Investigation of Capability of Multilayer Perceptron and Time Lag Neural Networksfor Surface Flow Estimation in the Zayandehrud Basin

Runoff is a key factor in watershed management and it is valuable for suitable programming to improve watershed condition. Also, runoff estimation in watersheds is important for water resources studies and design of the flood control structures. Flexibility of Artificial Neural Networks (ANNs), its low cost and high capability for monthly runoff estimation was considered to using capability of these networks for monthly runoff estimation in the upstream sub basins of the Zayandehrud Dam. Moreover, Zayandehrud Dam Basin has important role for water supply of urban, agriculture and industrial sectors. In this study, two types of neural networks; including multilayer perceptron (MLP) and time lag neural networks (TLNNs) was evaluated in basins of Samandegan, Plasjan and Zayandehrud. Results indicated that TLNN model has high capability for runoff estimation, although this structure involves high sensitivity during training stage. Also this model had lower error compared with MLP Model. These networks (TLNNs) can estimate monthly runoff with suitable training and implementing inputs of 3 or 4 rain gauges and 2 or 3 temperature stations. TLNNs are suitable simulator of watershed response to the inputs such as rainfall and temperature, due to its high flexibility.