Regional flood estimation technique for NSW : application of generalised least squares quantile regression technique

As a part of the ongoing revision of the Australian Rainfall and Runoff, a number of techniques have previously been tested for regional flood estimation using data from Victoria and Queensland. The results from these studies indicate that quantile regression technique (QRT) based on generalised least squares (GLS) regression can provide quite accurate design flood estimation. This paper develops and tests QRT-GLS method for NSW using data from 96 catchments. The paper initially presents the streamflow data preparation method including infilling gaps, outlier, trend and rating curve error analyses. Flood prediction equations are then developed using the QRT-GLS method. Independent testing showed that QRT-GLS method can provide design flood estimates quite accurately in that the average variance of prediction and standard error of prediction values are quite reasonable with median relative error values in the range of 13 to 40%. The QRT-GLS method on average predicted a quantile with an accuracy of prediction equivalent to an average record length of 74 years for eastern NSW and 65 years for western NSW. The developed prediction equations are relatively easy to apply, requiring data for three independent variables, catchment area, design rainfall intensity and stream density.