Systematic Approach for Variable Selection in the Weather Generator Model

This study provides an analytical procedure for selecting variables for the Weather Generator (WG) model to be used in the assessment of climatic change impacts at a local scale. The approach integrates the K-NN technique with the Principle Component Analysis (PCA), termed of WG-PCA. The Upper Thames River Basin in South-Western Ontario, Canada is selected as a case study. The outcome of the WG model includes 38 years of synthetically generated data. In the study we tested three models: (a) the WG-1Var employing only one variable (precipitation); (b) WG-3Var employing three variables (precipitation, maximum temperature, and minimum temperature); and (c) WG-PCA. The simulation results indicate that WG models employing three variables (WG-3Var and WG-PCA) produce the meteorological variables with better temporal and spatial correlation compared to the WG-1Var. More notable result is that the WG-PCA employing only the first principle component provides the comparable results to those of WG-3Var and much better results than WG-1Var.