A New X-band Weather Radar System with Distributed Phased-Array Front-ends: Development and Preliminary Observation Results

Weather radars are essential detection tools for severe storm monitoring and short-term weather forecasting, especially nowcasting, and are also used to increase the likelihood of recognizing and understanding severe convective weather systems such as supercells, tornadoes, and downbursts. The weather radar monitoring networks around the world, such as the Weather Surveillance Radar (WSR-88D; Crum et al., 1998) and the Terminal Doppler Weather Radar (TDWR; Istok et al., 2004) in the United States, the Operational Programme of the Exchange of Weather Radar Information (OPERA) in Europe (Köck et al., 2000), and the China Next Generation Weather Radar (CINRAD; Xu, 2003), all have played important roles in monitoring, forecasting, and research of disastrous weather events. Most of these weather radars have long detection ranges for better spatial coverage.

In spite of the success that these radar networks have achieved in assisting and advancing weather monitoring and forecasting (Lee and White, 1998; Stumpf et al., 1998; Serafin and Wilson, 2000; Vasiloff, 2001; Huuskonen et al., 2014), these weather radars also face challenges due to their limitations such as blind areas of detection, low spatiotemporal resolution (LaDue et al., 2010), and the inability to detect three-dimensional (3D) motion of precipitation particles (Wurman et al., 2007). The blind areas of detection, for example, miss detection of the bottom of precipitation clouds; they are due to the ground clutters/obstructions and the earth surface curvature (Wurman et al., 1997; Bharadwaj et al., 2010). The unavoidable radar beam broadening in cases of long-range detection results in low resolution, along with other factors including limited scanning speed (Zrnic et al., 2007). Moreover, each individual Doppler radar can only obtain the radial velocity of precipitation particles, which is projected along the direction of radar electromagnetic waves, limiting the detection of 3D motion within clouds (Chen and Chandrasekar, 2018).

To improve the lower atmospheric (0–3 km above ground level or AGL) detection, a concept of networked weather radars was put forward by the Collaborative Adaptive Sensing of the Atmosphere (CASA) project, which was carried out by the Engineering Research Center established by the National Science Foundation (NSF) of the United States in 2003 (Brotzge et al., 2006). According to this CASA concept, multiple X-band low-power short-range weather radars were used to form the CASA networked radar system. Then, by implementing the Distributed Collaborative Adaptive Sensing (DCAS) mode (McLaughlin et al., 2005), high-spatiotemporal resolution observation was realized, whereas the blind areas of detection of one radar can be detected by another (Salazar et al., 2009). In 2013, a CASA-like networked X-band weather radar system consisting of two solid dual-polarization radars and two magnetron radars was established in China (Li et al., 2015). Both the CASA and CASA-like projects showed that the networked radar had the potential to improve the accuracy of mesoscale severe weather forecast and to extend the leading time of forecast and early warning. For example, CASA has supplemented the WSR-88D weather radar network with valuable tornado detail evolution data (Banacos et al., 2012). Other studies from CASA field experiments have demonstrated improvements in ensemble probabilistic forecasts (Snook et al., 2012), short-term predictability of precipitation (Ruzanski and Chandrasekar, 2012), Doppler radar data assimilation (Putnam et al., 2014), nowcasting (Ruzanski et al., 2011), and forecaster wind assessment and warning (Rude et al., 2012). Furthermore, with the overlapped detection areas of adjacent radars in CASA, radial velocity data from multiple radars can be used to retrieve wind fields, in addition to the enhanced spatial resolution (Chen and Chandrasekar, 2018).

In wind retrieval, radial velocities at a given point in the detection space from multiple networked Doppler radars should ideally be at the same time, or at least the data time differences (DTD) among the data of different radars are small enough or negligible. An advection correction method has been used to improve the wind retrieval in cases of large DTD and fast developing weather systems (Chen and Chandrasekar, 2018). Meanwhile, reducing the DTD has been an important goal in many networked radar systems in the recent decades, so that the retrieved 3D motions are capable of capturing the internal structure and evolution of fast developing and moving storms (Wurman, 1994). Shapiro et al. (2009) analyzed pattern-translation components U and V, which depict the translational movement of an unchanging pattern, in dual-Doppler wind analysis. Their results indicated that the improvement in wind analysis can be significant if the volume scan is completed within 2 min. Potvin et al. (2012) further demonstrated that shorter volume-scan time leads to better retrieval results. Thus, rapid-scanning weather radars are desirable for wind field retrieval. However, although the volume-scan time has been reduced with various techniques, the current 1–2 min volume-scan times are still problematic for fast-changing convective systems. Both the reflectivity factor fusion and radial velocity synthesis suffer from large errors because of the large DTD. Therefore, developing rapid-scanning radar systems is still a critical need and challenge for high-resolution reflectivity and wind observations.

