Regional flood modelling in Western Australia : application of regression based methods using ordinary least squares

Flooding is one of Australia's costliest natural disasters, which on average costs around $AUD400 million annually. In 2010-2011 alone, the cost of flood damage has exceeded $AUD20 billion. Flood estimation is therefore crucial in assessing and managing flood risks. Flood Frequency Analysis (FFA) is one of the most commonly adopted techniques used to estimate floods with an associated frequency, known as design floods. These design floods are used in the planning and design of water infrastructure projects, along with various floodplain planning regulations. FFA largely relies on the existence of long recorded streamflow data. Due to the sheer size of Australia and the substantial costs involved, streamflow gauges are generally restricted to the highly populous and coastal regions of Australia. As design flood estimates are quite often needed in ungauged catchments, which have insufficient, unreliable or no streamflow data, an alternative method known as Regional Flood Frequency Analysis (RFFA) is generally adopted for design flood estimation. RFFA attempts to substitute the lack of temporal data with spatial data, to make more accurate flood estimates at ungauged sites. The most commonly adopted RFFA techniques include the Probabilistic Rational Method (PRM), the Index Flood Method (IFM) and the Quantile Regression Technique (QRT). The national guideline for design flow estimation, Australian Rainfall and Runoff (ARR) 1987, currently recommends a particular form of the PRM and the IFM for use in Western Australia (WA). The PRM has been widely criticised due to the simplistic assumption involved in plotting and interpolating the dimensionless runoff coefficients. At the same time, identifying homogeneous regions in connection with the IFM, has also proven to be problematic in Australia. Regression-based RFFA techniques have, however, shown promise in a number of recent studies across Australia and are the standard method in some other countries, such as the United States. The aim of the paper is to develop and compare two regression-based techniques: the QRT and the Parameter Regression Technique (PRT). In the QRT, individual flood quantiles are regressed against the catchment characteristics; while in the PRT, the parameters of a probability distribution (here the Log Pearson type 3 distribution is considered) are regressed against the catchment characteristics. An ordinary least squares regression method is used in this study for developing the prediction equations. The study uses streamflow and catchment characteristic data from 206 catchments across WA and part of the Northern Territory to develop regional prediction equations. The dataset was divided into three regions, according to the Australian Drainage Divisions (VI, VII and VIII), with 125, 12 and 69 catchments respectively. Independent testing of the QRT and the PRT was carried out on independent test catchments, which were randomly selected and were not used in the development of the prediction equations. It has been found that both the QRT and PRT provide reasonable predictions. The developed prediction equations are also relatively easy to apply as they only contain two predictor variables, being catchment area and design rainfall intensity. It should be noted that results from GLS regression for WA presented in Haddad et al. (2011b) should be more reliable than this study. The findings from this study and other relevant RFFA studies would form the basis of recommendation of a new RFFA for WA in the upcoming edition of ARR