Evaluating the QPE of a dual polarization S band weather radar in Brazil

The first operational Brazilian dual polarization S band Doppler weather radar has been deployed in November of 2014 in São Paulo state to upgrade the existing McGill S band weather radar installed in 1989. By incorporating this new dual polarization technology of SELEX with an antenna with 1 degree beam width and gate resolution of 125 meters, the Department of Water Resources and Energy (DAEE) of the State of São Paulo expected to improve the quantitative precipitation estimation (QPE) from 2x2km2 to 0.5x0.5 km2 in a radius of 150 km. This new radar was configured to make 8 elevations scans (1.0, 1.6, 2.4, 3.2, 4.2, 5.5, 6.9 and 8.6 degrees) with a gate resolution of 250 m, 49 time samples and PRF of 550 Hz every 5 minutes. The QPE employed by this radar is the Surface Rainfall Intensity (SRI) product that is based on Ryzhkov et al. (2015) polarimetric algorithm. The SELEX SRI was configured to produce rain estimates every 5 minutes with a grid resolution of 0.5 x 0.5, 1x1 and 2x2 km2. As this product is produced operationally, it is possible to evaluate the performance of this QPE by comparing with an existing rain gauge network operated by DAEE. This network is composed of 108 tipping buckets that accumulate rain every 10 minutes and are transmitting in real-time to DAEE central computer. This rain gauge network is under the radar domain and it has several gauges every 20 km. As a result, all 5 minutes QPE 0.5 x 0.5 km2 products were accumulated to produce rain maps with a temporal resolution of 10 minutes. Using initially the summer time of 2015/2016, we compared the collocated rain gauges observations with the 10 minutes QPEs over the domain of 150 km radius. In this comparison we found that errors improved with time integration, i.e., at 10 minutes interval the correlation coefficient (CC) is 0.79, at 1 hour is 0.87 and at daily accumulations it reaches 0.91, while the RMS varies from 4.21mm, 2.37mm and 0.96mm. The BIAS did not vary and it remained constant with an underestimation of 30%. As Ryzhkov et all (2015) algorithm is based on 3 rain rate intervals, we computed the same errors as before but for each rainfall rate threshold. For the interval of 0-6 mm/h, CC varied from 0.79 at 10 minutes to 0.87 and 0.92 for hourly and daily intervals respectively. The RMS followed the same behavior, it was initially 4.21 at 10 minutes, 2.36 mm at 1 hour and 0.99 mm on daily scale. The bias was constant at -32%. For the interval of 6-50 mm/h interval, CC varied from 0.78 at 10 minutes to 0.78 and 0.85 for hourly and daily intervals respectively. The RMS followed the same behavior, it was initially 0.96 at 10 minutes, 0.96 mm at 1 hour and 0.83 mm on daily scale. The bias was constant at 5%. These results indicate that polarimetric correction improves the QPE performance. Finally we decided to evaluate the performance of the QPE as a function of the radar distance in order to check the beam broadening and height of illuminated volume. First we computed the rain rate probability density function (PDF) for each grid cell. Next, we extracted the rainfall rate values for the percentiles of 50%,90%,95% and 99% and plotted as a function of radar distance. As expected, the rain rate increased with distance for all percentiles and it was possible to observe the bright band. Once, the rain rate depends on radar reflectivity (Z), differential radar reflectivity (ZDR) and specific differential phase (KDP), we read the first and second elevation sweeps and computed the correspondent PDF for every 1 km and we constrained the values for rain rate interval of 0.5 mm/h and 50 mm/h. KDP and ZDR did not present any variation with distance from the radar for any percentile. Nonetheless, Z showed an increase with distance from the radar for all the percentiles with a slope of 0.05 dB/km. As a next step, we will adjust a function to correct Z with distance and run again the QPE algorithm to a different time period and check if this distance dependence can be corrected.