Forecasting Water Yield Using Multivariate Analysis

In this paper, the effectiveness of principal-component regression in the forecasting of spring season water yield was investigated. A general model building scheme was adopted to aid in model development. A detailed residual analysis, adoption of split-sampling, analysis of the rationality of regression coefficients, and the analysis of the indicators of the relative importance of the independent/predictor variables were performed. Models developed by the regression of principal-components proved superior to the models derived by the traditional regression techniques. The study was applied to a multi-reservoir hydro-electric system in the Saguenay-Lac St-Jean region of Quebec, Canada.