Intercomparisons of CloudSat and Ground-Based Radar Retrievals of Rain Rate over Land

Sergey Y. Matrosov Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, and NOAA/Earth System Research Laboratory, Boulder, Colorado

Search for other papers by Sergey Y. Matrosov in
Current site
Google Scholar
PubMed
Close
Full access

Abstract

Experimental retrievals of rain rates using the CloudSat spaceborne 94-GHz radar reflectivity gradient method over land were evaluated by comparing them with standard estimates from ground-based operational S-band radar measurements, which are widely used for quantitative precipitation estimations. The comparisons were performed for predominantly stratiform precipitation events that occurred in the vicinity of the Weather Surveillance Radar-1988 Doppler (WSR-88D) KGWX and KSHV radars during the CloudSat overpasses in the vicinity of these ground radar sites. The standard reflectivity-based WSR-88D rain-rate retrievals used in operational practice were utilized as a reference for the CloudSat retrieval evaluation. Spaceborne and ground-based radar rain-rate estimates that were closely collocated in space and time were generally well correlated. The correlation coefficients were approximately 0.65 on average, and the mean relative biases were usually within ±35% for the whole dataset and for individual events with typical rain rates exceeding ~2 mm h−1. For events with lighter rainfall, higher biases and lower correlations were often present. The normalized mean absolute differences between satellite- and ground-based radar retrievals were on average ~60%, with an increasing trend for lighter rainfall. Such mean differences are comparable to combined retrieval errors from both ground-based and satellite radar remote sensing approaches. Evaluation of potential effects of partial beam blockage on the ground-based radar measurements was performed, and the influence of the choice of relation between WSR-88D reflectivity and rain rate that was utilized in the ground-based rain-rate retrievals was assessed.

Corresponding author address: Sergey Y. Matrosov, R/PSD2, 325 Broadway, Boulder, CO 80305. E-mail: sergey.matrosov@noaa.gov

Abstract

Experimental retrievals of rain rates using the CloudSat spaceborne 94-GHz radar reflectivity gradient method over land were evaluated by comparing them with standard estimates from ground-based operational S-band radar measurements, which are widely used for quantitative precipitation estimations. The comparisons were performed for predominantly stratiform precipitation events that occurred in the vicinity of the Weather Surveillance Radar-1988 Doppler (WSR-88D) KGWX and KSHV radars during the CloudSat overpasses in the vicinity of these ground radar sites. The standard reflectivity-based WSR-88D rain-rate retrievals used in operational practice were utilized as a reference for the CloudSat retrieval evaluation. Spaceborne and ground-based radar rain-rate estimates that were closely collocated in space and time were generally well correlated. The correlation coefficients were approximately 0.65 on average, and the mean relative biases were usually within ±35% for the whole dataset and for individual events with typical rain rates exceeding ~2 mm h−1. For events with lighter rainfall, higher biases and lower correlations were often present. The normalized mean absolute differences between satellite- and ground-based radar retrievals were on average ~60%, with an increasing trend for lighter rainfall. Such mean differences are comparable to combined retrieval errors from both ground-based and satellite radar remote sensing approaches. Evaluation of potential effects of partial beam blockage on the ground-based radar measurements was performed, and the influence of the choice of relation between WSR-88D reflectivity and rain rate that was utilized in the ground-based rain-rate retrievals was assessed.

Corresponding author address: Sergey Y. Matrosov, R/PSD2, 325 Broadway, Boulder, CO 80305. E-mail: sergey.matrosov@noaa.gov

1. Introduction

Although the primary goal of the CloudSat nadir-pointing W-band (94 GHz; wavelength λ = 3.2 mm) Cloud Profiling Radar (CPR) is to collect global information on clouds, this spaceborne radar proved to be a useful tool for observing and quantifying precipitation (e.g., Stephens et al. 2008). Traditional nonpolarimetric radar approaches for quantitative precipitation estimation (QPE) are based on relating the equivalent radar reflectivity factor Ze (hereinafter, just reflectivity) to rain rate R. These approaches, however, are of limited use for interpreting CloudSat data in liquid precipitation because of a number of factors including high attenuation of W-band signals in rain, multiple-scattering (MS) effects that are due to the geometry of observations (e.g., Battaglia et al. 2010), and non-Rayleigh-scattering effects, which usually result in saturation of nonattenuated W-band reflectivities in rainfall at a level of ~25–27 dBZ (Matrosov 2007).

CloudSat methods for rain-rate estimation have to utilize radar signal-attenuation effects as useful information for liquid precipitation retrievals (e.g., Matrosov et al. 2008; Haynes et al. 2009; Mitrescu et al. 2010; Lebsock and L’Ecuyer 2011). These effects are used with CloudSat data in different ways. Most existing methods use the path-integrated-attenuation (PIA) constraint, which is determined by assuming that radar returns from the surface in clear air are known with a reasonable accuracy. Application of the PIA-based retrieval approaches, such as those used to derive the existing CloudSat rainfall “2C-PRECIP-COLUMN” (Haynes et al. 2009) and “2C-RAIN-PROFILE” (Lebsock and L’Ecuyer 2011) products, is currently limited to CPR measurements over water, when would-be surface returns in the absence of hydrometeors in a vertical atmospheric column are approximated on the basis of a priori information about surface wind speed and temperature (e.g., Tanelli et al. 2008). Use of the PIA constraint is not generally available for heavier rainfall when the surface returns are not detected because of a combination of very strong attenuation and MS effects (e.g., Battaglia et al. 2008).

A gradient method to retrieve rain rates from CloudSat data (e.g., Matrosov 2007, 2013) estimates the W-band attenuation coefficient (i.e., specific attenuation) in rain from the vertical gradients of observed CPR reflectivities and then relates this coefficient to R. This method does not use the PIA information from surface returns and therefore is applicable to observations above any surfaces. The main assumption of this method is that after accounting for gaseous attenuation the vertical gradients of observed reflectivities are primarily caused by liquid hydrometer attenuation and, for heavier rainfall, by MS enhancement. Changes in a vertical profile of nonattenuated reflectivities contribute to the uncertainty of estimates of the attenuation coefficient (and therefore rain rate). Given this assumption, the gradient method is best suited for stratiform precipitation for which vertical variations in nonattenuated reflectivity profiles are relatively small even at centimeter wavelengths (e.g., Bringi and Chandrasekar 2001), which are typically used in ground-based precipitation sensing radars. At W band the vertical variability of nonattenuated reflectivity in stratiform rain is further suppressed relative to longer wavelengths because of strong non-Rayleigh-scattering effects.

