Attachment D - Towards the Next Generation Operational Meteorological Radar .pdf

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1357

Towards the Next Generation Operational Meteorological Radar Mark Weber, Kurt Hondl, Nusrat Yussouf, Youngsun Jung, Derek Stratman, Bryan Putnam, Xuguang Wang, Terry Schuur, Charles Kuster, Yixin Wen, Juanzhen Sun, Jeff Keeler, Zhuming Ying, John Cho, James Kurdzo, Sebastian Torres, Chris Curtis, David Schvartzman, Jami Boettcher, Feng Nai, Henry Thomas, Dusan Zrnić, Igor Ivić, Djordje Mirković, Caleb Fulton, Jorge Salazar, Guifu Zhang, Robert Palmer, Mark Yeary, Kevin Cooley, Michael Istok, and Mark Vincent

ABSTRACT: This article summarizes research and risk reduction that will inform acquisition decisions regarding NOAA’s future national operational weather radar network. A key alternative being evaluated is polarimetric phased-array radar (PAR). Research indicates PAR can plausibly achieve fast, adaptive volumetric scanning, with associated benefits for severe-weather warning performance. We assess these benefits using storm observations and analyses, observing system simulation experiments, and real radar-data assimilation studies. Changes in the number and/ or locations of radars in the future network could improve coverage at low altitude. Analysis of benefits that might be so realized indicates the possibility for additional improvement in severe-weather and flash-flood warning performance, with associated reduction in casualties. Simulations are used to evaluate techniques for rapid volumetric scanning and assess data quality character-istics of PAR. Finally, we describe progress in developing methods to compensate for polarimetric variable estimate biases introduced by electronic beam-steering. A research-to-operations (R2O) strategy for the PAR alternative for the WSR-88D replacement network is presented.

KEYWORDS: Radars/Radar observations; Weather radar signal processing; Mesoscale forecasting;

Nowcasting; Numerical weather prediction/forecasting; Operational forecasting

PUBLISHER'S NOTE: This article was modified on 12 August 2021 to correct the affiliation for coauthor Derek Stratman.

https://doi.org/10.1175/BAMS-D-20-0067.1 Corresponding author: Mark Weber, mark.weber@ll.mit.edu In final form 21 March 2021 ©2021 American Meteorological Society For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy.

Article

Brought to you by NOAA Central Library | Unauthenticated | Downloaded 01/10/22 09:07 PM UTC

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1358

AFFILIATIONS: Weber,* Yussouf , Stratman, Schuur, Kuster, Wen, Torres, Curtis, Schvartzman, Boettcher, Nai, Ivić, and Mirković—Cooperative Institute for Mesoscale Meteorological Studies, University of Okla-homa, and NOAA/Office of Oceanic and Atmospheric Research/National Severe Storms Laboratory, Nor-man, Oklahoma; Hondl and Zrnić—NOAA/Office of Oceanic and Atmospheric Research/National Severe

Storms Laboratory, Norman, Oklahoma; Jung, Putnam, and Wang—School of Meteorology, University of Oklahoma, Norman, Oklahoma; Sun, Keeler, and Ying—National Center for Atmospheric Research, Boulder, Colorado; Cho, Kurdzo, and Thomas—MIT Lincoln Laboratory, Lexington, Massachusetts;

Fulton, Salazar, and Yeary—Advanced Radar Research Center, University of Oklahoma, Norman, Oklahoma; Zhang and Palmer—School of Meteorology, and Advanced Radar Research Center, University of Oklahoma, Norman, Oklahoma; Cooley and Istok—NOAA/National Weather Service, Silver Spring, Maryland; Vincent—NOAA/Office of Oceanic and Atmospheric Research, Silver Spring, Maryland

* Current affiliation: MIT Lincoln Laboratory, Lexington, Massachusetts

T he National Oceanic and Atmospheric Administration’s (NOAA) current operational meteorological radar is the Weather Surveillance Radar-1988 Doppler (WSR-88D). The WSR-88D was developed and is maintained by the Next Generation Weather Radar

(NEXRAD) program tri-agencies—the Departments of Commerce, Defense, and Transportation (Crum and Alberty 1993). The first of these radars became operational in 1993 with full deployment taking place over the following 5 years. Three additional radars were added after 2000, and coverage will be improved at select locations by lowering the elevation angle of the lowest tilts. Figure 1 shows the network’s coverage for heights of 3000 ft above ground level (AGL) and greater (1 ft ≈ 0.305 m).

The WSR-88D provides high sensitivity, angular resolution, and transmit–receive stability, all of which support high-quality observations of weather variables. In 2013, the system was upgraded to a dual-polarization configuration. This enables discrimination between different hydrometeor types (Straka et al. 2000) and identification of nonmeteorological scatterers such as insects, birds, and chaff (Krause 2016), while also improving the accuracy of quantitative

Fig. 1. WSR-88D CONUS coverage.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1359

precipitation estimation (QPE) (Cocks et al. 2019). NOAA’s Observing System Integrated Analysis (NOAA 2016) indicates that the WSR-88D has the highest operational impact of NOAA’s weather observing systems for three critical mission areas—severe thunderstorms and tornadoes, hydrology, and fire weather.