Fortunately, the phased-array technology has been recognized to be feasible for building rapid-scanning weather radars (Ortiz et al., 2020). In 1997, a volume-imaging radar wind profiler was developed by the University of Massachusetts-Amherst in the United States for atmospheric boundary layer turbulence studies (Mead et al., 1998). In 2002, the National Severe Storms Laboratory (NSSL) and National Weather Service (NWS) of the National Oceanic and Atmospheric Administration (NOAA), Federal Aviation Administration (FAA), and the University of Oklahoma in the United States developed a National Weather Radar Testbed (NWRT) (Zrnic et al., 2007) for detecting severe convective weather systems and carried out field experiments in May 2009 (Heinselman and Torres, 2010). Compared with the WSR-88D, the NWRT has demonstrated the advantages of the phased-array antenna, which can capture rapid-changing weather system with 30 s to 1 min volume-scan time (Bluestein et al., 2010). Following the NWRT, several other phased-array weather radars were developed (Kurdzo et al., 2017), such as the Multi-mission Phased-array Radars (MPAR; Weber et al., 2007), the Rapid Doppler On Wheels at the Center for Severe Weather Research (DOW; Wurman and Randall, 2001), the Meteorological Weather Radar 2005 X-band Phased Array (MWR-05XP; Bluestein et al., 2010), the Atmospheric Imaging Radar (AIR; Isom et al., 2013), and the Cylindrical Polarimetric Phased-Array Radar (CPPAR; Zhang et al., 2011; Fulton et al., 2017) in the United States. These phased-array radars reduced the volume-scan time to less than 1 min. In 2012, under a grant from the National Institute of Information and Communications Technology (NICT), an X-band Phased-array Weather Radar (PAWR) was developed by Toshiba Corporation in Japan and installed at Osaka University (Suita Campus) for field experiments (Mizutani et al., 2015; Adachi et al., 2016a, b; Ushio et al., 2017). In 2017, in combination with other detection instruments, two X-band PAWRs, serving as a phased-array weather radar network, completed a field experiment in Japan (Shimamura et al., 2016; Yoshida et al., 2017). In the common detection area of the PAWRs, the volume-data update occurs every 30 s. Thus, the phased-array technology, which has already been used in the field of weather radars, can definitely shorten the volume-scan time. Benefits of the rapid scanning and adaptive volumetric scanning of the PAR have been clarified for improving the severe weather and flash-flood warning performance, compensation methods have been developed for the polarimetric products to solve the beam-steering challenge caused by the PAR technology (Weber et al., 2021), and a rotating PAR (RPAR) using a distributed beam technique has been introduced as a potential PAR which has a good cost performance compared with other candidate PARs (Schvartzman et al., 2021a).

Inspired by the phased-array technology and the distributed weather radar systems such as the CASA with the DCAS collaborative scanning mode for more rapid scanning, a new weather radar system with distributed phased-array front-ends was developed and then initiated field experiments in China. Because of the triangular deployment requirement of multiple phased-array front-ends and a unique synchronized azimuthal scanning (SAS) rule of the front-ends, this new weather system is named Array Weather Radar (AWR). Different from the other radar systems, the AWR comprises one radar back-end and three or more phased-array radar front-ends. The prototype of the AWR was designed in 2015, the first X-band single-polarized AWR with three phased-array front-ends was manufactured in cooperation with relevant companies in 2017 (Ma et al., 2019), and a field experiment was carried out in April 2018. The most outstanding advancements of the AWR are the achievement of 12 s rapid volume-scan time and 2 s DTD which are forced by the SAS rule, expandable large spatial coverage of one AWR (one back-end with multiple distributed front-ends), and the enhanced spatiotemporal resolution.

The following sections present, in order, the design concept of the AWR system and feasibility of achieving the above advancements (section 2), the front-ends that achieve the rapid volume scan (section 3), the deployment of the front-ends that form fine detection areas (FDAs) (section 4), and the synchronized azimuthal scanning (SAS) mode that achieves 2 s DTD (section 5). A brief description of calibration procedures, actual deployment, field experiments, and some preliminary experimental results from the first X-band single-polarized AWR are presented in section 6, followed by main conclusions and future discussions that are summarized in section 7.

With the advent of Doppler weather radar, not only the detection of echo intensity (reflectivity factor), but also the detection of radial velocity is realized. Although the radial velocity contains only one component of the precipitation particle motion, which is projected along the direction of radar electromagnetic waves, it provides meteorologists with atmospheric kinematic information for understanding the development of weather systems. However, to fully describe the atmospheric dynamics, wind fields at different heights in the precipitation clouds are needed, with which the internal development and evolution of weather systems can be monitored and more accurate early warning of severe storms would be possible. Since a single radar only detects one radial velocity for a given point in space, an accurate detection of the 3D motion of precipitation particles in a concerned area requires two conditions to be met: 1) at least three non-coplanar radial velocities are detected; and 2) the DTD among these radial velocities must be small enough or negligible so that the full motion vector can be composed. The AWR focuses on satisfying these two conditions: detecting three non-coplanar radial velocities, while keeping the DTD among the radial velocities smallest.

