Accurate hydrologic forecasts can inform and potentially improve water management decisions. This paper presents the development of an ensemble forecasting approach using a set of recurrent dynamic artificial neural network models having the Nonlinear AutoRegessive with eXogenous input (NARX) architecture. The multiple decisions that were made during the model-development process are summarized in a suggested methodology for practitioners. Application of the k-fold cross-validation and ensemble methodology to forecasting net inflow to Lake Okeechobee, Florida, is exhibited. The ensemble forecast method is demonstrated to provide better generalization performance and more-accurate forecasts than can be provided by a single NARX model.
