2 and 10 mm h−1, respectively. The blue solid and dashed lines have the same respective thresholds but for R(ZHH, ZDR) QPE.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
The higher RRMSE caused by the ZDR observation error can be minimized by applying various filtering techniques (Blackman and Illingworth 1997; Lee 2006). For example, the NEXRAD operational QPE algorithm smooths out nonmeteorological variability in the measurements using the most frequent value in the nine adjacent radar sample bins along each radial. This technique is referred as a “Mode-9 filter” (Istok et al. 2009). These filtering techniques are customized to a particular radar and scan strategy, however. There is no consensus in the research community as to the usage of one filtering technique versus another. Since evaluation of the performance of the filtering technique is not the main goal of this study, it will not be discussed in this paper. Note also that, without any filtering, the variational QPE outperforms the R(ZHH, ZDR) algorithm consistently in the presence of large ZDR errors.
4. Data
In the previous section, accuracy of the variational and the power-law dual-pol QPE algorithms has been compared using the OSSE studies. To demonstrate performance of the variational QPE, the measurements collected during an epic Colorado flood event that occurred between 11 and 12 September 2013 (Hamill 2014) were analyzed. This event produced nearly 300 mm of accumulated rainfall in 48 h. The results from the variational QPE were compared with NEXRAD operational products and rain gauge data.
a. Rain gauge data
The rain gauge data from an “automated local evaluation in real time” (ALERT; http://alert5.udfcd.org) network, maintained by the Urban Drainage and Flood Control District (UDFCD) of Denver, Colorado, was used for comparison. Raw ALERT gauge data were converted from tip counts to physical precipitation amounts. Then time-history plots for individual ALERT gauge sites were carefully reviewed to flag sites with data spikes. Hourly rainfall maps that are based on these data were also examined to detect anomalous data values. In situ measurements from 100 rain gauges were used for validation.
b. Parsivel disdrometer data
Data were analyzed from two Particle Size Velocity (Parsivel) disdrometers (Löffler-Mang and Joss 2000) installed in Fourmile Canyon, Colorado, and owned by NCAR (their locations are marked in Fig. 9). The RSD measurements by the Parsivel instruments during the flood event were compared with the nearby rain gauges, operated by UDFCD, located less than 2 km from the Parsivel. In general, the peak rainfall intensity and the total accumulation during the flood event were consistent with the operational gauge measurements.
c. KFTG level-2 data
The level-2 data at 0.9° elevation scan from KFTG were processed by the variational algorithm. The ground-clutter contamination at 0.9° was less than at 0.5° elevation scan. Radar measurements associated with ρhv < 0.8 were flagged as nonmeteorological data, and they were deleted. Some poor-quality and noisy suspicious ZDR measurements remained even after the aforementioned ρhv threshold.
d. KFTG level-3 products
The dual-pol QPE algorithm in the NEXRAD Radar Product Generator (RPG) processes the data from the Radar Data Acquisition computer at “super resolution” of 0.5° by 0.25 km. The RPG then low-pass filters the data and decreases the spatial resolution to 1° by 0.25 km before further processing. The NEXRAD dual-pol QPE level-3 rain product—namely, the Digital Precipitation Rate (DPR)—is the final instantaneous precipitation rate as based on the dual-pol QPE algorithm. The DPR was downloaded from the National Climatic Data Center data archive. The hourly QPE product from the Z–R legacy Precipitation Processing Subsystem was also considered for comparison.
5. Comparison between radar-based QPEs and rain gauge
As discussed in section 3a, the 
(a) The values of background term a (
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
The NEXRAD was operated in “volume coverage pattern” 12, with 15 sample pairs of simultaneously transmitted polarimetric radar returns in 4.2 min per cycle during this event. The ZDR had a higher observation error because of the lower number of samples collected during the fast scanning strategy. Figure 8b shows the time series of NDF of the theoretically derived standard deviation of the ZDR observation error [Bringi and Chandrasekar 2001, their Eq. (6.142)]. The ZDR error varied between 0.1 and 1.0 dB. The mean standard deviations of the ZDR observation error estimated from the theoretical method were between 0.35 and 0.55 dB. This range is higher than the desired value of <0.2 dB recommended by Melnikov (2004). The ZDR observation error from the diagnostic method is also plotted in Fig. 8b. Even though the actual ZDR observation error is unknown, the closeness between the ZDR observation errors indicates that the diagnostic method has the potential for retrieving sufficiently accurate ZDR observation errors.
