1. Introduction

The term “bright band” used in radar meteorology refers to a layer of locally high reflectivity observations associated with the melting of aggregated snow. The phenomenon has been recognized since the very early ages of radar meteorology (e.g., Ryde 1946; Austin and Bemis 1950; Wexler and Atlas 1956; Lhermitte and Atlas 1963). The locally high reflectivity causes significant overestimation in radar precipitation estimates if an appropriate correction is not applied (e.g., Koistinen 1991; Joss and Lee 1995; Andrieu and Creutin 1995; Kitchen et al. 1994; Smyth and Illingworth 1998; Westrick et al. 1999; Vignal et al. 1999, 2000; Seo et al. 2000; Vignal and Krajewski 2001; Germann and Joss 2002; Bellon et al. 2005; and references therein). Thus it is important to identify those areas of radar observations that are affected by the brightband layer. Knowledge of the brightband layer can also provide information about microphysical processes in the precipitation (e.g., Takeda and Fujiyoshi 1978; Stewart et al. 1984; Willis and Heymsfield 1989) and can lead to more accurate rainfall estimation by using appropriate Z–R relationships (e.g., Smith 1986; Huggel et al. 1996). Further, the height of the brightband layer is an indication of 0°C isotherm and can be useful in identifying areas of potential icing for aviation and in the data assimilation for numerical weather prediction (NWP).

Various techniques have been developed for automated identification of brightband layer from radar observations. Fabry and Zawadzki (1995) studied brightband structure using a vertically pointing radar with very high temporal (2 s) and spatial (15 m) resolution. They analyzed reflectivity profiles from five different precipitation regimes in the area of Montreal, Canada, and found that the melting of ice particles is not the only mechanism for bright band. The shape, density, and fall speed of the ice particles also play important roles for the existence of a bright band. Sánchez-Diezma et al. (2000) examined impacts of radar volume scan sampling strategies on the observed brightband peak intensity and depth using simulated data, and, based on the simulation results, they developed a brightband identification (BBID) algorithm. Gourley and Calvert (2003) developed an automated BBID scheme for the base-level reflectivity data from Weather Surveillance Radar–1988 Doppler (WSR-88D) radars and compared the brightband top and bottom heights obtained from the BBID scheme with model 0°C temperature height and with observations from a vertically pointing radar. The BBID scheme by Gourley and Calvert (2003) applies to each bin column of the base-level reflectivity data and then averages the bin-by-bin brightband information (e.g., bottom and top heights) over space and time. The current paper presents an alternative BBID scheme for volumetric radar observations that is based on the mean vertical profiles of reflectivity (VPRs). By using the VPRs instead of the base data, the scheme is largely simplified and more computationally efficient. The new BBID scheme is similar to that proposed in Sánchez-Diezma et al. (2000) but is adapted to WSR-88D scan strategies and was initially evaluated using the radars in the conterminous United States (CONUS).

Section 2 provides details of the new automated BBID algorithm. Evaluation results from several radars and two precipitation events are shown in section 3, statistics of the BBID results from 134 radars in CONUS for a 7-month period in 2007 are presented in section 4, and a summary is provided in section 5.

2. Algorithm description

The automated BBID algorithm presented here comprises the following three steps:

1. convective and stratiform precipitation segregation;

2. computation of volume scan mean VPRs for different precipitation groups; and

3. brightband identification from stratiform VPRs.

b. Volume scan mean VPR computation

Two mean VPRs are computed for each radar volume scan, one for convective and another for stratiform precipitation. A volume scan of reflectivity data is quality controlled to remove nonprecipitation targets. The quality control scheme (Lakshmanan et al. 2007) uses a neural network approach that is based on horizontal and vertical reflectivity structure, in addition to pre- and postprocessing. The pre- and postprocessing utilize spatial and temporal reflectivity filters and heuristic rules based on radar scan mode and environmental data to remove specific nonprecipitation echoes such as speckles, sun strobes, biological returns, and anomalous propagations due to nocturnal radiation cooling near the surface. Details of the quality control scheme can be found in Lakshmanan et al. (2007).

