چکیده:
Rainfall-runoff relationship is very important in many fields of hydrology such as water supply and water resource
management and there are many models in this field. Among these models, the Artificial Neural Network (ANN) was
found suitable for processing rainfall-runoff and opened various approaches in hydrological modeling. In addition,
ANNs are quick and flexible approaches which provide very promising results, and are cheaper and simpler to
implement than their physically based models. Therefore, this study evaluated the use of ANN models to forecast
daily flows in Bar watershed, a semi-arid region in the northwest Razavi Khorasan Province of Iran. Two different
neural network models, the multilayer perceptron (MLP) and the radial basis neural network (RBF), were developed
and their abilities to predict run off were compared for a period of fifty-five years from 1951 to 2006. The best
performance was achieved based on statistical criteria such as RMSE, RE and SSE. It was found that MLP showed a
good generalization of the rainfall-runoff relationship and is better than RBF. In addition, 1-day antecedent runoff
affected river flow, such that the statistical criteria decreased but the 5-day antecedent rainfall remained unaffected.
Furthermore, considering MLP, RE and RMSE, the best model produced the values 46.21 and 0.75 while the RBF
model recorded 177.60 and 0.82, respectively.
خلاصه ماشینی:
"The rainfall-runoff relationship is one of the most complex hydrological events to comprehend, due to the tremendous spatial and temporal variability of watershed characteristics, precipitation patterns, and the number of variables involved in the modeling of the physical processes (Tokar and Johnson, 1999; Buch et al.
Nowadays, artificial neural networks (ANNs) have become one of the most promising tools in the modeling of complex hydrological processes, such as the rainfall- runoff process.
The particular advantage of the ANN is that, even if the exact relationship between sets of input and output data is unknown but is acknowledged to exist, the network can be trained to learn that relationship, without requiring prior knowledge of the catchment's characteristics (Minns and Hall, 1996; Dawson, 1996).
1. Study area Measured data from Bar-Arieh watershed were used to develop and compare the ability of both MLP and RBF models to predict stream flow.
In this research two models of artificial neural network MLP and RBF was considered and then compared XN= X max - Xi X max - X min (1) with evaluation criteria such as RE, RMSE, SSE and CR.
60 for training and testing data in multi-layer perceptron and radial basis function respectively while this was much lower than when 1-day antecedent runoff was not used so that the least error in MLP and RBF without 1- day run off are 710.
That showed that the ANN rainfall-runoff model trained using BPA and Multi-layer perceptron through statistical criteria do not perform well.
Rainfall-Runoff models using artificial neural networks for ensemble stream flow prediction, hydrological processes."