American Society of Civil Engineers


Comparative Studies in Problems of Missing Extreme Daily Streamflow Records


by W. W. Ng, (Grad. Student, Dept. of Civ. Engrg., Univ. of Waterloo, Waterloo, ON, Canada N2 L-5E1. E-mail: wng@lakeheadu.ca), U. S. Panu, (corresponding author), F.ASCE, (Prof., Dept. of Civ. Engrg., Lakehead Univ., 955 Oliver Rd., Thunder Bay, Canada ON, P7B-5E1 E-mail: umed.panu@lakeheadu.ca), and W. C. Lennox, (Prof. Emeritus, Dept. of Civ. Engrg., Univ. of Waterloo, Waterloo, ON, Canada N2 L-5E1. E-mail: wclennox@uwaterloo.ca)

Journal of Hydrologic Engineering, Vol. 14, No. 1, January 2009, pp. 91-100, (doi:  http://dx.doi.org/10.1061/(ASCE)1084-0699(2009)14:1(91))

     Access full text
     Purchase Subscription
     Permissions for Reuse  

Document type: Journal Paper
Abstract: This study evaluates the performance of different estimation techniques for the infilling of missing observations in extreme daily hydrologic series. Generalized regression neural networks (GRNNs) are proposed for the estimation of missing observations with their input configuration determined through an optimization approach of genetic algorithm (GA). The efficacy of the GRNN-GA technique was obtained through comparative performance analyses of the proposed technique to existing techniques. Based on the results of such comparative analyses, especially in the case of the English River (Canada), the GRNN-GA technique was found to be a highly competitive method when compared to the existing artificial neural networks techniques. In addition, based on the criteria of mean squared and absolute errors, a detailed comparative analysis involving the GRNN-GA, k-nearest neighbors, and multiple imputation for the infilling of missing records of the Saugeen River (Canada), also found the GRNN-GA technique to be superior when evaluated against other competing techniques.


ASCE Subject Headings:
Algorithms
Neural networks
Streamflow
Comparative studies