The main objective of this study was to evaluate CloudSat gradient-method rain-rate retrievals in stratiform rainfall over land using ground-based scanning S-band (~3 GHz) Weather Surveillance Radar-1988 Doppler (WSR-88D) measurements. The WSR-88D data are routinely used over the United States for QPE purposes. The uncertainty of WSR-88D QPE retrievals could be as high as 30%–40% as compared with rainfall accumulations that are directly observed by rain gauges, which are often considered to be the “ground truth” (e.g.,Krajewski et al.2010). Nonetheless, scanning-precipitation-radar-based QPE remains the main remote sensing tool for obtaining precipitation information in many practical applications. Earlier different ground-based radar measurements have been used to validate the CloudSat precipitation-occurrence algorithm (Hudak et al. 2008).

2. Precipitation events used for intercomparisons

With a few gaps in the western United States, the Next Generation Weather Radar (NEXRAD) network of WSR-88Ds, which have a 460-km nominal “long” range for reflectivity measurements, covers most of the area of the lower 48 states. Although the WSR-88D systems sample reflectivity data at a 1-km range by 1° azimuth grid in the legacy resolution mode and at a 0.25 km by 0.5° grid in the super resolution mode, which has been in use approximately from the summer of 2008, the actual cross-beam resolution degrades with range because of beam broadening and Earth sphericity effects. The WSR-88D data are collected in repetitive volume patterns, which consist of several azimuthal plan position indicator scans conducted at different elevation angles. A typical duration of a volume pattern is approximately 5 min.

The CloudSat CPR resolution volume is ~1.5 km across the satellite track, ~1.8 km along the track, and 0.5 km in the vertical direction (Tanelli et al. 2008). Vertical oversampling allows for providing vertical profiles of observed reflectivity with an increment of 0.24 km. Overall CPR and WSR-88D sampling volumes near the ground radar sites are not vastly different, which is favorable for comparisons of precipitation-retrieval results. Two consecutive CloudSat orbit ground tracks are spaced by ~24.7° in longitude. The orbits approximately repeat themselves each 16-day period.

Validation/evaluation of different remote sensing QPE methods is usually performed on the basis of the best available collocated comparisons of rainfall accumulations over a certain time interval (e.g., for hourly accumulations) or event total accumulations obtained from methods considered. One example of such an evaluation is comparisons of ground-radar-based QPE results with available gauge data (e.g., Krajewski et al. 2010). Meaningful comparisons with gauge accumulations, however, are not practical for CloudSat rainfall retrievals (except, maybe, for climatological snowfall comparisons in the polar regions where the ground separation between the orbits is much smaller than in the midlatitudes and tropics) because of such factors as fast satellite orbit speed, the long revisiting time period, a large ground-track separation between consecutive orbits, vastly differing resolution volumes, and the fact that most gauge types (unlike the CPR) are better suited to provide information about rainfall accumulation than about instantaneous rain rates.

Because of the impeding factors stated above, this study focuses on rain-rate comparisons as retrieved from CloudSat and WSR-88D measurements with the best possible collocation in space and time. Although WSR-88D-based QPE retrievals are obviously not exactly the ground truth, such comparisons have value because relatively novel CloudSat rain-rate retrievals over land are compared with the results from the ground-based meteorological-radar QPE approach, which has been in practical use for many years and is relatively well established. Realize, however, that instantaneous rain rates generally exhibit higher spatial and temporal variability than do rainfall accumulations.

Intercomparisons of WSR-88D and CloudSat CPR retrievals are most practical for precipitation events observed when the satellite crosses over the ground-based radar sites (or over locations in the vicinity of these sites), where ground-based and spaceborne retrievals could be better collocated. CloudSat ground tracks pass within a few kilometers of a number of the NEXRAD sites. This study focuses on the precipitation events observed in the vicinity of two WSR-88Ds: the KGWX Greenwood Springs, Mississippi, radar (33.8969°N, 88.3292°W) and the KSHV Shreveport, Louisiana, radar (32.4508°N, 93.8414°W). The site altitudes for these radars above mean sea level (MSL) are approximately 140 and 80 m for KGWX and KSHV, respectively.

The ground-based-radar choice was dictated, in part, by the fact that for such southern locations of the WSR-88D sites the freezing level in the atmosphere is relatively high, even during colder months, allowing for retrieval of rain rates by use of the gradient method, which requires at least several resolution gates that are free of ground clutter and melting-layer contamination. It corresponds to a conservative requirement of freezing-level (FL) heights being approximately 2 km above the radar site because CPR measurements in the first several gates, whose centers are nominally above the ground, could be contaminated by surface returns and because the melting layer in stratiform precipitation systems is approximately 500 m thick (e.g., Matrosov 2008). The relatively flat terrain around the KGWX and KSHV WSR-88D sites also minimizes influence of radar beam blockage and ground-clutter effects on precipitation observed with the ground-based radars.

CloudSat passage over a particular ground-based radar during a precipitation occurrence in the vicinity of that radar site is a relatively rare event. Of main interest to this study were precipitation events that consisted mostly of stratiform-rain regions covering relatively large areas (at least several dozens of kilometers). This situation allows for better collocation of satellite and ground-based retrievals and results in more data points for comparisons. Stratiform rainfall typically produces a radar bright band (BB) that is located just below the freezing level (e.g., Bringi and Chandrasekar 2001). Unlike the BB in longer-wavelength radar observations for which the BB is caused by reflectivity enhancement by melting snow/ice particles, the BB in CPR measurements is caused, in part, by strong signal attenuation in liquid hydrometeors (Sassen et al. 2007; Matrosov 2008).

