가뭄저유량의 기후영향을 고려하기 위한 KNN 비매개변수 시뮬레이션

Nonparametric simulation with KNN local linear regression for the climatic influence of low-flow extremes

Stochastic simulation of hydroclimatic variables reproducing irregular behaviors are beneficial in assessing their impacts to other regimes. In the current study, KNN-based local linear regression method (KLR), which is a nonparametric stochastic simulation model, is presented to reproduce nonlinear and heteroscedastic relation in hydroclimatic variables. The results show that the key statistics and nonlinear dependence of the tested data is well reproduced in the simulated data from the KLR model as well as heteroscedascity whereas the traditional resampling technique KNNR presents some biases in key statistics such as variance and lag-1 correlation. Furthermore, the Min7D flow of the Romaine River, which is critical in providing hydropower to users, was simulated with the KLR model for the observed period and the future period incorporating the observed and extended North Atlantic Oscillation index. The results of the observed period illustrates that the KLR model reproduces well the key statistical characteristics and nonlinear heteroscedastic relation.