Real-Time Flood Forecasting Using Hybrid Neural Networks

Document Type : Research Article

Authors

Abstract

Hybrid models which are based on methods which divide a complex simulation problem to several
simple local models and combine the results, potentially could result in different output. The input
space in this method is divided into subspaces, and then some single models are assigned to each
specific region of the space. In this research by using some floods generated by a hydrologic model,
advantages of hybrid models in real-time flood forecasting compared to global models was
investigated. To do this, the results of a global ANN model which simulates whole of the flood
processes using a single model, are compared with that of two hybrid models, one consisting of a 4
ANN and the other consisting of 8 ANN. The results shows that hybrid models have significantly
better results in flood forecasting specially in forecasting time and amount of peak discharges. This is
very important in flood forecasting in flood warning systems because of their important role in flood
mitigation activities.