A comparison of conventional and wavelet transform based methods for streamflow record extension

Abstract

Sufficient hydrological data, such as streamflow, are important to represent the long-term characteristics of a watershed in order to support decision-making, policy and management. Lack of data remains one of the main challenges of hydrological analyses. Therefore, the main goal of this study is to extend streamflow records using a proposed approach where the wavelet transform (WT) is incorporated as a pre-processing method into eight existing record extension techniques, namely the ordinary least-square regression (OLS), maintenance of variance (MOVE) types 1–4, Kendall-Theil robust line (KTRL), KTRL2 and robust line of organic correlation (RLOC). The performance of the WT-based methods in estimating individual data values, means and standard deviations of the extended records, and a series of percentiles of the extended records was then compared with that of the respective conventional methods (i.e., without the WT). The data used in the analysis consisted of 67 pairs of target-index stations that were obtained from Canada’s Reference Hydrometric Basin Network database, all of which contain outliers. As such, the results and discussions obtained are applicable for cases where the data contain outliers. The WT-based methods (particularly the KTRL-WT, KTRL2-WT and RLOC-WT) demonstrated consistent improvements in precision and accuracy compared with their conventional counterparts, especially in estimating the means and standard deviations of the extended records. In estimating individual data values, the WT-based methods showed inconsistent improvements. Finally, in terms of percentiles, greater improvements were seen in the estimation of higher percentiles, more specifically for the MOVE1-WT, MOVE2-WT, KTRL2-WT and RLOC-WT compared with their conventional counterparts.