The National Weather Service (NWS) initiated a WSR-88D Service Life Extension Program (SLEP) in 2013 to extend the system’s ability to operate reliably until approximately 2040 (Cook et al. 2014). The agency has recently begun evaluating alternative strategies for further sustaining or replacing the WSR-88D network after this date. High-level Radar Functional Requirements (RFRs) (NWS 2015) provide a starting point for determining the architecture for this next generation operational weather radar. Broadly, the RFRs are based on the observing capabilities of the WSR-88D, although NWS acknowledges that alternative technologies such as polarimetric phased-array radar (PAR) may enhance tornado and severe-weather warn-ing performance by means of rapid volumetric scanning (~once per minute). The RFR also discusses the operational value of expanded radar observations below 500 m AGL, noting NWS interest in data from Federal Aviation Administration (FAA) and private-sector radars to enhance coverage near the surface.

In 2017, meteorological PAR research at NOAA’s National Severe Storms Laboratory (NSSL) (e.g., Zrnić et al. 2007; Stailey and Hondl 2016) was augmented with additional funding to address requirements for a planned multi-agency Spectrum Efficient National Surveillance Radar (SENSR) acquisition (FAA 2019). NOAA ultimately withdrew from this program, based on a number of considerations including that the time available prior to SENSR’s planned contract award was insufficient for the research needed to determine whether PAR could meet NOAA’s future “high-resolution weather” requirements. Nevertheless, the modeling, analysis, and risk reduction conducted as part of the SENSR program remain highly relevant to NOAA’s analysis of technology and acquisition alternatives for its future meteorological radar system (NOAA 2020), which would be deployed beginning ca. 2035. In addition to the PAR alternative, NOAA is evaluating the feasibility of sustaining the WSR-88D beyond 2040, or replacing it using a similar mechanically scanned reflector-antenna radar.

Phased-array radar capabilities and challenges Potential mission benefits for the future radar system, discussed in the “Warning and forecast benefits” section, depend to a significant extent on the ability to perform rapid, flexible scan-ning. PAR provides two basic methods for achieving this. First, its beam(s) may be electroni-cally repositioned—essentially instantaneously—allowing for a decrease in average dwell time (e.g., Yu et al. 2007), or adaptive reduction in the number of beam positions to be serviced (Torres et al. 2016). Second, PAR architectures that digitally sample multiple subarrays (or individual transmit–receive elements) allow for simultaneously active receive beams to be formed. When coordinated pulse-transmission strategies are used (e.g., Zrnić et al. 2015;

Melnikov et al. 2015; Weber et al. 2017), these allow scan time to be reduced by a factor equal to the number of receive beams.

Figure 2 depicts a concept for how such techniques could enhance severe-weather warnings and forecasts.1 PAR-enabled volume-coverage patterns (VCPs) provide surface-to-storm-top ob-servations approximately once per minute, significantly improv-ing NWS forecasters’ cognitive models of severe-storm structure and thereby warning performance (see Heinselman et al. 2012, 2015; Bowden et al. 2015; Wilson et al. 2017). These rapid-update VCPs also improve model analysis of storm spinup (Yussouf and Stensrud 2010), which can improve performance of the Warn-on-Forecast System (WoFS) (Stensrud et al. 2009).

Intermixed with these VCPs, adaptively steered beams

1 This concept of operations does not assume a specific PAR configuration (e.g., a four-faced stationary array) or rapid-scan technique, just the capability to execute the concep-tual scanning. Practical considerations for rapid scanning are discussed in the “PAR rapid-scanning methods” section.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1360

measure “clear-air” winds in the near-storm environment by taking advantage of higher av-erage power radiated by PAR, and its capability to integrate successive processing intervals.

Appendix A describes how an operational PAR could plausibly measure winds using Bragg-scattered echoes to an altitude of 7 km AGL. Assimilation of these wind observations could further improve WoFS performance through better analysis of updraft forcing, size, strength, and the storm track. Finally, some scan time is allocated for adaptive, very-high-update scan-ning of meteorologically significant storm volumes.

Fundamental differences exist, however, between PAR and a mechanically scanned reflector antenna radar such as the WSR-88D, and these may change the characteristics of its meteorological observations. For example, antenna beamwidth increases and gain decreases as the beam is steered electronically away from the array normal. Achieving acceptable antenna sidelobe levels, and ensuring good beam pattern matching for the two polarizations, will require careful array design, calibration, and monitoring to identify degraded or failed transmit–receive elements. Rapid-scan methods that utilize multiple, simultaneous receive beams must minimize “cross-talk” between these beams. The “PAR rapid-scanning methods” and “PAR antenna sidelobes” sections describe our work to develop appropriate requirements and to assess the data-quality impact of candidate PAR configurations.

To be operationally useful, meteorological radar dual-polarization variables must be measured with high accuracy. For example, biases of a few tenths of a decibel in estimates of differential reflectivity (ZDR)—the ratio of the received horizontal (H) and vertical (V) signal components—may increase rain-rate estimate errors by 20% or more, which is significant in comparison to other sources of error (ZrniĆ et al. 2010).

