Streamflow data preparation for regional flood frequency analysis : important lessons from a case study

Regional flood frequency analysis requires flood data series of reliable quality and reasonable length at well distributed sites over a relatively homogeneous region. The challenge posed in collating a database for regional flood frequency analysis lies in maximising the amount of useful flood information, while minimising the 'noise' that may be present in some flood series. Flood data from a number of sites may be unrepresentative, the useful length of a flood series may be affected by missing records, and noise may have been introduced by outliers, trends and rating curve extrapolation errors. This paper presents a case study for the state of Victoria involving 415 stations across the state, initially selected based on catchment size, record length (>10 years), general streamflow data quality, degree of regulation, urbanisation and land use change. Further examination indicated that many stations did not satisfy the criteria of homogeneity and representativeness for the purposes of regional analysis. Furthermore, to reduce the potential effects of inter-decadal variability, the minimum length of records (after infilling of some missing data) was increased to 25 years. Trend analysis indicated the presence of a downward trend for many stations after the late 1980s. Finally, the influence of errors from the extrapolation of rating curves was minimised by placing limits on the degree of extrapolation required to estimate the largest observed events. At the conclusion of all these tests, only 133 stations were left in the database for development of a regional flood frequency method, which represents only one-third of the initially selected 415 stations. The paper highlights the importance of careful screening of flood data to ensure that each data series to be used satisfies the stringent requirements of the intended regional flood frequency analysis method.