a. Total accumulated rainfall validation
The 48-h total accumulated rainfall from Z–R, dual-pol, and the variational QPE is shown in Fig. 9. The Z–R QPE (Z = 250R1.2 during the 2013 Colorado flood event) in Fig. 9a shows the lowest accumulated rainfall estimation. In regions 1 and 2, the total accumulated rainfall is 250–300 mm (as measured by rain gauges); the Z–R QPE estimation is ~200 mm in region 1, however. The dual-pol QPE estimation of ~250 mm is higher than the Z–R QPE, as shown in Fig. 9b. Underestimation of dual-pol QPE in regions 1 and 2 is noticeable. On the other hand, the variational QPE of 300 mm is the closest to the rain gauge observation, as shown in Fig. 9c.
The 48-h accumulated rainfall map (mm) for 11–12 Sep 2013 from the KFTG NEXRAD level-3 Z–R (a) QPE product and (b) dual-pol QPE and from the (c) variational QPE. Two Parsivel disdrometer sites are shown as red triangles (a detailed comparison of radar-estimated rain and measurements from these stations is shown in Fig. 12, below). (d) Scatterplot of 48-h accumulation between rain gauges and the radar-based estimations shown in (a),(b), and (c). RRMSE and RBIAS values are included in the legend.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
Figure 9d compares 48-h accumulated radar rainfall estimations with data from rain gauges. The RBIAS and RRMSE were derived using Eqs. (14) and (15). The radar estimations were averaged within a 2-km radius of rain gauges. The results indicate that the variational QPE performed better than dual-pol and Z–R KFTG level-3 products. The variational QPE underestimates by 11% while the dual-pol and Z–R QPEs underestimate by 24% and 51%, respectively. Moreover, the RRMSE of the variational QPE is 29%, which is less than that for the KFTG level-3 dual-pol and Z–R QPEs.
Sensitivity of the variational QPE to fixed 
RRMSE and RBIAS of 48-h accumulated rainfall derived by the variational QPE algorithm using diagnosed and fixed ZDR errors.
b. Hourly rainfall validation
Figure 10 compares hourly rainfall from the variational and dual-pol QPE. The variational QPE has smaller values of RRMSE (2.15%) and RBIAS (−12.83%) than the dual-pol QPE RRMSE (2.45%) and RBIAS (−24.91%) for all rain intensities > 0.2 mm h−1. The RRMSE and RBIAS of hourly rainfall estimates were further examined by gradually increasing the rain-intensity threshold as shown in Fig. 11. In general, RRMSE decreases as rainfall intensity increases, as shown in Fig. 11a. The RRMSEs for rainfall rate > 20 mm h−1 are about 26% and 34% for the variational and dual-pol QPE, respectively. The variational QPE consistently outperforms dual-pol QPE. Figure 11b shows that RBIAS of the variational QPE is always smaller (from −9% to −14%) than that of the dual-pol QPE (from −24% to −29%). Overall, the results indicate that the variational QPE estimates more accurate hourly rainfall.
The NDFs of the hourly rainfall rate from rain gauge measurements and (a) dual-pol QPE or (b) variational QPE. The numbers are shown in logarithm scale.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
(a) RRMSE and (b) RBIAS of the variational (red lines) and dual-pol (blue lines) QPE as a function of hourly rainfall threshold.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
c. The validation of instantaneous rainfall rate
Comparisons of 5-min instantaneous QPEs from dual-pol and the variational method with the rain gauge data (5 min) are shown in Fig. 12. The RRMSEs are similar for both QPEs. One of the major reasons for high RRMSEs of instantaneous QPE is the large difference in sampling volumes between the radar and the rain gauges (Ryzhkov et al. 2005a).