While challenging, calibration to achieve such accuracy is performed for the WSR-88D, but will be more problematic using an electronically steered array. For practical polarimetric PAR antennas, the co- and cross-polarization patterns vary by many decibels as the beam is electronically steered in azimuth and elevation to angles well away from broadside. Unequal variations of the H- and V-polarized co-polar patterns cause biases in the estimates of radar products (referred to as co-polar biases). At the same time, significant cross-polar patterns produce variable levels of contamination of H by V returns and vice versa. These also introduce biases in polarimetric-variable estimates. Finally, close proximity of H and V transmit–receive electronics in the array can result in significant coupling between the H and V signals. Due to these issues, careful calibration and bias correction will be necessary to achieve polarimetric

Fig. 2. Notional PAR scanning concept of operations. (left) Full volume coverage patterns (VCPs) similar to those of the WSR-88D will be accomplished every 60 s. Different colors show groupings of beam elevation tilts (low-elevation-angle surveillance and Doppler scans, midlevel “batch-PRT” scans, and high-elevation-angle Doppler scans). (center) Environmental wind profiles will be measured to altitudes of ~7 km AGL every 300 s. (right) Very-high-update observa-tions (<30 s) of rapidly evolving severe storm volumes will be performed periodically.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1361

data quality comparable to that from reflector antenna radars such as the WSR-88D. This is one of the major challenges for adoption of PAR for weather observations. In the “Dual-polarization calibration and bias correction” section, we describe preliminary (and ongoing) efforts to demonstrate dual-polarization calibration and bias-correction methods using a 10-cm wavelength, polarimetric PAR deployed at NSSL.

Warning and forecast benefits Rapid-volume-update WoFS experiments. In this section, we illustrate benefits of rapid-update PAR observation to WoFS. We believe that increasing tornado-warning lead times will require both of these technologies to work synergistically. Our work substantiates previous observing system simulation experiments (OSSEs) by using real PAR observations of a severe thunderstorm over an extended time period.

The National Weather Radar Testbed (NWRT) PAR located in Norman, Oklahoma (Forsyth et al. 2011), collected observations from the 31 May 2013 El Reno, Oklahoma, torna-dic supercell event. The NWRT successfully interrogated the storm from convective initiation through the entire evolution of the tornado with the exception of one 5-min period. These frequent volumetric data provided an opportunity to evaluate the impact of temporal frequency of PAR observations in an experimental version of WoFS at 1-km horizontal grid spacing.

Stratman et al. (2020) conducted experiments using various data assimilation (DA) cycling intervals to produce very short-term forecasts of this storm. Experiments were conducted using the WRF-ARW model (Skamarock et al. 2008) and the 4DEnSRF DA system (Wang et al. 2013).

The experiments assimilated PAR data every 1, 3, 5, and 15 min, as shown in Fig. 3a. A final experiment (Cyc1 + Cyc15) explored the benefit of an adaptive DA cycling interval, where 1-min cycles were used only during storm initiation and early stages of development. Ensemble forecasts were initialized from each of those experiments every 15 min from 2200 to 2300 UTC to predict the tornadic supercell storm with WRF history files every 5 min through 0000 UTC the next day.

The authors compared these experiments using subjective evaluations of 0–1-h low-level reflectivity (Z) and midlevel updraft helicity (UH), and objective metrics including ensemble fractions skill score (eFSS) (Duc et al. 2013) and object-based detection/false-alarm scoring.

For all evaluation methods, the 1-min PAR data assimilation experiments outperformed those with lower DA-cycling frequencies.

As an example, for each experiment, Fig. 3b shows average eFSS as a function of neighbor-hood width (km) for the five 1-h forecasts of 2-km MSL reflectivity > 35 dBZ. For this case the “reference” and “random” eFSSs are approximately 0.65 (dashed line) and 0.15, respectively (see Roberts and Lean 2008). While forecasts at scales with eFSS values greater than the random eFSS are considered to be skillful, the reference eFSS is considered to be the target. The experiment using 1-min PAR DA generated the highest average eFSS for neighborhood widths larger than ~16 km, with the 3- and 5-min DA cycles providing similar eFSS values for neighborhood widths smaller than 16 km. The 15-min DA cycle—currently used in the real-time WoFS—resulted in the smallest eFSS values for all neighborhood sizes, which is at least partially the result of its predicting more spurious convection. Analysis indicates that more frequent DA improved the reflectivity forecasts by developing storms more quickly in the correct locations, while removing spurious storms. It is noteworthy that the adaptive cycling experiment (not shown), where more frequent DA occurred while storms were initiating, significantly reduced WoFS computation re-quirements while maintaining the benefits of a continuous 1-min DA cycle (Stratman et al. 2020).

For the same El Reno storm, we directly compared WoFS performance when DA used 1-min PAR data from the NWRT versus 5-min VCPs from the Oklahoma City WSR-88D (KTLX). The PAR-based forecasts exhibited higher probability of strong UH within the validated midlevel rotation track.

They also captured a secondary mesocyclone that developed behind the initial mesocyclone.

More accurate UH forecasts resulted because the PAR-based WoFS analysis produced stronger

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1362

updrafts and larger midlevel temperature anomalies that sustained these updrafts. This case study, which extends findings of Supinie et al. (2017), is discussed in more detail in NSSL (2020).

Finally, experiments were conducted to assess the impact of assimilating rapid-update ZDR observations, in addition to reflectivity and radial velocity (NSSL 2020). Because the NWRT PAR in 2013 was not polarimetric, we used rapid-update data from NSSL’s experimental WSR- 88D (KOUN) as a proxy for observations from a future polarimetric PAR. (This approach is also used in the “Rapid-update polarimetric radar observations of severe weather and flash flooding” section.) Assimilation of ZDR improved the accuracy of WoFS forecasts of Z and UH for the El Reno storm for the earliest-initialized forecasts. For later initialization times, forecasts of these variables were similar with and without ZDR assimilation. Substantially improved forecasts of Z, UH, and large hail for the northern storm (near Stillwater, Oklahoma) that formed later than the El Reno storm resulted when ZDR was assimilated. The improvement resulted from a better representation of the storm’s “cold pool,” with higher temperature.

