1. Introduction
a. Background
The estimation of raindrop size distribution (DSD) parameters, including the central tendency of the DSD (mean or median drop size), has been a primary objective of polarimetric radar since the pioneering theoretical study of Seliga and Bringi (1976). Using surface disdrometer measurements of a heavy-rain event, Seliga et al. (1986) demonstrated that simulated differential reflectivity Zdr could provide reasonably accurate estimates of the median volume diameter D0 when the D0 = F(Zdr) relation is derived from the disdrometer DSD observations and a radar scattering model. Observational studies focused on the estimation of D0 or the mass-weighted mean diameter Dm from radar measurements of Zdr alone and found reasonably good agreement with ground-based disdrometer (Goddard et al. 1982; Goddard and Cherry 1984; Aydin et al. 1987) and airborne particle imaging probe (Bringi et al. 1998) estimates. According to these empirical radar studies, Zdr-based estimates of D0 (or Dm) have an absolute normalized bias error ≤ 5% and a normalized standard error of 7%–15% relative to disdrometer or probe measurements of D0 (or Dm), which is consistent with the simulations of Seliga et al. (1986) and also Jameson (1994).
These early studies provide an analysis framework in which DSD parameters can be estimated directly from polarimetric radar observations using equations derived from disdrometer DSD measurements as input to a radar scattering model. Beyond DSD, the derived D0(Zdr) equation is dependent on other details of the rain model, including the assumed drop shape versus size relation (Goddard et al. 1982; Goddard and Cherry 1984; Bringi et al. 1998; Thurai and Bringi 2005). Drop oscillations and canting tend to bias the drop shape slightly toward a more spherical shape in a manner that is nonlinear with diameter (Chandrasekar et al. 1988; Beard et al. 1991; Bringi et al. 1998; Thurai and Bringi 2005; Thurai et al. 2009). The sensitivity of D0(Zdr) to uncertainty in the drop shape versus size relation led Bringi et al. (2002) and Gorgucci et al. (2002) to develop the so-called effective β method for deriving DSD parameters (D0, Nw, μ) of an assumed gamma distribution (Ulbrich 1983) from the triplet of observed horizontal reflectivity Zh, Zdr, and specific differential phase Kdp. The method takes advantage of the combined use of Kdp and Zdr to mitigate the effects of drop oscillation and canting by estimating the slope β of an assumed linear relationship between drop shape and size. The effective β method has been applied to polarimetric radar observations in a wide variety of climate regimes, and the retrieved DSD parameters were in general agreement with surface disdrometers (Bringi et al. 2002, 2003).
Brandes et al. (2004a) demonstrated that the estimation error for β due to measurement error in Kdp is very large for Kdp < 1.5° km−1, thereby limiting the practical utility of the effective β method for DSD retrieval to heavy rain (e.g., rain rate R > 70 mm h−1 at S band or R > 40 mm h−1 at C band). The retrieval of D0 must often default to a Zdr-only approach in many rainfall situations (Zhang et al. 2001; Bringi et al. 2002, 2006, 2009; Brandes et al. 2003, 2004a,b). Fortunately, significant progress has been made in the experimental measurement of drop shape over a wide range of sizes (Beard and Kubesh 1991; Andsager et al. 1999; Thurai and Bringi 2005; Thurai et al. 2007). The empirical results (e.g., Thurai and Bringi 2005; Thurai et al. 2007) are generally consistent with the range of axis ratios predicted by theory (Beard and Chuang 1987). Since there appears to be good agreement between recent empirical drop shape–size relations in the literature (Goddard et al. 1994; Thurai and Bringi 2005; Brandes et al. 2002; Thurai et al. 2007), the estimation of D0 from Zdr can likely be accomplished without significant error because of drop shape assumptions (Brandes et al. 2003, 2004a,b; Bringi et al. 2006, 2009).
b. Motivation
Another long-standing issue in polarimetric radar rainfall retrieval methods is their potential sensitivity to DSD truncation, including assumptions regarding both maximum (Dmax) and minimum (Dmin) diameters (Ulbrich and Atlas 1984; Ulbrich 1985, 1992). Note that all drop diameters are in terms of the equivalent spherical diameter. Truncation of the DSD at the large diameter end of the spectrum can influence the accuracy of the gamma model parameters fit to the DSD using the method of moments (Ulbrich and Atlas 1998). Developing relations between radar observables and rainfall properties, which are both calculated from integral moments of the DSD, requires assumptions regarding the limits (Dmin, Dmax) of those rainfall integral parameters. Bias errors in the derived radar–rainfall relations can result simply from inappropriate assumptions regarding the limits of the DSD integrals involved (Ulbrich 1985). DSD moment errors (especially for high moments) are more sensitive to the DSD uncertainty for large raindrops (Cao et al. 2008; Cao and Zhang 2009). Therefore, the uncertain range of maximum raindrop diameter (including the truncation) could affect the accuracy of DSD retrieval. Since both D0 and Zdr are calculated from high-order moments of the DSD, the development of an accurate D0(Zdr) relation also depends on acceptable DSD truncation assumptions, including the choice of Dmax (Ulbrich and Atlas 1984; Ulbrich 1992). Ulbrich and Atlas (1984) concluded that the relationship between D0 and Zdr depends strongly on Dmax only when D0 ≥ 2.5 mm for an assumed gamma DSD. Since Dmax and D0 appear to be proportional in DSD observations, Ulbrich and Atlas (1984) demonstrated that the relationship between D0 and Zdr for a gamma DSD is relatively insensitive to changes in Dmax/D0 for values of D0 ≤ 3.2 mm provided Dmax/D0 ≥ 2.5.
However, the measured ratio of Dmax/D0 rarely exceeds 2.5 in typical disdrometer sample volumes. In Keenan et al. (2001), the 95th percentile of Dmax/D0 in 1-min DSD data was 2.4 over Darwin, Australia. In a large sample of 1-min DSDs observed by 2D video disdrometers (2DVDs) over Huntsville, Alabama, the mean Dmax/D0 was 2.0 and 92.5% of the data were characterized by Dmax/D0 < 2.5 (Fig. 1) when the number of drops ≥ 300 and the rain rate ≥ 1 mm h−1. Using similar 2DVD instruments and thresholds, a similar result (i.e., mean Dmax/Dm = 1.9) was found by Gatlin et al. (2015), who analyzed an order of magnitude more 1-min DSD samples from a variety of locations around the globe in a wide variety of precipitation conditions. As noted by Ulbrich (1992), the average Dmax/D0, which is observed by a single disdrometer, increases with the integration time (i.e., 1.9, 2.5, and 2.9 over 1, 10, and 30 min, respectively). Assuming the longer integration periods provide DSDs that are consistent with those found within a typical radar resolution volume, natural rainfall may typically meet the Dmax/D0 ≥ 2.5 criteria required by Ulbrich and Atlas (1984) for D0(Zdr) estimates that are relatively insensitive to Dmax assumptions.
Frequency histogram of (a) the ratio of the maximum diameter over the median volume diameter (Dmax/D0) and (b) the maximum diameter Dmax for 7678 one-minute drop size distributions collected in a variety of rain types over Huntsville using 2DVDs. Drop size distributions were utilized when total drop concentration NT ≥ 300 drops, rain rate R ≥ 1.0 mm h−1, and no hail or other ice hydrometers were present in the sample volume. All diameters are equivalent spherical diameters.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
However, partially compensating volumetric sampling limitations of a single disdrometer with long integration periods likely mixes DSDs from a variety of rainfall types and microphysical processes that may not be representative of the DSDs affecting an instantaneous measurement by radar. For example, one can consider the horizontal scale of advection given various sample times. Assuming a conservative relative motion of 5 m s−1, a 1-km patch of rain will have moved by a disdrometer in 200 s (roughly 3 min of integration). One minute of integration results in a spatial scale of 300 m, which is about the size of a radar gate space. On the other hand, a 10–30-min integration period is equivalent to a horizontal scale of 3–9 km, which may be too large to be representative of instantaneous conditions in a radar sample. Hence, there is still some uncertainty regarding the appropriate Dmax/D0 in a sample volume consistent with radar applications and hence regarding the sensitivity of D0(Zdr) to various Dmax assumptions.
It is important to note that the results of Ulbrich and Atlas (1984) are specific to S band (λ = 10 cm). More recent S band studies by Brandes et al. (2003, 2004a,b) have demonstrated good agreement between polarimetric radar–based and surface disdrometer-based retrievals of DSD parameters while evaluating the constrained-gamma approach of Zhang et al. (2001). Brandes et al. (2003) suggested that their DSD parameter retrievals from S-band Zdr were fairly insensitive to Dmax assumptions. However, Brandes et al. (2003) noted that large drop regions characterized by Zdr > 3 dB (D0 > 3.2 mm) were likely not well represented by the constrained-gamma model and were therefore ignored in their studies. In that respect, Brandes et al. (2003) conclusions regarding the robustness of S-band D0(Zdr) retrievals are similar to Ulbrich and Atlas (1984).
