Statistical forecast model for Libby basin, Montana

A statistical procedure has been developed that significantly improves the inflow volume forecast for the basin above Libby Dam, Montana. The criteria used to assess the forecast performance is the standard error of the first-of-month forecast of the April-August runoff volume. The previous model fitted a multivariate linear regression model to the observed April-August runoff using lumped values derived from observed snow, rain, and antecedent runoff variables. During the forecast season station averages were used as surrogates for unavailable variables. The factors in the new procedure instrumental in reducing the standard errors were 1) restructuring from a single-season regression equation for two subbasins to a single basin best-fit model developed for each date-of-forecast, 2) employment of a 'best correlation to residuals' scheme to aid in variable selection, and 3) use of principal components regression to develop the final model, presented in terms (etc).