An examination of the available CloudSat overpasses, which occurred over the KGWX radar site during 2006–12, revealed six predominantly stratiform rainfall events, for which horizontal extents along the CloudSat ground track exceeded approximately 50 km and FL heights were greater than 2 km. These events, which are typically associated with the passage of frontal atmospheric systems, were observed on the dates shown in Table 1. The CloudSat orbit crossings over the vicinity of the KGWX ground-based radar site occur on the ascending satellite node at approximately 1915 UTC. It corresponds to 1415 central daylight time at the site location.

Table 1.

Statistical parameters characterizing comparisons of CloudSat and WSR-88D (Ze = 200R1.6) rain-rate retrievals. Means, standard deviations, and RMSE are in millimeters per hour, and RMB and NMAD are in percent.

Table 1.

The same number of precipitation events with the same characteristics (i.e., mostly stratiform rainfall with a detectable BB covering a horizontal range that is greater than approximately 50 km in the ground-based-radar coverage area) was observed also when the CloudSat crossed over the vicinity of the KSHV radar site during 2006–12. The dates of these KSHV crossings are also shown in Table 1. The KSHV site crossings occur on the descending satellite node. The crossing time is approximately 0820 UTC, which corresponds to nighttime at the KSHV location.

The relatively modest number (i.e., 12) of satellite overpasses in this study is explained by event requirements (i.e., the dominance of the stratiform regime, the spatial extent of precipitation, and the FL height), by the stoppage in CloudSat operations during April–October 2011, by the fact that only the orbits above the two WSR-88D locations were considered, and by the relatively long repetition time of the same orbits (i.e., 16 days). On the other hand, these events provided a relatively large amount of individual comparison data points, as shown in section 4. The fact that in general all of the available events that satisfied the requirements mentioned above were considered allows an assumption that the comparisons provided below are representative of mostly stratiform rainfall.

3. Intercomparison approaches and a case study

An example of the observed CloudSat CPR reflectivity cross section of the KGWX precipitation event that occurred on 4 October 2009 is shown in Fig. 1. The satellite ground track passed within 1 km of the KGWX radar. A region of rainfall extends to the both sides from the radar site. Near the KGWX site, the direction of the CloudSat orbit extends approximately in south–north direction along the azimuth of 348°. As seen in Fig. 1, the CPR BB features are prominent and are observed at an altitude of about 4 km MSL with a tendency of descending toward the ground as the satellite moves north.

Fig. 1.
Fig. 1.

A cross section of CloudSat CPR reflectivity during a precipitating event observed on 4 Oct 2009 (~1915 UTC) in the vicinity of the KGWX WSR-88D site.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

For the precipitation event of 4 October 2009, Fig. 2 shows the CloudSat ground track overlaid on the map of KGWX radar reflectivity measurements conducted at the lowest beam elevation angle (i.e., 0.48°) almost simultaneously with the satellite passage. The sequence of radar beam elevation angles (tilts), which are used for WSR-88D volume scans, is determined by the volume coverage patterns (VCPs). The VCP-11 was typically used for observations of precipitation events analyzed in this study, including the KGWX event of 4 October 2009 shown here in relative detail. This VCP has the best overall volume coverage. WSR-88D scanning during this VCP is performed for a sequence of 14 radar elevation angles β, ranging from ~0.5° to 19.5°. Typical values of the elevation angles for this VCP are shown in Fig. 3. The lowest elevation angle of about 0.5° approximately corresponds to the half beamwidth of the WSR-88D antennas. The radar elevation tilt angles in Fig. 3 represent the average values. The actual beam elevations usually slightly vary for individual volume scans. The heights of the radar beam centers above the radar site level for the VCP-11 elevation angle tilts as a function of the distance from the radar are also shown in Fig. 3. These heights increase with radar range because of Earth sphericity and refraction effects. During some of the experimental events analyzed in this study, the VCP-12 and VCP-121 were used for WSR-88D scanning instead of the VCP-11. These other VCPs have a slightly different selection of beam elevation tilts (http://en.wikipedia.org/wiki/NEXRAD). For all WSR-88D VCPs, however, the lowest and highest beam elevations are ~0.5° and 19.5°, respectively; therefore, the total space coverage is approximately the same.

Fig. 2.
Fig. 2.

CloudSat ground track (shown by a white line) for 1914 UTC 4 Oct 2009 overlaid on the lowest elevation tilt reflectivities from the KWGX radar, the location of which is shown by the black dot.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

Fig. 3.
Fig. 3.

Heights of the WSR-88D beam centers above ground level for elevation angles employed during VCP-11 scanning.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

a. The WSR-88D procedure for rain-rate retrieval

The WSR-88D volume-scan measurements can be used to reconstruct S-band reflectivities in different planes emulating horizontal and vertical cross sections. The data from the 1914 UTC KGWX radar volume scan, which closely corresponds to the time of CloudSat passage over the radar site shown in Fig. 1, were used to reconstruct the vertical cross section of observed KGWX reflectivities in the vertical plane of CPR measurements. Figure 4 shows this reconstruction of S-band reflectivities, which are practically unattenuated by moderate rain, in the plane of the CloudSat reflectivity cross section from Fig. 1. The exact latitude and longitude points of the CPR data were used to interpolate KGWX data presented in Fig. 4. In this figure the wedgelike region of no data just above the radar site, which is located at 0.14 km MSL, is not sampled by the KGWX radar because the elevation angle tilt is limited by 19.5°.

Fig. 4.
Fig. 4.

A cross section of KGWX radar reflectivities matching the CloudSat overpass on 4 Oct 2009 that is shown in Fig. 1. The lower and upper lines bounding the areas of no data correspond to the lower/upper edges of the first/last tilts (i.e., 0.0° and 20.0°) of the WSR-88D beam.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

When reconstructing the WSR-88D reflectivity cross sections, the heights of the upper and lower edges of the radar beam [hi(u) and hi(l), respectively] for a given elevation angle βi (where i is the beam tilt number) and the radar range were estimated for the spherical Earth geometry using the approach from Doviak and Zrnic (1993, their Eqs. 2.28) and accounting for the NEXRAD radar beamwidth of 0.96° [i.e., hi(u) and hi(l) correspond to the elevation angle tilts of βi + 0.48° and βi − 0.48°, respectively] with an assumption of the Gaussian antenna pattern. The WSR-88D reflectivity data were sampled at latitudes and longitudes of each CloudSat CPR data profile with averaging along the beam of ±0.9 km from the profile center. Such averaging approximately accounts for the CPR footprint along the satellite ground track. When sampling the ground-based radar data at CPR data geographical locations, the WSR-88D measurements from two neighboring azimuths that bracket the direction to the center of the CPR profile were linearly interpolated to match better the satellite data.