Several studies have noted the impact of Mie resonance associated with large raindrops on C band (λ = 5 cm) polarimetric radar observables and rainfall retrieval algorithms (Bringi et al. 1991; Meischner et al. 1991; Aydin and Giridhar 1992; Carey et al. 2000; Zrnić et al. 2000; Keenan et al. 2001). The behavior of Zdr in large (e.g., D > 5 mm) raindrops at C band in the Mie scattering regime is well known (e.g., Zrnić et al. 2000). Resonance occurs for drops larger than about 5 mm where C-band Zdr exhibits decidedly nonmonotonic behavior, especially relative to S-band (i.e., nonresonant) Zdr (Fig. 2). Deviations between C- and S-band Zdr reach 3.5 dB at drop diameters just below 6 mm. By comparison, X-band Zdr has a muted resonance response in raindrops larger than about 3 mm and is much closer to S-band Zdr, except from 3 to 4 mm where deviations can reach up to 0.74 dB. Despite the obvious potential impact of resonance, the fundamental importance of measuring DSD central tendency, and the growing numbers of C-band polarimetric radars worldwide, no study to date has investigated the detailed sensitivity of C-band retrieval of D0(Zdr) to assumptions regarding Dmax. We note that such Dmax sensitivity studies have been conducted on the C-band polarimetric retrieval of rain rate, attenuation, and differential attenuation (Zrnić et al. 2000; Keenan et al. 2001).
The differential reflectivity (Zdr; dB) of monodisperse raindrops of equivalent spherical diameter (D; mm) for X-band (red), C-band (blue), and S-band (dashed black) wavelengths. For the monodisperse simulations in this figure, the following assumptions were made: drop shape vs size relationship of Thurai et al. (2007), a drop temperature of 20°C, and a mean and standard deviation of the drop canting angle of 0° and 7.5° (Huang et al. 2008), respectively.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
In situ aircraft probe, videosonde, and surface disdrometer observations have demonstrated that large raindrops in the range of 5–8 mm, and possibly up to 9–10 mm, occur in a wide variety of rainfall regimes (Beard et al. 1986; Rauber et al. 1991; Takahashi et al. 1995; Schuur et al. 2001; Hobbs and Rangno 2004; Fujiyoshi et al. 2008; Gatlin et al. 2015). Nonetheless, large drop occurrence in disdrometer observations is rare. For example, 97% of all 1-min DSDs sampled by 2DVDs were characterized by Dmax < 5 mm over Huntsville, Alabama (Fig. 1). In a 2DVD study of large (D ≥ 5 mm) raindrop occurrence worldwide, Gatlin et al. (2015) found only 10 464 large raindrops in a total rain sample consisting of over 224 million drops (i.e., <0.004% occurrence on a per-drop basis). Because of sampling limitations, disdrometers likely undersample the number of large raindrops and hence Dmax (Ulbrich and Atlas 1984; Ulbrich 1992; Smith et al. 1993; Keenan et al. 2001; Brandes et al. 2003). As a result, there is still considerable uncertainty regarding the appropriate Dmax for any given rainfall situation. Reflecting this uncertainty, there is currently no consensus on the appropriate value of Dmax to use in a C-band radar analysis framework using disdrometer observations (e.g., Zrnić et al. 2000; Keenan et al. 2001).
In C-band radar studies, parameterized values of Dmax have been assumed to be constant in the range of 4–10 mm (Aydin and Giridhar 1992; Carey et al. 2000; Zrnić et al. 2000; Keenan et al. 2001; Tabary et al. 2009), a constant multiple C of the disdrometer observed central tendency (D0 or Dm) where the C has ranged from 2.5 to 3.5 (e.g., Dmax = CD0) (Keenan et al. 2001; Bringi et al. 2002, 2003, 2006, 2009; Thurai et al. 2007), and the actual disdrometer-measured maximum diameter despite recognized sampling limitations (Ryzhkov and Zrnić 2005; Tabary et al. 2009). Several of these studies have developed and applied C-band D0(Zdr)fit best-fit relations but no sensitivity test to the Dmax assumption has yet been conducted. As pointed out by Zrnić et al. (2000), the uncertainty in Dmax in the range of 5–8 mm is exactly where the maximum sensitivity to resonance is expected.
Because of the sampling limitations of disdrometers, Dmax is estimated or essentially parameterized in the studies above. In the instances where Dmax is parameterized as a multiple of D0 (Dmax = CD0), it is important to note that D0 here is the disdrometer observed D0. This approach was first taken by Keenan et al. (2001) to provide a physically realistic domain for Dmax in the polarimetric variable scattering calculations based on the observed DSD data even though the disdrometer typically underestimates Dmax. Multiple studies (Bringi et al. 2002, 2003, 2006, 2009; Thurai et al. 2007) have since employed Keenan et al.’s approach. To the extent that the parameterized Dmax in the resulting rain model is different than the observed Dmax, the resulting model D0 will be different than the disdrometer observed D0 even though the observed D0 is sometimes used to parameterize Dmax in the first place. In the studies where Dmax is parameterized, it is the model D0 that is used to develop the D0(Zdr)fit equation to estimate D0 from radar observations of Zdr. More details will be provided in section 2.
c. Objectives
Because of their relative affordability, C-band polarimetric radars are in common use worldwide for both research and operations. As such, it is critical to assess errors associated with raindrop size retrievals at C band because of the uncertainty in the maximum raindrop diameter, as has been accomplished for other C-band rain algorithms such as rain rate and propagation correction (Zrnić et al. 2000; Carey et al. 2000; Keenan et al. 2001). The fundamental definition of maximum drop diameter, its impact to precipitation remote sensing algorithms, and more specific to this study, a complete characterization of potential errors in C-band radar drop size retrievals are important for a number of applications, including global physical and statistical ground validation (Chandrasekar et al. 2008) of satellite precipitation remote sensing methods such as for the NASA Global Precipitation Measurement (GPM) mission (Hou et al. 2014). Given the current uncertainty in parameterizing the large drop tail of the DSD (Zrnić et al. 2000), we conduct a Dmax sensitivity test to assess the potential errors inherent in the development of a C-band D0(Zdr)fit relation using disdrometer data. Because Dmax variability is likely to have a more significant impact on D0 retrievals at larger Zdr and because of the exacerbating influence of resonance scattering at C band, errors in the retrieval of D0 as a function of Zdr are presented in addition to overall sample errors. Results at X and S band are also briefly compared with C band to highlight the important impact of resonance on these potential errors at C band. The potential effect of drop temperature is also explored.
2. Data and methodology
a. Raindrop model development
Data from the Colorado State University low-profile and NASA GPM Ground Validation compact 2DVDs (Schönhuber et al. 2008) located at the instrument berm of the National Space Science Technology Center (NSSTC) in Huntsville, Alabama, were utilized to develop an experimental raindrop model. Disdrometer data were collected in 1-min integration periods. To estimate robust DSD statistics and gamma fits to the DSD data, only 1-min periods with total drop concentration (NT) ≥ 300 drops and R ≥ 1.0 mm h−1 were used, providing 7678 one-minute-averaged samples of the binned DSD. DSD data from the low-profile and compact 2DVD units were available from 2007 to 2011 and from late 2009 to 2011, respectively. DSD data were binned at 0.25 mm through early 2010 after which time the bin size was reduced to 0.20 mm. A comparison of results showed no significant impact of the change in bin size to the goals of this study. A comparison of the side-by-side 2DVD units by Thurai et al. (2011) demonstrated excellent agreement in measuring DSD parameters. For example, the correlation coefficient and fractional standard error between the Dm values measured by the collocated low-profile and compact 2DVDs was 0.95 and 5%, respectively.
The DSD dataset contained a wide variety of rainfall types, including convection and stratiform precipitation within ordinary thunderstorms, tropical storms, mesoscale convective systems, and severe storms occurring across all seasons of the year. One-minute-averaged DSD samples with likely hail, snow, or mixed-phase precipitation contamination were removed through manual inspection of the 2DVD fall speed data, surface temperature, available sounding data, Advanced Radar for Meteorological and Operational Research (ARMOR; Petersen et al. 2005, 2007) polarimetric observations, and NOAA Storm Data.
