Comparison of parametric and nonparametric streamflow record extension techniques
MetadataShow full item record
The extension of short data records, based on information from long term data records, is a common procedure used in the planning and operation of many water resources systems. Alternative methods for extending the available streamflow data record at locations where the period of recorded data is considered too short are presented. Various deficiencies in existing regression-based parametric techniques related to the assumption of normally distributed and random residuals are id ntified. An alternate nonparametric approach which is not subject to the above assumptions is presented. The nonparametric method utilizes the relationship between the index and base record to identify similar flow patterns that can be used to generate streamflow data. The extension techniques were evaluated, and the results of the evaluation were verified, using monthly streamflow data from gauging stations in Manitoba and Ontario. The techniques are evaluated based on their relative performance in reproducing statistical features of the historic data. The parametric and nonparametric methods displayed comparable performance. The residual series of the parametric models did not follow the normal distribution, even though a data transformation was performed. Residual series from both techniques displayed autocorrelation, indicating the inability of the models in taking into account time varying relationships in the data. Model performance generally increased with common period of record. The nonparametric methods tended to improve as the available data increased. Recommendations are made as to the preferred approach under varying data availability conditions. The nonparametric techniques are recommended as a viable alternative in cases where the residual series obtained from the parametric models are not normally distributed. Using a nonparametric model as an alternative to a parametric model may involve a trade-off in terms of statistical performance under certain conditions. A procedure for implementing the record extension techniques is presented.