Mid-short-term daily runoff forecasting by ANNs and multiple process-based hydrological models

In recent decades, the daily runoff forecasting based on artificial neural network (ANN) models has become quite important to deliver sustainable use and effective planning and management of water resources. The performance of the existent ANN models for 1 day in advance forecasting are frequently reported. However, the mid-term forecasting by ANN is scarce in the literature. In this study, a feed forward network trained with a back-propagation learning algorithm (BP-ANN) is used to construct a mid-short-term daily runoff forecasting system. ANN models having various input variables were constructed and the best structure was investigated. Moreover, the performance of ANN models and multiple process-based rainfall-runoff models, including Xinanjiang, ESSI, SWAT and XXT, is compared. Baohe River basin, located in central China, is chosen as a case study area. The results show that in general the performance of ANN models decrease as the lead time increase when the lead time is less than 11 days; while it varies slightly with lead time when the lead time is larger than 11 days. The ANN model with an appropriate combination of stream flow, precipitation as input variables performs much better than all the process-based rainfall-runoff models in terms of Nash-Sutcliffe efficiency for mid-short-term daily runoff forecasting.