An Evaluation of the Performance of an Advanced Approach of the K-Nearest Neighbour in Simulating the Daily Meteorological Data

Weather data generators (WGs) have been developed for an extension of time series of such weather variables as rainfall, temperature and relative humidity to provide better understanding of systems affected by climatic factors. Different algorithms have been applied in these generators, broadly divided into parametric & non-parametric ones. In this study, the performance of non-parametric generator of K-nearest neighbor (KNN) with the capability of extrapolating data, has been evaluated in six synoptic stations of Iran namely; Tehran, Qazvin, Mashhad, Bushehr, Tabriz and Rasht for the period of 1961-2005. Besides, some of the obtained results have been compared with parametric generator of LARS-WG to show the priority of this approach to parametric methods. The results revealed that in most cases the KNN approach presents a better performance in simulating the parameters of observed series; however, LARS-WG better performs in simulating length of wet and dry spells but with minor differences.