Parametric and Non-Parametric Model Performance Differences in Estimation of Critical Stream Flow Discharge Levels

Kiarie F. K.

Abstract: Nonparametric estimation based on quantile regression methodology of Koenker and Basset (1978) and conventional parametric regression approaches were applied to a river regime to estimate volatility in streamflow discharge levels. Consistency and asymptotic normality properties of estimators obtained from both approaches were given. From the study results nonparametricquantile regression approach yielded better results than other methods. Other than for boundary effects which require boundary modifications, the model validation results implied good performance of the nonparametric model in estimating critical streamflow discharge levels.

Keywords: Kernel estimate, quantile autoregression, consistency, asymptotic normality, critical discharge level