More specifically, the evaporation of small raindrops is largely responsible for colder, more intense cold pools. The assimilation of medium to high ZDR associated with medium Z in the downdraft areas within the storm helped retrieve the drop size distribution (DSD) shift toward larger drops, which led to reduced evaporation. The more intense cold pools when ZDR was not assimilated cut off the inflow of high equivalent potential temperature (θe) air in the en-vironmental boundary layer and the forecast storms were less intense and began to dissipate.

Rapid-update polarimetric radar observations of severe weather and flash flooding. To further assess the impact of polarimetric PAR observations on the warning decision process, Fig. 3. (a) Schematic of the DA experiments. Red and blue vertical lines indicate the times PAR volumetric observations are assimilated. (b) Average eFSS as a function of neighborhood width

(km) for 1-h forecasts of reflectivity > 35 dBZ at 2 km MSL. The neighborhood-based verification is used to eliminate small displacement errors in the high-resolution model. Details are shown in the legend (courtesy: Stratman et al. 2020).

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1363

KOUN was operated using rapid-update (1–2 min) storm observations that were achieved by confining azimuthal scanning to a sector of approximately 90°. These data were then used to demonstrate potential operational applications of polarimetric variables, as well as to assess how rapid-update observations might provide further operational benefits.

One dual-polarization radar signature examined was the ZDR column (e.g., Illingworth et al.

1987; Tuttle et al. 1989; Kumjian et al. 2014). ZDR columns are a good indicator of updraft loca-tion and intensity and, therefore, severe-weather potential (e.g., Ryzhkov et al. 1994; Kumjian 2013). The ZDR-column-depth algorithm (Snyder et al. 2015) was run on 42 different storms and over 1400 volume scans across 13 rapid-update data cases collected using KOUN (Kuster et al. 2019). The analy-sis indicated that both ZDR col-umns and −20°C reflectivity cores (i.e., 50+ dBZ at −20°C;

Nelson 1983; Witt et al. 1998) are good indicators of storm intensity, and in many cas-es, local maxima in these signatures occurred prior to severe-wind and hail reports.

In these cases, ZDR columns developed 3.5–9.0 min ear-lier than −20°C reflectivity cores. Furthermore, the rapid-update KOUN data provided more complete information about ZDR column evolution and longer lead time than KOUN data that were downs-ampled to emulate 5-min VCPs with equivalent eleva-tion-angle coverage. For all severe-hail reports considered (n = 21), median lead times (i.e., times between severe-weather report and radar-signature local maxima) of ZDR column size and median depth were 4.0 and 5.3 min greater for rapid-update data when compared to traditional-update data, respectively (Fig.

4a). A similar pattern was found for severe-wind reports (n = 11), where median lead times were 7.5 and 7.0 min greater for rapid-update data when compared to traditional-update data (Fig. 4b).

Fig. 4. Boxplots showing amount of time signature local maxima (i.e., peaks) occurred prior to (a) severe-hail reports and (b) severe-wind reports for rapid-update (<2-min updates) KOUN data (pink) and traditional-update (5–6-min updates) KOUN data (yellow). Blue dots indicate data used in boxplot creation. “Column size” is cross-sectional area of the ZDR column at the 0°C isotherm. “Column median” is median depth above the 0°C isotherm for all grid cells within a column, with a corresponding definition for “Column maximum.”

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1364

Another dual-polarization radar signature examined was specific-differential-phase (KDP) cores. The KDP cores evident in dual-polarization radar observations provide important ad-ditional information, because they indicate areas of melting graupel and hail and, therefore, higher potential for water loading and evaporation if these hydrometeors fall through an unsaturated layer below cloud base (e.g., Srivastava 1985, 1987; Proctor 1988). Observing processes such as melting, water loading, and evaporation is important, because all three can increase the intensity of a downdraft and a subsequent downburst. Downbursts, which are a significant hazard to aviation and public safety, develop quickly and may be difficult to detect at low altitude owing to radar line-of-sight limitations and/or ground clutter. Previous research to identify radar precursor signatures aloft focused on descending reflectivity cores and convergent radial-velocity signatures (e.g., Isaminger 1988; Roberts and Wilson 1989).

To examine KDP cores as a potential downburst-precursor signature, we examined the size, magnitude, and vertical gradient of KDP cores (KDP ≥ 1° km−1) for 24 storm cases containing 81 different wet downbursts (i.e., downbursts associated with heavy rain at the surface) across the Southeast, Great Plains, and Southwest. Of these, 4 cases and 16 downbursts had rapid-update KOUN data available. The analysis showed that KDP cores developed and intensified prior to every downburst in the dataset. While KDP cores were a reliable indicator of downburst development, using them to anticipate the intensity of an impending downburst may be more challenging. Distributions of KDP-core characteristics between strong and weak downbursts had significant overlap (not shown). However, some distributions of KDP-core characteristics, such as maximum KDP near the environmental melting layer, had greater differences and less overlap between strong and weak downbursts with rapid-update versus 5-min-update data (Fig. 5).

Rapid-update volumetric radar data may therefore be important to take full advantage of this downburst-precursor signature in an operational setting.

