ABSTRACT
This chapter analyzes the assumption that data in a given system arise from a single-parent distribution may be questionable when data from large watersheds. The concept of robust flood frequency models is introduce by Kuczera found that regionalized estimates were preferable to estimates based on short record lengths and estimates which combined both site and regional information were preferable for larger record lengths. Cox regression model was used by J. A. Smith and A. F. Karr for flood frequency analysis. These models allowed incorporation of time varying exogenous information into flood frequency analysis. The Hydrocomp Simulation Program was most successful in reproducing flood frequency curves determined from historic streamflow records. The distribution of simulated floods match that of observed floods. The bias in computed flood risk was discussed by Hardison and Jennings recommended that the accuracy of procedures used in flood frequency analysis be appraised in terms of standard errors.
