Synthetic Streamflow Forecast Generation

Abstract

Simulation models have proven a durable and useful tool for the development of water resource system operating policies. One important component of any operating policy is streamflow forecasting. Incorporation of streamflow forecasts in a simulation model can be difficult because of the stochastic nature of forecasts and the range of forecast models in use. In this work, a forecast generation algorithm is developed which is applicable for any forecast model, and requires as input only the coefficient of prediction, a measure of forecast accuracy, and the first three moments of the forecast period flows. The algorithm uses a lag one Markov structure for forecast errors, where the correlation coefficient is a function of the forecast model coefficient of prediction. The generation algorithm is demonstrated for simulation of 7 and 120 day forecasts for the Skykomish River, Washington.