From comparisons of Figs. 1 and 4, it can be seen that, despite its coarser resolution, the WSR-88D reflectivity cross section reproduces the general vertical features of the precipitating system seen in the CPR data, including higher cloud tops between latitudes of about 33.2° and 34.5°, and some isolated reflectivity spikes (e.g., near 35.5°). Except the vicinity of the radar site (i.e., approximately between 33.4° and 34.3°), the radar BB are not very well pronounced in the KGWX data. This can be explained by the beam-broadening effects, which degrade the actual WSR-88D resolution with increasing range. The closely collocated (in space and time) CloudSat CPR and KGWX WSR-88D measurements, such as those shown in Figs. 1 and 4, allow for detailed intercomparisons of the spaceborne and ground-based rain-rate retrievals with proper matching satellite and ground-based radar estimates.

Although the NEXRAD network was polarimetrically upgraded during 2012–13 and new dual-polarization radar QPE approaches are being developed, the WSR-88D data collocated with CloudSat measurements available to this study include only the WSR-88D reflectivity and Doppler velocity measurements. Therefore, the conventional WSR-88D rain-rate retrievals that are based on S-band reflectivity (Ze) are considered here. Such retrievals are used in the National Mosaic and QPE system (NMQ; e.g., Zhang et al. 2011). The standard NEXRAD ZeR relation for stratiform rainfall, which is given by
e1
was primarily utilized with ground-based WSR-88D measurements in this study.

As mentioned above, the precipitation event shown in Fig. 1 is generally of the stratiform type. For the most part it exhibits a pronounced BB except in an area of warm rain (i.e., at latitudes south from approximately 33°), which has echo tops that are generally lower than the environmental freezing level. There are also some small regions, where rainfall can be interpreted as convective (e.g., near latitudes of 33.25°, 33.35°, and 35.3°). The CPR BB is elevated in these regions (relative to the stratiform areas), which is likely due to some upward air motions. The CPR ground returns are not clearly seen in convective regions, nor in some stratiform areas (e.g., around 34°), because the W-band signal attenuation by rainfall there is strong and MS effects are significant. Overall, the BB separates the predominantly ice-phase precipitating cloud above and the liquid hydrometeor layer below. The rain-rate retrievals from the WSR-88D were performed using the reflectivity measurements observed when corresponding interpolated resolution volumes were fully within the rainfall layer.

In addition to the spatial interpolation and averaging of the WSR-88D data, the linear time interpolation of the ground-based radar measurements was performed using two consecutive volume scans that bracket the CPR profile time. All of the interpolation/averaging procedures for the given tilt data as mentioned above were performed for the S-band reflectivity measurements in units of millimeters to the sixth power per meter cubed. Interpolating introduces some uncertainties in WSR-88D reflectivities matched with CPR measurements. These uncertainties might contribute to the scatter in the retrieved rain-rate data, which is analyzed in section 4.

The resulting mean WSR-88D reflectivities were then converted to rain-rate estimates using the Ze–R relation. For each CPR profile, the mean NEXRAD derived rain rate RN was calculated as an average of all rain rates at elevation tilts that are fully within the rainfall layer. The upper boundary of this layer hr was determined using the CPR BB height hBB:
e2
where the 0.6-km term conservatively accounts for a thickness of the melting layer, the influence of which should be minimized when calculating WSR-88D rain-rate estimates. The CPR BB heights were used because WSR-88D BBs are often not very pronounced (especially at longer radar ranges) because of the beam-broadening effects.

b. The CloudSat procedure for rain-rate retrievals

The CloudSat reflectivity-gradient method estimates the W-band attenuation coefficient in rain α(h) at a height h from the vertical gradients of measured W-band Zeo reflectivity as (Matrosov 2007, 2013)
e3
where the term Go(h) accounts for model attenuation in atmospheric gases and nonprecipitating liquid. The subscript ss for the gradient denotes the single-scattering-assumption values, which are obtained from the observed-gradients values affected by MS {i.e., [∂Zeo(h)/∂h]ms} as
e4
In (4) the dimensionless coefficient γ ≤ 1, and the values of this coefficient for stratiform rainfall with different FL heights are adopted from Matrosov et al. (2008). Because γ also depends on rain rate, correcting for MS requires iterations.

After estimating the α(h) and correcting for the air-density height changes, the retrieved CloudSat rain-rate values are derived from the linearized relation between rain rate and attenuation coefficient: R (mm h−1) = 1.2α (dB km−1) (Matrosov 2007). A typical estimated retrieval uncertainty is approximately 30%–40% for rain rates between approximately 5 and 15 mm h−1, and it increases for lower rain rates. Multiple scattering overwhelms CloudSat measurements for heavy rainfall, and retrievals of rain rates that are higher than about 20–25 mm h−1 using any CloudSat method are problematic.

c. Intercomparisons for the case study

For the event of 4 October 2009, Fig. 5 shows WSR-88D rain rates RN sampled along the CloudSat ground track according to the procedures described above. The data are not shown for latitudes that correspond to the KGWX radar ranges when the upper edges of the lowest tilt WSR-88D beam were higher than the upper boundary of the rain layer [i.e., h1(u) > hr]. For those latitudes, WSR-88D-based rainfall QPE is contaminated by melting and ice hydrometeors. The comparison data affected by the so-called QPE cone of silence (see Fig. 4) near the radar site, within which no WSR-88D data are available because of the tilt limitations, were excluded.

Fig. 5.
Fig. 5.

