Regional flood estimation in Australia: Application of gene expression programming and artificial neural network techniques

Flood damage can be minimised by ensuring optimum capacity to drainage structures. An underdesign of these structures increases flood damage cost whereas an overdesign incurs unnecessary expenses. The optimum design of water infrastructures depends largely on reliable estimation of design floods which is a flood discharge associated with a given annual exceedance probability. For design flood estimation, the most direct method is flood frequency analysis which requires long period of recorded streamflow data at the site of interest. This is not a feasible option at many locations due to absence or limitation of streamflow records; hence regional flood estimation methods are preferred. Regional flood frequency analysis (RFFA) involves transfer of flood characteristics from gauged to ungauged catchments. The RFFA methods are widely used in practice. In the past, different RFFA methods have been proposed for Australia, which are based on linear models such as Probabilistic Rational Method (PRM) and index flood method. More recently, regression-based methods have been investigated for Australia, which are also log-linear models. There have been successful application of non-linear models like Artificial Neural Networks (ANN), Gene Expression Programming (GEP) and Fuzzy based methods in hydrology in some other parts of the world. However, there has not been any notable application of these methods in RFFA study in Australia. This paper focuses on the application of the ANN and GEP to regional flood estimation problems in Australia. The GEP approach used in this study provides an integrated mechanism for the identification of the optimum hydrological regions for RFFA study in eastern Australia. In the preliminary study, optimum regions were obtained based on geographic and state boundaries, climatic conditions and catchment attributes. The proposed approaches were applied to 452 stations in the eastern Australia. Results depict that the GEP and ANN approach have a much better generalization capability of RFFA problems. An independent test has shown that the ANN based model provides more accurate flood quantile estimates than the GEP. Overall, the best ANN-based RFFA model is achieved when all the data set of 452 catchments are combined together to form one region, which gives an ANN-based RFFA model with median relative error of 35% to 44% and median ratios (of predicted and observed values) of 0.99 to 1.14.