An AWR is comprised of one radar back-end and three or more phased-array radar front-ends. In order to obtain, for example, three non-coplanar radial velocities, three front-ends of the AWR are deployed as a group in a triangular placement and operate in the SAS mode. While an equilateral triangle is the ideal deployment for maximum FDA and minimum DTD, an acute triangular placement is more common and practical due to limitations of physical locations. Nevertheless, the equilateral triangular placement is used in most of the discussions herein, which can be easily extended to an acute triangular placement. The DTD among the front-ends is reduced by the SAS mode and a phased-array fast scanning technology. The AWR consisting of one back-end and three or more phased-array front-ends is a complete weather radar system. Even though multiple phased-array front-ends are deployed at different locations, they are controlled by only one back-end.

In brief, AWR is not a phased-array radar (PAR) network comprising several independent PARs and a central control site, of which the latter is built as an additional component to accomplish the coordination of each individual PAR. At the same time, it is also not a regular PAR, if the AWR comprises one radar back-end and only one radar front-end, then this AWR is actually the regular PAR.

The block diagram of the AWR comprising one back-end and three or more phased-array front-ends at different locations is shown in Fig. 1. The primary technical specifications of a single-polarized three-front-end AWR system are given in Table 1.

Figure 1.  The block diagram of an AWR. The AWR consists of one radar back-end (left) and more than three radar front-ends (right) at locations different from the back-end. The back-end of the AWR comprises a data server, a monitor server, and a process server; and every front-end comprises 64 channel processors, 64 channel TR modules, a 64-channel antenna array, an azimuth rotation servo, and a power supply. Three front-ends of the AWR are deployed as a group in a triangular placement and operate in the SAS mode. The communication between the back-end and the front-ends mainly includes SAS controlling and radar data transmitting.

The AWR back-end (Fig. 1) comprises a data server, a monitor server, and a process server. The data server is used for collecting, storing, and managing radar products and equipment status data from the radar front-ends; the monitor server monitors the operational status of the radar front-ends, dispatches control commands to the radar front-ends for completing the SAS, analyzes the status data, and displays radar error warnings and various radar products; and the process server processes the radar products from the radar front-ends and completes data quality control processes. The outputs from the process server comprise various radar products such as wind fields, intensity fusion products, and other physical data.

Each radar front-end comprises 64 channel signal processors (Fig. 2), 64 channel transmit-receive (TR) modules (Fig. 3), a 64-channel antenna array, a servo, and a power supply. The antenna array (1.45 m × 1.2 m) uses an array of slot waveguides. The waveguide array is composed of 64 slot waveguides, which are wire-connected to the 64 channel TR modules. The azimuth rotation servo drives the antenna to rotate mechanically in horizontal directions for completing azimuthal scanning. The power supply powers all components in each front-end. Figure 4 is a picture showing the five main components of the phased-array front-end discussed above.

Figure 2.  The detailed block diagram of the 64-channel signal processors. As shown in the figure, multiple signal processing procedures are finished in an A/D acquisition unit, an FPGA unit and multiple DSP units in sequence.

Figure 3.  The detailed block and functional diagram of one-channel transmit-receive (TR) module. The basic components and part of the technical specifications are given in the figure.

Figure 4.  Picture showing five main components of one front-end of the first X-band single-polarized AWR deployed at the Huanghua International Airport, Changsha, China. The five components include (1) phased-array antenna and radome, (2) TR modules, (3) power supply, (4) signal processors, and (5) servo.

The communication between the back-end and the distributed radar front-ends at different locations are completed via optical fibers, wherein the SAS controlling commands and basic physical products such as reflectivity factor, radial velocity, and spectrum width from the multiple radar front-ends are transmitted during the communication.

Figure 3 shows one channel TR module block and functional diagram. Each 64-channel transmitter sends signals to the antenna and sequentially forms four wide beams with an elevation width of 22.5° and an azimuth width of 1.6° covering the 0°–90° elevation by controlling the phases of the 64 channels. During receiving, the 64-channel receiver receives target echo signals. The echo signals are processed through a low-noise amplifier, a mixer, and an intermediate frequency amplifier firstly; secondly, they are converted into digital signals through a 64-channel analog-to-digital converter (ADC); then they are fed into the 64 channel signal processors. As shown in Fig. 2, in the 64 channel signal processors, the digital signals are processed through down-conversion, extraction, filtering, digital beamforming, linear pulse compression, ground clutter suppression, spectrum analysis, parameter estimation, and other algorithms. Part of the outputs from the signal processor, which include the reflectivity, the radial velocity, and the spectrum width are finally fed to the radar back-end. The primary technical specifications of every front-end of the X-band single-polarized AWR are shown in Table 2.

