Evaluation of Streamflow Forecasting Models

Abstract

For application purposes, it is no longer a sound investment to develop a streamflow forecasting model from basics. Currently, streamflow forecasting models are available for nearly every scenario one can imagine. A model could be stochastic or conceptual; lumped parameter or distributed parameter. The task of developing a model has been transferred to one of evaluation and selection since no single model can be applied universally without sacrificing some element of its performance. Therefore, it is necessary to have some kind of consensus as to how forecasting models are evaluated and selected for each individual application. In the past, the evaluations were often conducted by comparing the forecasted and the observed streamflows with numeric and/or graphic criteria with little consideration given to the specific application. However, in this study, forecasting models are evaluated through simulated real-time applications to investigate which one maximizes the system performance.