Rain rate along the satellite-orbit ground track retrieved from CPR and KGWX WSR-88D measurements during the CloudSat overpass at 1914 UTC 4 Oct 2009.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

Figure 5 also depicts the CloudSat rain-rate RC results retrieved from CPR measurements shown in Fig. 1. To closely match the WSR-88D retrievals, RC data depicted in Fig. 5 represent average values for the rain layer resolved by the ground-based radar measurements. The CPR data that were suspected to be contaminated by surface returns were not used in the retrievals. The CloudSat results are shown for the same latitude interval as the KGWX rain-rate estimates.

As seen from Fig. 5, the collocated WSR-88D and CPR retrievals of rain rate are in overall agreement, except for a small area of warm rain observed to the south of 33°. The rainfall echo in this area does not generally extend above the freezing level, and therefore no radar BB signatures are present. Overall, results for both ground-based and satellite retrieval exhibit characteristic maxima of rain rates in the regions of isolated convective activity. Some underestimation of CloudSat retrievals relative to the ground-based estimates is present for the highest rain-rate peak (i.e., near 33.25°), which corresponds to the squall line seen to the south from the KGWX radar site in Fig. 2 and is slightly mismatched in satellite and ground-based data.

4. Statistical results of intercomparisons

The intercomparison data from all 12 aforementioned precipitation events of collocated WSR-88D (including KGWX and KSHV) and CPR rain-rate retrievals are shown on a scatterplot in Fig. 6a. The presented data correspond to the rainfall exhibiting the BB as detected by the CloudSat measurements. The Ze–R relation in (1) was used for WSR-88D retrievals. Although most such rainfall can be considered to be stratiform, some local areas of convective-rain elevated BB features may be present as discussed in the analysis of Fig. 1.

Fig. 6.
Fig. 6.

The scatterplots between approximately collocated rain-rate retrievals from the CloudSat CPR and KGWX/KSHV radar measurements for the (a) Ze = 200R1.6 and (b) Ze = 300R1.4 relations. Combined data from 12 observational events are shown. Black, red, green, blue, cyan, and purple square/plus symbols correspond to the first, second, third, fourth, fifth, and sixth KGWX/KSHV events as marked in Table 1.

Citation: Journal of Applied Meteorology and Climatology 53, 10; 10.1175/JAMC-D-14-0055.1

To assess intercomparisons of rain-rate retrievals in a quantitative sense, the correlation coefficient r between RC and RN as well as different statistical parameters characterizing the retrieval intercomparisons were calculated. These statistical parameters include the normalized relative mean bias (RMB),
e5
the root-mean-square error (RMSE),
e6
and the normalized mean absolute difference (NMAD),
e7
where angle brackets denote averaging with respect to the dataset, which, for all of the events considered in this study, consisted of 1349 closely collocated ground-based and satellite rain-rate estimates.

Table 1 presents the statistical parameters of the ground-based and spaceborne rain-rate retrievals for the 12 individual events and for the whole dataset, the scatterplots of which are shown in Fig. 6. There is a relatively good average correlation between RN and RC values, which is characterized by a correlation coefficient of 0.67 for the whole dataset. On average, the CloudSat estimates are biased slightly positive (higher; ~10%) in comparison with the WSR-88D retrievals. The NMAD value of 58% for the whole dataset is comparable to estimated uncertainties of retrievals. As with CloudSat retrievals, reflectivity-based WSR-88D rain-rate retrievals also have substantial errors. Krajewski et al. (2010) showed that these errors for event total accumulations are on the order of 25%–40%. Higher errors are expected for WSR-88D instantaneous rain-rate retrievals, which are compared here with satellite retrievals.

An analysis of the statistical scores for individual events (Table 1) shows that larger CloudSat rain-rate biases (relative to the WSR-88D retrievals) are observed for those events characterized by lighter rainfall with mean RC ≤ 2 mm h−1. One possible reason for this fact is that, as mentioned previously, the uncertainty of the CPR gradient method retrievals degrades as rain rate diminishes and signal attenuation in rain becomes less pronounced. For the rest of the events with mean RC > 2 mm h−1, the relative biases between satellite- and ground-based radar rain-rate estimates are generally within ±34%. These results are consistent with previous comparisons between CloudSat and WSR-88D rain-rate estimates in landfalling Hurricanes Gustav and Ike, which were characterized by relative biases in the range from −8% to 31% and standard deviations of ~54%–56% (Matrosov 2011).

Overall, it can be concluded that the CloudSat–WSR-88D retrieval differences are within uncertainties caused by joint errors of ground-based and satellite remote sensing approaches, because the variance of the difference between two independent estimates of rain rate is the sum of their variances and the variance can be considered as the expected error (which is defined in terms of the standard deviation) squared.

Although, for the purpose of intercomparison consistency, the results in Table 1 and Figs. 5 and 6a were obtained using the WSR-88D retrievals utilizing the relation in (1), an effect of using another common NEXRAD Ze–R relation, namely,
e8
was evaluated. Even though the relation in (8) is considered to be convective in the NMQ system (e.g., Zhang et al. 2011), it is sometimes used as a default relation in many WSR-88D-based QPE schemes. A scatterplot of spaceborne and ground-based radar rain-rate retrievals when (8) was used for deriving WSR-88D estimates is shown in Fig. 6b. This change of the ZeR relation did not significantly affect intercomparison results. The corresponding statistical scores for the whole dataset were 0.66, 15%, 2.0 mm h−1, and 61% for r, RMB, RMSE, and NMAD, respectively, which are relatively close to the scores for the WSR-88D relation in (1) that are given in the first line of Table 1.

There is not much difference between the results when either (1) or (8) is used, even though the overwhelming majority of retrievals in this study represent stratiform precipitation regions with low-to-moderate rain rates. This is because both stratiform and convective default NEXRAD ZeR relations produce very similar (~10%–15%) rain rates when reflectivities are within the range between 25 and 45 dBZ. This was a typical range of WSR-88D reflectivities that were observed during the observational events considered in this study.