Finally, we examined the potential im-pact of rapid-update radar observations on flash-flood warnings. High-spatiotemporal-resolution QPE is essential for accurate flash-flood warning and forecasting, especially in urban environments and headwater areas.

Wen et al. (2021) assessed PAR benefits by means of case studies of two central Oklahoma flooding events with KOUN rapid-update data. Rain rates derived from the KOUN data and proxies for lower-resolution QPE products were used to force the Ensemble Framework for Flash Flood Forecasting (EF5) (Flamig et al. 2020). Estimated peak rain rate decreased with coarser temporal (and spa-tial) resolution with corresponding impacts on the streamflow simulations. Monte Carlo simulations using the KOUN data indicate

Fig. 5. Violin plots showing the maximum KDP values within the KDP core at the elevation angle closest to the environ-mental melting layer for (a) rapid-update (~2-min) KOUN data and (b) traditional update (~5–6-min) WSR-88D data.

Strong downbursts had a maximum radial velocity of

19.5 m s−1 or higher, while weak downbursts had a maxi-mum radial velocity of less than 19.5 m s−1. The red area of the plot shows the probability density with a greater width indicating a higher frequency of occurrence. Associated boxplots are included within each violin plot for reference.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1365

that the medians of the peak stream discharge estimates were approximately 10% lower with 5- versus 1-min radar observations. In one of the cases, this was sufficient to signifi-cantly reduce objective skill metrics for flash-flood warnings because the discharge peak exceeded the flash-flood threshold by only a modest amount. The EF5 is an integral aspect of the Flooded Locations and Simulated Hydrographs (FLASH) system (Gourley et al. 2017), which is used operationally by the National Weather Service to support flash-flood warning decision-making by forecasters. Hence, these results show where direct improvement to NWS flash-flood warning products could be made.

Targeted observations and environmental wind measurements. We used OSSEs to explore additional short-range forecasting benefits of the PAR scanning capabilities illustrated in Fig. 2. Kerr and Wang (2020) present an ensemble-based “targeted-observation” algorithm.

The algorithm predicts the impacts of supercell radial-velocity observations before the obser-vations are made, thus providing guidance for which azimuth–elevation sectors should be scanned. The forecast metric of interest is the low-level rotation forecast (0–1-km UH). They show that the algorithm can accurately predict true-error-variance reduction, and provide examples where the automated scanning guidance improves UH forecasts relative to pre-defined full-volume scans.

Huang et al. (2020) explore the benefit of assimilating enhanced low- and midlevel clear-air radial-velocity observations from PAR (see appendix A) to improve supercell thunderstorm prediction. Synthetic PAR observations of a splitting supercell were assimilated using an ensemble Kalman filter approach. Experiments showed that assimilation of environmental clear-air radial velocity can reduce wind-analysis errors in both the near-storm environment and within the storm, which improves resulting forecasts. This was especially the case for forecast lead times greater than 30 min. As a result of assimilating clear-air radial-velocity data, the probabilities of strong UH and precipitation within “truth” swaths increased by as much as 30%–40%. Diagnostics suggest that analysis and prediction of linear and nonlinear dynamic forces improves when clear-air radial velocity is assimilated. This in turn contributes to more accurate forecasts of supercell track, vertical motion, maintenance, and structure. We note that more frequent assimilation (1 vs 5 min) of clear-air radial velocity did not systemati-cally improve forecast performance in this experiment.

Thunderstorm nowcasts for aviation. The WSR-88D network provides critical information on convective storms that affect the safety and efficiency of aviation operations. Of interest here are 0–2-h “nowcasts” of thunderstorm location, extent, and intensity that depend to a significant extent on radar observations (Evans and Ducot 2006; Moosakhanian 2016).

NSSL (2020) describes OSSEs assessing aviation nowcast improvements that might result from a future operational weather radar system. Radar observations of thunderstorms affecting the New York City (NYC) airport complex (JFK, EWR, and LGA) and the NYC Terminal Control Area (TCA) on 7–8 August 2018 were used to initialize the experiments. The Variational Doppler Radar Analysis System (VDRAS) assimilated synthetic observations from three dif-ferent radar network configurations: 1) the two operational WSR-88Ds—KDIX and KOKX, 2) these WSR-88Ds plus the LGA and JFK Terminal Doppler Weather Radars (TDWRs), and 3) these four operational radars plus a hypothetical “gap filler” near the northwest New Jersey border that improves low-altitude coverage in the western part of the experiment domain.

Experiments were performed for each network configuration using both 5-min VCPs and 1-min PAR VCPs.

Analyzed and forecast fields of reflectivity, vector wind, rainwater mixing ratio, and aviation-relevant variables were visually compared, and objective performance metrics were calculated. The experiments indicate that assimilation of rapid-update PAR data improves

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1366

model analysis and subsequent 0–2-h forecasts. The improvement increases for experiments that included more radars (i.e., TDWRs and the gap filler), indicating that rapid-update ob-servations and increased low-altitude coverage work synergistically. In these experiments, PAR 1-min VCPs are especially beneficial during convective initiation when discrete storms are evolving rapidly, and for wind forecasts.

Radar network configuration impact on QPE. Kurdzo et al. (2020) assess potential improve-ments in QPE accuracy for a future dual-polarization weather radar network where changes in the number and/or locations of radars could improve low-altitude coverage. They analyzed the statistical accuracy of QPE estimates as a function of range from the radar, corresponding beam height, and coincident Automated Surface Observing System (ASOS) gauge rain rate. Their analysis used the specific-attenuation-based rain-rate estimator R(A) (Ryzhkov et al. 2014), since this is the technique the NWS is adopting for the Multi-Radar Multi-Sensor (MRMS) system in late 2020. A total of 4750 one-hour cases across three consecutive summer seasons (May–August of 2015–17) were processed.