NDFs of the 5-min rainfall rate from rain gauge measurements and (a) dual-pol QPE or (b) variational QPE. The numbers are shown in logarithm scale.
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
The RBIAS of the variational QPE is approximately −18%, whereas the RBIAS of dual-pol QPE is approximately −29%. The underestimation of dual-pol QPE is more pronounced with increasing rainfall rate. Significant overestimation of dual-pol QPE is associated with low rain gauge values, as shown in the red-outlined box of Fig. 12a. This mismatch is caused by noisy ZDR due to the low sampling rate of KFTG in the regions of weak dual-polarimetric radar measurements.
6. Summary
Detailed investigations of variational-based QPE algorithm using OSSE studies and NEXRAD S-band polarimetric radar in a Colorado flood event between 11 and 12 September 2013 are presented in this paper. The results indicate that the variational algorithm retrieves more accurate rain estimation than does a conventional dual-pol power-law algorithm. Because the variational method dynamically combines the forward model, measurements, and background information using appropriate error variance, its retrieval is more accurate.
The OSSE studies show that, even though 
Even in the presence of a large observation error as high as 0.5 dB in ZDR, the RBIAS of the variational QPE remains < 15%. On the other hand, the RBIAS of conventional dual-pol QPE [i.e., R(ZHH, ZDR)] is >100%, and it is highly vulnerable to the ZDR measurement noise. Smoothing the data helps to reduce measurement noise, but it degrades spatiotemporal resolutions of precipitation and underestimates the heavier rainfall.
The variational QPE was compared with NEXRAD level-3 Z–R and dual-pol QPE operational products for the epic 2013 Colorado flood event. The results indicate that the dual-pol QPE has significantly improved the rainfall estimation relative to the Z–R QPE primarily because of the utilization of DPMs. The dual-pol QPE operational radar products underestimated the rain rate, however. It is postulated that empirical power-law relations that are based on predetermined rain microphysics are not appropriate to this particular “tropical rain” event. The variational QPE retrieved more accurate rainfall estimation than did dual-pol QPE in this particular event despite the fact that both algorithms used the same dual-polarization radar measurements from the KFTG radar.
The advantage of the simple but robust operational dual-pol QPE power-law algorithm is its capability to retrieve rain in mixed-phase precipitation such as rain–hail or rain–graupel. In contrast, the variational algorithm presented in this paper is applicable only to rain that is below the melting layer. One of the limitations of the variational algorithm is that it cannot be implemented for real-time operational application in the current configuration, because it is computationally expensive relative to the conventional dual-pol QPE algorithm. In the future, possible ways to reduce the computation time will be investigated by thinning the radar data. Also, performance of the variational algorithm in a variety of rain events that include cold- and warm-season events at various geographical locations will be investigated.
Acknowledgments
The authors thank Jim Wilson and Scott Ellis of NCAR for fruitful discussions. The assistance of Daniel Megenhardt of NCAR in providing the KFTG radar data is gratefully acknowledged. The project is funded by the National Science Council of Taiwan under Grants NSC 100-2119-M008-041-MY5. Any opinions, findings, and conclusions, or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.
APPENDIX
Implementation of the Discrete-Approximation and Statistical Diagnostic Methods
The flowchart of processing procedures of the variational QPE algorithm is shown in Fig. A1. The data quality-control processes, including system bias in ZHH and ZDR, and nonmeteorological-signal removal, are performed before applying the variational QPE algorithm (step 1). Subsequently, the first guess of 
Flowchart describing the various numbered steps (as numbered outside the top-right corners of the boxes) in the procedure for estimating discrete-approximation-based 
Citation: Journal of Applied Meteorology and Climatology 55, 7; 10.1175/JAMC-D-15-0222.1
REFERENCES
Anagnostou, M., E. Anagnostou, and J. Vivekanandan, 2006: Correction for rain path specific and differential attenuation of X-band dual-polarization observations. IEEE Trans. Geosci. Remote Sens., 44, 2470–2480, doi:10.1109/TGRS.2006.873204.
