Development of a regional flood frequency estimation model for Pilbara, Australia

![CDATA[This paper presents the development of regional flood frequency estimation (RFFE) models for Pilbara, Western Australia as a part of Australian Rainfall and Runoff (ARR) revision. The Pilbara is an economically important arid region, which is rich in natural resources and mining activities. A total of 11 catchments are selected from the Pilbara region to develop the RFFE models. The selected catchments are arid with average annual rainfall values are in the range of 200 and 350 mm. For each of the selected catchments, flood quantiles are estimated for 6 annual exceedance probabilities (AEPs), which are 50%, 20%, 10%, 5%, 2% and 1% using two estimation approaches: (i) annual maximum flood series (AMS): Log Pearson Type 3 (LP3) distribution-Bayesian procedure and (ii) partial duration series (PDS) (with average number of events per year equal to 0.5): Generalised Pareto distribution and L moments procedure. The AMS consists of maximum discharge value in each calendar year; however, the PDS consists of all the independent peak flows over a certain threshold. Two different RFFE models are developed (using AMS and PDS data) using an index flood method where the 10% AEP flood quantile (Q10) is used as the index variable and a dimensionless growth factor for AEP of x% (GFx) is used to estimate Qx (where, x = 50%, 20%, 10%, 5%, 2% and 1%). The prediction equation for the index flood is developed using an ordinary least squares regression with two predictor variables (catchment area and design rainfall intensity). The results demonstrate that the performance of the two RFFE models is quite similar. For AEPs of 50% to 1%, the absolute median relative error values of the estimated flood quantiles are found to be in the range of 23% to 46% and 42% to 56% for AMS and PDS based RFFE models, respectively. The results reveal that both the models are able to generate flood quantiles with a good accuracy; however the AMS-based RFFE model outperforms the PDS-based model.]]