On average, at low rain rates and for observation volumes close to the radar, there is a slight overestimate of rain rates. For the more operationally significant scenario of heavy rain with flash-flood potential, rain rate is underestimated and the bias increases with range. The authors apply their analysis geospatially using NOAA Atlas-derived (https://hdsc.nws.noaa.gov /hdsc/pfds/index.html) 2-yr/1-h rainfall return rates and current WSR-88D locations. Maps of significant QPE error show the largest errors in areas of the south-central, southeastern, and mid-Atlantic United States where peak rainfall return rates are high and distances from the nearest WSR-88D are significant. The mean QPE error across the United States for the selected rainfall return rate is an underestimate of approximately 6 mm h−1.

In evaluating candidate future operational radar network configurations, the authors hypothesize scenarios (e.g., Weber et al. 2007) where some or all FAA-operated aircraft sur-veillance radars would supply high-quality dual-polarization precipitation observations. They show that the contiguous United States (CONUS) mean “peak-rate” QPE underestimate could be reduced by as much as 60% with the additional radar observations, and that many of the geospatial areas where current QPE biases are significant would be filled.

Since the FAA does not currently plan to implement polarimetric weather observation capability on their aircraft surveillance radars, these scenarios must be viewed as hypotheti-cal. The authors note, however, that a relatively small number of additional weather radars could have a significant impact. They define a geospatial Possible Improvement Factor (PIF) field based on estimated peak-rain-rate QPE errors with “perfect” radar coverage across the CONUS. (PIF includes a factor related to population density.) As expected, areas of large PIF are highly localized. One-quarter of the associated benefits pool could be claimed with the addition of only 25 weather radars.

Econometric benefits. We conclude this section with an analysis of monetized benefits for tornado and flash-flood warnings for today’s radar network and compare this to what might be realized with a future system where PAR enables rapid scanning, and/or the density of the network is greater. This work is illustrative of methods needed to support a more com-prehensive cost–benefit analysis of candidate future operational radar system architectures.

Regression analysis on many years of historical data (Cho and Kurdzo 2019a,b, 2020) show that better radar coverage and performance improve warning statistics (e.g., detection probability, lead time, false-alarm ratio), which, in turn, reduces casualties. Radar coverage was quantified with a fraction of vertical-space-observed (FVO) metric that includes terrain blockage, Earth’s curvature effects, and the “cone of silence” above each radar that results from limited elevation scan angles. As the key radar performance metric, they used the

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1367

cross-radial horizontal resolution (CHR). For tornadoes, they also included volume-scan up-date rate, because the PAR innovative sensing experiments (PARISE) (Heinselman et al. 2015;

Wilson et al. 2017) enabled modeling the dependence of tornado warning performance on update rate. Radar network geospatial benefit models were developed based on these statistical relationships, and benefit estimates were computed for different operational radar architectures.

A fine-resolution (1/120° × 1/120°) latitude–longitude CONUS grid was used for all model development and usage computations. Fatalities were monetized according to the Department of Transportation’s value of statistical life (VSL), which was $11.6 million in 2019 dollars.

Injuries were valued as fractions of VSL at $3.1 million (hospitalized) and $0.55 million (treated and released).

Tornadoes. Using 21 years (1998–2018) of data, Cho and Kurdzo (2019a,b) showed that better radar coverage/performance (as measured by FVO and CHR) improves tornado warning per-formance (detection probability and false-alarm ratio). They augmented these dependencies with the results from PARISE indicating that faster volume-scan updates also enhance tornado-warning statistics (detection probability, false-alarm ratio, and lead time). In combination, these relationships allowed for generation of geospatial maps of estimated tornado-warning performance for a given weather radar system architecture.

The authors then showed that tornado casualty rate is statistically dependent on popu-lation inside the tornado path, tornado surface dissipation energy, fraction of population residing in mobile housing, local historical false-alarm ratio, and warning lead time. (There are other factors that are thought to impact casualty rate, but only those that could be straightforwardly quantified geospatially and relevant to benefit estimation were tested.)

The resulting regression model can be used to generate casualty-rate estimates on a geospa-tial grid, given the outputs of the tornado-warning performance model, population, tornado-occurrence rate (parsed by EF number), and mobile-housing statistics. In addition to casualty cost, they also estimated the cost of time lost by people taking shel-ter on false tornado warnings, since they saw that false-alarm rates could be reduced by better radar coverage/performance.

The tornado benefit model was run on the existing CONUS weather radar network as well as on hypothetical configurations.

For the current case, in addition to the WSR-88D, Cho and Kurdzo (2019a,b) included the TDWR in response to a survey that they conducted showing that NWS forecasters rely heavily on its data (where available) in making tornado-warning decisions.

Key results are:

• A tornado-based benefit of $535 million per year is provided by today’s radars.

• The remaining benefit pool is $676 million per year. Figure 6 shows the geospatial distri-bution of this benefit.

Fig. 6. Density map of the remaining benefits pool for tornado warnings.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1368

• Nearly half ($333 million per year) of this potential benefit could be reaped if the existing radars had rapid-update (1-min) capability. The modeling methods and assumptions used here suggest that a denser radar network might contribute toward claiming the remaining $343 million per year benefit. Obviously, other factors such as warning dissemination and public compliance are also critical in minimizing casualties.

