The theory of chaos which deals with unpredictable complex nonlinear systems had its breakthrough in the past decades. The aim of this study was to employ chaos methods, including phase space reconstruction and local approximation method, to examine the existence of chaos in streamflow dynamics. This study also applied ARIMA time series model to predict streamflow data in the series. The major objective was to investigate and compare the prediction accuracy of daily streamflow time series models from the gauging station in the northeastern Taiwan. The gauging station was at the Lanyang River. The observed streamflow time series spanned a time period of 70 years from 1950 to 2019. The first 50-year streamflow data were used as the training data set, and the last 20-year data were used as the testing data set. This study applied efficiency criteria: coefficient of variation, Nash-Sutcliffe efficiency, and persistence index. The results showed that the local approximation prediction models had better prediction performance than the ARIMA models at the gauging station used in this study. Prediction performance was much better in the recession periods than in the rising periods. The results concluded that the existence of chaos might exist only in the recession periods. In the rising periods, streamflow dynamics were more stochastic.
Key Words: Chaos theory, local approximation method, Nash-Sutcliffe efficiency, persistence index, ARIMA,
streamflow prediction.