This study combines neural networks and fuzzy arithmetic to present a count
erpropagation fuzzy-neural network (CFNN) for streamflow reconstruction. Th
e CFNN has a rule-based control, a modified self-organizing counterpropagat
ion network, and a fuzzy control predictor. It can generate rules automatic
ally by increasing the training data to improve the accuracy or streamflow
reconstruction. The CFNN establishes the input and output relationship of a
watershed without set-up parameters. The parameters are estimated systemat
ically by the approach converging to an optimal solution. One sequence of d
ata generated by the Monte Carlo method is used to demonstrate the accuracy
of the CFNN. The streamflow data of the Da-chia River, in central Taiwan,
is also used to evaluate the performances of the CFNN. The results indicate
the reliability and accuracy of the CFNN for streamflow reconstruction. Co
pyright (C) 2001 John Wiley & Sons, Ltd.