Stochastic Analysis of Daily River Flow

Abstract

Optimum water resources design may be achieved using data generation techniques based on models statistically derived from historical records of a hydrologic variable. However, certain statistical techniques of time series analysis are restricted to series possessing the time-invariance property which hydrologic data might not possess. The analysis of river flow records in the light of this limitation is presented. To obtain a mathematical representation of daily streamflow, a trend-free model made up of an oscillatory component and an autoregressive process is postulated to apply to the records of five rivers. The oscillatory component is detected and isolated using spectral and fourier analyses, and Markov schemes are fitted to the standardized residual series. The adequacy of the models fitted is examined through a comparison of the theoretical variance with the computed variance resulting from the application of the model. The importance of the underlying series, which is the sequence generated in Monte Carlo techniques, is discussed.