Robust Techniques for Extension of Streamflow Records

by Sri Rangarajan,

Document Type: Proceeding Paper

Part of: WRPMD'99: Preparing for the 21st Century


Longer hydrological records are required for the planning of water resource systems. Sensitivity of a system is studied under various scenarios of hydrological inputs to understand its performance, and diligently select the operation parameters. For Manitoba Hydro, a hydraulic-based electrical utility, availability of longer records is critical for assessing the system reliability, and in establishing firm export contracts with other companies. Some of the major rivers have records dating back to 1912. However, some of the tributaries, that contribute about 20% of inflow, do not have concurrent records. Another interesting aspect is that the Environment Canada has begun since 1994 to reduce the number of stations or to transfer the monitoring responsibility to water users such as Manitoba Hydro. This situation, which will be the reality of the 21st century, poses the following challenges: (1) developing concurrent records in all the tributaries; and (2) assessing the value of information obtained from the hydrometric stations, and selecting the vital stations. Three robust statistical procedures are proposed to assist the utility in predicting relevant hydrological information. Two of these procedures, namely the spatial regression (SR) and generalized MOVE (GMOVE), are parameter-based, and the third, the artificial neural network (ANN), is a non-parametric procedure. All the three procedures use the data of an index hydrometric station for extending the flow records in other stations. The three procedures were compared in terms of the mean-squared error between the observed and predicted records. ANN was superior for predicting flow records in some tributaries, whereas the other two procedures performed well in other stations. Guidelines are provided for selecting the appropriate extension procedure(s) for individual tributaries in the system.

Subject Headings: Streamflow | Hydrology | Parameters (statistics) | Neural networks | Water resources | Utilities | System reliability | Hydraulics | Canada | Manitoba

Services: Buy this book/Buy this article


Return to search