A new method to produce categorical streamflow forecasts

Abstract

Categorical forecasts of streamflow are important for effective water resources management. Typically, these are obtained by generating ensemble forecasts of streamflow and counting the proportion of ensembles in the desired category. Here we develop a simple and direct method to produce categorical streamflow forecasts at multiple sites. The method involves predicting the probability of the leading mode (or principal component) of the basin streamflows above a given threshold and subsequently translating the predicted probabilities to all the sites in the basin. The categorical probabilistic forecasts are obtained via logistic regression using a set of large-scale climate predictors. Application to categorical forecasts of the spring (April-June) streamflows at six locations in the Gunnison River Basin exhibited significant long-lead forecast skill. Copyright 2006 by the American Geophysical Union.