• Greater than 99% of the tornado benefit exists east of the Rockies.

• Northern Alabama has the highest potential single-site benefit of $4 million per year

($7 million per year with rapid update).

Flash Floods. In the United States, operational flash-flood warning decisions are based on numerous products from the FLASH system, including radar QPE and flash-flood guidance (FFG). There are various sources of QPE and FFG errors, and the situation is further complicated because forecasters do not utilize a uniform set of data products and decision-support tools.

Using 11 years (October 2007 to December 2018) of historical data, Cho and Kurdzo (2020) showed that better weather radar coverage (as measured by FVO and CHR) led to improved flash-flood warning performance (detection probability and false-alarm ratio). By linking radar coverage directly to warning performance, they bypassed the very complex problem of characterizing QPE and FFG product errors, considerably streamlining the analysis.

The authors showed that flash-flood casualty rate was meaningfully dependent on popu-lation, fraction of population residing in mobile housing, and the presence of a warning.

The resulting regression model could then be used to generate casualty-rate estimates on a geospatial grid, given the outputs of the flash-flood warning performance model, population, flash-flood occurrence rate, and mobile-housing statistics.

The flash-flood benefit model was run on the existing CONUS WSR-88D network as well as on candidate future scenarios. Key results are:

• Current weather radars provide a flash-flood-based benefit of $316 million per year.

• The remaining radar-based benefit pool is $13 million per year.

• The current flash-flood warnings provide a benefit of $392 million per year.

• The remaining warnings-based benefit pool (obtained with 100% of flash floods warned) is $69 million per year.

The remaining radar-based benefit pool of $13 million per year is modest, indicating that the WSR-88D network coverage for flash floods is effective. Note that this model does not consider the enormous benefit state, local, and tribal groups get by managing water resources using weather radar. This is particularly important in the western United States (the Rio Grande is a prime example) and has motivated recent projects for states to purchase and support their own radars.

PAR rapid-scanning methods A significant increase in scan rate relative to that of the WSR-88D must be achieved to enable scanning concepts such as depicted in Fig. 2 and invoked in the preceding section. Figure 7 shows volume-scan time for the precipitation-mode coverage pattern VCP-212 as a function of the “speedup factor”—relative to the WSR-88D average scan rate for this VCP—and the frac-tion of the overall scan timeline allocated to the VCP. Achieving once per minute volume scans while reserving 20%–40% of the timeline for long-dwell environmental wind measurements and/or focused observations will require that scanning be accomplished at an average rate 6–8 times faster than that of the WSR-88D.

A planar PAR antenna with multiple faces or a cylindrical PAR with multiple, simultane-ously active sectors (Zhang et al. 2011) “automatically” provides a speedup factor equal to

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1369

the number of faces or sectors.

Additional scan-rate increase can be achieved using the methods referenced in the “Phased-array radar capabili-ties and challenges” section.

Zrnić et al. (2015), Zhang et al.

(2015), and Weber et al. (2017) discuss scanning concepts for four-faced PARs that realize speedup factors of approxi-mately 8. A PAR with a single planar face that rotates in azi-muth may be significantly less expensive, however, as the cost of a modern PAR depends strongly on the area of its active aperture (Herd and Conway 2016). Achieving the desired speedup factor using such a con-figuration will require that multiple digitally formed receive beams be used, complemented by adaptive scanning and dwell-time reduction where appropriate. Weber (2019), Weber et al.

(2020), and Schvartzman et al. (2021) discuss scanning concepts and technical methods for a single-face rotating PAR. Weber (2019) includes a recurring-cost estimate of $13 million per radar for a configuration consistent with NWS requirements for angular resolution and sensitivity.

The scanning concepts of operation (CONOPs) cited in the preceding paragraph were presented at a high level and, as with other CONOPs under development, will require rigor-ous validation. Here we discuss a method to evaluate any such CONOPs using data-driven simulation. Figure 8 depicts the Signal Processing and Radar Characteristics (SPARC) simula-tor (Schvartzman and Curtis 2019). The input to SPARC is WSR-88D level-2 data, which are quality-controlled and resampled onto a 3D grid of finely spaced “scattering centers.” For the modeled radar configuration (characterized by parameters such as antenna pattern, range weighting function, spatial sampling grid, and transmit-ted waveform), simulated returns from the scattering centers within each resolu-tion volume are weighted and coherently summed to emulate the time series sig-nals received by the radar.

These data are processed using the modeled radar’s signal processing methods to simulate its output.

A Command and Control (CC) simulator (Torres and Schvartzman 2020; NSSL 2020) builds on SPARC by applying temporal interpolation to the input WSR-88D data and adding an adaptive scan-control function that can modify the scan parameters of the modeled radar. It thus offers the capability to study interactions between radar system characteristics, Fig. 7. VCP-212 scan time vs speedup factor and radar timeline fraction allocated to volume scanning.

Fig. 8. Signal Processing and Radar Characteristics (SPARC) simulator.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1370

signal processing techniques, and scanning strategies in determining the properties of radar-variable estimates that would be output from the modeled radar.

