Prediction of the level of Paraguay river using neural networks

Abstract

Backpropagation neural networks are implemented for prediction of the level of Paraguay River at Ladário city, MS. Using 274 monthly mean values, the trained network predicts the levels of the four next months with relative errors smaller than 17%. For some special points, the prediction results also show that the neural network method seems to be useful to predict time series related to phenomena influenced by complex climatic and geophysical processes, and it does not deal directly with causal relationships involved in the phenomena studied. A discussion about the variability of the estimation errors for different predicted data is carried out here.