Seasonal Streamflow Forecasts Based Upon Regression

by Jery R. Stedinger, Cornell Univ, United States,
Jan Grygier, Cornell Univ, United States,
Hongbing Yin, Cornell Univ, United States,



Document Type: Proceeding Paper

Part of: Computerized Decision Support Systems for Water Managers

Abstract:

Many western water-supply system operators employ forecasts of seasonal inflows for the snow-melt season to plan spring and summer operations. This study considers the simple statistical forecasting models employed by most agencies. For several streams in Northern California, linear functions of snowpack water content, precipitation, and previous flows did as well at forecasting seasonal streamflows as models which used nonlinear functions of those variables. Linear regression models did better than W.V. Tangborn and L.A. Rasmussen's hydrometerological (HM) model at seasonal runoff prediction. Also considered was the common practice of including in regression forecasting models 'future' precipitation variables (whose values would not be known on the forecast date), and then substituting mean values in their place. The honest adjusted R² for those models were generally lower than those of the corresponding regression models derived without future precipitation variables; in several cases, use of future precipitation variables led to selection of models without the best variables for prediction, and hence to forecasts with larger prediction errors.

Subject Headings: Regression analysis | Forecasting | Errors (statistics) | Seasonal variations | Mathematical models | Hydrologic models | Streamflow | California | United States

Services: Buy this book/Buy this article

 

Return to search