Application of Linear Random Models to Four Annual Streamflow Series

A simple method for describing random time series is illustrated by application to the annual streamflow data of the St. Lawrence, the Missouri, the Neva, and the Niger rivers. The technique is illustrated for identifying the appropriate form of the general autoregressive-moving average model by use of the sample autocorrelation function of each series. The values of the parameters of the suggested model of each series are estimated and the results checked to suggest further modification of the model. The best model for each of the four samples showed a reduction in the variance value to 0.49, 0.64, 0.59, and 0.62 of the original variance with the use of one or two parameters. The use of the model for one year ahead forecasting on the Missouri River data is shown.