Battan, L. J., 1973: Radar Observation of the Atmosphere. University of Chicago Press, 324 pp.
Berkowitz, D. S., J. A. Schultz, S. Vasiloff, K. L. Elmore, C. D. Payne, and J. B. Boettcher, 2013: Status of dual pol QPE in the WSR-88D network. 27th Conf. on Hydrology, Austin, TX, Amer. Meteor. Soc., 2.2. [Available online at https://ams.confex.com/ams/93Annual/webprogram/Manuscript/Paper221525/AMS27thHydro_2.2Berkowitz_et_al.pdf.]
Blackman, T. M., and A. J. Illingworth, 1997: Examining the lower limit of KDP rainrate estimation including a case study at S-band. Preprints, 28th Conf. on Radar Meteorology, Austin, TX, Amer. Meteor. Soc., 117–118.
Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar: Principles and Applications. Cambridge University Press, 636 pp.
Bringi, V. N., V. Chandrasekar, N. Balakrishnan, and D. S. Zrnić, 1990: An examination of propagation effects in rainfall on radar measurements at microwave frequencies. J. Atmos. Oceanic Technol., 7, 829–840, doi:10.1175/1520-0426(1990)007<0829:AEOPEI>2.0.CO;2.
Bringi, V. N., G.-J. Huang, V. Chandrasekar, and E. Gorgucci, 2002: A methodology for estimating the parameters of a gamma raindrop size distribution model from polarimetric radar data: Application to a squall-line event from the TRMM/Brazil campaign. J. Atmos. Oceanic Technol., 19, 633–645, doi:10.1175/1520-0426(2002)019<0633:AMFETP>2.0.CO;2.
Cao, Q., G. Zhang, and M. Xue, 2013: A variational approach for retrieving raindrop size distribution from polarimetric radar measurements in the presence of attenuation. J. Appl. Meteor. Climatol., 52, 169–185, doi:10.1175/JAMC-D-12-0101.1.
Chang, W.-Y., J. Vivekanandan, and T.-C. C. Wang, 2014: Estimation of X-band polarimetric radar attenuation and measurement uncertainty using a variational method. J. Appl. Meteor. Climatol., 53, 1099–1119, doi:10.1175/JAMC-D-13-0191.1.
Cifelli, R., W. A. Petersen, L. D. Carey, S. A. Rutledge, and M. A. F. da Silva Dias, 2002: Radar observations of the kinematic, microphysical, and precipitation characteristics of two MCSs in TRMM LBA. J. Geophys. Res., 107, 8077, doi:10.1029/2000JD000264.
Cifelli, R., V. Chandrasekar, S. Lim, P. C. Kennedy, Y. Wang, and S. A. Rutledge, 2011: A new dual-polarization radar rainfall algorithm: Application in Colorado precipitation events. J. Atmos. Oceanic Technol., 28, 352–364, doi:10.1175/2010JTECHA1488.1.
Cocks, S. B., D. S. Berkowitz, R. Murnan, J. A. Schultz, S. Castleberry, K. Howard, K. L. Elmore, and S. Vasiloff, 2012: Initial assessment of the dual-polarization quantitative precipitation estimate algorithm’s performance for two dual-polarization WSR-88Ds. 28th Conf. on Interactive Information Processing Systems (IIPS), New Orleans, LA, Amer. Meteor. Soc., 7B.2. [Available online at https://ams.confex.com/ams/92Annual/webprogram/Manuscript/Paper203568/Initial%20Analysis%20of%20DP%20QPE%20Performance%20Final%2017%20Jan%202012.pdf.]
Desroziers, G., L. Berre, B. Chapnik, and P. Poli, 2005: Diagnosis of observation, background and analysis error statistics in observation space. Quart. J. Roy. Meteor. Soc., 131, 3385–3396, doi:10.1256/qj.05.108.
Figueras i Ventura, J., F. Kabeche, B. Fradon, R. Hogan, A. A. Boumahmoud, A. Illingworth, and P. Tabary, 2010: Extensive evaluation of polarimetric quantitative precipitation estimations (QPE) in ideal and less ideal conditions. Sixth European Conf. on Radar in Meteorology and Hydrology, Sibiu, Romania, ERAD, 6 pp. [Available from the corresponding author.]
