Autocorrelation and Spectral Analyses in Hydrology

Forty years of daily flow records of a river are analyzed using autocorrelation and spectral techniques. The two methods of analysis are compared relative to economy and extracted information. For statistical prediction, autocorrelation is shown to be more economical. The discriminatory power of the spectral density function, which allows the detection of noncommensurable harmonics, did not carry very much weight for the time series studied. A reduction in calculations can be achieved by examining the behavior of the correlogram only in the vicinity of suspected periods. In modeling, autoregressive techniques may be readily extended to the multivariate case. The intended use of the results is also a factor in the choice of methodology in data analysis.