After the quality control (QC), reflectivity observations from all tilts in an annular region between two predefined ranges (r₁ and r₂; see Fig. 2) are divided into two groups based on precipitation types (see section 2a). The annular region is chosen to be sufficiently close to the radar so that high vertical resolution of reflectivity can be obtained in the final VPRs. And the region needs to be away from the radar to avoid the cone of silence in the immediate vicinity of the radar. In addition, the number of data samples within a 20-km radius of the radar is limited because of a terrain clearance rule in the QC scheme, whereby any radar bin whose bottom is within 50 m of the ground is removed. Empirical values of 20 and 80 km are currently used for r₁ and r₂, which will give 194 400 data points, assuming a volume scan of 9 tilts and 360 radials. The gate spacing of the reflectivity data used in this study is 1 km.

Reflectivity data for each precipitation type are further grouped into evenly spaced vertical layers according to the central height of the reflectivity bins. The height of each vertical layer h(k) is defined as the following:

View Expanded

Here k is the layer index, Δh (default = 200 m) is the thickness of each layer, N (default = 98) is the number of vertical layers, and h₀ (default = 0.5 km above radar level) is the height of the lowest level. Within each layer the mean and standard deviation of all the reflectivity observations (in dBZ) are computed as follows:

View Expanded

View Expanded

Here Z(k) (in dBZ) is the mean log-scale reflectivity in the kth layer, σ_Z(k) is the standard deviation, M is the total number of reflectivity observations in the kth layer, i is the index of reflectivity observations, and Z(i) is an observed reflectivity value in log scale (dBZ) within the kth layer. Note that all the reflectivities referred to in this paper are in dBZ, all the empirical parameters have been adapted to reflectivity data that were averaged in log units, and the use of vertical profiles of reflectivity for quantitative purposes (e.g., for extrapolating reflectivities below the lowest radar beam) is not warranted.

Two rules are applied to assure a representative and robust VPR: (i) only reflectivities higher than a threshold (Z₀) are included in VPR, and (ii) a minimum number (M₀) of reflectivity observations with Z(i) ≥ Z₀ are required within each height layer to get a valid mean reflectivity for the VPR. Both Z₀ and M₀ are adaptable parameters (defaults are 10 dBZ and 10, respectively). If at any given layer a valid Z(k) cannot be obtained, then a linear interpolation using valid Z values from layers above and below is applied to get an alternative Z(k). The interpolation is limited within a depth of Δh_intp = ±1 km (adaptable). A three-point running mean is applied to the final VPR to reduce random fluctuations.

Figure 3 shows example VPRs from the same squall line event as shown in Fig. 1. The stratiform VPRs (Figs. 3a,b) are different from the convective VPRs (Figs. 3c,d) because of the different microphysical processes in the two precipitation regimes. The brightband feature is apparent in the stratiform VPRs (Figs. 3a,b), while there is no brightband feature in the convective VPRs (Figs. 3c,d). The automated BBID algorithm is developed based on the mean stratiform VPRs. The details of the algorithm are presented in the next section.

c. Brightband identification

Brightband identification is based on the volume mean stratiform VPRs and a temperature profile at the radar site. The VPRs are not averaged in time over multiple volume scans. The single-volume-scan VPRs, instead of temporally averaged VPRs, are used for the brightband identification because the former is more closely related to the instantaneous rain-rate field than is the latter. The brightband layer identified in the current work will be used to delineate regions of potential brightband contaminations in the rain-rate field. Figure 4 shows a simplified conceptual model of bright bands that was proposed in previous studies such as Fabry and Zawadzki (1995) and Sánchez-Diezma et al. (2000). Above the freezing level (i.e., 0°C height), the cloud contains mostly snow. When snowflakes fall and enter an environment warmer than 0°C they start to melt (h_a, Fig. 4). With the wet surface, the snowflakes become sticky and easily aggregate into larger pieces. These large snowflakes with wet surfaces are highly reflective objects to centimeter- (and longer) wavelength radars and result in a level of maximum reflectivity (h_m, Fig. 4). When the ice completely melts, the volume of each hydrometeor becomes smaller, and large raindrops break and shed. Therefore, reflectivity decreases quickly below the maximum level until the raindrop size distribution reaches a balance (h_b, Fig. 4). The automated BBID builds upon this conceptual model and encompasses three objectives:

