Streamflow data preparation for trend analysis : a case study for Australia

![CDATA[The accuracy of the outcome of flood quantile estimation and trend analysis largely depends on the quantity and quality of the available streamflow data. Various types of quality issues in streamflow data can be found such as gaps in the data series, a short record length of the data, and outliers in the data. The outcome of streamflow data analysis can be influenced by the ways in which the missing values and outliers are processed. Therefore, it is of utmost importance to scrutinise the collected streamflow data series to ensure that they are suitable for trend analysis. This study considered Australian catchments and collected annual maximum flood (AMF) series data from a large number of stations all over Australia. This study uses several steps to check data quality to minimise errors in collected AMF data. The collected AMF data has the quality code against each data point assigned by the streamflow gauging authority. These data points are checked and if appropriate, gauging stations with poor-quality coded data are excluded from this study. The presence of any outliers in the AMF series is identified. In this study, multiple Grubbs-Beck tests are adopted. Gaps in the AMF series are infilled using different methods based on the availability of the monthly instantaneous maximum (IM) data, monthly maximum mean daily (MMD) data of the missing year and the availability of the missing year’s AMF data of nearby stations. Where IM and MMD data for the missing year are available, the gap is filled by comparing the IM data with MMD data of the same station for the year with data gaps. Where annual maximum mean (AMD) flow is available and AMF data is missing then missing AMF data is estimated using regression between the AMD series against the AMF series of the same station. The coefficient of determination R2 of this regression is found between 0.9 to 0.99. Where IM and MMD are not available, simple linear regression is used between the station of missing AMF and the nearby station’s AMF where the missing year's AMF is available to fill up the gap. A total of 676 stations are initially selected each having a minimum record length of 20 years. For in-filling gaps in AMF series, priority is given to the first approach followed by the second and third as appropriate. Finally, 307 stations are selected with minimum record of 50 years. Twenty-three (7%) out of 307 stations are found to have missing points/gaps. These data will be used to examine trends and in non-stationary flood frequency analysis.]]