The method of truncated moments of Ulbrich and Atlas (1998) was utilized to fit a gamma distribution model (Ulbrich 1983) to the 1-min 2DVD DSD data. The method of truncated moments accounts for the finite Dmax in the retrieval of the gamma model parameters: N0 (intercept parameter), D0 (median volume diameter), and μ (shape parameter). The triplet of gamma fit parameters along with the Dmax assumed for each sensitivity test and a constant Dmin fully characterized the gamma rain DSD for input into the radar scattering model, which is detailed in the next paragraph. The value of Dmin was fixed at 0.4 mm since the first bin of the 2DVD was not utilized in this study because of drop undercounting. Consistent with Ulbrich and Atlas (1998), the choice of Dmin had little impact on the outcome of this study. The parameterized Dmax was varied to encompass the variety of assumptions currently found in the literature, as discussed in more detail below. Note that all drop diameters (e.g., Dmin, Dmax, D0) are in terms of equivalent spherical diameters. As found by Ulbrich and Atlas (1998), it is important to note that a change in DSD shape is associated with the selection of a parameterized Dmax. In fact, that is why we derive the gamma DSD triplet of (N0, D0, μ) using the truncated method of moments after assuming the parameterized Dmax. Other rain model assumptions required for input into the radar scattering model were 1) the recommended drop shape versus diameter relationship of Thurai et al. (2007), 2) a Gaussian canting angle distribution with mean of 0° and a standard deviation of 7.5° (Huang et al. 2008), and 3) a drop temperature of 20°C. The drop temperature T of T = 20°C was assumed throughout most of the study except when T = 10°C or T = 30°C was required for comparison.
b. Radar scattering model
The 
c. Median volume diameter retrieval experiments
A series of experiments to estimate D0 from Zdr [i.e., D0(Zdr)fit] were conducted by varying the parameterized Dmax through the range of assumptions found in the literature discussed in section 1, including 1) constant Dmax = 4, 5, 6, 7, 8, 9, and 10 mm; 2) actual 2DVD-measured Dmax; and 3) Dmax = CD0 where C = 2.0, 2.5, 3.0, 3.5, and 4.0 and D0 here is the observed D0 from the 2DVD observations of drop size and counts. Although likely physically unrealistic, the constant Dmax assumptions provide a direct way to explore the impact of resonance on the behavior of D0(Zdr)fit. In the latter case of Dmax = CD0, Dmax is capped at a maximum of 8 mm consistent with the practice of recent polarimetric studies (Bringi et al. 2006, 2009; Thurai et al. 2007) and with the idea that spontaneous raindrop breakup would typically occur around this diameter (e.g., Kamra et al. 1991). Note that constant Dmax of 9 and 10 mm were explored to understand the sensitivity of this assumption. All other rain and radar model characteristics were fixed for calculating the intrinsic Zdr associated with each DSD. To simulate the retrieval of D0 from observed Zdr in the presence of radar measurement error, Gaussian noise was added to the simulated intrinsic Zdr values with a mean of 0 dB and a standard deviation of 0.25 dB. Vertically pointing scans (Gorgucci et al. 1999) can be used to mitigate bias error in Zdr, thus justifying the 0 dB mean. The standard deviation of Zdr is based on vertically pointing ARMOR scans of drizzle (e.g., 0.20–0.25 dB).
Simulated median volume diameter (D0; mm) vs the differential reflectivity (Zdr; dB) at C band (black diamonds) and the best-fit fourth-order polynomial, D0(Zdr)fit (blue asterisks), assuming (a) (Dmax)fit = 3.5D0 and (b) (Dmax)fit = 2D0. Note that the presence of negative Zdr is associated with the inclusion of simulated Gaussian noise in Zdr.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Last, for each experiment, a (Dmax)truth was postulated as the true maximum diameter for developing the truth dataset of (D0)truth against which the polynomial fit in Eq. (1) was evaluated. The bias and standard errors of Eq. (1) were assessed as a function of the mismatch between the assumed (Dmax)fit for Eq. (1) and the postulated (Dmax)truth. In this manner, the sensitivity of D0(Zdr)fit to Dmax assumptions was assessed. For both values of Dmax, the truncated method of moments is used to develop the gamma triplet of parameters (N0, D0, μ). The quantity (Dmax)fit is used to derive (N0, D0, μ)fit, which are input to a radar scattering model to derive (Zdr)fit for the development of Eq. (1) or D0(Zdr)fit. Meanwhile, (Dmax)truth is used to derive (N0, D0, μ)truth, which are also input to the radar scattering model to derive (Zdr)truth. To determine bias and standard error as a function of Dmax assumptions, the performance of the D0(Zdr)fit equation is evaluated against the (D0)truth data.
3. Results and discussion
The results of the D0 retrieval experiments are first overviewed by presenting the family of D0(Zdr)fit polynomials at C band associated with varying the (Dmax)fit assumption through values typically found in the literature, as reviewed in section 1b. The potential overall bias and standard (i.e., scatter) errors of the D0 retrievals and their sensitivity to Dmax assumptions are then assessed by assuming a (Dmax)fit and evaluating the associated D0(Zdr)fit polynomial against various truth datasets, (D0)truth, associated with a different postulated (Dmax)truth. The D0 retrieval errors associated with an incorrectly postulated Dmax are also evaluated as a function of Zdr and radar wavelength (X, C, and S band) to highlight the importance of resonant scattering. Finally, the potential impact of drop temperature is explored by conducting sensitivity tests at various drop temperatures.
a. Sensitivity of polarimetric D0 retrieval to maximum diameter at C band
Examples of simulated pairs of (D0, Zdr) data at C band and the associated D0(Zdr)fit best-fit polynomials for the assumptions of Dmax = 3.5D0 and Dmax = 2D0 can be found in Figs. 3a and 3b, respectively. As expected, D0 increases monotonically with increasing C-band Zdr with some scatter about a best-fit polynomial. The primary effect of increasing the assumed Dmax from 2D0 to 3.5D0 is to flatten the D0(Zdr)fit polynomial at moderate to large Zdr. In other words, the simulated D0 are systematically smaller at a given Zdr for Dmax = 3.5D0 relative to Dmax = 2D0, particularly at moderate-to-large Zdr.
To better understand the effect of Dmax and the impact of resonant scattering at C band, the best-fit polynomials D0(Zdr)fit assuming constant Dmax varying from 4 to 10 mm are shown in Fig. 4a. At most values of Zdr, the D0 inferred from the best-fit polynomials decreases with the assumed (Dmax)fit. A significant transition in the functionality of D0 with respect to Zdr occurs between the polynomials associated with (Dmax)fit = 5 and 6 mm. The difference between the two polynomials is most obvious at moderate-to-large Zdr (e.g., 1.5 < Zdr < 3 dB) where the inferred D0 from the polynomial assuming (Dmax)fit = 5 mm becomes increasingly larger than the D0 inferred from the polynomial with (Dmax)fit = 6 mm. The D0(Zdr)fit polynomials for (Dmax)fit ≥ 6 mm tend to be much flatter with Zdr > 1.5 dB than those for (Dmax)fit < 6 mm. This significant transition in the D0(Zdr)fit polynomial behavior with (Dmax)fit is associated with resonance scattering and the rapid increase in Zdr for drops between 5 and 6 mm in diameter (Fig. 2).
The simulated median volume diameter (D0; mm) vs the simulated differential reflectivity (Zdr; dB) at C band derived from observed DSD data. Each curve represents a fourth-order polynomial fit to simulated pairs of (Zdr, D0) with a different assumption regarding the maximum raindrop size (Dmax)fit. (a) Dmax is assumed constant [(Dmax)fit = 4, 5, 6, 7, 8, 9, or 10 mm as shown]. For (Dmax)fit = 4 and 5 mm, markers not accompanied by a plotted curve represent an extrapolation of the polynomial fit beyond the maximum simulated Zdr. (b) Dmax is assumed to be a multiple of D0 [(Dmax)fit = 2D0, 2.5D0, 3D0, 3.5D0, and 4D0 as shown]. In both (a) and (b), the dashed line represents the polynomial fit of D0 vs Zdr for the actual observed 2DVD maximum. Polynomial coefficients for the curves in (b) can be found in Table 1. The remaining assumptions and methods for deriving the simulated pairs of (Zdr, D0) at C band from observed DSD data are discussed in section 2.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Note that the D0(Zdr)fit polynomial for (Dmax)fit = 5 mm is only strictly valid to about Zdr = 3 dB as values larger than this do not occur when (Dmax)fit = 5 mm (Fig. 4a). Extrapolation of the best-fit polynomial associated with (Dmax)fit = 5 mm to larger values of Zdr would result in drastically larger estimated D0 than the polynomial associated with (Dmax)fit = 6 mm (or larger Dmax). Similar conclusions can be drawn for the D0(Zdr)fit polynomial associated with (Dmax)fit = 4 mm, which is much closer in behavior to the polynomial for (Dmax)fit = 5 mm than to (Dmax)fit ≥ 6 mm. Similarly, the D0(Zdr)fit polynomial for (Dmax)fit = 6 mm is much closer to those polynomials associated with (Dmax)fit ≥ 7 mm. In fact, there is very little difference in the estimated D0 from the polynomials associated with (Dmax)fit ≥ 7 mm except at very large Zdr > 4 dB.
It is worth noting that the behavior of the D0(Zdr)fit polynomial for a (Dmax)fit given by the actual disdrometer-measured Dmax is very close to the polynomials for (Dmax)fit = 4–5 mm for Zdr ≤ 2.25 dB and then rapidly deviates as those polynomials curve upward to large D0 for increasing Zdr (Fig. 4a). For Zdr > 2.25 dB, the D0(Zdr)fit polynomial associated with the actual 2DVD-measured Dmax remains relatively flat similar to the polynomials for (Dmax)fit ≥ 6 mm but falling at a noticeably larger D0 for a given Zdr up to about 4.5 dB.
By assuming a constant (Dmax)fit varying from 4 to 10 mm, it is clear that 1) increasing (Dmax)fit results in generally smaller estimated D0(Zdr)fit for a given Zdr (up to about 4.5 dB), and 2) resonance scattering causes a dramatic decrease in the inferred D0(Zdr)fit at Zdr > 1.5 dB for all polynomials having (Dmax)fit ≥ 6 mm. However, assuming a constant Dmax for all DSDs is not physically realistic, is not consistent with the 2DVD measurements in Fig. 1 and is not in keeping with many previous studies (section 1b).