Furness, G., 2005: Using optimal estimation theory for improved rainfall rates from polarization radar. Masters dissertation, Dept. of Mathematics, University of Reading, 66 pp.
Giangrande, S. E., and A. V. Ryzhkov, 2008: Estimation of rainfall based on the results of polarimetric echo classification. J. Appl. Meteor. Climatol., 47, 2445–2462, doi:10.1175/2008JAMC1753.1.
Giangrande, S. E., R. McGraw, and L. Lei, 2013: An application of linear programming to polarimetric radar differential phase processing. J. Atmos. Oceanic Technol., 30, 1716–1729, doi:10.1175/JTECH-D-12-00147.1.
Goodall, C., H. Austin, R. Frederick, D. A. Glenn, and L. E. Pagano, 2012: Experiences with dual polarization radar in operations at Newport/Morehead City NC. 28th Conf. on Interactive Information Processing Systems (IIPS), New Orleans, LA, Amer. Meteor. Soc., 5B.2. [Available online at https://ams.confex.com/ams/92Annual/webprogram/Manuscript/Paper200269/DualPol_MHX_AMS.pdf.]
Gorgucci, E., and V. Chandrasekar, 2005: Evaluation of attenuation correction methodology for dual-polarization radars: Application to X-band systems. J. Atmos. Oceanic Technol., 22, 1195–1206, doi:10.1175/JTECH1763.1.
Gorgucci, E., V. Chandrasekar, and G. Scarchilli, 1995: Radar and surface measurement of rainfall during CaPE: 26 July 1991 case study. J. Appl. Meteor., 34, 1570–1577, doi:10.1175/1520-0450-34.7.1570.
Gorgucci, E., G. Scarchilli, and V. Chandrasekar, 1999: Specific differential phase estimation in the presence of nonuniform rainfall medium along the path. J. Atmos. Oceanic Technol., 16, 1690–1697, doi:10.1175/1520-0426(1999)016<1690:SDPEIT>2.0.CO;2.
Gorgucci, E., G. Scarchilli, V. Chandrasekar, and V. N. Bringi, 2001: Rainfall estimation from polarimetric radar measurements: Composite algorithms immune to variability in raindrop shape–size relation. J. Atmos. Oceanic Technol., 18, 1773–1786, doi:10.1175/1520-0426(2001)018<1773:REFPRM>2.0.CO;2.
Gorgucci, E., V. Chandrasekar, and L. Baldini, 2006: Correction of X-band radar observation for propagation effects based on the self-consistency principle. J. Atmos. Oceanic Technol., 23, 1668–1681, doi:10.1175/JTECH1950.1.
Gorgucci, E., R. Bechini, L. Baldini, R. Cremonini, and V. Chandrasekar, 2013: The influence of antenna radome on weather radar calibration and its real-time assessment. J. Atmos. Oceanic Technol., 30, 676–689, doi:10.1175/JTECH-D-12-00071.1.
Hamill, T. M., 2014: Performance of operational model precipitation forecast guidance during the 2013 Colorado Front-Range floods. Mon. Wea. Rev., 142, 2609–2618, doi:10.1175/MWR-D-14-00007.1.
Hogan, R. J., 2007: A variational scheme for retrieving rainfall rate and hail reflectivity fraction from polarization radar. J. Appl. Meteor. Climatol., 46, 1544–1564, doi:10.1175/JAM2550.1.
Hubbert, J., and V. N. Bringi, 1995: An iterative filtering technique for the analysis of copular differential phase and dual-frequency radar measurements. J. Atmos. Oceanic Technol., 12, 643–648, doi:10.1175/1520-0426(1995)012<0643:AIFTFT>2.0.CO;2.
Hubbert, J., V. Chandrasekar, V. N. Bringi, and P. Meischner, 1993: Processing and interpretation of coherent dual-polarized radar measurements. J. Atmos. Oceanic Technol., 10, 155–164, doi:10.1175/1520-0426(1993)010<0155:PAIOCD>2.0.CO;2.
