Predicting snowmelt runoff using a deterministic watershed model with stochastic precipitation inputs

Abstract

The accuracy of currently used long-term runoff forecasting techniques, such as used by the Soil Conservation Service, are limited because of their inability to deal with the uncertainty in the amount of precipitation expected to fall after the forecast date. The basis for a simulation-based, long-term runoff forecasting technique is developed to overcome this problem by simulating future precipitation events. The technique utilizes a deterministic watershed snowmelt model and a sequence, event-based stochastic precipitation model to provide daily precipitation data inputs for the watershed model. A number of sets of inputs are run through the watershed model to produce an equal number of predictions of total seasonal runoff. A relative frequency distribution of total seasonal runoff is then plotted to which a PDF may be fitted. Various criteria were used to test the precipitation model. The majority showed no significant differences between the observed and simulated data. The lack of data prevented reasonable watershed model optimization and testing. Taking into consideration the poor watershed model response the forecasting technique responded well to the uncertainty in future precipitation and to abnormal monthly precipitation.