Improvement of accuacy of radar rainfall rate by real-time ClassZR

Quantitative rainfall intensity of Mt. Gwangdeok radar in Korea was estimated using real-time optimum Z-R relationship according to cloud types for several cases with rainfall over 30 mm for one day at observatory. For radar QPE, reflectivity was classified in the basis of the rainfall intensity of 10 mm/hr with vertical and horizontal reflectivity gradients and BBF(bright band fraction). Rainfall rate estimated by Marshall-Palmer relationship (MP) and rainfall rate by optimum Z-R relationship depending on precipitation type (ClassZR) were compared to investigate the reliability. As a result of correlation and RMSE for MP and ClassZR, it was shown that ClassZR had higher correlation and lower RMSE than MP. Through analysis of error distribution for MP and ClassZR, it was proved that underestimation rate was about 63% for ClassZR and 92% for MP and the reliability of ClassZR was higher than that of MP from distribution of average bias and scatter. For the evaluation of the accuracy improvement and validity of ClassZR, objective verification was performed and compared with RAR(Radar and AWS Rainrate). The performance of ClassZR was assessed by using five factors(Hit rate, CSI, POD, FAR and Bias) for judgment of rain/no rain and comparing time-series of rainfall intensity. In analysis of hit of rain/no rain, capability of ClassZR was highly evaluated as compared with MP. In analysis depending on threshold, the rate of decrement of ClassZR was lower than at MP. ClassZR with heavy rain over 300 mm/hr for one day had higher correlation and lower average bias and scatter as compared with RAR. The ratio of underestimation and overestimation was 6:4 for ClassZR and 3:7 for RAR. According to rainfall intensity, the reliability of ClassZR was highly proved in light rain and moderate rain, whereas, the reliability of RAR was higher in heavy rain. From the result of this study, rainfall estimation by ClassZR showed higher correlation with raingauge rainfall as compared with MP and RAR. ClassZR compensated the underestimation of MP and the overestimation of RAR. Therefore, algorithm of precipitation estimation by ClassZR considering precipitation type reduced errors from underestimation or overestimation and upgraded the quantitative accuracy of rainfall amount.