The ZeR relations are determined by the raindrop size distributions (DSDs). While there is not much difference in the intercomparisons resulting from the choice between the WSR-88D mean relations in (1) and (8), the variability among such relations for individual events that is due to DSD differences can be significant (e.g., Doviak and Zrnic 1993). The DSD-driven variability in CloudSat αR relations used for the satellite retrievals here is expected to be more modest (in relative terms) than that for S-band ZeR relations because of significant non-Rayleigh-scattering effects that dampen the DSD influence at W band. It is suggested that the varying biases between ground-based and satellite radar retrievals of rain rate for different events are caused, in part, by DSD-induced differences in event-specific ZeR and αR relations. Specific DSDs and relations, however, are typically not known a priori for particular events, and the radar-based methods usually use mean ZeR relations as presented in this study.

The CloudSat gradient rain-rate retrieval method is immune to the CPR absolute calibration, but WSR-88D reflectivity-based QPE retrievals are sensitive to the reflectivity measurement errors. Although the NEXRAD radars undergo regular absolute calibration checks, some calibration biases are possible (e.g., Gourley et al. 2003). A 1-dB bias in Ze measurements, for example, would result in an ~15%–20% shift in retrieved rain rates when a ZeR relation with an exponent similar to those in (1) and (8) is used. This will affect statistical parameters of intercomparisons (except for the correlation coefficient). The detailed calibration analysis of the WSR-88Ds is, however, beyond the scope of the current research. The standard quality-controlled level-II NEXRAD data were used in this study.

Another potential factor that can negatively affect the WSR-88D reflectivity-based rain-rate retrievals is partial beam blockage, which could result in underestimation of reflectivity values and, hence, reduce rain rates retrieved from ground-based radar data. The lower beam edge of the lowest elevation tilt for NEXRAD scanning routines is close to 0° and shielding by nearby tall terrain features or objects can cause partial beam blockage for some azimuthal directions. Although the KGWX and KSHV radar coverage maps from the National Climatic Data Center (obtained online from http://www.ncdc.noaa.gov/nexradinv/) do not indicate any significant beam blockage even for the lowest beam tilt at these radar locations, influences of potential beam-blockage effects on the derived rain-rate retrieval intercomparison statistics were evaluated as part of this study.

During this evaluation, in addition to the intercomparisons described above when data from all WSR-88D tilts were used, the intercomparisons of spaceborne and ground-based radar rain retrievals were conducted with WSR-88D data when the 0.5° elevation tilt data were excluded. As a result of this exclusion, the RMB, RMSE, NMAD, and r values corresponding to the Ze–R relation in (1) and the whole dataset changed to 19%, 2.2 mm h−1, 67%, and 0.6, respectively. Similar relative changes were present when the Ze–R relation in (8) was used (data are not shown). The overall change in statistical scores is relatively modest, and their variability may also (i.e., in addition to potential beam-blockage effects) be influenced by some vertical variability in rain rates because the data at lower altitudes were omitted as a result of the exclusion of the lowest tilt measurements. Thus it can be concluded that the WSR-88D partial beam-blockage effect (if any) did not influence the results of this study intercomparison in a significant way.

5. Summary and conclusions

Attenuation-based approaches to retrieve rain rates over water surfaces have been used with the CloudSat spaceborne 94-GHz radar measurements for a number of years, and the corresponding data products exist. This study was focused on evaluating CloudSat experimental retrievals of rain rates over land surfaces using reflectivity gradients observed in liquid hydrometeor layers of predominantly stratiform precipitating systems exhibiting radar bright band. The evaluation was achieved by intercomparisons of CloudSat rain-rate retrievals with conventional estimates from WSR-88Ds using data collected during 2006–12 from satellite overpasses in the vicinity of the WSR-88D sites.

The WSR-88D KGWX and KSHV radars were chosen for these intercomparisons because of the existence of CloudSat overpasses over them during precipitation events and also because of the relatively flat terrain in their vicinity, which minimizes beam-blockage effects in ground-based radar measurements. Rain-rate retrievals using the CloudSat attenuation-based reflectivity-gradient method were compared with estimates from the KGWX and KSHV radars that are closely collocated in space and time. The standard NEXRAD QPE approach, which is based on utilizing the ZeR relations, was used in these intercomparisons. This approach is widely used in many hydrological and meteorological applications, and it was considered in this study as a reference for testing a relatively novel satellite technique such as the CloudSat rain-rate retrieval method over land. The rain-rate retrievals from the satellite and ground-based radars used in this study were performed for rainfall layers that were free from contamination from melting and ice hydrometeors. The BB and CPR surface return requirements limited retrieval datasets to events with freezing levels that were not lower than approximately 2 km above radar sites.

The intercomparison results showed that rain rates obtained from CloudSat and WSR-88D measurements generally were in good agreement. Depending on the choice of the WSR-88D relation and the radar tilts used, correlation coefficients for the whole dataset consisting of 12 observational events were in a range between 0.6 and 0.67. The satellite retrievals were biased on average by about 10%–20% higher (positive) relative to ground-based radar estimates; for individual events with mean rain rates exceeding about 2 mm h−1, the biases were within ±34%. Events with lighter rainfall generally exhibited higher biases and lower correlations, in part because of larger satellite-retrieval uncertainties. The normalized mean absolute differences between satellite and ground-based retrievals were on average ~60%. These differences are on the order of magnitude of joint rain-rate retrieval uncertainties from both the ground-based and satellite remote sensing approaches. Estimating potential partial beam-blockage effects on the ground-based radar measurements in the dataset considered in this study indicated that these effects (if present) were not likely to significantly change the results.

Although the majority of the compared rain-rate data corresponded to stratiform rainfall regime with a stable radar BB, rainfall peaks associated with isolated convective regions were also identified by CloudSat, and a reasonable agreement was present between WSR-88D and CloudSat rain-rate retrievals in these regions. Spaceborne retrievals for isolated regions of warm rain were not very favorable when compared with ground-based radar data, which might indicate limitations of the CloudSat reflectivity-gradient method for this type of rainfall. The intercomparison results are encouraging for the use of the CloudSat gradient method over land surfaces, but in the future more intercomparisons are needed with a larger variety of sensors to better understand accuracy and limitations of this method in precipitation of various types. A comparison of ground-based and satellite radar data reveals that WSR-88Ds observe ice regions of stratiform precipitating systems well. It would be useful in future to evaluate whether operational weather radars can provide reliable quantitative information on ice contents in such stratiform systems using simple approaches that are based on reflectivity data (e.g., Matrosov 1997).