To simulate more realistic D0(Zdr)fit polynomials and test a range of assumptions utilized in the literature, (Dmax)fit was started at 2D0, which is the mean value for the DSD dataset utilized in this study, and increased to 2.5D0, then 3D0 and finally 3.5D0, which have all been utilized in the literature to retrieve polarimetric radar-based DSD equations (e.g., Fig. 1; Keenan et al. 2001; Bringi et al. 2002, 2003, 2006, 2009; Thurai et al. 2007). The resulting D0(Zdr)fit polynomials are provided in Fig. 4b along with the D0(Zdr)fit polynomials associated with 4D0 and the actual 2DVD-measured Dmax for reference. The coefficients of the corresponding fourth-order polynomials [Eq. (1)] can be found in Table 1.
Coefficients (a, b, c, d, and e) of the fourth-order polynomial for D0(Zdr)fit as shown in Eq. (1) for C band. The coefficients were obtained from a nonlinear least squares curve fit to pairs of simulated (Zdr, D0) for various Dmax assumptions at C band. Methods and assumptions for developing the simulation dataset are discussed in section 2.
As noted for constant Dmax, the estimated D0 from the D0(Zdr)fit polynomials for a given Zdr decreases with increasing (Dmax)fit = CD0 (i.e., with increasing C) (Fig. 4b). The differences between the various D0(Zdr)fit polynomials are particularly noticeable at Zdr > 1.5 dB. As (Dmax)fit = CD0 (i.e., C) increases, the D0(Zdr)fit polynomials become increasingly flatter at 1.5 < Zdr < 4 dB, resulting in a significantly lower estimated D0 for a given Zdr. The difference between adjacent D0(Zdr)fit polynomials in the family of (Dmax)fit = CD0 curves becomes smaller as C increases. In other words, the difference between (Dmax)fit = 2D0 and 2.5D0 is larger than the difference between 3D0 and 3.5D0. In fact, there is very little difference between assuming (Dmax)fit = 3.5D0 and 4D0. The D0(Zdr)fit polynomial associated with a (Dmax)fit equal to the measured 2DVD Dmax is more similar to (Dmax)fit = 2D0 at small Zdr (although it is difficult to see that in Fig. 4b) and most similar to (Dmax)fit = 2.5D0 through a broad range of Zdr from 1 to 4 dB.
b. Overall D0 retrieval error
To assess the overall D0 retrieval error, each D0(Zdr)fit polynomial associated with an assumed (Dmax)fit was evaluated against a (D0)truth dataset associated with a postulated (Dmax)truth. To visualize this kind of test in Fig. 5a, the pairs of (Zdr, D0)fit data used for derivation of the D0(Zdr)fit polynomial assuming (Dmax)fit = 3.5D0 are accompanied by pairs of true (Zdr, D0)truth assuming (Dmax)truth = 2D0, which is close to observed (Fig. 1). In this example, it is clear that the D0(Zdr)fit polynomial assuming (Dmax)fit = 3.5D0 is underestimating the true D0, (D0)truth, at a given Zdr assuming (Dmax)truth = 2D0. The “fit” and “truth” datasets are then reversed in Fig. 5b. Correspondingly, it is easy to see how the D0(Zdr)fit polynomial assuming (Dmax)fit = 2D0 is overestimating the true D0, (D0)truth, at a given Zdr assuming (Dmax)truth = 3.5D0.
As in Fig. 3, but (a) the pairs of (Zdr, D0)fit data (black diamonds) used for derivation of the best-fit fourth-order polynomial [D0(Zdr)fit; asterisks] assuming (Dmax)fit = 3.5D0 are accompanied by pairs of true (Zdr, D0)truth (red pluses) assuming (Dmax)truth = 2D0, and (b) the pairs of (Zdr, D0)fit data (black diamonds) used for derivation of the best-fit fourth-order polynomial [D0(Zdr)fit; asterisks] assuming (Dmax)fit = 2D0 are accompanied by pairs of true (Zdr, D0)truth (red pluses) assuming (Dmax)truth = 3.5D0. Note that the “fit” (black diamonds) and “truth” (red pluses) (Zdr, D0) datasets are simply reversed between (a) and (b).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Normalized bias errors for various D0(Zdr)fit polynomials assuming constant (Dmax)fit from 4 to 10 mm across the entire sample of (Zdr, D0)truth assuming various (Dmax)truth are provided in Fig. 6. For (Dmax)fit ≥ 6 mm and (Dmax)truth ≥ 6 mm, the absolute normalized bias errors of D0(Zdr)fit are small (<3%). For (Dmax)fit ≥ 6 mm and (Dmax)truth < 6 mm, the normalized bias errors of D0(Zdr)fit range from −4.2% to −5.9%. Compared to a (Dmax)truth equal to the 2DVD-measured Dmax, the normalized bias errors for D0(Zdr)fit assuming (Dmax)fit ≥ 6 mm are approximately −7%. If (Dmax)fit < 6 mm and (Dmax)truth < 6 mm, then the absolute normalized bias errors of D0(Zdr)fit are fairly small (<4%). If (Dmax)fit < 6 mm and (Dmax)truth is equal to the 2DVD-measured Dmax, then the normalized bias errors of D0(Zdr)fit range approximately from 3% to 9%. On the other hand, if (Dmax)fit < 6 mm and (Dmax)truth > 6 mm, then the normalized bias errors of D0(Zdr)fit are extremely large, ranging from 26% to 75%, and are largely the result of resonant scattering on the truth dataset and of extrapolating the D0(Zdr)fit polynomials to larger values of Zdr that are not present in the fit dataset but are in the truth dataset (Fig. 4a). These very large values of normalized bias error are likely not realistic since Dmax is likely not constant for all DSD but Fig. 6 does emphasize the potential effect of resonance on the bias error associated with retrieving D0 using C-band Zdr associated with assuming an inappropriate Dmax.
The NB (%) of the median volume diameter estimated from the best-fit polynomial, D0(Zdr)fit, assuming the (Dmax)fit in the abscissa, relative to a true dataset, (Zdr, D0)truth, with an assumed (Dmax)truth. Each colored curve represents the NB of D0(Zdr)fit relative to a different truth dataset, (Zdr, D0)truth, assuming the indicated (Dmax)truth while all else is held equal. The quantity Dmax is assumed constant at different values (Dmax = 4, 5, 6, 7, 8, 9, and 10 mm) or set to the actual 2DVD-measured Dmax as shown.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
To provide a more realistic assessment of bias error, the normalized bias errors for various D0(Zdr)fit polynomials assuming (Dmax)fit = CD0 (C = 2–4) across the entire sample of (D0)truth assuming various (Dmax)truth = CD0 (C = 2–4) are provided in Fig. 7. Despite the known sampling limitations of disdrometers, the actual 2DVD-measured Dmax is also utilized as a potentially realistic (Dmax)fit and (Dmax)truth. As shown by the statistics of Dmax/D0 in Fig. 1a and other studies (Keenan et al. 2001; Gatlin et al. 2015), this range of C and hence Dmax should adequately represent a potential realistic range of Dmax when considering potential undersampling of large drops by the 2DVD. As a result, the range of normalized bias errors in Fig. 7 should bracket the overall bias errors potentially present in recent D0 retrieval studies using Zdr associated with potential misalignment of the assumed and actual Dmax. Because the (Dmax)fit = 2D0 polynomial is the most different than the others (Fig. 4b), its range of possible normalized bias errors for D0(Zdr)fit is the largest (0%–16%). The possible range of normalized bias errors for D0(Zdr)fit decreases as (Dmax)fit increases from 2D0 to 4D0. For (Dmax)fit = 2.5D0, the possible normalized bias error for D0(Zdr)fit ranges from −2% to 9%. As expected from Fig. 4b, the range of possible bias error for D0(Zdr)fit assuming a (Dmax)fit of the measured 2DVD Dmax falls between (Dmax)fit = 2D0 and 2.5D0. For (Dmax)fit = 3D0 (4D0), the possible normalized bias error ranges from −6% to 4% (−8% to 0%). As summarized in section 1a, the absolute normalized bias error for D0 in past studies has typically been estimated to be less than 5%. The results herein demonstrate that the absolute normalized bias error for estimating D0 using C-band Zdr could be 2–3 times as large (i.e., 10%–15%) as previously estimated (section 1a) because of a potential error in the assumed maximum drop diameter. Of course, the overall bias error depends on the degree of mismatch between the assumed and true Dmax and many (61%) of the tested scenarios in Fig. 7 have normalized bias errors falling within ±5%.