By using the phased-array technology, each front-end of the AWR completes 0°–90° elevation scans electronically; at the same time, the front-end mechanically rotates 360° in azimuth by using the servo. For one front-end, all 4096 elements participate to produce the spoiled beams by phase controlling and amplitude weighting. Each front-end transmits four 22.5°-width transmit beams sequentially covering the elevations of 0°–22.5°, 22.5°–45°, 45°–67.5°, and 67.5°–90°, respectively. Corresponding to each 22.5°-width beam, a digital beamforming processor in the signal processor forms sixteen 1.6°-width receive beams simultaneously, as shown in Fig. 5. Thus, a total of 64 receive beams are obtained from the four transmit beams covering the 0°–90° elevations, and the elevation interval between any two adjacent receive beam centerlines is about 1.4°.

Figure 5.  Schematic diagram of transmit and receive beam coverage of 0°–22.5° elevations of the AWR digital beamforming coverage pattern. One 22.5°-width transmit beam (light gray color) and corresponding sixteen 1.6°-width receive beams (light purple color) are shown in the figure. Actually, each front-end transmits four transmit beams sequentially covering the elevations of 0°–22.5°, 22.5°–45°, 45°–67.5°, and 67.5°–90°, respectively. Corresponding to each 22.5°-width beam, the signal processor forms sixteen 1.6°-width receive beams simultaneously.

The volume-scan time can be calculated as follows: the 0°–22.5° elevation scanning comprises the following two steps: 1) transmitting 64 pulses with 4 μs pulse width and a pulse repetition period of 50 μs to detect targets within the 0.6–3 km range; and 2) transmitting 64 pulses with 20 μs pulse width and a pulse repetition period of 133 μs to detect targets within the 3–20 km range. The time for completing the 0°–22.5° elevation scanning (units: μs) is:

The processes for completing the scanning of 22.5°–45°, 45°–67.5°, and 67.5°–90° elevations are the same as those for the 0°–22.5° elevation scanning. Therefore, the time for completing the entire 0°–90° elevation scanning (units: μs) is:

The horizontal beamwidth of the front-end antenna is also 1.6°. If this horizontal beamwidth is regarded as azimuth resolution, 225 horizontal beams are needed to complete the 0°–360° azimuth scanning. Therefore, the volume-scan time (units: s) is:

In practical operation of the AWR, for ensuring that the three-front-end AWR can meet a rapid volume scan requirement, short pulse repetition times (PRTs) are required, therefore, each front-end of the AWR sacrifices the detection range which can be expanded by increasing the number of the front-ends. According to the volume-scan time calculated above, the servo rotation speed is set as 30° s⁻¹; accordingly, the time to complete one volume scan is 12 s.

According to the velocity measurement requirements described in section 2.1, the three front-ends of the AWR are deployed at the vertices of a triangle, as shown in the plane layout of the first X-band single-polarized three-front-end AWR deployed at the Huanghua International Airport, Changsha, China (Fig. 6). The baseline of the triangle is about 20 km. Each radar front-end is in the center of its detection space, which is a hemispheric volume with a radius equaling the maximum detection range. In the overlapping detection area of the three radar front-ends, each spatial point has three reflectivity observations Z₁ (x, y, z), Z₂ (x, y, z), and Z₃ (x, y, z) and three radial velocity observations V₁ (x, y, z), V₂ (x, y, z), and V₃ (x, y, z), where x, y, z are Cartesian coordinates.

Figure 6.  The plane layout of the first X-band single-polarized three-front-end AWR and the horizontal projection of the detection areas at Huanghua International Airport, Changsha, China. In the figure, each radar front-end is in the center of its detection area, the locations of the three front-ends (red stars), the detection range of each front-end (black circles), as well as the actual distances (words in light gray) between the front-ends are shown. The overlapped detection area of the three front-ends is the fine detection area (FDA in red words); the areas that are detected by two front-ends are enhanced detection areas (EDAs in green words); and the areas detected by only one front-end are normal detection areas (NDAs in blue words). The picture of the radar front-end 3 of the AWR installed on an iron tower is also shown in the figure. In addition, an L-band boundary layer wind profile radar (black star) is located at (28.17°N, 113.23°E), the data of which can be used to complete a cross check and validation process with the data of the AWR. When there are more than three front-ends, all of the FDAs will form a continuous larger FDA without gaps (Figs. 9 and 12).

Figure 7.  The AWR synchronized azimuthal scanning (SAS) procedure flowchart. The three synchronization parameters calculated by the back-end include the antenna rotation speed, the entry azimuth, and the time of radar beams entering the FDA, if the azimuth synchronization error is more than 0.05 s (Table 1). The back-end of the AWR transmits synchronous control commands to the multiple distributed phased-array front-ends to adjust the rotation speed so that the SAS can be realized.

The 3D volume (more accurately, a scalene Reuleaux tetrahedron with a flat base) formed by the overlapped detection areas of the three front-ends is the 3D FDA, of which the intersection with the ground surface is a scalene Reuleaux triangle (with curved sides). However, it is sufficient in the following discussion and actual applications to consider only the triangular area (straight sides) as the FDA. Figure 6 shows the actual layout of a typical three-front-end AWR and the horizontal projection of the detection areas. The areas that are detected by two overlapping front-ends are enhanced detection areas (EDA), and the areas detected by only one front-end are normal detection areas (NDA).

