Analysis of radar-rainfall error characteristics and implications for streamflow simulation uncertainty

Due to the inherent indirect nature of radar-rainfall measurements, hydrologists have been interested in understanding the characteristics of radar-rainfall estimation errors and how they propagate through hydrological simulations. This study implements an observation-based empirical approach to analyse different characteristics of the total radar-rainfall estimation error such as overall and conditional bias, random error and spatio-temporal dependence. The implications of the radar error characteristics for streamflow simulations and the estimation of their uncertainty are examined using a physically-based distributed rainfall-runoff model. An empirical error model is used to generate several realizations of probable surface rainfall fields that reflect the identified characteristics of the radar error. These realizations are used to generate ensemble of streamflow predictions. The main conclusions are that: (a) radar errors have complex spatio-temporal characteristics that exhibit significant sampling and natural variations; (b) adjustment of overall and conditional radar biases results in the most significant improvements in runoff predictions; (c) radar random errors have non-negligible correlations both in time and space; and (d) the simulated runoff hydrographs are sensitive to the assumed degree of correlation in the radar errors fields. This study is an initial step toward developing more rigorous approaches for accounting for the effects of radar error on hydrological predictions.