Artificial Neural Networks Based Regional Flood Estimation Methods for Eastern Australia: Identification of Optimum Regions

摘要

In Australia, design flood estimation in smaller ungauged catchments is often carried out using the rational method. Recently, application of quantile regression technique has been investigated in Australia. In contrast to these traditional methods, Artificial Neural Networks (ANNs) can be applied to regional flood frequency analysis (RFFA). The ANNs do not impose a model structure on the data and can better deal with non-linearity of the input and output relationship. This paper focuses on the development and testing of the ANNs based RFFA methods for eastern Australia. A number of alternative regions are tested e.g. (a) each of the states of NSW, Victoria, Queensland and Tasmania is considered to be a separate region; (b) all these states form one region; and (c) summer and winter dominated parts of these states form two separate regions. Independent testing shows that option (b) is the best performing and can provide quite accurate design flood estimates with a median relative error values in the range of 39% to 56%.