Compared with the long-range weather radars, a three-front-end AWR has limited spatial coverage. However, the total coverage area as well as the FDA of the AWR can easily be augmented by increasing the number of the radar front-ends. Figure 6 shows the most basic deployment of the typical three-front-end AWR system.

Weather radars perform volume scans centered on the location of the radar antennas. Given the fact that it takes time (2 mins for example) for a regular weather radar to complete one volume scan, no two radars at different locations can detect all spatial points within the common detection area at the same time. That is to say, DTD commonly exists between the data at the same spatial point detected by any two radars. For a certain point (x, y, z) in space, three weather radars can obtain three observations of, say radial velocities, at three different times t, t+a, and t+b:

where, a and b are the DTD between V_r,2 and V_r,1, and V_r,3 and V_r,1, respectively. When every weather radar scans in an unsynchronized manner, the maximum DTD for the points within their common detection area is:

which can reach up to 2 mins if each radar has a 2 min volume-scan time. For a developing weather system, the DTD for the same spatial point translates to different parts of the weather system at a given time point. In other words, when one radar detects one part (spatial point) of the weather system, another radar detects this same spatial point after a few minutes, but the part of the weather system detected by the previous radar has moved away during the time delay of the second radar. The second radar actually detects a different part of the weather system that moves into this spatial point. Therefore, if the DTD between two radars is 2 mins and the weather system moves at a velocity of, say, 20 m s⁻¹, the detected data of the two radars are actually information of two locations that are 2.4 km apart, relative to the weather system. If these data are treated as same location at same time, in the case of synthesizing radial velocities, for example, the result is likely to have significant errors.

In order to reduce the DTD, the AWR uses an entry azimuth synchronization method during the SAS. The entry azimuth is the azimuth angle of a front-end that starts scanning the FDA.

The SAS is a unique feature of the AWR, which is completed by the cooperation between the front-ends and the back-end. The monitor server in the back-end (Fig. 1) is responsible for controlling and monitoring the SAS mode of the phased-array front-ends. The monitor server sends synchronous scanning commands, checks the synchronization status of each front-end, determines the starting time and rotation speed of each front-end, and controls antenna rotation of each front-end, so that each front-end enters the FDA at the same time. The servo in each front-end (Fig. 1) drives the front-end to complete the SAS.

The AWR SAS procedure is briefly summarized in a flowchart in Fig. 7. As soon as the AWR starts, the back-end calculates three synchronization parameters including an antenna rotation speed ω (in ° s⁻¹), an entry azimuth φ (in °), and the time of radar beams entering the FDA (i.e., the time of the antenna reaching the entry azimuth); and sends the synchronous scanning commands and the three synchronization parameters to each front-end. Each front-end of the AWR has a servo control unit that controls the operation of a servo motor and communicates with the back-end about the servo status and antenna orientation information. The servo motor drives the radar antenna while the azimuth sensor provides the azimuth angle status of the front-end.

Upon the receipt of the synchronous scanning commands, each front-end servo control unit calculates the time to start the servo rotation according to the three synchronization parameters. After starting to rotate, the servo controls the rotation speed so that the time error of radar beams entering the FDA is no more than 0.05 s (Table 1). The entry azimuth φ is acquired continuously, and azimuth synchronization errors are continuously calculated, monitored and displayed. If the beam enters the FDA too early, the rotation speed of the front-end is reduced by at most 1%; likewise, if the radar beam is delayed in entering the FDA, the rotation speed of the front-end is increased by at most 1%. These adjustments ensure that the fluctuation of beam scanning speed is small and the average time error for the front-end to enter the FDA is smaller than 0.05 s.

As shown in Fig. 8a, three front-ends A, B, and C are arranged at the vertices of an equilateral triangle ABC with baseline L. The rotation directions of the three front-ends can be either clockwise or counter-clockwise. We first discuss in Fig. 8a where all three front-ends rotate clockwise.

Figure 8.  Synchronized azimuthal scanning (SAS) scheme of a three-front-end AWR. (a) Layout of the three front-ends A, B, and C and their scanning scheme; (b) DTD distribution within the FDA when all rotate in the same direction (clockwise); (c) DTD distribution when B rotates in the opposite direction (counterclockwise) from A and C.

After the rotation adjustments and all the front-ends reach a stable scanning state as detailed in the previous subsection, every front-end should theoretically rotate at the same speed ω and reached their respective entry azimuth at the same time t₀. Assuming the front-end A is located at coordinate (x_A, y_A) and its initial entry azimuth at time t₀ is φ_A=φ_0A, then the location of front-ends B and C are:

respectively. The initial entry azimuth of front-ends B and C are:

respectively. Since all three front-ends are rotating at the same speed ω = 30° s⁻¹ and they scan 60° azimuths to complete scanning the FDA, every front-end completes scanning the FDA in 2 s, ending at t₀ + 2 s. Therefore, the maximum DTD for all the points in the FDA is 2 s.