Acknowledgments

This study was funded in part by NASA Project NNX13AQ31G and the NOAA Hydrometeorology Testbed Project.

REFERENCES

  • Battaglia, A., J. Haynes, T. L’Ecuyer, and C. Simmer, 2008: Identifying multiple-scattering-affected profiles in CloudSat observations over the oceans. J. Geophys. Res., 113, D00A17, doi:10.1029/2008JD009960.

    • Search Google Scholar
    • Export Citation
  • Battaglia, A., S. Tanelli, S. Kobayashi, D. Zrnic, R. J. Hogan, and C. Simmer, 2010: Multiple scattering in radar systems: A review. J. Quant. Spectrosc. Radiat. Transfer, 111, 917947, doi:10.1016/j.jqsrt.2009.11.024.

    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar. Cambridge University Press, 636 pp.

  • Doviak, R. J., and D. S. Zrnic, 1993: Doppler Radar and Weather Observations. Academic Press, 562 pp.

  • Gourley, J. J., B. Kaney, and R. A. Maddox, 2003: Evaluating the calibrations of radars: A software approach. 31st Int. Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., P3C.1. [Available online at https://ams.confex.com/ams/pdfpapers/64171.pdf.]

  • Haynes, J. M., T. S. L’Ecuyer, G. L. Stephens, S. D. Miller, C. Mitrescu, N. B. Wood, and S. Tanelli, 2009: Rainfall retrieval over the ocean with spaceborne W-band radar. J. Geophys. Res., 114, D00A22, doi:10.1029/2008JD009973.

    • Search Google Scholar
    • Export Citation
  • Hudak, D., P. Rodriguez, and N. Donaldson, 2008: Validation of the CloudSat precipitation occurrence algorithm using the Canadian C band radar network. J. Geophys. Res., 113, D00A07, doi:10.1029/2008JD009992.

    • Search Google Scholar
    • Export Citation
  • Krajewski, W. F., G. Villarni, and J. A. Smith, 2010: Radar-rainfall uncertainties: Where are we after thirty years of effort? Bull. Amer. Meteor. Soc., 91, 8794, doi:10.1175/2009BAMS2747.1.

    • Search Google Scholar
    • Export Citation
  • Lebsock, M. D., and T. S. L’Ecuyer, 2011: The retrieval of warm rain from CloudSat. J. Geophys. Res., 116, D20209, doi:10.1029/2011JD016076.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 1997: Variability of microphysical parameters in high-altitude ice clouds: Results of the remote sensing method. J. Appl. Meteor., 36, 633648, doi:10.1175/1520-0450-36.6.633.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2007: Potential for attenuation-based estimations of rainfall rate from CloudSat. Geophys. Res. Lett., 34, L05817, doi:10.1029/2006GL029161.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2008: Assessment of radar signal attenuation caused by the melting hydrometeor layer. IEEE Trans. Geosci. Remote Sens., 46, 10391047, doi:10.1109/TGRS.2008.915757.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2011: CloudSat measurements of landfalling Hurricanes Gustav and Ike (2008). J. Geophys. Res., 116, D01203, doi:10.1029/2010JD014506.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2013: Characteristics of landfalling atmospheric rivers inferred from satellite observations over the eastern North Pacific Ocean. Mon. Wea. Rev., 141, 37573768, doi:10.1175/MWR-D-12-00324.1.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., A. Battaglia, and P. Rodriguez, 2008: Effects of multiple scattering on attenuation-based retrievals of stratiform rainfall from CloudSat. J. Atmos. Oceanic Technol., 25, 21992208, doi:10.1175/2008JTECHA1095.1.

    • Search Google Scholar
    • Export Citation
  • Mitrescu, C., T. L’Ecuyer, J. Haynes, S. Miller, and J. Turk, 2010: CloudSat precipitation profiling algorithm—Model description. J. Appl. Meteor. Climatol., 49, 9911003, doi:10.1175/2009JAMC2181.1.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., S. Matrosov, and J. Campbell, 2007: CloudSat spaceborne 94 GHz radar bright bands in the melting layer: An attenuation-driven upside-down lidar analog. Geophys Res. Lett., 34, L16818, doi:10.1029/2007GL030291.

    • Search Google Scholar
    • Export Citation
  • Stephens, G. L., and Coauthors, 2008: CloudSat mission: Performance and early science after the first year of operation. J. Geophys. Res., 113, D00A18, doi:10.1029/2008JD009982.

    • Search Google Scholar
    • Export Citation
  • Tanelli, S., S. L. Durden, E. Im, K. S. Pak, D. G. Reinke, P. Partain, J. M. Haynes, and R. Marchand, 2008: CloudSat’s cloud profiling radar after two years in orbit: Performance, calibration, and processing. IEEE Trans. Geosci. Remote Sens., 46, 35603573, doi:10.1109/TGRS.2008.2002030.

    • Search Google Scholar
    • Export Citation
  • Zhang, J., and Coauthors, 2011: National Mosaic and Multi-Sensor QPE (NMQ) System: Description, results, and future plans. Bull. Amer. Meteor. Soc., 92, 13211338, doi:10.1175/2011BAMS-D-11-00047.1.

    • Search Google Scholar
    • Export Citation
Save
  • Battaglia, A., J. Haynes, T. L’Ecuyer, and C. Simmer, 2008: Identifying multiple-scattering-affected profiles in CloudSat observations over the oceans. J. Geophys. Res., 113, D00A17, doi:10.1029/2008JD009960.

    • Search Google Scholar
    • Export Citation
  • Battaglia, A., S. Tanelli, S. Kobayashi, D. Zrnic, R. J. Hogan, and C. Simmer, 2010: Multiple scattering in radar systems: A review. J. Quant. Spectrosc. Radiat. Transfer, 111, 917947, doi:10.1016/j.jqsrt.2009.11.024.

    • Search Google Scholar
    • Export Citation
  • Bringi, V. N., and V. Chandrasekar, 2001: Polarimetric Doppler Weather Radar. Cambridge University Press, 636 pp.