As in Fig. 6, but Dmax is assumed to be a multiple of D0 (Dmax = 2D0, 2.5D0, 3D0, 3.5D0, 4D0) or set to the actual 2DVD-measured Dmax as shown.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Since the major effect of varying C from 2 to 4 in (Dmax)fit = CD0 on the D0(Zdr)fit polynomials is to shift the D0(Zdr)fit polynomial upward and downward (Fig. 4b), it was hypothesized that the normalized standard error would not vary dramatically because of a mismatch between the assumed (Dmax)fit and (Dmax)truth. Because there is a minor change in the shape of the polynomial with varying (Dmax)fit that is most apparent for (Dmax)fit = 2D0 relative to the others (Fig. 4b), it was anticipated that its potential range of normalized error would be the largest. Results for the normalized standard error generally confirm these expectations (Fig. 8). For (Dmax)fit = 2D0, the normalized standard error for D0(Zdr)fit ranges from 14% to 22%, which is somewhat larger than found in past studies (7%–15%) as noted in section 1a. For the other tested (Dmax)fit in Fig. 8, the normalized standard error for D0(Zdr)fit varied between 14% and 17%, which is on the high end but generally consistent with these past studies. In this study, the standard error of Zdr is assumed to be 0.25 dB, which is consistent with the standard deviation of Zdr in vertically pointing ARMOR scans of drizzle. While reasonable for ARMOR, this standard error of Zdr may be slightly higher than assumed in some prior studies, which could account for some of the difference in the normalized standard error for D0(Zdr)fit.
As in Fig. 7, but the NSE (%) is depicted.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
c. D0 retrieval error as a function of differential reflectivity
Of course, bias and standard errors over the entire sample only tell part of the story. It is important to understand how errors in the estimated D0 might vary as a function of the independently measured radar property, in this case Zdr. As such, the normalized bias error is presented in Figs. 9a–e as a function of Zdr for the same range of assumptions for (Dmax)fit = CD0 (C = 2–3.5) or the measured 2DVD Dmax. In each panel of Figs. 9a–e, the (Dmax)truth, each represented by a different curve, is varied through a similar range of Dmax [i.e., (Dmax)truth = CD0 (C = 2–4) or the measured 2DVD Dmax]. The largest normalized bias errors for D0(Zdr)fit can be found at the two extremes of the assumed (Dmax)fit, which are (Dmax)fit = 2D0 (Fig. 9a) and 3.5D0 (Fig. 9d). For (Dmax)fit = 2D0, the normalized bias error for D0(Zdr)fit can exceed 0.1 for Zdr as low as 1–1.5 dB and can increase from there to values as large as 0.23–0.47 for larger Zdr, depending on the degree of mismatch between the assumed (Dmax)fit and (Dmax)truth. Of course, if (Dmax)fit and (Dmax)truth are well aligned, then the bias errors with Zdr are generally smaller. Similarly, for (Dmax)fit = 3.5D0, the absolute normalized bias error for D0(Zdr)fit can reach 0.1 for Zdr as low as 1–1.75 dB and can increase from there to values as large as 0.17–0.29 for larger Zdr when the mismatch between the assumed (Dmax)fit and (Dmax)truth is large [e.g., see the curves for (Dmax)truth = 2D0, 2.5D0 and the 2DVD-measured value in Fig. 9d].
Normalized bias of the median volume diameter estimated from the best-fit polynomial, D0(Zdr)fit, as a function of binned Zdr assuming the (Dmax)fit of (a) 2D0, (b) 2.5D0, (c) 3D0, (d) 3.5D0, and (e) the actual 2DVD-measured Dmax, relative to a true dataset, (Zdr, D0)truth, with an assumed (Dmax)truth (colored lines as shown). The Zdr bins start at 0.25 dB, are separated by 0.5 dB, and end at 4.75 dB. Note that the first Zdr bin centered at 0.25 dB encompasses Zdr < 0.5 dB, including some negative Zdr (e.g., Figs. 3 and 5), and the last bin centered at 4.75 dB encompasses Zdr > 4.5 dB, including a few Zdr over 5 dB.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Absolute normalized bias errors for D0(Zdr)fit larger than 0.1 can be found at Zdr > 1.5 dB for all assumed (Dmax)fit in Figs. 9a–e depending on the misalignment with (Dmax)truth. Even if it is assumed that Dmax/D0 ≥ 2.5 always, which may or may not be realistic as highlighted in Fig. 1a and earlier discussion, the absolute normalized bias error can still exceed 0.1 at Zdr as low as 1.75 dB (Fig. 9b), which is associated with a D0 of 1.7–2.0 mm (Fig. 4b). Clearly, the relationship between D0 and C-band Zdr for a gamma DSD is not relatively insensitive to changes in Dmax/D0 for values of D0 ≤ 3.2 mm provided Dmax/D0 ≥ 2.5 as found by Ulbrich and Atlas (1984) for S-band Zdr in their study. This difference with the S band results in Ulbrich and Atlas (1984) is due in part to the effect of resonance on C-band Zdr in large drops greater than 5 mm in diameter (Fig. 2), although differences in methodology and data may also play some role.
As shown in Fig. 10, the normalized standard error for D0(Zdr)fit tends to be a maximum at both small Zdr (<1.5 dB) and very large Zdr (>4 dB). In between a Zdr of 1.5 and 4 dB, the normalized standard errors of D0(Zdr)fit tend to be <0.1. One exception is for an assumed (Dmax)truth equal to the 2DVD-measured Dmax. The elevated values (>0.1) of normalized standard error at Zdr < 1.5 dB are due in large part to the effect of random noise on Zdr. For Zdr > 4 dB, the sample size is relatively small and the parameterization error for D0(Zdr)fit is likely larger (e.g., Fig. 3). The assumed Dmax does not generally have a large effect on the variation of the normalized standard error with Zdr.
As in Fig. 9a, but for the NSE of the median volume diameter estimated from the best-fit polynomial, D0(Zdr)fit. Results for only (Dmax)fit = 2D0 are given because the results for other assumed (Dmax)fit do not vary noticeably from what is presented here. In other words, the NSE of D0(Zdr)fit as a function of Zdr is not strongly dependent on the assumed (Dmax)fit.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
d. Wavelength dependence of D0 retrieval error
To explore the wavelength dependence of these results while holding all else equal, the normalized bias of D0(Zdr)fit as a function of Zdr for two different extreme Dmax mismatch scenarios is shown in Fig. 11 for X, C, and S band: (i) (Dmax)fit = 3.5D0 and (Dmax)truth = 2D0 (Fig. 11a), and (ii) (Dmax)fit = 2D0 and (Dmax)truth = 3.5D0 (Fig. 11b). In both scenarios, the absolute normalized bias for D0(Zdr)fit at X band slightly exceeds that of C and S band for Zdr < 1.5 dB, which is associated with the slight impact of resonance on X-band Zdr between a drop diameter of 3 and 4 mm (Fig. 2). For Zdr ≥ 1.5 dB, the absolute normalized bias for D0(Zdr)fit at C band exceeds X and S band (with one minor exception at Zdr = 3.75 dB in Fig. 11a). In fact, the absolute normalized bias error for D0(Zdr)fit at C band can be significantly larger than at S band by as much as 0.12–0.13 for Zdr > 3 dB in Fig. 11a and 0.13–0.32 for Zdr > 2.5 dB in Fig. 11b. This difference in the bias error of the D0 estimate between C-band Zdr and S-band Zdr is due to the effect of resonance on C-band Zdr in large (>5 mm) raindrops (Fig. 2).
As in Fig. 9, but adding results at X band and S band to the C band shown earlier for (a) (Dmax)fit = 3.5D0 and (Dmax)truth = 2D0 (cf. Fig. 9d) and (b) (Dmax)fit = 2D0 and (Dmax)truth = 3.5D0 (cf. Fig. 9a).
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
e. Temperature dependence of D0 retrieval error
The impact of T was explored by conducting additional sensitivity tests at drop temperature of T = 10°C and T = 30°C in addition to the standard temperature of T = 20°C. The values of Zdr at T = 10°C and Zdr at T = 30°C are now compared with Zdr at T = 20°C in Figs. 12a and 12b, respectively. At larger values of Zdr, (Zdr at T = 10°C) < (Zdr at T = 20°C) while it is reversed when the temperature is increased [i.e., (Zdr at T = 30°C) > (Zdr at T = 20°C)]. In other words, resonance has a larger impact on increasing C-band Zdr at warmer temperatures. The impact of varying temperature on D0(Zdr)fit was explored for a variety of Dmax assumptions. The results are shown in Fig. 13 for Dmax = 3.5D0 since the results for other Dmax assumptions were comparable. As expected from Fig. 12, at larger Zdr, there is a negative bias in D0(Zdr)fit for a temperature that is colder in truth (Ttruth = 10°C) than what was assumed in developing the fit equation (Tfit = 20°C) (Fig. 13) while there is a positive bias in D0(Zdr)fit for a temperature that is warmer in truth (Ttruth = 30°C) than what was assumed in developing the fit equation (Tfit = 20°C) (Fig. 13). For a fairly large temperature mismatch of 10°C, the overall bias of D0(Zdr)fit is negligible (≈0 dB) in the mean over the full range of Zdr (i.e., for all DSDs). For larger Zdr values, the D0(Zdr)fit bias error can approach ±5%–10% for a temperature bias of ±10°C (Fig. 13). While these errors are not trivial, it is worthwhile to note that the impact of Dmax uncertainty is likely often larger than the impact of temperature uncertainty on the accuracy of the D0(Zdr)fit estimator (cf. Figs. 9 and 13).
