Roy W. Koch

Date of Award


Document Type


Degree Name

Doctor of Philosophy (Ph.D.) in Systems Science / Civil Engineering


Systems Science

Physical Description

3, xi, 133 leaves: ill. 28 cm.


Water-supply -- West (U.S.) -- Forecasting -- Simulation methods, Runoff -- West (U.S.) -- Forecasting -- Simulation methods




Water supply forecasting in the western United States is the prediction of the volume of water passing a given point on a stream during the primary snowmelt runoff season. Most water supply forecasts are produced from multiple linear regression models using snowpack, precipitation, and streamflow measurements as independent variables. In recent years, conceptual watershed simulation models, typically using a time step of one day, have also been used to produce these forecasts. This study examines model usage for: water supply forecasting in the West and has three specific purposes. The first is to examine the traditional usage of multiple linear regression and develop improved regression techniques to overcome several recognized weaknesses in traditional practice. Four techniques have been used in this study to improve water supply forecasts based on regression. They are: (1) basing the regression model only on data: known at forecast time (no future data); (2) principal components regression; (3) cross-validation; and (4) systematic searching for optimal or near-optimal combinations of variables. The second purpose of the study is to develop a monthly streamflow simulation model suitable for use in water supply forecasting. Such a model has not previously been used in this application, and it provides a forecasting tool midway in complexity between regression procedures and conceptual watershed simulation models. The third purpose of the study is to compare the accuracy of forecasts from regression, the monthly model developed here, and two conceptual watershed simulation models. It has generally been assumed, but not tested, that complex simulation models will give more accurate forecasts than simpler models. This study attempts to begin determining if this is true. Conceptual modeling results from previous studies on three basins in Idaho and Montana were obtained to represent current practice in the usage of this type of model. The results of the study led to the following conclusions: (1) significant improvements in forecast accuracy over past practice with regression can be obtained by the use of the four techniques developed here; (2) the monthly model performed better than the conceptual watershed models most of the time, for both seasonal volumes and monthly flows; (3) for the three test watersheds, regression provided the best forecast accuracy among the three modeling techniques most of the time, for both seasonal volumes and monthly flows; (4) optimal use of conceptual watershed models requires automated calibration schemes; and (5) in basins of complex orography, denser data networks will be required to calculate meaningful values of mean areal precipitation. This study has contributed to the practice of water supply forecasting by providing improvements to regression techniques, providing a new monthly model, developing a mean areal precipitation and temperature procedure based on kriging, and giving some initial direction for further investigations in the use of conceptual watershed models. The inability of the two simulation approaches to surpass regression in forecast accuracy brings up several issues with respect to modeling. These issues are in the areas of model calibration, model conceptualization, spatial and temporal aggregation, and areal averaging of input data. Further investigation is required to elucidate these issues before clear conclusions can be made about the relative forecasting abilities of simple and complex models. Further investigation is also required to study water management decision making and the kinds and accuracies of forecast information required to optimize these decisions.


If you are the rightful copyright holder of this dissertation or thesis and wish to have it removed from the Open Access Collection, please submit a request to and include clear identification of the work, preferably with URL

Persistent Identifier