For any given point D (x, y) within the FDA, the time it takes for each front-end beam to reach this point can be found from the angle each front-end is scanned from its initial entry azimuth at t₀ to the azimuth of D with respective to each front-end. This scanned angle divided by ω is the time (t_A, t_B, t_C) it took for each front-end to reach point D. The scanned angles (S_A, S_B, and S_C as shown in Fig. 8a) for the three front-ends are found as:

The scanned angles should be between 0°–60°. Thus, different integer number of 180° may need to be added to S_A, S_B, or S_C to keep them between 0°–60°. Then, the scan times for the three front-ends are:

The DTD for point D is now:

Figure 8b shows the DTD distribution within the FDA. The minimum DTD is zero and the maximum DTD is 2 s. The average DTD of the entire FDA is 1.3 s.

Consider the case where the rotation directions of the three front-ends are not uniform, for example, keeping A and C clockwise rotation but letting B to rotate counter-clockwise. The scanned angle S'_B is labeled in Fig. 8a, which is simply S'_B = 60–S_B. The DTD distribution in this case is shown in Fig. 8c. Although the minimum and maximum DTD are still the same as in Fig. 8b, the average DTD of the entire FDA is reduced to 0.9 s.

Without the SAS mode, the front-end may fall out of synchronization with other front-ends. Assuming that the rotation speed ω' of the front-end A is actually ω' = 0.999ω = 29.97° s⁻¹, meaning that the error is equivalent to 0.1% of ω, then the time it takes for the front-end A to complete the scanned angle S_A (i.e., to reach point D) is now:

This means that the time to reach point D is increased by 0.1%. Likewise, for the entire volume scan, the time delay is 0.012 s (i.e., 0.1% of 12 s). If this continues for the lifetime of an average storm, say one hour, then the time delay will be 3.6 s. The maximum DTD would be 5.6 s. In the worst case when time delays are more than 12 s, then the maximum DTD would still be within 12 s, given the volume-scan time is still 12 s. This falls back to the uncoordinated scan mode. Therefore, the SAS is the key to keep DTD below 2 s.

Figure 9 is the deployment diagram of seven radar front-ends. Seven radar front-ends can form six FDAs and expand six times the size of FDA coverage of the basic three-front-end deployment. It shows a layout and scan sequence diagram of a seven-front-end AWR, with the front-ends labels as A, B, C, D, E, F, and G. These seven front-ends form a hexagonal area with six triangular FDAs inside. Each FDA is covered by three adjacent front-ends. In order that the DTD in each FDA be less than 2 s, the rotation direction of the central front-end A should be opposite from the other six front-ends.

Figure 9.  Synchronized azimuthal scanning (SAS) of a seven-front-end AWR that forms a hexagonal region with six FDAs. The front-end A rotates clockwise while B, C, D, E, F, and G rotate counterclockwise. The handle attached to each front-end in the figure represents the initial entry azimuth. The numbers surrounding each front-end are times (in seconds since the beginning of the synchronous scanning) when each front-end arrives that azimuthal position. Deployment distance between every two radar front-ends (i.e., the maximum detection range of one radar front-end) is about 20 km, the horizontal range of the FDA product is approximately triangular area with a baseline of about 20 km. Actually, the horizontal FDA of the three-front-end AWR is larger than a triangle. It is actually a scalene Reuleaux triangle (an irregular triangle with the sides of arcs instead of straight lines).

Upon receiving the synchronous scanning commands from the back-end, the servos of the seven front-ends drive the antennas to rotate while adjusting the rotational speed so that every front-end reaches its corresponding entry azimuth at time t₀ and starts rotating at the constant speed ω. As illustrated in Fig. 9, if front-end A starts its scanning clockwise from φ_A = φ_0A = −30 at t₀, then the other six front-ends B through G should start scanning counter-clockwise at the same time from

φ_B = φ_0A + 180, φ_C = φ_0A + 300, φ_D = φ_0A + 60, φ_E = φ_0A + 180, φ_F = φ_0A + 300, φ_G = φ_0A + 60,

respectively. The instants (in seconds since t₀) of each front-end entering each respective FDA are labeled in Fig. 9. It is indicated that, in this SAS mode, scanning of each FDA can be completed by all three relevant front-ends within 2 s, namely, the DTD is ≤ 2 s as discussed in section 5.3. Scanning of the entire hexagonal region is completed in 12 s, with all seven front-ends starting rotation at the same time t₀ and completing one volume scan at the same time t₀ + 12 s. In addition, this scanning scheme also assures the optimal overall DTD as shown in Fig. 8c.

Before the deployments of the first X-band single-polarized three-front-end AWR, metal sphere far-field calibration procedures including an azimuth angle and elevation angle calibration procedure, a velocity calibration procedure, and a reflectivity calibration procedure were completed in clear sky conditions and are described here.