1. Find the local maximum in the VPR that is near the background freezing level.

2. Check for the existence of a bright band.

3. If a bright band exists, find the bottom and top heights of the brightband layer.

The search for the local maximum in the VPR starts from 500 m above the model analysis 0°C height at the radar site and continues downward. A 500-m cushion is used to account for uncertainties in the model 0°C height due to infrequent and sparse upper-air sounding observations. Once the local maximum is identified, the algorithm finds the height above (below) the maximum level where reflectivity monotonically decreases by a given percentage (default = 10%) of the maximum reflectivity (in dBZ). A bright band exists if the following criteria are met:

View Expanded

View Expanded

View Expanded

Here h_m is the height of the maximum reflectivity; h_a (h_b) is the height above (below) the maximum reflectivity level where the reflectivity decreases by 10% of the maximum; and D₀ and D₁ are adaptable parameters that are constrained by the depth and symmetry of the brightband layer. The parameter D₁ is equivalent to the parameter ΔH in Fabry and Zawadzki (1995). Based on their simulation results and our studies of several thousands of VPRs from the WSR-88Ds, an empirical value of 1.0 km is used for D₁, and 1.5 km is used for D₀. Note that these values are dependent on radar scan strategies and the vertical resolution of reflectivity observations. The D₀ and D₁ parameters will require tuning if the algorithm is used on radars other than the WSR-88Ds.

If a bright band exists, its top and bottom heights are defined according to the following:

View Expanded

View Expanded

Here D_t and D_b serve as caps on the brightband (BB) top and bottom, and the default values for D_t and D_b are 500 and 700 m, respectively. Different values are used for D_t and D_b to account for the different slopes of VPRs above and below the brightband peak level (Fig. 4). The increasing rate of reflectivity above the brightband peak is usually larger than the decreasing rate of reflectivity below (Fabry and Zawadzki 1995).

Previous studies with very high-resolution vertically pointing radar data showed that the brightband layer thickness is usually less than a few hundred meters (Fabry and Zawadzki 1995). However, simulation results from Sánchez-Diezma et al. (2000) using a 10-cm radar with 1° beamwidth showed that the impact of the brightband layer on radar observations can be as thick as 2 km. This brightband expanding effect due to the radar beam spreading with range is illustrated in Fig. 2. When the radar is operating in volume scan pattern (VCP) 21, a brightband layer thickness of 500 m can impact radar bins over a depth of 1 to 1.5 km within the range of 80 km (see Fig. 2). Because of the beam-spreading effect, radar-observed BB tops could start a few hundred meters above the 0°C height (Figs. 3a,b) instead of right at (or slightly below) the 0°C height as the conceptual model (Fig. 4) indicated. Beyond the range of 100 km, the expansion of the brightband influence is even larger (Fig. 2). Meanwhile the peak intensity will decrease with range because of the smoothing effect of the radar power density function (Sánchez-Diezma 2000).

Since the current BBID scheme uses reflectivities near the radar to identify the brightband layer, the resultant depth of the bright band will be smaller than the depth influenced by the bright band at the far range (Fig. 2). In addition, if a brightband layer is not distributed evenly in space and only exists outside the annular region where the VPRs are computed, the current BBID scheme would not identify it. If the freezing level is near the radar horizon, and the radar only observes the brightband top and perhaps the peak, but not the bottom, then the current scheme would not be able to identify the bright band. Therefore, the brightband existence and depth information should be used with caution when applied at far ranges. Note that the values of D_t and D_b are also adaptive parameters depending on the radar data resolution and need to be retuned for different radars.

