Towards a new regional flood frequency analysis method for Western Australia

The development of a regional flood frequency analysis (RFFA) method for Western Australia (WA) poses a number of challenges: the region is very large and sparsely gauged, except for the South-west of the State, and in this more densely gauged part there is a significant variation of hydrological characteristics from coastal to inland areas. This paper describes how these challenges can be overcome by dividing the state into three different regions and by selecting the most appropriate RFFA method from a range of candidate methods. The database adopted for WA comprises data from 146 catchments. The State was divided into three distinct regions since there are concentrations of stream gauging stations in these three parts, which are separated by long distances. These three regions are (i) Kimberley region, with 14 stations (top part of WA from Drainage Division VIII) (ii) Pilbara region, with 12 stations (middle western part of WA, Drainage Division VII) and (iii) South-west WA, with 120 stations (Drainage Division VI). The analysis adopted in this study includes a Bayesian Generalised Least Squares (BGLS) regression approach in developing prediction equations for selected flood quantiles (Quantile Regression Technique – QRT) or for the first 3 moments of the LP3 distribution (Parameter Regression Technique – PRT). The BGLS regression model explicitly accounts for the sampling variability in the dependent variable data e.g. inter-station correlation and variation in record lengths from site to site, and offers a better alternative to the most commonly adopted Ordinary Least Squares regression