Abstract
This paper analyzes errors of X-band MP radar rainfall and proposes a method to stochastically generate rainfall fields reflecting the error structures. By introducing multivariate AR model with Cholesky decomposition, the presented method reflects both space and time correlations, the latter of which could not be represented in a previous method. The proposed method is applied to Typhoon events in the Chikusa River Basin to generate conditioned random rainfall fields to analyze the uncertainty in rainfall-runoff predictions. The analysis indicated that the average relative errors in X-band MP radar rainfall are about 0.3 to 0.4 within 40 km range from radar sites. The errors propagate into the uncertainty in peak discharge estimates with about 0.1 of coefficient of variation. Further analysis with assumed different errors in random rainfall fields and their impact on streamflow predictions is also presented.
