Precipitation Simulation Based on k-Nearest Neighbor Approach Using Gamma Kernel

Abstract

This paper presents a weather generator that produces new values of precipitation to generate realistic weather sequences. The model has been applied to a network of 14 meteorological stations around the Upper Thames River Basin (UTRB), Ontario, Canada. We developed a simple model that employs the k-nearest neighbor resampling approach with gamma kernel perturbation. This gamma kernel perturbation enables the production of new values rather than merely reshuffling the historical data to generate realistic weather sequences. Daily precipitation was simulated at all the locations in and around the considered basin. The comparison of simulated data to the observed data led to the conclusion that the proposed perturbation algorithm performs quite well at preserving the monthly and annual historical statistics. The improved model was shown to produce precipitation amounts different from those observed in the past record.