In the azimuth angle and elevation angle calibration procedure, GPS data from a drone which carries the metal sphere and moves at different places in the overlapped detection range of the three front-ends of the AWR were compared with the location data of the metal sphere detected by the AWR to obtain the deviations between the two kinds of location data. According to the results, the azimuth deviations are compensated during data processing and the elevation deviations are compensated by adjusting the antenna elevations.

In the velocity calibration procedure, the drone carried the metal sphere above the connection line between two AWR front-ends and then moved back and forth between the two AWR front-ends; at this time, the velocities detected by the two AWR front-ends should be in opposite directions but have the same value. The results show that the velocity errors of the three front-ends are within 0.3 m s⁻¹, which meet the design requirements.

In the reflectivity calibration procedure, based on parameters such as distance between the metal sphere and each AWR front-end, the frequency, the receiving and transmitting beam widths, and the range resolution of the front-end, the diameter of the metal sphere, etc., the standard reflectivity of the metal sphere was calculated according to the radar equation, and the standard reflectivity was compared with the reflectivity detected by the front-end to obtain reflectivity deviations. Finally. the reflectivity values were calibrated by adding or subtracting the deviations.

In April 2018, Eastone Washon Science and Technology Ltd. installed the first X-band single-polarized three-front-end AWR at Changsha Huanghua International Airport, Hunan province, China. The three phased-array front-ends of the AWR deployed used horizontal polarization antennas. The layout of the three front-ends and the actual distances (words in light gray) between adjacent front-ends are shown in Fig. 6. The FDA of the AWR covers the airfield runway of the airport. As shown in the Fig. 9 and discussed in section 5.4, the front-ends A and B should be pointing at each other at 0 second time. In practical application, since it is important to avoid direct transmission into the other front-end, the three front-ends of this first AWR were set to different frequencies to avoid frequency interference, and the frequency differences among the front-ends were 10 MHz.

Precipitation events have been captured by the AWR since its deployment. Here we present one precipitation event to demonstrate the high-resolution wind field synthesis and retrieval capabilities of the AWR preliminarily, which is a precipitation event that occurred at 2320:48 UTC 21 April 2019. Given the small DTD of the AWR, the data from one front-end could be used to compensate the low azimuth and elevation resolutions of other front-ends. The final horizontal and vertical resolutions of intensities and wind fields in the FDA of the AWR were set to interpolated grid sizes of 100 m × 100 m in this event. The 100 m × 100 m × 100 m gridded wind fields at 12 s temporal resolution were directly calculated from the radial velocities from the three front-ends using a wind synthesis method (Armijo, 1969; Bousquet and Chong, 1998; Chen and Chandrasekar, 2018).

Figures 10a–c show the reflectivity and synthesized horizontal wind field grid products (wind barbs from u and v components) at heights of 0.7, 1.0, 2.1, and 5.1 km AGL. Evident are moderate northwest winds near the surface, weak mid-level southerly winds, and strong southwesterly winds at upper-levels. For this northeastward moving system, the low-level scan sees light precipitation toward the downwind direction and moderate precipitation on the western side. The detection results including wind directions and wind speeds from an L-band boundary layer wind profile radar (WPR) at the Changsha Huanghua International Airport, which is shown in Fig. 6 (black star), are shown in a red rectangle at the upper right corner of each panel in Fig. 10. From the qualitative comparison between the AWR and the WPR, at the heights of 1.0, 2.1, and 5.1 km, the tendency of the wind directions and wind speeds from the two radars are basically the same, except for the results at the height of 0.7 km where the wind directions from the two radars are opposite. It is likely that the WPR is greatly influenced by the ground clutter near the ground and thus the measurement results from the WPR are not representative. In addition, a more detailed comparison between the WPR data and the AWR retrieval has been completed (Li et al., 2020). The comparison results show that the synthesized wind fields of the AWR are more consistent with that of the WPR in stable precipitation events such as stratiform cloud precipitation, while larger differences are shown in convective precipitations, which may be caused by the different detection principles between the WPR and the AWR.

Figure 10.  Reflectivity (color) and wind fields (barbs for horizontal wind) from the three-front-end AWR and the wind fields from an L-band wind profile radar (WPR) at the heights of (a) 0.7 km, (b) 1.0 km (c) 2.1 km, and (d) 5.1 km AGL are shown; all fields are valid at 2320:48 UTC 21 April 2019; the geographic position of the horizontal origin is at (28.16°N, 113.21°E); and the wind speed and direction result from the L-band WPR (barb for horizontal wind in a red rectangle at the upper right corner of each sub-figure are shown at the same heights as the AWR.

Figures 11a–h show the continuous reflectivity and synthesized horizontal wind fields CAPPIs every 12 s in this precipitation event at the height of 2.1 km AGL from 2321:12 UTC to 2322:36 UTC. These preliminary results show that the AWR can continuously provide the wind field for the evolving precipitation event every 12 s.

Figure 11.  Reflectivity (color) and wind fields (barbs for horizontal wind) grid product of the AWR every 12 s at the height of 2.1 km AGL from 2321:12 UTC to 2322:36 UTC 21 April 2019 (Fig. 11a to Fig. 11h; the geographic position of the horizontal origin is at (28.16°N, 113.21°E).

