Regional Flood Relationships by Nonparametric Regression

Since some theoretical assumptions needed in linear regression are not always fulfilled in practical applications, nonparametric regression was investigated as an alternative method in regional flood relationship development. Simulation studies were developed to compare the bias, the variance and the root-mean-square-errors of nonparametric and parametric regressions. It was concluded that when an appropriate parametric model can be determined, parametric regression is preferred over nonparametric regression. However, where an appropriate model cannot be determined, nonparametric regression is preferred. It was found that both linear regression and nonparametric regression gave very similar regional relationships for annual maximum floods from New Brunswick, Canada. It was also found that nonparametric regression can be useful as a screening tool able to detect data deficient relationships.