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 relationsh
ip development. Simulation studies were developed to compare the bias,
the variance and the root-mean-square-errors of nonparametric and par
ametric regressions. It was concluded that when an appropriate paramet
ric model can be determined, parametric regression is preferred over n
onparametric 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, C
anada. It was also found that nonparametric regression can be useful a
s a screening tool able to detect data deficient relationships.