Hubbert, J., S. M. Ellis, M. Dixon, and G. Meymaris, 2010a: Modeling, error analysis, and evaluation of dual-polarization variables obtained from simultaneous horizontal and vertical polarization transmit radar. Part I: Modeling and antenna errors. J. Atmos. Oceanic Technol., 27, 1583–1598, doi:10.1175/2010JTECHA1336.1.
Hubbert, J., S. M. Ellis, M. Dixon, and G. Meymaris, 2010b: Modeling, error analysis, and evaluation of dual-polarization variables obtained from simultaneous horizontal and vertical polarization transmit radar. Part II: Experimental data. J. Atmos. Oceanic Technol., 27, 1599–1607, doi:10.1175/2010JTECHA1337.1.
Illingworth, A. J., and R. J. Thompson, 2005: The estimation of moderate rain rates with operational polarisation radar. Proc. 32th Conf. on Radar Meteorology, Albuquerque, NM, Amer. Meteor. Soc., P9R.1. [Available online at https://ams.confex.com/ams/pdfpapers/96204.pdf.]
Istok, M. J., and Coauthors, 2009: WSR-88D dual polarization initial operational capabilities. 25th Conf. on Interactive Information Processing Systems (IIPS) for Meteorology, Oceanography, and Hydrology, Phoenix, AZ, Amer. Meteor. Soc., 15.5. [Available online at https://ams.confex.com/ams/pdfpapers/148927.pdf.]
Lee, G. W., 2006: Sources of errors in rainfall measurements by polarimetric radar: Variability of drop size distributions, observational noise, and variation of relationships between R and polarimetric parameters. J. Atmos. Oceanic Technol., 23, 1005–1028, doi:10.1175/JTECH1899.1.
Löffler-Mang, M., and J. Joss, 2000: An optical disdrometer for measuring size and velocity of hydrometeors. J. Atmos. Oceanic Technol., 17, 130–139, doi:10.1175/1520-0426(2000)017<0130:AODFMS>2.0.CO;2.
Marshall, J. S., and W. M. K. Palmer, 1948: The distribution of raindrops with size. J. Meteor., 5, 165–166, doi:10.1175/1520-0469(1948)005<0165:TDORWS>2.0.CO;2.
Martner, B. E., 1977: A field experiment on the calibration of radars with raindrop disdrometers. J. Appl. Meteor., 16, 451–454, doi:10.1175/1520-0450(1977)016<0451:AFEOTC>2.0.CO;2.
Melnikov, V. M., 2004: Simultaneous transmission mode for the polarimetric WSR-88D. NOAA/NSSL Rep., 84 pp. [Available online at http://www.nssl.noaa.gov/publications/wsr88d_reports/SHV_statistics.pdf.]
Pepler, A. S., and P. T. May, 2012: A robust error-based rain estimation method for polarimetric radar. Part II: Case study. J. Appl. Meteor. Climatol., 51, 1702–1713, doi:10.1175/JAMC-D-11-0159.1.
Pepler, A. S., P. T. May, and M. Thurai, 2011: A robust error-based rain estimation method for polarimetric radar. Part I: Development of a method. J. Appl. Meteor. Climatol., 50, 2092–2103, doi:10.1175/JAMC-D-10-05029.1.
Rodgers, C. D., 2000: Inverse Methods for Atmospheric Soundings: Theory and Practice. World Scientific, 238 pp.
Ruzanski, E., and V. Chandrasekar, 2012: Nowcasting rainfall fields derived from specific differential phase. J. Appl. Meteor. Climatol., 51, 1950–1959, doi:10.1175/JAMC-D-11-081.1.
Ryzhkov, A. V., 2007: The impact of beam broadening on the quality of radar polarimetric data. J. Atmos. Oceanic Technol., 24, 729–744, doi:10.1175/JTECH2003.1.
Ryzhkov, A. V., and D. S. Zrnić, 1995: Comparison of dual-polarization radar estimators of rain. J. Atmos. Oceanic Technol., 12, 249–256, doi:10.1175/1520-0426(1995)012<0249:CODPRE>2.0.CO;2.