A new weather radar system (AWR) which consists of one radar back-end and distributed phased-array radar front-ends has been developed, inspired by both existing weather radar networks and the phased-array technology. Different from other weather radar systems, the AWR uses the phased-array technology and the SAS to achieve 12 s rapid volume scan with 3D fine detection of microscale features of convection events. With the unique SAS rule for three to seven front-ends, the AWR achieves < 2 s DTD while completing the full regional volume scan within 12 s.

Considering the small FDA limitation of the three-front-end AWR, we discuss the generalized AWR deployment solution below and the DTD for this deployment. As mentioned in section 4, an AWR can be built with more front-ends to cover a large region, yet still one AWR system with only one radar back-end. Based on the seven-front-end AWR discussed in section 5.4, a simple arrangement as shown in Fig. 12 can be realized for larger coverage. Figure 12 shows a 37-front-end AWR, with the center front-end of each group of seven AWRs (F, H, Q, S, U, D₁, F₁) rotating clockwise and the others rotating counter-clockwise. From the above discussion of the seven-front-end AWR in section 5.4, the DTD in each group of the seven front-ends is still smaller than 2 s, as shown by the darker yellow FDAs. In the light-yellow-colored FDAs, the maximum DTD is 10 s and the minimum is 2 s. This is determined by the seven-front-end AWR scanning scheme and the fact that the three front-ends surrounding these FDAs are all rotating in the same counter-clockwise direction. Taking the FDA G-L-M as an example, the front-end G starts scanning this FDA at second 6, L starts at second 2, and M starts at second 10. In this 37-front-end AWR, scanning of the entire region is still completed in one volume-scan time (12 s) with every front-end starting and ending one volume scan at the same times. The DTD is less than 2 s in 75% of the region, and 2–10 s in the remaining 25% of the region. In this way, 37 radar front-ends can form 42 FDAs, which can solve the small FDA limitation, meanwhile, the DTD of the 75% of the AWR coverage region can be less can 2 s. The AWR can cover a city which has the area similar to the region covered by the AWR with this kind of deployment solution in the future.

Figure 12.  Layout of a 37-front-end AWR and the synchronized azimuthal scanning (SAS) scheme (labels are the same as in Fig. 9). In the darker yellow FDAs, the DTD in each group of seven front-ends is less than or equal to 2 s; and in the light-yellow-colored FDAs, the maximum DTD is 10 s and the minimum is 2 s.

In order to improve the speed of complex computations, suitable server configurations should be provided for the three-front-end AWR. Taking the three-front-end AWR in actual operation for example, nine servers including one control server, four normal product servers, three GPU product servers, and one data storage server are configured for one three-front-end AWR; the detailed server configurations are given in Table 3. In order to avoid data stacking, the control server controls the three-front-end AWR to complete one volume scan within 12 s, the volume-scan data is transmitted to and stored in the data storage server by using the optical fiber communication at the speed of 10 G s⁻¹ within 12 s, and the radar products are generated in the GPU product servers and displayed in the normal product servers within 12 s. The volume-scan data quantity and the actual computational complexity are different under different weather conditions. Considering the huge amount of computation, most product computation algorithms are completed in the three GPU product servers. For the seven-front-end AWR, there are more challenges for the server configurations.

With the small DTD achievement, wind field synthesis and/or retrieval at different heights has been achieved from the field experiments of the first X-band single-polarized three-front-end AWR deployed at the Changsha Huanghua International Airport, China. Synthesized 100 m × 100 m × 100 m gridded wind fields at 12 s temporal resolution reveals fine structural and evolutional features inside convective clouds.

However, it should be noted that in-depth wind field comparison and the reflectivity comparison are not considered in this paper; in addition, attenuation in the convective precipitation areas is problematic, because of the X-band limitations. Our future work will focus on these aspects. Moreover, future work could also focus on some advanced PAR scanning techniques which can be used to enhance azimuth resolution (Bluestein et al., 2010; Schvartzman et al., 2021b).

Upon further refining of the wind field retrieval, this new radar system offers new capabilities for detecting fine-scale wind and intensity fields of severe convective events, which is promising to advance our understanding and nowcasting of severe storms, as well as the development of numerical weather prediction models.

Acknowledgements. Thanks are due to Professor Zhenhui WANG from Nanjing University of Information Science and Technology and Professor Xiaoyang LIU from Peking University for valuable suggestions; to Eastone Washon Science and Technology Ltd. for providing the AWR for field experiments; to graduate students Fangping LI, Wanyi WEI, and Yu LI from Chengdu University of Information Technology for their dedicated work on data collection and graphic production; and to Chuan LUO, Caiwen REN, Jingyi SUN, Shuyu ZHANG, Siwei LV, Wen YANG and others from Eastone Washon Science and Technology Ltd. for their dedicated work on AWR data collection and processing. This work is supported by Natural Science Foundation of China (NSFC) (Grant No. 31727901).