Real-time Quality Control of Phased Array Weather Radar Data Observed Every 30 Seconds

The X-band phased array weather radar (PAWR) installed at Osaka, Kobe, and Okinawa in Japan are steadily operated 24 hours a day. The aim of the PAWR is to detect a warning sign of torrential rainfall and to predict weather hazard using the fine and dense three-dimensional (3D) radar data observed every 30 seconds. We have been conducting researches of the PAWR data assimilation (DA) for weather forecast and 3D nowcast. We are also publishing real-time quick-look images on a web page (http://pawr.nict.go.jp/), and are providing the real-time observation data for a smartphone application (http://pawr.life-ranger.jp/). It is important to process data quality control (QC), because the observation data includes various unnecessary echoes, which are surface clutter, interference noise, and other low-quality data due to rain attenuation etc. Although we have already developed the QC algorithm of PAWR data for the DA experiment, the calculation takes about 40 seconds. We need faster QC algorithm for not only DA but also other real-time application, the required computation time is less than 10 seconds.
The QC information of PAWR data are provided as a 8-bit (1-byte) flag file in the same format as the original raw data. The QC flag file in the polar coordinate can be used for various general purposes. The QC flags are [0] Valid data, [1] Shadow, [2] Clutter possible, [3] Clutter certain, [4] Noise, [5] Rain attenuation, [6] Range side-lobe, and [7] (Reserve). The first flag [0] shows valid data for both reflectivity and velocity. The shadow flag [1] is calculated from Digital Elevation Map (DEM) of GSI of Japan using a 4/3 equivalent earth radius model. The surface clutter flags [2] and [3] are main results of this QC algorithm. These two kinds of the clutter flags will be used for different purposes. The clutter possible flag [2] excluding doubtful data is useful for DA, while the clutter certain flag [3] with few missing data is useful for nowcast and visualization. Surface clutter echoes are judged by a statistical clutter map, speckle filter, texture of the reflectivity, and Doppler velocity data. The different coefficients of texture and velocity for convective and stratified rain are used to distinguish clutter from meteorological echoes. Information of vertical gradient and time correlation of reflectivity are not used in the current program to save calculation time. The noise flag [4] in the current version corresponds to only interference noise echo, which penetrates to a radial direction. The rain attenuation flag [5] is made from the typical X-band K-Z relationship of convective and stratiform rainfall respectively. The flag [6] is to indicate range side-lobe contamination. The false echoes sometimes appear at the back and forth of a very strong echo in the long-pulse region of the PAWR.
Figure 1 shows an example of reflectivity distribution in a PPI of a low elevation angle (2.0 degrees) and the QC flags corresponding to the observation data. Although the original QC flags are independent of one another, large number flags in the figure overwrites small number flags. In this example, the clutter possible flag (>4) is the same as the statistical clutter map. Therefore, only the clutter certain flag (>8) distribution shows actual judgment clutter. Most of the clutter possible region must be real rainfall echoes, but a circular region on the northwest side may be surface clutter. The calculation processing time of the current QC program is 5 to 8 seconds using one CPU core. Since the processing time depends on the echo area, the computation time is required for stratiform rain cases. By using multiple CPU cores, we have allowance to add new algorithm and parameters to improve the QC accuracy. Ultimately, it is necessary to balance computation time with QC accuracy, but the prospect of real-time processing has been achieved.