3. Case studies

Figures 5 and 6 show BBID results from a widespread wintertime stratiform precipitation event that occurred on 24 January 2007 in the south Texas area (Fig. 5). The brightband layer was detected at three radars (KHGX, KEWX, and KLCH) in the region over a 24-h period (Fig. 6). The brightband top heights from radar data are available every 5 min, while the model analysis is only available every hour, and the upper-air data are even less frequent (every 12 h). The brightband top height identified from the KHGX site showed similar trend to the brightband top from the KEWX site, and both time series had physically realistic smooth transitions over time. However, the model’s 0°C height near the KHGX site showed some large temporal variations at times around 0300 UTC and 1100 UTC, and after 1600 UTC. These variations appeared to be less physically realistic than those in the brightband top time series. Therefore the brightband top height identified from radar data could potentially be used to improve the model temperature and cloud analysis.

Figure 7 shows time series of brightband top, bottom, and peak-level heights from the same event as in Fig. 6. The average difference between brightband top and bottom heights is around 1 km, indicating that the constraint for brightband depth (i.e., D_t + D_b = 1.2 km) is sufficient to encompass the BB layer with the 10% drop-off of the peak reflectivity. The top and bottom heights delineate areas where radar reflectivities are potentially inflated. Radar precipitation estimation is usually based on the reflectivity observations in the lowest tilt. If the lowest radar tilt intersects this layer, then an adjustment (reduction) is necessary to mitigate the potential overestimation when the data are used for precipitation estimation.

Figure 8 shows a squall line that passed across Oklahoma and Kansas on 24 May 2007 (Fig. 8a) and the associated ratio bias map of the hourly radar precipitation estimation against rain gauge observations (Fig. 8b). Overestimations of 50%–100% occurred within the trailing stratiform region as outlined by the red polygon in Figs. 9a and 9b. The BBID results from four different radars (Fig. 9) showed that a brightband layer was detected in the stratiform region. The brightband bottom height ranges from around 3.1 km (Fig. 9b) above mean sea level in the northern part of the squall line to 3.5 km (Figs. 9c,d) in the southern part. The center of the polygon region is on the average about 125 km away from the surrounding radars. At this distance, the top of the lowest tilt is ∼3 km above radar level (Fig. 2), or ∼3.4 km above mean sea level (the average terrain height is about 400 m in the area). Therefore, the lowest tilts from the surrounding radars were affected by the brightband layer, and the subsequent radar precipitation estimates were inflated due to the bright band’s high reflectivity values. This case demonstrates the importance of the brightband information and the value of the BBID scheme in diagnosing areas of low radar quantitative precipitation estimation (QPE) quality.

All the brightband depths derived from the four radars in the squall line case are plotted in Fig. 10. Among 118 volume scans in which a bright band was identified, only three BB depths reached the upper limit of 1.2 km (Fig. 10a). Most of the BB depths are between 800 and 900 m (Fig. 10b), and the total average is 925 m. As discussed in section 2c, these BB depths are based on radar observations in the close ranges where the VPRs are computed. At far ranges, the depth of radar observations being potentially affected by the bright band will be larger.

4. Long-term statistics of BBID results

The BBID scheme has been tested on 134 radars (Table 1) in the CONUS domain for a period of 7 months from March to September 2007. Figure 11 shows the monthly averaged brightband top and bottom heights, height of the peak reflectivity in VPR, and the RUC 0°C level. All the heights had a minimum in April, which is likely associated with an unusual cold wave that occurred across much of the central plains, Midwest, and into the southeast United States (more information available from the National Climatic Data Center at http://www.ncdc.noaa.gov/oa/climate/research/2007/apr/apr-cold-event.php) in early April 2007. The heights increased after April and reached a maximum in August. The peak reflectivity was below the RUC 0°C height for all months with an average distance of ∼100 m, and the root-mean-square difference between the two heights was 386 m. The brightband top appeared to be always higher than the RUC 0°C level because of the beam-spreading effect. The brightband top and bottom heights were slightly nonsymmetric with respect to the peak reflectivity level, and the BB top was a little closer to the peak than was the bottom (the average height difference between BB top and peak is 376 m, and the average difference between BB peak and bottom is 443 m).