A comparison of differential reflectivity (Zdr; dB) simulated at different drop temperatures: (a) Zdr at T = 10°C and (b) Zdr at T = 30°C vs Zdr at T = 20°C, which was the standard drop temperature assumption throughout the rest of the study. The maximum drop diameter was assumed to be Dmax = 3.5D0 for this temperature sensitivity test. The radar wavelength was C band. All other rain properties were as discussed in section 2a.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
Normalized bias (%) of the median volume diameter (D0) estimated from the best-fit polynomial, D0(Zdr)fit, as a function of binned Zdr assuming a drop temperature for the fit equation of Tfit = 20°C relative to a truth dataset with temperatures varying as shown by the colored lines in the figure key (i.e., Ttruth = 10°C: solid red, Ttruth = 20°C: dashed black, Ttruth = 30°C: dashed blue). The maximum drop diameter was assumed to be Dmax = 3.5D0 for this temperature sensitivity test. The radar wavelength was C band. All other rain properties were as discussed in section 2a.
Citation: Journal of Applied Meteorology and Climatology 54, 6; 10.1175/JAMC-D-14-0079.1
4. Conclusions
Estimating the raindrop size, including median volume diameter, has been a long-standing objective of polarimetric radar–based precipitation retrieval methods, particularly those using the differential reflectivity. Theoretically, Zdr is a measure of the reflectivity-factor-weighted mean axis ratio and therefore indirectly the reflectivity-factor-weighted drop size via the assumption of a drop size relation (Jameson 1983; Bringi and Chandrasekar 2001, p. 398). The theoretical relationship between Zdr and D0 (or Dm) is more complex and requires more assumptions regarding the DSD (e.g., Bringi and Chandrasekar 2001, p. 398). From a practical perspective, the relationship between Zdr and D0 is typically derived empirically using rain DSD observations from a disdrometer such as the 2DVD, a raindrop physical model, and a radar scattering model. Because disdrometers are known to undersample large raindrops and therefore underestimate the maximum raindrop size, Dmax is often an assumed parameter in the rain physical model. Because there is remaining uncertainty regarding the appropriate Dmax for a given DSD (Fig. 1), there have been a wide variety of assumptions for Dmax. Because Dmax affects the tail of the DSD and Zdr is reflectivity weighted (i.e., D6 for Rayleigh–Gans scattering), variability in the Dmax assumption can affect the relationship between Zdr and D0, as was noted in early studies such as Ulbrich and Atlas (1984) at S band.
Although there have been a number of DSD retrieval application studies at C band, the sensitivity of the relationship between Zdr and D0 to Dmax and the associated potential error in the retrieved D0 have not been investigated in any detail. C-band polarimetric radars are commonly used in research and operations worldwide because of their decreased cost relative to S band. Compared to S band, C band has some complicating factors to consider before analysis of raindrop properties, including increased propagation effects and resonance scattering (or non-Rayleigh–Gans scattering) in large raindrops (i.e., diameter > 5 mm). Resonance scattering can complicate the relationship between the Zdr and diameter of individual raindrops (Fig. 2) and between Zdr and D0 (Fig. 4) of realistic DSDs with an assumed Dmax.
In this study, a series of experiments were conducted that postulate a (Dmax)fit for the fitting of a fourth-order polynomial to (Zdr, D0)fit data derived from 2DVD data, a rain model, and a radar scattering model at C band. The resulting D0(Zdr)fit polynomial, which is associated with an assumed (Dmax)fit, was then compared with a truth dataset, (D0)truth, which is associated with an assumed (Dmax)truth (Fig. 5). The normalized bias and scatter errors for D0(Zdr)fit relative to (D0)truth were then computed while varying both (Dmax)fit and (Dmax)truth through the range of assumptions found in the literature.
It is found that the overall absolute normalized bias errors for D0(Zdr)fit can be as high as 10%–15% at C band depending on the degree of mismatch between the postulated (Dmax)fit and (Dmax)truth. Normalized bias errors for D0(Zdr)fit at C band ranged from −8% to 15%, again depending on the postulated error in Dmax (Fig. 7). The magnitude of the potential bias error in the estimated D0 from C-band Zdr is larger than has been noted in a number of early studies at S band (i.e., 5%). Importantly, the absolute normalized bias error for D0(Zdr)fit at C band increases with Zdr, can reach 10% for Zdr as low as 1–1.75 dB and can increase from there to values as large as 15%–45% for larger Zdr when the mismatch between the assumed (Dmax)fit and (Dmax)truth is large (Fig. 9). For Zdr ≥ 1.5 dB, the magnitude of the potential bias error in the estimated D0 is larger for C-band Zdr than it is for S-band and X-band Zdr (Fig. 11) because of resonance scattering effects in large drops >5 mm (Fig. 2). For Zdr < 1.5 dB, the bias errors in D0 retrievals for X-band Zdr are larger than for C and S band because of resonant scattering at X band in smaller drops (e.g., 3–4 mm as seen in Fig. 2). As expected, the normalized scatter error is not as greatly affected by variability in the Dmax assumption (Figs. 8 and 10) and is largely controlled by the assumed standard error in Zdr. For a fairly large temperature mismatch of 10°C, the overall bias of D0(Zdr)fit is negligible (≈0 dB) in the mean over the full range of Zdr (i.e., for all DSDs). For larger Zdr values, the D0(Zdr)fit bias error can approach ±5%–10% for a temperature bias of ±10°C (Fig. 13). While these errors are not trivial, it is worthwhile to note that the impact of Dmax uncertainty is likely often larger than the impact of temperature uncertainty on the accuracy of the D0(Zdr)fit estimator (cf. Figs. 9 and 13).
As noted by others (e.g., Ulbrich 1992; Smith et al. 1993), estimating the appropriate maximum drop diameter of a DSD from a single disdrometer measurement over a period that is consistent with the spatial scale of radar data is difficult if not futile based on sampling limitations and may often underestimate Dmax. On the other hand, using larger Dmax (e.g., ≥2.5D0) for all DSD that are exceedingly rare (i.e., occurrence at 95th to 99th percentile or larger) in single disdrometer measurements consistent with radar resolution volume spatial scales could potentially overestimate Dmax. As shown in this study, this uncertainty in Dmax has implications for the estimated bias error in retrieved D0 using Zdr, especially at C band because of resonance.
Without reducing this uncertainty, Dmax is currently a tunable parameter in the radar retrieval model that can be adjusted to maximize agreement between polarimetric radar and independently observed (e.g., disdrometers, wind profilers) estimates of D0, assuming independent measurements of drop size are available. If adjustments to Dmax are made to optimize agreement, then this uncertainty in Dmax could be masking other potential sources of polarimetric radar bias error. Regardless of whether independent measurements of drop size are available or not, it would be highly desirable to reduce uncertainty in Dmax associated with a given DSD to increase the robustness of the polarimetric radar estimate of D0, especially at C band. Ongoing efforts to reduce uncertainty in Dmax include dense networks of many disdrometers distributed within a typical radar footprint (e.g., Jaffrain and Berne 2012; Petersen et al. 2013). Including many disdrometers within a dense network will increase the instantaneous DSD sample size in a typical radar footprint and potentially reduce the uncertainty in estimating Dmax relative to a single disdrometer, which is what has been typically done.
Until methods for estimating Dmax and knowledge for parameterizing Dmax improve, we recommend using the Dmax = 3D0 assumption as it minimizes the error relative to the possible upper range of C that is typically observed in disdrometer data and assumed in the literature, as can be seen in Fig. 7. We also strongly recommend that investigators note the uncertainty in their Dmax parameterization and resulting impact on their results based on this study.
Future work will include detailed intercomparisons of very large samples of polarimetric radar and 2DVD disdrometer estimated D0 values using both targeted NASA GPM Ground Validation (GV) field campaign (i.e., several weeks to several months with ancillary precipitation observations) and multiyear fixed site observational datasets such as the ARMOR C-band radar and the CSU 2DVD in Huntsville. In this way, methods for estimating D0 from C-band radar will continue to be improved and uncertainty reduced.
Acknowledgments
This research is funded by Dr. Ramesh Kakar, NASA Precipitation Measurement Mission (PMM), Dr. Gail Skofronick-Jackson, NASA GPM Project Scientist, and Dr. Mathew Schwaller, NASA GPM Project Office under the following NASA Contracts: NNM05AA22A, NNM11AA01A, and NNX13AI89G. We also acknowledge the late Dr. Arthur Hou, whose leadership as the former NASA GPM Project Scientist was essential to the success of the GPM program and the realization of this study. We thank Patrick Gatlin, Matt Wingo, and Chris Schultz for their careful operation and maintenance of the 2DVD units at the NSSTC berm in Huntsville over many years. We thank Dr. V. N. Bringi for the use of the CSU 2DVD data, close collaboration, scientific leadership in the retrieval of raindrop characteristics from polarimetric radar, and his insight into this research. We acknowledge the many members of the NASA PMM DSD Working Group who have provided ideas, insightful comments, and encouragement during the conduct of this research, including Mr. Patrick Gatlin, Dr. Ali Tokay, Dr. Merhala Thurai, and Dr. Chris Williams. Constructive comments from four anonymous reviewers greatly improved the clarity of the text and figures.