To illustrate application of the CC simulator, we consider a four-faced planar array that uses overlapped subarrays (Herd et al. 2005) to form clusters of up to five simultaneous-receive beams in azimuth. Three adaptive scanning algorithms are implemented to reduce the VCP time: 1) the Adaptive Digital Signal Processing Algorithm for PAR Timely Scans (ADAPTS;

Torres et al. 2016), 2) an adaptive dwell-determination technique that adjusts the pulse-repetition times (PRTs) and number of samples to be used at a particular beam position (Torres and Schvartzman 2020), and 3) a multibeam algorithm that varies the transmit-beam-spoil factor and receive-beam-cluster size based on the reflectivity gradients observed (Weber et al. 2017). The most appropriate technique for each azimuth–elevation segment of a VCP is selected using information from a periodic fast-surveillance scan.

Simulated weather observations from this configuration are shown in Torres and Schvartzman (2020) and NSSL (2020). Other than reduction in the areal coverage of boundary layer clear-air echoes, differences between the simulated output and the WSR-88D truth are minimal, and were judged to be insignificant for operational decision-making. (Full-sensitivity boundary layer observations would be obtained less frequently than the precipitation-mode scans, using periodic clear-air scans.)

The scan-time reduction when the adaptive techniques are integrated is shown in Fig. 9 for each face of the modeled PAR. From the radar timeline perspective, the most stressing storm scenario occurs between scan numbers 150 and 180, when most of the 90° azimuth sector allocated to “Face #1” contains strong weather returns. This requires that a signifi-cant fraction of the beam positions in this sector be surveilled, and limits the ability to employ multiple-receive beams without incurring data-quality impacts from cross-coupling among these beams. Even so, with the assumed four-faced PAR architecture, this “worst case” speedup factor (1.4 × 4 = 5.6) approaches the desired operating range indicated in Fig. 7.

PAR antenna sidelobes The WSR-88D antenna’s sidelobes are very low, exceeding the original NWS specification. At angles not affected by the feed-support struts, the envelope of the two-way sidelobe level decreases from approximately −65 dB for near sidelobes to −100 dB or less at angles more than 10° from the peak (see Fig. 13.4 in Doviak 2017). Achieving such low sidelobes with PAR is feasible, but will require a large aperture, careful calibration, and regular array maintenance to mini-mize the number of failed transmit–receive elements.

Figure 10 models a PAR con-figuration with a first sid-elobe at about −60 dB and sidelobes at greater angles that approach the WSR-88D sidelobe envelope (black line) if no more than 1%–3% of the array elements fail.

Operators of military PARs, however, typically allow up to 10% of the elements to fail

Fig. 9. Scan time for a single elevation as a function of scan number using three integrated adaptive-scanning techniques. The corresponding WSR-88D scan time (for a 90° azimuth sector) is the dashed line at the top.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1371

before maintenance is scheduled. With this fraction of failed elements, associated “spurs” in the modeled two-way antenna pattern at large angles approach −80 dB.

It is reasonable to ask then whether the benefit of enhanced temporal resolution provided by PAR would be at least somewhat offset by higher antenna sidelobes. NSSL’s experimental NWRT PARs—the SPY-1A used until 2016 and the current Advanced Technology Demonstra-tor (ATD) (Hondl and Weber 2019)—do not have the large apertures and array configurations needed to approach NWS requirements. The ATD’s first two-way sidelobe, for example, is −42 dB and other two-way sidelobes within 10° of the main beam approach −60 dB. In the near term, therefore, we are relying on simulations to examine this issue. Nai et al. (2020a,b) used SPARC to simulate PAR output data for 10 simulated antenna configurations with differing sidelobe levels. The metric used to quantify the sidelobe levels for each antenna pattern was the increase in integrated sidelobe level relative to the WSR-88D antenna pat-tern (∆ISL). They analyzed 12 storm cases and scored the simulated data for each radar on a scale of 1 = fully acceptable to 5 = unacceptable. This scoring was performed by a vet-eran NWS forecaster with experience in both operations and training, and was based on her judgment as to the “distraction” to forecasters caused by sidelobe-compromised data.

Examples of such distractions are noisy radial velocity and/or false circulation signatures at lower-elevation angles. When these are present, the forecaster must examine other data fields [typically spectrum width (σv) and/or ρHV] to ascertain whether the velocity data are plausible. Nai et al. (2020a,b) detail the scoring process and provide examples of simulated variable fields for the modeled antenna patterns.

Fig. 10. (top left) Modeled antenna pattern for the array depicted. The receive aperture is 8 m in diameter, approximately 50% larger than the subaperture used to transmit. (bottom left) Amplitude taper on receive drives the sidelobes to the desired level. (right) The two-way antenna pattern is plotted including random amplitude and phase errors (0.6 dB, 6° RMS) and the indicated percentage of failed transmit–receive elements. The black “stated requirement” was developed for the SENSR program and corresponds to the approximate antenna pattern envelope of the WSR-88D.

A M E R I C A N M E T E O R O L O G I C A L S O C I E T Y J U LY 2 0 2 1 E1372

The cur ves in Fig. 11 show the fitted data-quality scores as a function of the antenna pattern ∆ISL for the separate storm cases, as well as an overall best fit. Overlaid colored areas depict qualitatively how this analysis could be used to assess the operational im-pacts of PAR antenna side-lobes for industry-proposed designs. As an example, we consider the antenna pattern modeled in Fig. 10 with 10% failed elements.

Its numerically calculated integrated sidelobe level (ISL) is −50.7 dB for the first three sidelobes and −48.6 dB for all sidelobes.