  • Doviak, R. J., and D. S. Zrnic, 1993: Doppler Radar and Weather Observations. Academic Press, 562 pp.

  • Gourley, J. J., B. Kaney, and R. A. Maddox, 2003: Evaluating the calibrations of radars: A software approach. 31st Int. Conf. on Radar Meteorology, Seattle, WA, Amer. Meteor. Soc., P3C.1. [Available online at https://ams.confex.com/ams/pdfpapers/64171.pdf.]

  • Haynes, J. M., T. S. L’Ecuyer, G. L. Stephens, S. D. Miller, C. Mitrescu, N. B. Wood, and S. Tanelli, 2009: Rainfall retrieval over the ocean with spaceborne W-band radar. J. Geophys. Res., 114, D00A22, doi:10.1029/2008JD009973.

    • Search Google Scholar
    • Export Citation
  • Hudak, D., P. Rodriguez, and N. Donaldson, 2008: Validation of the CloudSat precipitation occurrence algorithm using the Canadian C band radar network. J. Geophys. Res., 113, D00A07, doi:10.1029/2008JD009992.

    • Search Google Scholar
    • Export Citation
  • Krajewski, W. F., G. Villarni, and J. A. Smith, 2010: Radar-rainfall uncertainties: Where are we after thirty years of effort? Bull. Amer. Meteor. Soc., 91, 8794, doi:10.1175/2009BAMS2747.1.

    • Search Google Scholar
    • Export Citation
  • Lebsock, M. D., and T. S. L’Ecuyer, 2011: The retrieval of warm rain from CloudSat. J. Geophys. Res., 116, D20209, doi:10.1029/2011JD016076.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 1997: Variability of microphysical parameters in high-altitude ice clouds: Results of the remote sensing method. J. Appl. Meteor., 36, 633648, doi:10.1175/1520-0450-36.6.633.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2007: Potential for attenuation-based estimations of rainfall rate from CloudSat. Geophys. Res. Lett., 34, L05817, doi:10.1029/2006GL029161.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2008: Assessment of radar signal attenuation caused by the melting hydrometeor layer. IEEE Trans. Geosci. Remote Sens., 46, 10391047, doi:10.1109/TGRS.2008.915757.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2011: CloudSat measurements of landfalling Hurricanes Gustav and Ike (2008). J. Geophys. Res., 116, D01203, doi:10.1029/2010JD014506.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., 2013: Characteristics of landfalling atmospheric rivers inferred from satellite observations over the eastern North Pacific Ocean. Mon. Wea. Rev., 141, 37573768, doi:10.1175/MWR-D-12-00324.1.

    • Search Google Scholar
    • Export Citation
  • Matrosov, S. Y., A. Battaglia, and P. Rodriguez, 2008: Effects of multiple scattering on attenuation-based retrievals of stratiform rainfall from CloudSat. J. Atmos. Oceanic Technol., 25, 21992208, doi:10.1175/2008JTECHA1095.1.

    • Search Google Scholar
    • Export Citation
  • Mitrescu, C., T. L’Ecuyer, J. Haynes, S. Miller, and J. Turk, 2010: CloudSat precipitation profiling algorithm—Model description. J. Appl. Meteor. Climatol., 49, 9911003, doi:10.1175/2009JAMC2181.1.

    • Search Google Scholar
    • Export Citation
  • Sassen, K., S. Matrosov, and J. Campbell, 2007: CloudSat spaceborne 94 GHz radar bright bands in the melting layer: An attenuation-driven upside-down lidar analog. Geophys Res. Lett., 34, L16818, doi:10.1029/2007GL030291.

    • Search Google Scholar
    • Export Citation
  • Stephens, G. L., and Coauthors, 2008: CloudSat mission: Performance and early science after the first year of operation. J. Geophys. Res., 113, D00A18, doi:10.1029/2008JD009982.

    • Search Google Scholar
    • Export Citation
  • Tanelli, S., S. L. Durden, E. Im, K. S. Pak, D. G. Reinke, P. Partain, J. M. Haynes, and R. Marchand, 2008: CloudSat’s cloud profiling radar after two years in orbit: Performance, calibration, and processing. IEEE Trans. Geosci. Remote Sens., 46, 35603573, doi:10.1109/TGRS.2008.2002030.

    • Search Google Scholar
    • Export Citation
  • Zhang, J., and Coauthors, 2011: National Mosaic and Multi-Sensor QPE (NMQ) System: Description, results, and future plans. Bull. Amer. Meteor. Soc., 92, 13211338, doi:10.1175/2011BAMS-D-11-00047.1.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    A cross section of CloudSat CPR reflectivity during a precipitating event observed on 4 Oct 2009 (~1915 UTC) in the vicinity of the KGWX WSR-88D site.

  • Fig. 2.

    CloudSat ground track (shown by a white line) for 1914 UTC 4 Oct 2009 overlaid on the lowest elevation tilt reflectivities from the KWGX radar, the location of which is shown by the black dot.

  • Fig. 3.

    Heights of the WSR-88D beam centers above ground level for elevation angles employed during VCP-11 scanning.

  • Fig. 4.

    A cross section of KGWX radar reflectivities matching the CloudSat overpass on 4 Oct 2009 that is shown in Fig. 1. The lower and upper lines bounding the areas of no data correspond to the lower/upper edges of the first/last tilts (i.e., 0.0° and 20.0°) of the WSR-88D beam.

  • Fig. 5.

    Rain rate along the satellite-orbit ground track retrieved from CPR and KGWX WSR-88D measurements during the CloudSat overpass at 1914 UTC 4 Oct 2009.

  • Fig. 6.

    The scatterplots between approximately collocated rain-rate retrievals from the CloudSat CPR and KGWX/KSHV radar measurements for the (a) Ze = 200R1.6 and (b) Ze = 300R1.4 relations. Combined data from 12 observational events are shown. Black, red, green, blue, cyan, and purple square/plus symbols correspond to the first, second, third, fourth, fifth, and sixth KGWX/KSHV events as marked in Table 1.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 211 79 3
PDF Downloads 137 42 2