Ryzhkov, A. V., and D. S. Zrnić, 1996: Assessment of rainfall measurement that uses specific differential phase. J. Appl. Meteor., 35, 2080–2090, doi:10.1175/1520-0450(1996)035<2080:AORMTU>2.0.CO;2.
Ryzhkov, A. V., S. E. Giangrande, and T. J. Schuur, 2005a: Rainfall estimation with a polarimetric prototype of WSR-88D. J. Appl. Meteor., 44, 502–515, doi:10.1175/JAM2213.1.
Ryzhkov, A. V., T. J. Schuur, D. W. Burgess, P. L. Heinselman, S. E. Giangrande, and D. S. Zrnić, 2005b: The Joint Polarization Experiment: Polarimetric rainfall measurements and hydrometeor classification. Bull. Amer. Meteor. Soc., 86, 809–824, doi:10.1175/BAMS-86-6-809.
Seliga, T. A., and V. N. Bringi, 1976: Potential use of radar differential reflectivity measurements at orthogonal polarizations for measuring precipitation. J. Appl. Meteor., 15, 69–76, doi:10.1175/1520-0450(1976)015<0069:PUORDR>2.0.CO;2.
Seliga, T. A., V. N. Bringi, and H. H. Al-Khatib, 1981: A preliminary study of comparative measurements of rainfall rate using the differential reflectivity radar technique and a raingage network. J. Appl. Meteor., 20, 1362–1368, doi:10.1175/1520-0450(1981)020<1362:APSOCM>2.0.CO;2.
Smyth, T. J., and A. J. Illingworth, 1998: Correction for attenuation of radar reflectivity using polarization data. Quart. J. Roy. Meteor. Soc., 124, 2393–2415, doi:10.1002/qj.49712455111.
Steiner, M., and J. A. Smith, 2000: Reflectivity, rain rate, and kinetic energy flux relationships based on raindrop spectra. J. Appl. Meteor., 39, 1923–1940, doi:10.1175/1520-0450(2000)039<1923:RRRAKE>2.0.CO;2.
Steiner, M., J. A. Smith, and R. Uijlenhoet, 2004: A microphysical interpretation of radar reflectivity–rain rate relationships. J. Atmos. Sci., 61, 1114–1131, doi:10.1175/1520-0469(2004)061<1114:AMIORR>2.0.CO;2.
Ulbrich, C. W., 1983: Natural variations in the analytical form of the raindrop size distribution. J. Climate Appl. Meteor., 22, 1764–1775, doi:10.1175/1520-0450(1983)022<1764:NVITAF>2.0.CO;2.
Vivekanandan, J., W. M. Adams, and V. N. Bringi, 1991: Rigorous approach to polarimetric radar modeling of hydrometeor orientation distributions. J. Appl. Meteor., 30, 1053–1063, doi:10.1175/1520-0450(1991)030<1053:RATPRM>2.0.CO;2.
Vivekanandan, J., S. M. Ellis, R. Oye, D. S. Zrnić, A. V. Ryzhkov, and J. Straka, 1999: Cloud microphysics retrieval using S-band dual-polarization radar measurements. Bull. Amer. Meteor. Soc., 80, 381–388, doi:10.1175/1520-0477(1999)080<0381:CMRUSB>2.0.CO;2.
Wang, Y., and V. Chandrasekar, 2009: Algorithm for estimation of the specific differential phase. J. Atmos. Oceanic Technol., 26, 2565–2578, doi:10.1175/2009JTECHA1358.1.
Yoneyama, K., C. Zhang, and C. N. Long, 2013: Tracking pulses of the Madden–Julian oscillation. Bull. Amer. Meteor. Soc., 94, 1871–1891, doi:10.1175/BAMS-D-12-00157.1.
Zhang, G., J. Vivekanandan, and E. Brandes, 2001: A method for estimating rain rate and drop size distribution from polarimetric radar. IEEE Trans. Geosci. Remote Sens., 39, 830–840, doi:10.1109/36.917906.