REFERENCES
Andsager, K., K. V. Beard, and N. F. Laird, 1999: Laboratory measurements of axis ratios for large raindrops. J. Atmos. Sci., 56, 2673–2683, doi:10.1175/1520-0469(1999)056<2673:LMOARF>2.0.CO;2.
Aydin, K., and V. Giridhar, 1992: C-band dual-polarization radar observables in rain. J. Atmos. Oceanic Technol., 9, 383–390, doi:10.1175/1520-0426(1992)009<0383:CBDPRO>2.0.CO;2.
Aydin, K., H. Direskeneli, and T. Seliga, 1987: Dual-polarization radar estimation of rainfall parameters compared with ground-based disdrometer measurements: October 29, 1982 central Illinois experiment. IEEE Trans. Geosci. Remote Sens., GE-25, 834–844, doi:10.1109/TGRS.1987.289755.
Barber, P., and C. Yeh, 1975: Scattering of electromagnetic waves by arbitrary shaped dielectric bodies. Appl. Opt., 14, 2864–2872, doi:10.1364/AO.14.002864.
Beard, K. V., and C. Chuang, 1987: A new model for the equilibrium shape of raindrops. J. Atmos. Sci., 44, 1509–1524, doi:10.1175/1520-0469(1987)044<1509:ANMFTE>2.0.CO;2.
Beard, K. V., and R. J. Kubesh, 1991: Laboratory measurements of small raindrop distortion. Part II: Oscillation frequencies and modes. J. Atmos. Sci., 48, 2245–2264, doi:10.1175/1520-0469(1991)048<2245:LMOSRD>2.0.CO;2.
Beard, K. V., D. B. Johnson, and D. Baumgardner, 1986: Aircraft observations of large raindrops in warm, shallow, convective clouds. Geophys. Res. Lett., 13, 991–994, doi:10.1029/GL013i010p00991.
Beard, K. V., R. J. Kubesh, and H. T. Ochs, 1991: Laboratory measurements of small raindrop distortion. Part I: Axis ratios and fall behavior. J. Atmos. Sci., 48, 698–710, doi:10.1175/1520-0469(1991)048<0698:LMOSRD>2.0.CO;2.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2002: Experiments in rainfall estimation with a polarimetric radar in a subtropical environment. J. Appl. Meteor., 41, 674–685, doi:10.1175/1520-0450(2002)041<0674:EIREWA>2.0.CO;2, Corrigendum, 44, 186, doi:10.1175/1520-0450(2005)44<186:C>2.0.CO;2.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2003: An evaluation of a drop distribution-based polarimetric radar rainfall estimator. J. Appl. Meteor., 42, 652–660, doi:10.1175/1520-0450(2003)042<0652:AEOADD>2.0.CO;2.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2004a: Comparison of polarimetric radar drop size distribution retrieval algorithms. J. Atmos. Oceanic Technol., 21, 584–598, doi:10.1175/1520-0426(2004)021<0584:COPRDS>2.0.CO;2.
Brandes, E. A., G. Zhang, and J. Vivekanandan, 2004b: Drop size distribution retrieval with polarimetric radar: Model and application. J. Appl. Meteor., 43, 461–475, doi:10.1175/1520-0450(2004)043<0461:DSDRWP>2.0.CO;2.
Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar: Principles and Applications.Cambridge University Press, 636 pp.
Bringi, V. N., V. Chandrasekar, P. Meischner, J. Hubbert, and Y. Golestani, 1991: Polarimetric radar signatures of precipitation at S- and C-bands. IEE Proc. F Radar Signal Process., 138, 109–119.
Bringi, V. N., V. Chandrasekar, and R. Xiao, 1998: Raindrop axis ratios and size distributions in Florida rainshafts: An assessment of multiparameter radar algorithms. IEEE Trans. Geosci. Remote Sens., 36, 703–715, doi:10.1109/36.673663.
Bringi, V. N., G.-J. Huang, and V. Chandrasekar, 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.
Bringi, V. N., V. Chandrasekar, J. Hubbert, E. Gorgucci, W. L. Randeu, and M. Schoenhuber, 2003: Raindrop size distribution in different climatic regimes from disdrometer and dual-polarized radar analysis. J. Atmos. Sci., 60, 354–365, doi:10.1175/1520-0469(2003)060<0354:RSDIDC>2.0.CO;2.
Bringi, V. N., M. Thurai, K. Nakagawa, G. J. Huang, T. Kobayashi, A. Adachi, H. Hanado, and S. Sekizawa, 2006: Rainfall estimation from C-band polarimetric radar in Okinawa, Japan: Comparisons with 2D-video disdrometer and 400 MHz wind profiler. J. Meteor. Soc. Japan, 84, 705–724, doi:10.2151/jmsj.84.705.
Bringi, V. N., C. R. Williams, M. Thurai, and P. T. May, 2009: Using dual-polarized radar and dual-frequency profiler for DSD characterization: A case study from Darwin, Australia. J. Atmos. Oceanic Technol., 26, 2107–2122, doi:10.1175/2009JTECHA1258.1.
Cao, Q., and G. Zhang, 2009: Errors in estimating raindrop size distribution parameters employing disdrometer and simulated raindrop spectra. J. Appl. Meteor. Climatol., 48, 406–425, doi:10.1175/2008JAMC2026.1.
Cao, Q., G. Zhang, E. Brandes, T. Schuur, A. Ryzhkov, and K. Ikeda, 2008: Analysis of video disdrometer and polarimetric radar data to characterize rain microphysics in Oklahoma. J. Appl. Meteor. Climatol., 47, 2238–2255, doi:10.1175/2008JAMC1732.1.
Carey, L. D., S. A. Rutledge, D. A. Ahijevych, and T. D. Keenan, 2000: Correcting propagation effects in C-band polarimetric radar observations of tropical convection using differential propagation phase. J. Appl. Meteor., 39, 1405–1433, doi:10.1175/1520-0450(2000)039<1405:CPEICB>2.0.CO;2.
Chandrasekar, V., W. A. Cooper, and V. N. Bringi, 1988: Axis ratios and oscillations of raindrops. J. Atmos. Sci., 45, 1323–1333, doi:10.1175/1520-0469(1988)045<1323:ARAOOR>2.0.CO;2.
Chandrasekar, V., A. Hou, E. Smith, V. N. Bringi, S. A. Rutledge, E. Gorgucci, W. A. Petersen, and G. Skofronick Jackson, 2008: Potential role of dual-polarization radar in the validation of satellite precipitation measurements: Rationale and opportunities. Bull. Amer. Meteor. Soc., 89, 1127–1145, doi:10.1175/2008BAMS2177.1.
Fujiyoshi, Y., I. Yamamura, N. Nagumo, K. Nakagawa, K. Muramoto, and T. Shimomai, 2008: The maximum size of raindrops—Can it be a proxy of precipitation climatology? Extended Abstracts, Int. Conf. on Clouds and Precipitation, Cancun, Mexico, International Commission on Clouds and Precipitation, P1.31. [Available online at http://cabernet.atmosfcu.unam.mx/ICCP-2008/abstracts/Program_on_line/Poster_01/Fujiyoshi_extended.pdf.]
Gatlin, P. N., M. Thurai, V. N. Bringi, W. Petersen, D. Wolff, A. Tokay, L. Carey, and M. Wingo, 2015: Searching for large raindrops: A global summary of two-dimensional video disdrometer observations. J. Appl. Meteor. Climatol., 54, 1069–1089, doi:10.1175/JAMC-D-14-0089.1.
Goddard, J. W. F., and S. M. Cherry, 1984: The ability of dual-polarization radar (copolar linear) to predict rainfall rate and microwave attenuation. Radio Sci., 19, 201–208, doi:10.1029/RS019i001p00201.
Goddard, J. W. F., S. M. Cherry, and V. N. Bringi, 1982: Comparison of dual-polarization radar measurements of rain with ground-based disdrometer measurements. J. Appl. Meteor., 21, 252–256, doi:10.1175/1520-0450(1982)021<0252:CODPRM>2.0.CO;2.
Goddard, J. W. F., J. D. Eastment, and J. Tan, 1994: Self-consistent measurements of differential phase and differential reflectivity in rain. Proc. 1994 Int. Geoscience and Remote Sensing Symp., Pasadena, CA, IEEE, 369–371.
Gorgucci, E., G. Scarchilli, and V. Chandrasekar, 1999: A procedure to calibrate multiparameter weather radar using properties of the rain medium. IEEE Trans. Geosci. Remote Sens., 37, 269–276, doi:10.1109/36.739161.
Gorgucci, E., V. Chandrasekar, V. N. Bringi, and G. Scarchilli, 2002: Estimation of raindrop size distribution parameters from polarimetric radar measurements. J. Atmos. Sci., 59, 2373–2384, doi:10.1175/1520-0469(2002)059<2373:EORSDP>2.0.CO;2.
Hobbs, P. V., and A. L. Rangno, 2004: Super-large raindrops. Geophys. Res. Lett., 31, L13102, doi:10.1029/2004GL020167.