The BBID results were further stratified based on latitude zones, as listed in Table 1. Numbers of volume scans that had brightband identifications for each latitude zone and for each month are listed in Table 2. Figure 12 shows average BB top heights derived from radars located in different latitude zones. The variations of the BB top heights in each latitude zone were similar to those in the CONUS domain average (Fig. 12), except that the BB top height for the highest latitude zone (45°–50°N) reached maximum in July instead of August. These results are consistent with the surface temperature records (assuming that the higher the surface temperature, the higher the freezing level) in the National Climatic Data Center’s U.S. National Overview Reports (available online at http://www.ncdc.noaa.gov/oa/climate/research/2007/aug/national.html), which show that the average August temperatures set new high records in eight central and southeastern states. In July, the average temperatures set new records in three of the northern mountain states (Idaho, Montana, and Wyoming), while much of the nation, from the Deep South to the Northeast, remained cool (http://www.ncdc.noaa.gov/oa/climate/research/2007/jul/national.html). The BB top height for the lowest latitude zone (20°–25°N) had small changes from month to month, indicating relatively small seasonal variations of freezing level in this zone. The largest change of BB top height occurred in the midlatitude zone (35°–40°N), where the monthly mean BB top varied from 3 km (MSL) in April to 5 km (MSL) in August.

The brightband layer depths derived from the BBID scheme for the whole CONUS domain and for different latitude zones are shown in Fig. 13. On the CONUS average, the monthly mean BB layer depth was ∼820 m, and it changed little from month to month (Fig. 13a). The standard deviation of the BB depth was ∼200 m and was slightly larger in the summer (June, July, and August) than in other months. The small variability of the monthly mean BB depth may be due to two factors: 1) the BB information is derived from a prespecified annular region, and it is constrained by the radar sampling geometry in addition to the 10% reflectivity drop-off criteria, all of which are independent of time; and 2) vertical distributions of the microphysical processes (i.e., melting of aggregated snow) that define the BB depth do not have significant time dependency, so the monthly average BB depth remains similar. The small variations of the mean BB layer depth are also shown for different latitude zones (Fig. 13b), except for the lowest latitude zone (25°–30°N). The relatively large variations of BB layer thickness for that zone were probably due to the fact that the majority of precipitation systems in the region are convective or warm rain processes, which do not have persistent brightband features.

5. Summary

A new automated brightband identification technique has been developed and evaluated across seasons and geographical regions. The new technique is based on vertical profiles of reflectivity from WSR-88D radar data and a background atmospheric temperature profile. The VPRs are derived from volumetric data from WSR-88Ds and are segregated for convective and stratiform precipitation, and the brightband layer information is derived from stratiform VPRs only. The new BBID scheme has been evaluated using about 7 months of WSR-88D data in the CONUS domain. The results indicate that the BBID algorithm can provide physically realistic BB layer information that is consistent with the conceptual model and matches the background freezing level field from a model analysis. The brightband layer can provide information for identifying potential icing hazards for aviation and for NWP model data assimilation. The bright band is also closely related to the microphysical processes in precipitation and is a very important factor for creating accurate radar QPEs.

The new automated BBID algorithm is stand-alone and computationally efficient (average CPU for VPR computation and BBID is ∼0.5 s for processing one volume scan data on a Dell PC with a 700-Mhz processor and 3.6 GB RAM). Forthcoming studies using the BBID information include quantifying and mitigating overestimation of radar QPEs due to the bright band and rain/snow delineation in complex terrain.