NEAREST-NEIGHBOR METHODS FOR NONPARAMETRIC RAINFALL-RUNOFF FORECASTING

This paper introduces a nonparametric regression method for time series to the stochastic hydrology literature. The technique, known as the nearest-neighbor (NN) method, is motivated by the classical rainfall-runoff forecasting problem. We have shown in a companion paper (Yakowitz, 1987) that in the time series setting it has attractive large-sample properties. As summarized herein, these include asymptotic convergence to the optimal forecaster under many circumstances in which the forecaster is not a linear function of the data. The technique is applied to a sequence of about 3000 seasonalized rainfall-runoff pairs from the Coshocton Watershed.