A statistical approach to radar rainfall estimates using polarimetric variables
2018
Abstract To improve the accuracy of radar rainfall estimates, this study examines rainfall relationships derived from polarimetric variables calculated from Drop Size Distributions (DSDs) measured by POSS (Precipitation Occurrence Sensor System) and PARSIVEL (PARticle Size and VELocity) disdrometers for eight different rainfall events in Korea associated with the Changma front, low pressure systems, typhoons, or the indirect effects of typhoons. Analysis of the correlation coefficients between polarimetric variables as independent parameters shows that multicollinearity is expected for Z–K DP , Z–A H , and K DP –A H . Of these, R(Z, K DP ) is the only relationship that had too low an accuracy for application to radar rainfall estimation. R(Z, Z DR , K DP ) and R(Z, K DP , A H ) were also omitted from this analysis because their intercept coefficients were too large. Analysis of the sensitivity of radar rainfall estimation to DSDs variation shows that the latest observed DSDs perform well, as much as 2.4 mm h −1 for RMSE (Root Mean Square Error) and 0.23 for NE (Normalized Error) in maximum. The statistical scores of each radar rainfall estimator vary between different rainfall events. This paper examines a new approach to radar rainfall estimation that is similar to the ensemble technique widely used in numerical prediction models. The ensemble members were chosen based on the average and standard deviation of their RMSE and NE for eight rainfall events. Two different weighting schemes were applied to each ensemble and the members were weighted equally or, alternatively, weighted based on their statistical scores. The performance of eight ensemble sets was examined using four independent rainfall events. There is little difference in the accuracy of each ensemble with respect to the weighting scheme applied. An ensemble composed of R(Z,Z DR ), R(Z), and R(K DP ), all given an equal weighting, was the most accurate.
Keywords:
- Correction
- Source
- Cite
- Save
- Machine Reading By IdeaReader
31
References
3
Citations
NaN
KQI