Hou, A. Y., and Coauthors, 2014: The Global Precipitation Measurement (GPM) Mission. Bull. Amer. Meteor. Soc., 95, 701–722, doi:10.1175/BAMS-D-13-00164.1.
Huang, G.-J., V. N. Bringi, and M. Thurai, 2008: Orientation angle distributions of drops after an 80-m fall using a 2D video disdrometer. J. Atmos. Oceanic Technol., 25, 1717–1723, doi:10.1175/2008JTECHA1075.1.
Jaffrain, J., and A. Berne, 2012: Quantification of the small-scale spatial structure of the raindrop size distribution from a network of disdrometers. J. Appl. Meteor. Climatol., 51, 941–953, doi:10.1175/JAMC-D-11-0136.1.
Jameson, A. R., 1983: Microphysical interpretation of multi-parameter radar measurements in rain. Part I: Interpretation of polarization measurements and estimation of raindrop shapes. J. Atmos. Sci., 40, 1792–1802, doi:10.1175/1520-0469(1983)040<1792:MIOMPR>2.0.CO;2.
Jameson, A. R., 1994: An alternative approach to estimating rainfall rate by radar using propagation differential phase shift. J. Atmos. Oceanic Technol., 11, 122–131, doi:10.1175/1520-0426(1994)011<0122:AAATER>2.0.CO;2.
Kamra, A. K., R. V. Bhalwankar, and A. B. Sathe, 1991: Spontaneous breakup of charged and uncharged water drops freely suspended in a wind tunnel. J. Geophys. Res., 96, 17 159–17 168, doi:10.1029/91JD01475.
Keenan, T. D., L. D. Carey, D. S. Zrnić, and P. T. May, 2001: Sensitivity of 5-cm wavelength polarimetric radar variables to raindrop axial ratio and drop size distribution. J. Appl. Meteor., 40, 526–545, doi:10.1175/1520-0450(2001)040<0526:SOCWPR>2.0.CO;2.
Levenberg, K., 1944: A method for the solution of certain non-linear problems in least squares. Quart. J. Appl. Math., 2, 164–168.
Marquardt, D., 1963: An algorithm for least-squares estimation of nonlinear parameters. SIAM J. Appl. Math., 11, 431–441, doi:10.1137/0111030.
Meischner, P. F., V. N. Bringi, D. Heimann, and H. Höller, 1991: A squall line in southern Germany: Kinematics and precipitation formation as deduced by advanced polarimetric and Doppler radar measurements. Mon. Wea. Rev., 119, 678–701, doi:10.1175/1520-0493(1991)119<0678:ASLISG>2.0.CO;2.
Petersen, W. A., and Coauthors, 2005: The UAH-NSSTC/WHNT ARMOR C-band dual-polarimetric radar: A unique collaboration in research, education and technology transfer. Extended Abstracts, 32nd Conf. on Radar Meteorology, Albuquerque, NM, Amer. Meteor. Soc., 12R.4. [Available online at https://ams.confex.com/ams/pdfpapers/96524.pdf.]
Petersen, W. A., K. R. Knupp, D. J. Cecil, J. R. Mecikalski, C. Darden, and J. Burks, 2007: The University of Alabama Huntsville THOR center instrumentation: Research and operational collaboration. Extended Abstracts, 33rd Int. Conf. on Radar Meteorology, Cairns, Australia, Amer. Meteor. Soc., 5.1. [Available online at https://ams.confex.com/ams/pdfpapers/123410.pdf.]
Petersen, W. A., V. N. Bringi, V. Chandrasekar, S. A. Rutledge, D. B. Wolff, A. Tokay, L. D. Carey, and M. Thurai, 2013: Radar deployments and field sampling for GPM ground validation field campaigns. Proc. 36th Conf. on Radar Meteorology, Breckenridge, CO, Amer. Meteor. Soc., 179. [Available online at https://ams.confex.com/ams/36Radar/webprogram/Handout/Paper228660/Petersen_etal_RadConf2013_Poster.pdf.]
Rauber, R. M., K. V. Beard, and B. M. Andrews, 1991: A mechanism for giant raindrop formation in warm, shallow convective clouds. J. Atmos. Sci., 48, 1791–1797, doi:10.1175/1520-0469(1991)048<1791:AMFGRF>2.0.CO;2.
Ryzhkov, A., and D. Zrnić, 2005: Radar polarimetry at S, C, and X bands: Comparative analysis and operational implications. Extended Abstracts, 32nd Conf. on Radar Meteorology, Albuquerque, NM, Amer. Meteor. Soc., 9R.3. [Available online at https://ams.confex.com/ams/pdfpapers/95684.pdf.]
Schönhuber, M., G. Lammer, and W. L. Randeu, 2008: The 2D-video-disdrometer. Precipitation: Advances in Measurement, Estimation and Prediction, S. Michaelides, Ed., Springer, 3–31.
Schuur, T. J., A. Ryzhkov, D. S. Zrnić, and M. Schönhuber, 2001: Drop size distributions measured by a 2D video disdrometer: Comparison with dual-polarization radar data. J. Appl. Meteor., 40, 1019–1034, doi:10.1175/1520-0450(2001)040<1019:DSDMBA>2.0.CO;2.
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., K. Aydin, and H. Direskeneli, 1986: Disdrometer measurements during an intense rainfall event in central Illinois: Implications for differential reflectivity radar observations. J. Climate Appl. Meteor., 25, 835–846, doi:10.1175/1520-0450(1986)025<0835:DMDAIR>2.0.CO;2.
Smith, P. L., Z. Liu, and J. Joss, 1993: A study of sampling-variability effects in raindrop size distributions. J. Appl. Meteor., 32, 1259–1269, doi:10.1175/1520-0450(1993)032<1259:ASOSVE>2.0.CO;2.
Tabary, P., G. Vulpiani, J. J. Gourley, A. J. Illingworth, R. J. Thompson, and O. Bousquet, 2009: Unusually high differential attenuation at C-band: Results from a two-year analysis of the French Trappes polarimetric radar data. J. Appl. Meteor. Climatol., 48, 2037–2053, doi:10.1175/2009JAMC2039.1.
Takahashi, T., K. Suzuki, M. Orita, M. Tokuno, and R. de la Mar, 1995: Videosonde observations of precipitation processes in equatorial cloud clusters. J. Meteor. Soc. Japan, 73, 509–534.
Thurai, M., and V. N. Bringi, 2005: Drop axis ratios from 2D video disdrometer. J. Atmos. Oceanic Technol., 22, 966–978, doi:10.1175/JTECH1767.1.
Thurai, M., G. J. Huang, V. N. Bringi, W. L. Randeu, and M. Schönhuber, 2007: Drop shapes, model comparisons, and calculations of polarimetric radar parameters in rain. J. Atmos. Oceanic Technol., 24, 1019–1032, doi:10.1175/JTECH2051.1.
Thurai, M., M. Szakáll, V. N. Bringi, K. V. Beard, S. K. Mitra, and S. Borrmann, 2009: Drop shapes and axis ratio distributions: Comparison between 2D video disdrometer and wind-tunnel measurements. J. Atmos. Oceanic Technol., 26, 1427–1432, doi:10.1175/2009JTECHA1244.1.
Thurai, M., W. A. Petersen, A. Tokay, C. Schultz, and P. Gatlin, 2011: Drop size distribution comparisons between Parsivel and 2-D video disdrometers. Adv. Geosci., 30, 3–9, doi:10.5194/adgeo-30-3-2011.
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.
Ulbrich, C. W., 1985: The effects of drop size distribution truncation on rainfall integral parameters and empirical relations. J. Climate Appl. Meteor., 24, 580–590, doi:10.1175/1520-0450(1985)024<0580:TEODSD>2.0.CO;2.
Ulbrich, C. W., 1992: Effects of drop-size-distribution truncation on computer simulations of dual-measurement radar methods. J. Appl. Meteor., 31, 689–699, doi:10.1175/1520-0450(1992)031<0689:EODSDT>2.0.CO;2.
Ulbrich, C. W., and D. Atlas, 1984: Assessment of the contribution of differential polarization to improved rainfall measurements. Radio Sci., 19, 49–57, doi:10.1029/RS019i001p00049.
Ulbrich, C. W., and D. Atlas, 1998: Rainfall microphysics and radar properties: Analysis methods for drop size spectra. J. Appl. Meteor., 37, 912–923, doi:10.1175/1520-0450(1998)037<0912:RMARPA>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.
Waterman, P. C., 1969: Scattering by dielectric obstacles. Alta Freq., 38(Speciale), 348–352.
Zhang, G., J. Vivekanandan, and E. Brandes, 2001: A method for estimating rain rate and drop size distribution from polarimetric radar measurements. IEEE Trans. Geosci. Remote Sens., 39, 830–841, doi:10.1109/36.917906.
Zrnić, D. S., T. D. Keenan, L. D. Carey, and P. May, 2000: Sensitivity analysis of polarimetric variables at a 5-cm wavelength in rain. J. Appl. Meteor., 39, 1514–1526, doi:10.1175/1520-0450(2000)039<1514:SAOPVA>2.0.CO;2.
